Nature Reviews - Molecular Cell Biology - February 2001 - Vol 2 No 2 by Nature Magazine

toc_r

Page 1 of 2

CONTENTS

February 2001 Vol 2 No 2

79 | In this issue doi:10.1038/35052024

89 | HOW DROSOPHILA APPENDAGES DEVELOP Ginés Morata doi:10.1038/35052047

Highlights PDF

[597K]

[4K]

81 | DNA REPLICATION Licence to kill doi:10.1038/35052008

82 | IN BRIEF TRANSCRIPTION | IMMUNOLOGY | APOPTOSIS doi:10.1038/35052015

82 | CHOLESTEROL BIOSYNTHESIS Unpicking NPC doi:10.1038/35052002

83 | PHAGOCYTOSIS Eating well doi:10.1038/35052029

83 | WEB WATCH Nuclear fuel doi:10.1038/35052004

84 | WEB WATCH Let's get together

98 | SNARE-MEDIATED MEMBRANE FUSION Yu A. Chen & Richard H. Scheller doi:10.1038/35052017 [2829K]

107 | RAB PROTEINS AS MEMBRANE ORGANIZERS Marino Zerial & Heidi McBride doi:10.1038/35052055 [2970K]

118 | PRIONS: HEALTH SCARE AND BIOLOGICAL CHALLENGE Adriano Aguzzi, Fabio Montrasio & Pascal S. Kaeser doi:10.1038/35052063 [957K]

127 | UNTANGLING THE ERBB SIGNALLING NETWORK Yosef Yarden & Mark X. Sliwkowski doi:10.1038/35052073 [2386K]

doi:10.1038/35052032

84 | CELL MOVEMENT Take steroids to move faster doi:10.1038/35052034

84 | PROTEIN TRAFFICKING Made for export doi:10.1038/35052037

85 | APOPTOSIS Into the groove

138 | FILAMINS AS INTEGRATORS OF CELL MECHANICS AND SIGNALLING Thomas P. Stossel, John Condeelis, Lynn Cooley, John H. Hartwig, Angelika Noegel, Michael Schleicher & Sandor S. Shapiro doi:10.1038/35052082 [2715K]

doi:10.1038/35052013

85 | IN BRIEF MEMBRANE FUSION | CELL DEATH | DEVELOPMENT | NEUROTRANSMISSION doi:10.1038/35052039

86 | PLANT PHYSIOLOGY Ironing out a yellow stripe doi:10.1038/35052041

147 | OPINION CONFOUNDED CYTOSINE! TINKERING AND THE EVOLUTION OF DNA Anthony Poole, David Penny & Britt-Marie Sjöberg doi:10.1038/35052091 [218K]

151 | OPINION

http://www.nature.com.ezproxy.library.wisc.edu/nrm/journal/v2/n2/toc_r.html

12/3/2006

toc_r

Page 2 of 2

86 | MOLECULAR MOTORS On track with kinesin doi:10.1038/35052011

86 | MEMBRANE FUSION Pore performance doi:10.1038/35052044

87 | DNA REPAIR Motif with a motive

RETROVIRAL RECOMBINATION: WHAT DRIVES THE SWITCH? Matteo Negroni & Henri Buc doi:10.1038/35052098 [797K]

157 | NatureView doi:10.1038/35052105

doi:10.1038/35052000

NATURE REVIEWS | MOLECULAR CELL BIOLOGY

http://www.nature.com.ezproxy.library.wisc.edu/nrm/journal/v2/n2/toc_r.html

12/3/2006

HIGHLIGHTS HIGHLIGHTS ADVISORS JOAN S. BRUGGE HARVARD MEDICAL SCHOOL, BOSTON, MA, USA PASCALE COSSART INSTITUT PASTEUR, PARIS, FRANCE GIDEON DREYFUSS UNIVERSITY OF PENNSYLVANIA, PHILADELPHIA, PA, USA PAMELA GANNON CELL AND MOLECULAR BIOLOGY ONLINE JEAN GRUENBERG UNIVERSITY OF GENEVA, SWITZERLAND ULRICH HARTL MAX-PLANCK-INSTITUTE, MARTINSRIED, GERMANY NOBUTAKA HIROKAWA UNIVERSITY OF TOKYO, JAPAN

D N A R E P L I C AT I O N

Licence to kill Cells are faced with a problem each time they divide — how can they ensure that their chromosomal DNA is duplicated only once per cell cycle? The first clues came five years ago, when components of the complex that ‘licenses’ the DNA for a single round of replication were identified. But this still left the question of how licensing is regulated. Reports in Science and Nature Cell Biology now fill in this piece of the puzzle, connecting the activities of two compo-

STEPHEN P. JACKSON WELLCOME/CRC INSTITUTE, CAMBRIDGE, UK ROBERT JENSEN JOHNS HOPKINS UNIVERSITY, BALTIMORE, MD, USA VICKI LUNDBLAD BAYLOR COLLEGE OF MEDICINE, HOUSTON, TX, USA TONY PAWSON SAMUEL LUNENFELD RESEARCH INSTITUTE, TORONTO, CANADA NORBERT PERRIMON HARVARD MEDICAL SCHOOL, BOSTON, MA, USA THOMAS D. POLLARD THE SALK INSTITUTE, LA JOLLA, CA, USA JOHN C. REED THE BURNHAM INSTITUTE, LA JOLLA, CA, USA KAREN VOUSDEN NATIONAL CANCER INSTITUTE, FREDERICK, MD, USA JOHN WALKER MRC DUNN HUMAN NUTRITION UNIT, CAMBRIDGE, UK

nents with positive and negative effects on the replication of DNA. Replication follows a stepwise pattern of protein assembly. Binding of the ‘origin-recognition complex’ at initiator elements in the DNA is followed by attachment of two further proteins, Cdc6 and Cdt1. Next, during G1, these proteins recruit the socalled MCM complex to form the ‘pre-replication complex’ (pre-RC). At this stage the cell is licensed for replication, probably because the MCM complex — which has helicase activity — opens up the chromatin and allows the replication machinery access to the DNA. Crucially, though, once the replication origin has fired, the pre-RC disassembles and cannot reform until the cell has gone through mitosis and entered G1 of the next cell cycle. Presumably nuclei in G2 either lack a factor needed to initiate replication, or contain an inhibitor that blocks it. A protein called geminin had previously been shown to block DNA replication by preventing loading of the MCM complex. So Wo h l s c h l e g e l

and colleagues have asked what its target might be. They used immunoprecipitation to show that geminin interacts with the MCM-loader Cdt1 in human cells. They then used an in vitro system of DNA replication to test whether geminin inhibits replication by targeting Cdt1. They found that geminin-dependent inhibition of pre-RC formation can be counteracted by excess Cdt1, supporting the idea that geminin targets Cdt1 to inhibit re-replication. Tada and co-workers approached the problem using fractionated egg extracts from Xenopus laevis. They showed that geminin blocks the licensing ability of one fraction, termed RLF-B, and purification of RLF-B revealed the active ingredient to be Cdt1. The authors confirmed that geminin and RLF-B interact and antagonize one another. Finally they found that the geminin present during metaphase is enough to completely block origin assembly. Geminin, then, has a licence to kill DNA replication — at least temporarily. Whether it acts only through Cdt1, and whether it is redundant with other regulatory mechanisms, are just some of the next questions to be tackled. Alison Mitchell References and links ORIGINAL RESEARCH PAPERS Wohlschlegel, J. A. et al. Inhibition of eukaryotic DNA replication by geminin binding to Cdt1. Science 290, 2309–2312 (2000) | Tada, S. et al. Repression of origin assembly in metaphase depends on inhibition of RLF-B/Cdt1 by geminin. Nature Cell Biol. 3, 107–113 (2001) WEB SITE Xenbase

The Kobal Collection

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 2 | FEBRUARY 2001 | 8 1

HIGHLIGHTS

IN BRIEF TRANSCRIPTION

Direct observation of DNA rotation during transcription by Escherichia coli RNA polymerase. Harada, Y. et al. Nature 409, 113–115 (2001) CHOLE STE ROL B IOSYNTH E S IS

Several years ago Kazuhiko Kinosita Jr and colleagues used an elegant single-molecule imaging system to show direct rotation of the F1-ATPase. Now they’ve done it again, this time with a DNAbased motor — RNA polymerase. They show, in real time, that single RNA polymerase molecules attached to a glass surface can rotate DNA for over 100 revolutions, and predict that this technique could help in resolving the individual steps of transcription. I M M U N O LO G Y

ICOS co-stimulatory receptor is essential for T-cell activation and function. Dong. C. et al. Nature 409, 97–101 (2001)

ICOS is critical for CD40-mediated antibody class switching. McAdam, A. J. et al. Nature 409, 102–105 (2001)

ICOS is essential for effective T-helper-cell responses. Tafuri, A. et al. Nature 409, 105–109 (2001)

Optimal T-cell activation requires an antigen-specific signal as well as a co-stimulatory signal which influences T-cell proliferation, cytokine secretion and the development of effector functions. ICOS is a member of the CD28/CTLA4 family of costimulatory molecules, and is expressed on activated T cells. Its ligand, B7RP1/B7H, is expressed on B cells and macrophages. So this pair of molecules does not seem to be involved in initial T-cell activation, but at a later stage during T-cell interactions with B cells and macrophages. To characterize the function of ICOS, three groups have now generated ICOS-deficient mice. They show that ICOS is essential for efficient T/B-cell interactions and for antibody production in response to T-cell-dependent antigens. Knockout mice have defective germinal centre formation and immunoglobulin class-switching, as well as decreased T-cell production of interleukin-4 and interleukin-13. Further work is required to understand the complex nature of costimulation so that therapies can be designed to treat immune-mediated diseases. A P O P TO S I S

An alternative, nonapoptotic form of programmed cell death. Sperandio, S. et al. Proc. Natl Acad. Sci. USA 97, 14376–14381 (2000)

The terms ‘apoptosis’ and ‘programmed cell death’ are often used interchangeably, but this paper describes a form of programmed cell death that fails to fulfil the criteria for apoptosis. Christened paraptosis (‘next to’ or ‘related to’ apoptosis), this form of cell death has a distinct morphology and biochemistry to apoptosis, it shows no response to caspase inhibitors or Bcl-xL, and it is mediated by an Apaf-1-independent caspase-9 activity.

Unpicking NPC Niemann–Pick disease type C is a hereditary disorder that leads to progressive neurodegeneration and death during early childhood. More than 95% of patients have defects in the NPC1 gene, whereas a small subset of patients have defects in a second gene, NPC2. But in both cases, cholesterol accumulates aberrantly in the lysosomes. Two reports in Science now bring us a step closer to understanding the processes behind both forms of this debilitating condition. Yiannis Ioannou and colleagues investigated the function of the protein that is mutated in NPC1 disease by looking for conserved motifs in its sequence. They found that it contains six copies of a lipid-attachment site also found in members of the resistance-nodulation-division (RND) family of prokaryotic permeases. These proteins are involved in pumping hydrophobic compounds such as antibiotic drugs, detergents and fatty acids from the cytosol of Gram-negative bacteria. Two members of the RND family — Escherichia coli AcrB and Pseudomonas aeruginosa MexD — show weak sequence homology to the NPC1 protein throughout their entire sequences. Moreover, NPC1 and MexD have similar membrane topology and secondary structures. So could human NPC1 be an RND permease? Ioannou and colleagues tested this by looking at accumulation of a fluorescent dye called acriflavine in normal and NPC1-deficient fibroblasts. They found that efflux of acriflavine from the endosomal/lysosomal system is blocked in the absence of NPC1, and that this efflux is an active process, requiring a proton-motive force. Next the authors engineered E. coli to express human NPC1. Then,

given that the defect in NPC1 disease is the accumulation of cholesterol, they wondered whether this might be a substrate for NPC1. Ioannou and colleagues could observe no build-up of cholesterol in NPC1-expressing E. coli cells, but they did see an accumulation of oleic acid indicating that, like its prokaryotic homologues, NPC1 might transport fatty acids across a membrane. In the second paper, Naureckiene et al. describe the protein responsible for NPC2. While trying to characterize the lysosome proteome they identified a human protein called HE1. The pig homologue of this protein is known to bind cholesterol so, given its lysosomal location, the authors wondered whether HE1 might be involved in NPC2 disease. HE1 could not be detected by Western blotting in fibroblasts from two patients with the condition, and sequence analysis revealed that both patients had mutations in the HE1 gene. Studies with a cholesterol-binding antibiotic, filipin, indicated the abnormal accumulation of cholesterol in fibroblasts from the NPC2 patients. But when Naureckiene and co-workers treated these fibroblasts with recombinant HE1, this build-up was reduced. The HE1-conditioned medium had no effect on cholesterol accumulation in NPC1-deficient fibroblasts, however, confirming that the defects in the two forms of NPC disease are different. Alison Mitchell References and links ORIGINAL RESEARCH PAPERS Davies, J. P.,

Chen, F. W. & Ioannou, Y. A. Transmembrane molecular pump activity of Niemann–Pick C1 protein. Science 290, 2295–2298 (2000) | Naureckiene, S. et al. Identification of HE1 as the second gene of Niemann–Pick C disease. Science 290, 2298–2301 (2000) FURTHER READING Marx, J. Disease genes clarify cholesterol trafficking. Science 290, 2227–2229 (2000)

www.nature.com/reviews/molcellbio

| FEBRUARY 2001 | VOLUME 2

HIGHLIGHTS

P H A G O C Y TO S I S

Eating well Remarkably little is known about the signalling pathways that drive phagocytosis. Once its prey is bound to the cell surface, what exactly causes a phagocyte to extend pseudopodia around its dinner and swallow it? In the Journal of Cell Biology, Roberto Botelho and colleagues describe the spatial and temporal regulation of two signals involved in phagocytosis, providing both a visual and an intellectual feast. The authors made use of fluorescent probes that specifically recognize membrane lipids. Using a macrophage cell line, they first turned their attention to the phospholipid phosphatidylinositol-4,5bisphosphate (PtdIns(4,5)P2). Transfection with a PtdIns(4,5)P2 probe (based on the plextrin homology (PH) domain of phospholipase Cδ fused to green fluorescent protein (PLCδ-PH-GFP)) allowed them to track the whereabouts of PtdIns(4,5)P2 in cells fed with antibody-coated erythrocytes. PtdIns(4,5)P2 accumulated in the phagocytic ‘cups’ as they extended round the erythrocytes, but then disappeared as soon as engulfment was complete and the vesicle membrane had fused. This was not due to membrane turnover because an acylated GFP that partitions into the inner leaflet of the plasma membrane neither accumulated as much in the phagocytic cup, nor disappeared as rapidly upon vesicle fusion. So can the transient appearance of PtdIns(4,5)P2 in the phagosome be accounted for by its local production and destruction? Apparently so: immunofluorescence revealed that type 1α phosphatidylinositol 4-phosphate 5-kinase, the enzyme that catalyses the synthesis of PtdIns(4,5)P2 from PtdIns(4)P, migrates to phagosomes as they form, then dissociates once phagocytosis is complete. But what accounts for the disappearence of PtdIns(4,5)P2? This is a tricky question as PtdIns(4,5)P2

PtdIns(4,5)P2 can be either cleaved by PLC or phosphorylated by phosphatidylinositol-3-OH kinase (PI(3)K), and both enzymes have been implicated in phagocytosis. The authors reasoned that PLCγ would be involved because activation of the Fc receptor, which stimulates phagocytosis, also activates PLCγ. Immunofluorescence with antibodies against PLCγ2 supported this hunch, which was then backed up using a probe for one of the products of PLC, diacylglycerol (DAG). This probe, based on the C1 domain of protein kinase Cδ (Cδ-GFP), showed that DAG is produced locally at the phagosome. Transfection of the two lipid probes together (one fused to yellow fluorescent protein (Cδ-YFP), the other fused to cyan fluorescent protein (PLCδ-PHCFP)) allowed the two lipids to be tracked simultaneously (see picture: 1, early phagocytic cup; 2, later phagocytic cup; 3, early phagosome; 4, mature phagosome): as PLCδ-PH-CFP dissociated from the phagosome, Cδ-YFP became associated, reaching a peak during vesicle closure. DAG therefore seems to be produced from PtdIns(4,5)P2 by phospholipase C at the phagosome. But are PtdIns(4,5)P2 and DAG needed for phagocytosis? Sequestration or removal of PtdIns(4,5)P2 reduced phagocytosis, as did treatment with PLC inhibitors. The PLC inhibitors prevented macrophages from forming proper phagocytic cups by blocking actin remodelling at sites of erythrocyte attachment. The next course in this phagocytic feast might explain how changes in the levels of PtdIns(4,5)P2 and DAG affect cytoskeletal dynamics, to translate regulation into the mechanics of engulfment. Another important question concerns PtdIns(4,5)P2 metabolism. Is it all hydrolysed by PLC, or is some of it converted to PtdIns(3,4,5)P3? Probes for this lipid might soon provide an answer. Cath Brooksbank

References and links

ORIGINAL RESEARCH PAPER Botelho, R. J. et al. Localized biphasic

changes in phosphatidylinositol-4,5-bisphosphate at sites of phagocytosis. J. Cell Biol. 151, 1353–1367 (2000) FURTHER READING Kwiatkowska, K. & Sobota, A. Signalling pathways in phagocytosis.Bioessays 21, 422–431 (1999)

DAG

Merge

3 2

2 3

WEB WATCH Nuclear fuel When is a ‘dot com’ not a ‘dot com’? When it’s cellnucleus.com, a new community-driven web site for those interested in the structure and function of the nucleus. The aim is not to make money, but to provide an up-to-date resource for all cell nucleus researchers. Established and maintained by Michael J. Hendzel from the University of Alberta, cellnucleus.com was officially launched last December. “The principal goal … is to freely disseminate information on the cell nucleus in order to stimulate interest in the topic”, says Hendzel. A core component of the site is an evolving textbook, with several prominent names signed up to contribute chapters. But at the moment, only two ‘temporary’ reviews are posted. The site also features an antibody database and a list of links to the home pages of various cell nucleus researchers. One useful tool is a literature update, listing recently published papers in the field. Papers can be viewed by date or by category, and there are links to PubMed. Other features that will be added to the site include job and conference listings, posters and Powerpoint presentations from meetings, as well as a live-cell imaging database. This database should ultimately become a forum for “descriptive observations obtained in livecell experiments that may be of value to the community but do not fit nicely into any manuscripts derived from ongoing research projects within the laboratories involved”. To get to the site at the moment you’ll need to go to http://www.cellnucleus.org, but it should move to www.cellnucleus.com around the middle of the year. In the meantime, the site will continue to evolve, and the idea is that it will fuel discussion in the field and provide a useful teaching resource.

Alison Mitchell

Image courtesy of Sergio Grinstein, University of Toronto, Ontario, Canada.

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 2 | FEBRUARY 2001 | 8 3

HIGHLIGHTS

WEB WATCH Let’s get together Protein–protein interactions literally hold cellular processes together, but keeping up with the ever-expanding literature on what might bind what is enough to make you fall apart, and it’s likely to get worse as whole-genome scans for protein–protein interactions churn them out in bulk (see the review by Albertha J. M. Walthout and Marc Vidal on page 55 of our January issue). PreBIND — a new adjunct to the Biomolecular Interaction Network Database (BIND) built by researchers at the Samuel Lunenfeld Research Institute in Toronto — helps you to search the literature for possible interactions and then submit bona fide interactions to BIND. The version now available is a prototype that searches for protein–protein interactions in the yeast Saccharomyces cerevisiae, using proteins described in the NCBI’s RefSeq database, but there are plans to add more organisms. PreBIND is simple to use, and the help page is a model of clarity. You can search PreBIND by entering a protein name or RefSeq accession number. PreBIND then returns a list of papers that discuss other molecules that interact with your protein of interest. The power behind PreBIND is a trained alogorithm that allows it to recognize papers that discuss interactions. For that reason, it searches much more thoroughly than you could doing a simple PubMed search. For each paper that it finds, PreBIND lists the proteins discussed in the paper and provides a score of the likelihood that the paper contains interaction information. You can then review the paper, decide whether or not it discusses a true interaction, and submit your response to BIND. This will then be refereed by a moderator before being added to BIND. Over time, PreBIND and BIND should help us to gain valuable and reliable insights into molecular interactions.

CE LL MOVE M E NT

Take steroids to move faster As well as being important regulators of developmental and physiological processes, steroid hormones have a darker side — in the development and progression of breast, ovarian and prostate cancer. This was believed to be due to stimulation of proliferation, but evidence presented by Jianwu Bai and colleagues in Cell suggests a new function for steroid hormones: making cells move. Using the Drosophila melanogaster egg chamber, this group has identified mutants that block movement towards the oocyte of a group of cells called border cells. Genes previously identified using this system include slow border cells (slbo), which encodes a transcription factor, and the Drosophila E-cadherin gene. Their latest screen has identified taiman (tai) — meaning ‘too slow’: mutant clones typically remain stuck at the anterior end of the egg chamber instead of moving towards the oocyte. tai mutants had normal expression levels of SLBO, indicating that TAI might be involved

in a different pathway from SLBO, but E-cadherin was mislocalized. So what is TAI? It turns out to belong to a family of steroid-receptor co-activator (SRC) proteins previously not thought to exist in Drosophila. Its closest relative in mammals is AIB1, a SRC that is amplified in breast and ovarian cancer. Which steroid hormone receptor does TAI interact with? Several genes for steroid hormone receptors have been identified in the Drosophila genome but only one of these, the ecdysone receptor (a hetrodimer of the USP and EcR proteins), has a known ligand, which is synthesized by the ovary. USP and EcR were expressed together with TAI in wild-type border cells, and TAI, USP and EcR colocalized exactly on Drosophila polytene chromosomes, indicating that the three proteins form a complex. Furthermore, provided that SLBO was also expressed, treatment with ecdysone caused precocious migration of border cells whereas migration was abrograted in an

P R OT E I N T R A F F I C K I N G

Made for export Without plasma membrane proteins, every cell would be an island, unable to communicate with the outside world. So how does the cell ensure that newly synthesized plasma membrane proteins reach their destination? Proteins that don’t fold properly or aren’t correctly glycosylated are held back in the endoplasmic reticulum (ER), and an emerging mechanism that prevents the untimely escape of proteins from the ER is an ER-retention signal, RXR(R), which has to be masked by another protein to allow exit from the ER. But Ma and colleagues, reporting in the 12 January issue of Science, have found a new signal that, instead of holding proteins back, pushes them out.

The authors stumbled on this mechanism when they were trying to express different K+ channels in Xenopus oocytes. Two of the channels, Kir1.1 and Kir2.1, were efficiently expressed at the surface but others were not. So they swapped the carboxyl termini of the poorly expressed channels for those of the efficiently expressed channels, and got efficient expression. Conversely, removal of the C terminus of Kir2.1 beyond residue 374 reduced trafficking to the surface. This did not seem to be due to a fault in folding or assembly, because the small amount of surfaceexpressed truncated Kir2.1 had identical conductance properties to wild-type Kir2.1. Surface levels

ecdysoneless mutant. What next? As well as identifying the targets of the ecdysone receptor–taiman complex, it will be important to determine whether AIB1 has similar effects on cell motility in vertebrates. If so, it could explain the increased invasiveness of

could also be restored by expression of full-length Kir2.1 together with truncated Kir2.1, indicating that the two forms could assemble to create functional channels. Scanning mutagenesis narrowed the signal down to FCYENE: even conservative mutations in the italicized residues abolished surface expression of green-fluorescentprotein (GFP)-tagged Kir2.1. The position of the sequence seemed unimportant: it worked when inserted between GFP and Kir2.1, or at the extreme C terminus of Kir2.1. But does the export sequence work for other proteins? A truncated form of Kir3.1 normally remains stuck in the ER but adding FCYENE to its C terminus allowed it to escape. Likewise, FCYENE facilitated the export of a distantly related K+ channel, Kv1.2, which is normally helped to the surface by its partner, Kvβ2.

Cath Brooksbank

www.nature.com/reviews/molcellbio

| FEBRUARY 2001 | VOLUME 2

HIGHLIGHTS

IN BRIEF

A P O P TO S I S

Into the groove

AIB1-overexpressing cancer cells and suggest new therapeutic avenues. Cath Brooksbank References and links ORIGINAL RESEARCH PAPER Bai, J., Uehara,

Y. & Montell, D. Regulation of invasive cell behaviour by taiman, a Drosophila protein related to AIB1, a steroid receptor coactivator amplified in breast cancer. Cell 103, 1047–1058 (2000)

So which signal is the strongest if a protein contains both a FCYENE signal and an ER retention signal? On the cell’s factory floor, safety and quality control are of paramount importance, so the ER retention signal overrides the export signal, presumably to prevent the plasma membrane from being flooded with misfolded proteins. Is FCYENE the only signal for export? Apparently not; Kir1.1 has an unrelated sequence at its C terminus — VLSEVDETD — that also does the trick, and the vesicular stomatitis virus Gprotein has a DXE motif that is thought to speed its transit to the cell surface. No doubt these hint at more to come. Cath Brooksbank References and links ORIGINAL RESEARCH PAPER Ma, D. et al.

Role of ER export signals in controlling surface potassium channel numbers. Science 291, 316–319 (2001)

One way to inhibit apoptosis is by blocking the activity of caspases, a job done by the inhibitor of apoptosis (IAP) proteins. Last year saw the identification of a mitochondrial protein called Smac/DIABLO, which has the opposite effect — it counters the inhibitory effects of IAPs by physically interacting with them. The structural basis of this interaction is discussed by two papers in Nature, and the results might have implications for the design of small molecules to treat cancer. Liu and colleagues solved the solution structure of the BIR3 domain of one IAP (X-linked IAP; XIAP) complexed with a functionally active peptide derived from the amino terminus of Smac/DIABLO. Similarly, Wu and co-workers report the high-resolution crystal structure of Smac/DIABLO in a complex with the XIAP BIR3 domain. The BIR3 domain is a potent inhibitor for caspase-9, and wild-type Smac/DIABLO is known to interact with both the BIR2 and BIR3 domains of XIAP. The amino-terminal residues of Smac/DIABLO are needed for its function, and both papers identify the first four amino acids (Ala–Val–Pro–Ile) as being critical — they recognize a surface groove on the BIR3 domain. This recognition specificity is achieved through a series of hydrogen bonding and van der Waals contacts between the BIR3 and Smac/DIABLO residues. How, then, will this information aid in drug design? Some cancers overexpress IAPs, preventing apoptotic removal of the cancerous cells. Small molecules that mimic the effects of Smac/DIABLO could therefore be good potential drugs for killing these cells. And armed with the knowledge of the structure of the Smac/DIABLObinding groove, it should be possible to design high-affinity compounds that can slot into this groove. Alison Mitchell References and links ORIGINAL RESEARCH PAPERS Liu, Z. et al. Structural basis for binding of Smac/DIABLO to the XIAP BIR3 domain. Nature 408, 1004–1008 (2000) | Wu, G. et al. Structural basis of IAP recognition by Smac/DIABLO. Nature 408, 1008–1012 (2000)

MEMBRANE FUSION

Sequential action of two GTPases to promote vacuole docking and fusion. Eitzen, G. et al. EMBO J. 19, 6713–6720 (2000)

Vacuole fusion in Saccharomyces cerevisiae can be dissected into three steps: priming, docking and fusion. The Rab protein Ypt7p regulates vacuole fusion after the priming step – it binds in its GTP-bound form to the HOPS complex, which docks the membranes together. Ypt7p is then no longer needed for fusion to proceed. Surprisingly, fusion requires another, unknown GTPase that might regulate the calcium release necessary for fusion. C E L L D E AT H

Wee1-regulated apoptosis mediated by the Crk adaptor protein in Xenopus egg extracts. Smith, J. J. et al. J. Cell Biol. 151, 1391–1400 (2000)

The protein kinase Wee1 inhibits the cell-cycle kinase Cdc2, but Smith and colleagues now find a new function for it. They identified Wee1 in a screen for proteins that interact with the adaptor protein Crk, which is needed for apoptotic signalling in Xenopus egg extracts. Wee1, like Crk, is needed for apoptosis in this system and addition of Wee1 markedly accelerates apoptosis. By contrast, other Cdc2 inhibitors have no effect on apoptosis in this system. D E V E LO P M E N T

Wasp, the Drosophila Wiskott–Aldrich syndrome gene homologue, is required for cell fate decisions mediated by Notch signalling. Ben-Yaacov et al. J. Cell Biol. 152, 1–136 (2001)

WASP, first identified because its mutation causes Wiskott–Aldrich syndrome, is involved in translating signals into cell movement. Ben-Yaacov and colleagues now identify a single WASP-like gene in Drosophila and find that cell-fate decisions are disrupted in WASP mutants: the phenotype is enhanced by intoducing a weak mutant of the Notch receptor into the WASP mutants, and suppressed when the Notch pathway is elevated, indicating that WASP might be involved in Notch-dependent cell-fate decisions. N E U R OT R A N S M I T T E R R E L E A S E

Complexins regulate a late step in Ca2+-dependent neurotransmitter release. Reim, K. et al. Cell 104, 71–81 (2001)

Complexins bind to SNARE complexes in the brain. This study establishes their function as positive regulators of neurotransmission, acting at or after the Ca2+-dependent step. Whereas complexin-I knockout mice show no phenotype, complexin-I and -II double deletion mutants die a few hours after birth. Spontaneous neurotransmitter release is normal in these mice, but they fail to respond to action potentials. Complexins are unlikely to be calcium sensors, but probably interact with the calcium sensor (which might or might not be synaptotagmin).

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 2 | FEBRUARY 2001 | 8 5

HIGHLIGHTS

P L A N T P H Y S I O LO G Y

Ironing out a yellow stripe Iron presents a quandary in biology. Its chemical properties have made it an essential metal for life, forming a vital part of cytochromes, globins and various redox enzymes. However, in its inorganic state it is relatively insoluble, resulting in the development of specialized systems for iron uptake and transport around living organisms. Iron is mainly taken up into the biosphere through the roots of plants, and in the 18 January issue of Nature, Elsbeth Walker and colleagues report the identification of a protein that is responsible for moving iron across this barrier in the world’s major crop plants. One strategy adopted by plants to absorb iron is to extrude from their roots small organic compounds called phytosiderophores. These bind tightly to the sparingly soluble iron in the soil, and are then reabsorbed into the roots, taking their chelated iron with them. Walker and colleagues have discovered the transporter for phytosiderophores back into the root in the classic mutant of maize known as yellow stripe 1 (ys1). Yellow stripe 1 was first identified in 1929 by George W. Beadle, through its distinctive phenotype in which iron deficiency leads to incomplete pigmentation of leaves and alternating yellow and green stripes running their length

(bottom in the figure, under a wild-type leaf). By the early 1990s it was known that ys1 plants are defective in iron–phytosiderophore uptake, making the YS1 gene product a good candidate for the elusive phytosiderophore transporter. To identify the YS1 gene, Walker’s group used the Ac transposon to produce a library of mutant plants, one of whose disruptions proved allelic with ys1. The Ac-disrupted gene was then sequenced and its identity as yellow stripe confirmed by sequencing the allele from three separate ys1 mutants. The gene encodes a protein that contains twelve putative transmembrane regions, but is this really a phytosiderophore transporter? As a test, the YS1 gene was introduced into a mutant yeast (Saccharomyces cerevisiae) strain defective in extracting iron from the surrounding medium. This did not, on its own, rescue the yeast phenotype unless maize phytosiderophores were added to the growth medium. At least one mystery remains. Maize’s YS1 shows similarity to genes of unknown function from species throughout the plant kingdom — monocots, dicots, gymnosperms and mosses — many of which do not use phytosiderophores to absorb iron. Among those lacking phy-

M O L E C U L A R M OTO R S

On track with kinesin Like a miniature railway, conventional kinesin can transport molecular cargoes over long distances. It does this without dissociating from its microtubule rails — a property known as processivity. But how does kinesin stay on track? In the Journal of Cell Biology, Ronald Vale and colleagues describe two studies aimed at finding out. Conventional kinesin is a dimer, with two catalytic motor domains connected through a stalk to the cargo-binding carboxy-terminal tail. Each motor head is joined to the stalk by a flexible ‘neck’ region consisting of two parts — a ‘neck linker’ that interacts with the motor and an adjacent ‘neck coiled-coil’. The neck linker drives the characteristic ‘stepping’ movement of the two kinesin heads, ensuring both heads do not dissociate simultaneously. This mechanism relies on a conformational change, and there are two theories for how this might occur — that the neck linker is ‘unzippered’ from the motor, or that there is partial

unwinding of the neck coiled-coil. Tomishige and Vale have now tried to distinguish these possibilities by manipulating movement of the neck region using disulphide crosslinking of cysteine residues engineered into recombinant kinesin motors. Using this system to immobilize the neck linker, the authors found that kinesin no longer moved in only one direction. But crosslinking to permit limited movement allowed biased unidirectional diffusion of the kinesin. This indicates that partial movement of the neck linker is enough to determine directionality, but that full movement is required for active, processive movement. In support of the argument that conformational changes in the neck linker are needed for processivity, immobilization of the neck coiled-coiled region had very little effect on motor activity. So what does this region do? Tomishige and Vale noticed that the extent of processivity — or

tosiderophores is Arabidopsis thaliana, but its recently sequenced genome contains no fewer than eight homologues of YS1. Perhaps these cousins are involved in the transport of iron within plants; for example by acting as a receptor for nicotianamine, an iron chelator found in all plants, which has been proposed to aid the longrange movement of iron through phloem tubes. Iron deficiency is a leading nutritional disorder in the developing world, and this study brings us one step closer to staple crop varieties better able to accumulate bioavailable iron. Coming so close on the heels of the completion of the Arabidopsis genome sequence it also highlights the valuable genetic heritage of other, intensely studied plants. After all, Barbara McClintock discovered transposons, on which this and many other molecular biology studies are reliant, from work done entirely in maize. Christopher Surridge Senior Editor, Nature References and links ORIGINAL RESEARCH PAPER Curie, C. et al. Maize yellow stripe 1 encodes a membrane protein directly involved in Fe(III) uptake. Nature 409, 346–349 (2001)

‘run length’ — was decreased by up to 50% when the neck coiled-coil was immobilized. In the second study, Thorn et al. investigated this further by adding positive charge to the neck coiled-coil to generate ‘ultra-processive’ kinesin mutants. The gain in processivity was diminished by high salt concentrations or by cleaving off the negative carboxyl terminus of the microtubule protein tubulin, indicating that there might be an electrostatic interaction between this region of tubulin and the neck coiled-coil. This interaction is probably weak, however, as it was abolished by adding relatively low loads to the kinesin in an optical-trap assay. The current train of thought, then, is that the neck coiled-coil tethers kinesin near the microtubule surface, whereas the neck linker is involved in the conformational change behind processive movement. Alison Mitchell References and links ORIGINAL RESEARCH PAPERS Tomishige, M. & Vale, R. D. Controlling kinesin by reversible disulfide cross-linking: identifying the motility-producing conformational change. J. Cell Biol. 151, 1081–1092 (2000) | Thorn, K. S., Ubersax, J. A. & Vale, R. D. Engineering the processive run length of the kinesin motor. J. Cell Biol. 151, 1093–1100 (2000) WEB SITE The kinesin home page

www.nature.com/reviews/molcellbio

| FEBRUARY 2001 | VOLUME 2

MEMBRANE FUSION

Pore performance Transport vesicles continuously shunt molecules around the cell, ensuring delivery to the correct intracellular destination. Essential to this process are the SNARE proteins, which mediate docking and attachment of vesicles with their target membrane. But how, once a vesicle contacts the target membrane, are the repulsive forces between the membranes overcome to allow fusion? Scientists remain divided on this point. Whereas some believe that SNAREs, by bringing membranes into close proximity, are sufficient for this step, others propose a fusion pore model, in which a proteineous channel forms. Now, reporting in Nature, Peters and colleagues reveal the identity of such a channel, and it turns out to be a familiar one. Calcium and calmodulin can trigger the final step of fusion, but their downstream targets were not known. So, to identify calmodulin targets, Peters and colleagues used label-transfer analysis to isolate nearby factors on vacuolar membranes. Intriguingly, when they then purified these factors, mass spectrometry identified them as components of the V0 integral membrane sector of V-ATPase — a membrane complex that functions as a proton pump. The relevant calmodulin-binding partner was then found to be the V0 proteolipid. So can these proteolipids trigger fusion? To test this, the authors monitored their ability to trigger choline release in reconstituted liposomes. And they found, consistent with the channel model, that they triggered release in a calcium/calmodulin-dependent manner. Membrane docking requires formation of an electrochemical membrane potential, which is mainly generated by the V-ATPase. But if V0 is important for fusion itself then it should have a direct role in fusion, independent of proton pump activity. The authors confirmed this, showing that fusion could occur once they had knocked out proton pump activity. So where do SNAREs come in? By inhibiting different stages of fusion, the authors showed that trans-SNARE pairing mediates channel formation, but not maintenance. One possibility, then, is that SNAREs concentrate V0 sectors at the site of close membrane contact. What remains to be seen is how, once formed, these channels allow bilayer mixing. The authors’bets are hedged on a model in which channels expand radially to form an aqueous pore but, as is often the case, more specific analytical tools are needed before we can dig any deeper. Alison Schuldt References and links ORIGINAL RESEARCH PAPER Peters, C. et al. Trans-complex formation by proteolipid channels in the terminal phase of membrane fusion. Nature 1 February (2001)

D N A R E PA I R

Motif with a motive Scientists, like policemen, often have to work backwards — start with the motive to track down the culprit. Reporting in Nature, Downs et al. describe an example of such molecular detective work. They have linked phosphorylation of the Ser–Gln–Glu (SQE) motif in the yeast core histone H2A to alterations in chromatin — alterations that might facilitate the repair of damaged DNA. The SQE motif is phosphorylated in vitro by members of the phosphatidylinositol-3-OH kinaserelated kinase (PIKK) family. These include human ATM and ATR, and their Saccharomyces cerevisiae homologues Mec1p and Tel1p, all of which are central in eukaryotic responses to DNA damage. Because the carboxyl terminus of histone H2A contains an SQE motif, Downs et al. wondered whether phosphorylation of this sequence might be involved in signalling DNA damage. They generated a strain in which the SQE motif was deleted, and found that it was hypersensitive to chemicals that induce DNA damage, such as methyl methane-sulphonate (MMS). When S, Q and E were mutated individually, the phenotypes correlated with the relative importance of each residue in defining an optimal PIKKrecognition motif. So could the hypersensitivity to MMS reflect loss of recognition by a PIKK? And if so, which one? Strains with mutations in the

MEC1 gene could not phosphorylate the SQE motif of H2A in the presence of MMS, indicating that H2A might be a direct target for Mec1p. Moreover, although phosphorylation of the SQE motif had no effect on Mec1pdependent transcriptional or cellcycle responses to DNA damage, it did seem to be necessary for Mec1p-dependent repair of doublestranded DNA breaks. Given that H2A is a histone protein, one way of facilitating repair would be to alter the structure of chromatin around the broken DNA. This would allow the repair machinery easy access to the lesion. Consistent with this idea, Downs et al. found that in a mutant strain that mimicked the effect of phosphorylated SQE, the regions between individual nucleosomes were more sensitive to digestion by micrococcal nuclease — that is, compaction of the chromatin had been decreased. The carboxyl terminus of H2A sits at the part of the nucleosome where DNA enters and exits, so it’s ideally placed to influence the higher order structure of chromatin. How phosphorylation of the H2A carboxyl terminus leads to this effect is just one of the next lines of enquiry that the authors will be following up. Alison Mitchell References and links ORIGINAL RESEARCH PAPER Downs, J. A.,

Lowndes, N. F. & Jackson, S. P. A role for Saccharomyces cerevisiae histone H2A in DNA repair. Nature 408, 1001–1004 (2000)

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 2 | FEBRUARY 2001 | 8 7

REVIEWS HOW DROSOPHILA APPENDAGES DEVELOP Ginés Morata Just a glance at the body of the fruitfly Drosophila reveals that it has a main body part — the trunk — and a number of specialized appendages such as legs, wings, halteres and antennae. How do Drosophila appendages develop, what gives each appendage its unique identity, and what can the fruitfly teach us about appendage development in vertebrates?

IMAGINAL DISCS

Sac-like structures present in the larvae and composed of cells destined to form the different adult cuticular structures. They are named after the adult part that they make: for example, wing, leg, eye-antennal, genital. SEGMENT

The insect body is divided along the anteroposterior axis into a number of individual units or segments. This organization is visible in embryos, larvae and adult insects. LINEAGE

The cellular ancestry of a given structure. Two structures share the same lineage if the progeny of a single cell (a clone) can contribute to both. Otherwise they have different lineages.

Centro de Biología Molecular, Consejo Superior de Investigaciones Cientificas, Universidad Autónoma de Madrid, Madrid 28049, Spain. e-mail: gmorata@cbm.uam.es

Drosophila has a complicated life cycle that includes an embryonic phase, three larval periods (termed instars), a pupal stage and finally adulthood. Unlike in more primitive insects, these stages do not represent developmental forms that gradually evolve towards the adult stage, but are more like polymorphic forms of the same organism. During the pupal stage there is a complete metamorphosis in which nearly all larval structures disintegrate and are replaced by the structures of the adult fly. The latter are differentiated by groups of ‘imaginal’ cells that were present and had proliferated in the larva but did not contribute to the larval patterns. The best-characterized imaginal cells are those of the IMAGINAL DISCS — sac-like structures that form the adult cuticular structures (the hard, external skeleton of arthropods). It is from these imaginal discs that appendages arise, in a SEGMENT-specific manner, along with the cells that form the trunk corresponding to the segment. Different imaginal discs have their particular size and shape, and are named after the part of the body they form: wing disc, leg disc, eye-antennal disc, genital disc and so forth (FIG. 1). It is not always obvious where to draw the border between appendage and trunk. In the leg, for example, there is no clear morphological landmark separating the two parts, either in the differentiated structure or in the imaginal disc. Also, the trunk and appendages do not originate from different LINEAGES of cells: CLONAL 1,2 ANALYSES have shown that at least until late in development the descendants of a single cell can contribute to both structures. So what, exactly, defines an appendage? Some structures are not easily recognized as appendages at first sight (FIG. 1). It is obvious that a

leg, a wing or an antenna are appendages, but to call the mouthparts or the external analia appendages requires some elaboration. Ultimately, the distinction between appendage and trunk rests on different genetic and developmental mechanisms operating in the appendages and in the trunk. Some of the genetic mechanisms involved in the development of Drosophila appendages seem also to function in vertebrate limbs, suggesting that there is a universal appendage design common to all animals. Homeobox genes and morphogens

To appreciate how appendages subdivide and how they develop from the imaginal cells, it is necessary to consider some general processes that are not specific to appendages but have critical roles in their development: the process of compartmentalization and the function of two groups of molecular players — HOMEOBOX genes and morphogens. These molecules orchestrate the developmental signalling pathways that both differentiate limb from trunk, and make one limb different from another. Compartments. Originally defined in the wing disc3, compartments are basic components of the Drosophila body plan. They are parts of the body that originate from the same lineage of cells. From the beginning, the disc primordia in all segments contain two separate cell lineages, which form the anterior and posterior compartments3–5 of each adult segment. This is a consequence of the function of the homeobox gene engrailed (en), which segregates anterior and posterior compartment cells 6. In the imaginal

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 2 | FEBRUARY 2001 | 8 9

REVIEWS

CLONAL ANALYSIS

A type of study based on experiments in which individual cells are marked genetically during development. In the final structure the progeny of each of these cells will appear as a group of marked cells — a clone. HOMEOBOX

A 180-bp sequence present in many developmental genes of animals and plants. It encodes a DNA-binding helix–turn–helix motif, indicating that homeobox-containing gene products function as transcription factors.

discs, en behaves as a classical SELECTOR GENE6,7; it specifies the identity of the posterior compartment cells by selecting between anterior or posterior developmental programmes. The activity of en in the posterior compartment cells triggers a genetic and signalling cascade that patterns the appendage (reviewed in REFS 8,9). en activates the hedgehog (hh) gene, which encodes a secreted protein. The effect of the Hh protein in the posterior compartment cells is blocked by en, but the Wing

Haltere

Leg

SELECTOR GENES

The concept of selector genes is intimately linked to that of compartments — body regions of fixed lineage. Selector genes become activated precisely within compartments, where they ‘select’ specific developmental routes. Classical examples of selector genes are engrailed, apterous and the BX-C genes.

Labial Wing

HALTERE

A small dorsal appendage in the third thoracic segment, thought to be involved in flight control.

Leg 3 Haltere

Genital Leg 1

Leg 2

Prothoracic Clypeal Wing

LIM DOMAIN

A repeat of about 60 amino acids containing cysteine and histidine residues. It is thought to be involved in protein–protein interactions.

Antenna

HOMEOTIC MUTATIONS

A class of mutations in which a given organ or a segment develops in the same way as one normally present in another part of the body.

III Haltere

Analia

Leg 3 Mouthparts Leg 1

Leg 2

Figure 1 | Imaginal discs and the structures that develop from them. a | A third-instar larva showing the position of all imaginal discs (colour coded according to the appendages that they will develop into). Photographs of wing, haltere and leg discs are shown. In the leg discs the presumptive regions corresponding to trunk and appendage are indicated. b | The body of the adult Drosophila, indicating the trunk and the different appendages. In addition to wings, halteres, antennae and legs, the mouthparts and analia can also be considered as appendages. Appendages are segmentspecific structures: the second thoracic segment (II) develops the wing in the dorsal and the second leg in the ventral region. The third thoracic segment (III) develops the haltere and the third leg.

protein can move across the anterioposterior (A/P) border to the anterior compartment (FIG. 2). In the anterior cells close to the border, Hh induces the activity of the signalling genes wingless (wg, which encodes a member of the Wnt family of proteins) and/or decapentaplegic (dpp, which encodes a transforming growth factor-β homologue), which are largely responsible for patterning the appendages (see below). In addition, the activity of en in the posterior cells confers on them specific adhesion properties that make them immiscible with anterior cells. Expression of en sets up the expression of a series of subsidiary genes, some of which are presumed to encode specific cell-adhesion molecules. There is recent evidence that this function is mediated in part by the signalling protein Hh10–12 in cells close to the A/P border, as indicated in FIG. 2. This suggests that there might be an abrupt discontinuity of affinities at the A/P border, a useful property to keep the two compartments separate. In the wing disc, and probably also in the HALTERE disc, a second compartment subdivision appears during larval development1,3 separating the dorsal from the ventral regions of the pre-existing anterior and posterior compartments. This subdivision was ignored for a long time until it was found that the LIM-homeobox gene apterous (ap) is expressed in the dorsal compartment as precisely defined by the dorsoventral (D/V) border13,14. ap, like en, has the features that are normally associated with selector genes. It is involved in a binary switch that selects alternative pathways: the ON state specifies dorsal and OFF specifies ventral identity. ap activity is also important for the growth and patterning of the wing. It activates the gene fringe (fng), which modulates the ability of two ligands, Delta and Serrate, to activate their receptor Notch15,16. In turn, Notch induces vestigial (vg, which encodes a nuclear protein presumed to be a transcription factor17) and wg activity at the D/V border. Wg acts as an organizer of wing development18,19, expanding the vg domain and activating other response genes. Notch activation by ap is also necessary for the maintenance of the D/V lineage segregation; cells lacking either Notch or fng activity fail to recognize the D/V boundary20,21, even though they contain normal ap activity. Thus, there seems to be a selector-gene-driven signalling mechanism necessary for maintenance of the A/P and the D/V compartment borders. The A/P and D/V subdivisions are of great developmental importance as they establish fixed developmental borders — sources of the morphogenetic signals Hh, Dpp and Wg (see below) — that are largely responsible for patterning the appendages8,22. Homeobox genes. The terms homeobox and HOMEOTIC are often, erroneously, used interchangeably, so some clarification is required. Homeotic genes control the development of entire body regions along the A/P axis. Also, the borders of homeotic gene expression often correspond to the borders of compartments23. In Drosophila, homeotic genes are arranged in two separate clusters of tightly linked genes: the Antennapedia www.nature.com/reviews/molcellbio

| FEBRUARY 2001 | VOLUME 2

REVIEWS

BITHORAX COMPLEX (BX-C) GENES

Anterior compartment

Posterior compartment en

A group of three adjacent homeotic genes responsible for the identity of part of the thorax and the abdomen of the fly. Together with the antennapedia complex (ANTC), it forms the Hox gene cluster in Drosophila.

ci Ci(act)

Ci(rep)

Anterior-specific cell adhesion gene(s)

Posterior-specific cell adhesion gene(s)

Figure 2 | Generation of cell affinity differences between anterior and posterior compartment cells in the wing disc. The activity of engrailed (en) in the posterior cells induces hedgehog (hh), but the Hh pathway (vertical arrows in grey) cannot be activated in the posterior cells (probably because en represses cubitus interruptus (ci), the transcription factor that activates hh target genes in the cell nucleus). However, the Hh protein can travel across the anterioposterior (A/P) border to the anterior cells where the Hh pathway can be triggered. The end result of the Hh transduction cascade is that cleavage of the Ci protein is prevented. The uncleaved form Ci(act) becomes active. Ci(rep) represents the cleaved form, which acts as a repressor. The Ci(act) protein is thought to activate anterior-specific celladhesion genes. In the posterior compartment, en activates posterior-specific cell-adhesion genes, although it is possible that both en and ci induce different levels of expression in the same cell-adhesion gene12. Hh can only travel a short distance so the Ci(act) protein is only produced in the anterior cells close to the A/P border. This ensures that the maximal differences in cell affinities appear in the vicinity of the border.

Complex (ANT-C, named after the Antp mutation, which converts antennae to legs), which contains five homeotic genes, and the Bithorax Complex (BX-C, named after a mutation that converts halteres to wings), which contains three. The only homeotic gene that does not fall into one of these clusters is caudal (cad), which is responsible for analia development24.

Box 1 | The diverse functions of Hth and Exd

In addition to their interaction with Hedgehog, Decapentaplegic and Wingless signalling, the exd/hth genes have several important functions during development, some of which have not been considered in this review as they are not directly relevant to appendage development. A principal one is to confer in vivo specificity to Hox genes by acting as cofactors of Hox proteins: Hth and Exd modulate the affinity and specificity of the Hox proteins to their DNA target sites. The vertebrate homologue of Exd, Pbx, also associates with Hox proteins and presumably has a similar role in conferring Hox specificity (reviewed in REF. 96). hth and exd also have a Hox-like gene function to promote antennal development75; they participate in a binary switch that determines segment identity; the OFF state results in leg development, whereas the ON state promotes antennal development. Ectopic hth/exd activity can also induce antennal development, indicating that Hth/Exd can trigger a whole developmental programme. hth and exd are also involved in head development by repressing eye formation in the ventral head97, in the distinction of the dorsal and ventral identity in the adult abdomen, and in the positioning of thoracic bristles97,98.

In the absence of homeotic gene activity, all segments develop the same ‘ground’ pattern, a mixture of thoracic and cephalic pattern elements; no morphological diversity is generated along the A/P body axis. The ANT-C and BX-C genes encode transcription factors that contain a 180-base-pair (bp) stretch of sequence homology — the homeobox25,26 — coding for a helix–turn–helix DNA-binding motif, and are conserved throughout metazoans. In vertebrates, the number of homeotic genes is greater than in flies owing to cascade duplications that occurred during vertebrate evolution, but homologues of Drosophila’s ANT-C and BX-C complexes form single clusters, known as the Hox gene clusters. This name is now also used to describe the Drosophila ANT-C and BX-C genes. Drosophila and vertebrates also have homeoboxcontaining genes outside the Hox gene cluster that are not Hox genes (reviewed in REF. 27). There are more than 100 homeobox sequences in the Drosophila genome28. Of these, at least 25 represent true genetic functions as their mutations give rise to mutant phenotypes29. Many are involved in development and some, such as en, Distal-less (Dll), ap, extradenticle (exd) and homothorax (hth), are critical during appendage development. hth and exd deserve special mention as they are involved in several important developmental processes (see BOX 1). These two genes are highly conserved in insects and vertebrates30,31. exd is transcribed and translated all over the body but it is regulated at the subcellular level: the Exd protein is functional only when it localizes to the cell nucleus32,33. Hth behaves as a positive regulator of exd by promoting the transport of Exd to the cell nucleus31. Conversely, in the absence of exd activity, the Hth protein seems to be degraded34. Therefore hth and exd functions are always associated and can be considered as a single functional unit. Morphogens. The extracellular signalling molecules Hh, Dpp and Wg have critical functions in the processes of gene activation and pattern formation of the embryo, larvae and adult of Drosophila — and presumably throughout the animal kingdom. These molecules are morphogens (that is, carriers of positional information) that are secreted from a fixed source, establishing a concentration gradient from the origin. During the development of the appendages the sources are positioned along the A/P and D/V compartment boundaries. In their target cells, morphogens induce a cascade of events (reviewed in REFS 8,9) that, ultimately, are resolved in the nucleus of the receiving cell by transcription factors that regulate the activity of distinct response genes. The identity of these response genes often depends on the local concentration of morphogen, which is a measure of the physical distance from the source. Several developmentally important genes — many encoding transcription factors — have been identified that are activated by Wg and/or Dpp. There is good evidence that Wg and Dpp function as long-range signals; that is, they influence the gene activity of cells located many

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 2 | FEBRUARY 2001 | 9 1

REVIEWS

Dorsal Anterior

Posterior

Wing primordium

Leg primordium

Ventral

vg & hth/exd esg/sna

hth/exd esg

Dll

en/hh

Figure 3 | Gene activity in the leg and wing imaginal disc primordia in a late embryo (stage 14). Both primordia possess engrailed (en) and hedgehog (hh) activity (black lines) restricted to the posterior compartments. The leg primordium shows Distal-less (Dll ) activity (red) restricted to the central region and coincidental expression of homothorax (hth)/extradenticle (exd) and escargot (esg) (green) in a ring of cells surrounding the Dll domain. The wing primordium shows less diversity of gene expression. All cells express vestigial (vg), hth/exd , esg and snail (sna) (blue). The dotted line from the dorsal to the ventral of the embryo separates anterior and posterior compartments.

cell diameters away from the origin, whereas Hh, at least in the appendages, is a short-range morphogen. Allocation of appendage cells

The specification of the embryonic cells destined to form appendages occurs during embryogenesis. The primordia of the wing, haltere and leg imaginal discs originate in a lateral position of the embryo, at the intersection between two â&#x20AC;&#x2DC;stripesâ&#x20AC;&#x2122; expressing the Wg and Dpp morphogens35. A critical event is the activation of Dll36, which becomes a developmental and molecular marker of embryonic primordia. Dll expression is restricted to a lateral position in the embryo by the combined effects of Wg, Dpp and the Drosophila epidermal growth factor homologue spitz37. Studies on the expression of Dll homologues in many animal groups indicate that it has a basic function in the formation of body outgrowths throughout the animal kingdom38. In principle, all segments have the potential to develop appendages: in the absence of the BX-C genes, which repress Dll transcription39, Dll is activated in an equivalent site in all abdominal segments and appendage primordia are formed. This fits nicely with the idea that insects derive from multilegged ancestors40 that subsequently lost legs in their abdominal segments.

The initial leg disc is subdivided into central and peripheral domains (FIG. 3), with different patterns of gene expression that correspond to appendage and body trunk in adult flies. The cells in the central domain express Dll and those in the peripheral one express escargot37 (esg), which encodes a zinc-fingercontaining transcription factor41, and the two related homeobox genes hth and exd (FIG. 3). This probably reflects a genetic diversification that occurs during the early phase of disc development. During early embryogenesis, hth and exd are expressed highly and uniformly in the thorax, including the cells that originally express Dll. Later, Dll represses hth transcription (and as a consequence exd becomes inactive), thus establishing a distinction between hth/exd-expressing and nonexpressing cells. The existence of these two distinct domains is also indicated by the observation that Dll mutant cells can contribute only to ventral body wall and proximal leg structures42, that is, the hth/exd/esgexpressing domain. All these observations indicate that by the time it is set apart from other embryonic cells, the leg disc shows a complex pattern of gene expression: en, esg, Dll and hth are expressed in restricted domains (FIG. 3). The wing (and the haltere) disc originates from a subset of Dll-expressing cells that migrate dorsally. Three genes encoding putative transcription factors become transcriptionally active in these cells43: vg, snail (sna) and esg. Sna and Esg are related proteins that contain similar zinc-finger domains41,44. It is likely that these three genes are originally activated independently as a response to an unidentified induction, but later vg expression falls under the control of snail and esg43. Thereafter vg becomes a marker of wing and haltere development. Indeed, vg activity is essential to form the wing and haltere: in vg mutants these structures are not formed17. The wing and haltere are organized more simply than the leg disc at this stage (FIG. 3). For example, hth and exd, are uniformly expressed in all cells in the early discs45,46. Distinguishing trunk from appendage

The trunk and appendage regions do not derive from different lineages of cells but, from the beginning, they show different patterns of gene expression. How is the trunk/appendage distinction achieved? The critical element seems to be antagonism between the Hh/Wg/Dpp signalling pathways and hth/exd function (BOX 1). In the developing leg, the hh gene is activated by en in the posterior compartment cells47, but the secreted Hh protein induces the activation of dpp in anterior dorsal cells and of wg in anterior ventral cells close to the A/P boundary48. It is not clear how this difference arises, but wg is initially expressed in the ventral region of the early disc34,49 and this asymmetry might be maintained in later stages by the mutual antagonism between the Wg and Dpp pathways described below. The diffusion of Dpp and Wg from their origin in anterior and posterior directions organizes the leg pattern48. However, although Wg and Dpp are present in www.nature.com/reviews/molcellbio

| FEBRUARY 2001 | VOLUME 2

ÂŠ 2001 Macmillan Magazines Ltd

REVIEWS

Dorsal

Dpp

hth/exd

dac en

dac&Dll Posterior

Dll

hh Wg Ventral

hth/exd

Dll dac

Coxa Pleura

Femur Trochanter

Tibia

Tarsal segments

Figure 4 | Subdivision of the mature leg disc and the adult leg into distinct genetic subdomains. a | The disposition of the hedgehog (hh) (yellow), decapentaplegic (dpp) (dark blue) and wingless (wg) (red) domains. The thick arrows illustrate that the Dpp and Wg signals diffuse from their source to organize the pattern of the anterior and posterior compartments. b | Depending on their concentration they may activate Distal-less (Dll ) alone (red), Dll and dachshund (dac) (purple) or dac alone (blue). These appendage subdomains are further subdivided into smaller ones by the activity of genes that lie downstream of Dll or dac. The Dll/dac-expressing region would be the genuine appendage and the homothorax (hth)/extradenticle (exd )-expressing region an expansion of the trunk. c | The different subdomains in an adult leg. The dotted line extending from the Dll to the hth/exd domain is to illustrate that in young discs Dll and hth/exd define complementary domains, but this is later modified by the activation of dac.

both the presumptive appendage and trunk regions, they seem to be functional only in the appendage: the loss of activity of hh, wg or dpp results in the loss of the distal part of the leg (the tarsal segments, basitarsus, tibia and femur, FIG. 4), but the proximal segments (coxa and trochanter) and the body wall are unaffected50,51. This indicates a subdivision of the leg into two regions, an Hh-dependent and an Hh-independent one (FIG. 4). The Hh-dependent region is characterized during early disc development by the expression of Dll. In more advanced discs, another response gene, dachshund (dac), which encodes a nuclear protein necessary for eye and leg development52, is activated. Together, the expression domains of Dll and dac cover the entire Hh-dependent region in mature leg discs53,54. The Hh-independent region is genetically characterized by the activity of hth and exd, which behave as a single functional unit. How are the Hh-dependent and the hth/exd domains developmentally segregated? The current model is as follows34,54: the topology of the disc and the localization of the sources of Dpp and Wg ensure that cells in the central portion of the disc, which are destined to differentiate into distal structures, receive moderate to high levels of both Dpp and Wg. These levels are required for expression of Dll and dac31

which, in turn, repress hth/exd activity34,54,55. In the peripheral region, the levels of Dpp and Wg are lower and cannot activate either Dll or dac — but they could still activate other genes of the Hh cascade — and so hth/exd remains active. Where they are active, hth/exd completely blocks the response of Dpp and Wg target genes, in effect functionally eliminating Hh signalling. The antagonism between exd function and Hh signalling can be readily shown in experiments in which hth/exd function is forced in the Dll domain, which results in the loss of the appendage51. This mutual antagonism segregates two domains (FIG. 4): a central region where Dpp and Wg response genes are active and that will form the appendage, and a peripheral region where appendage development is prevented and that retains trunk features. In the wing disc the situation is more complicated. Whereas the disposition and function of Hh, Dpp and Wg remain constant during leg development, this is not the case during wing disc development, owing to additional regulatory mechanisms. For example, wg is expressed in the early wing disc in ventral anterior cells, just like in the leg disc, but later it falls under the control of Notch signalling56. This generates a new domain of wg expression in the D/V boundary. This secondary tier of Wg regulation is critical for establishing vg expression and the development of the appendage. Additionally, Hh has two distinct (dppmediated and non-dpp-mediated) functions57,58, a complication that does not occur in the leg. However, in the wing disc, as in the leg disc, there is antagonism between hth/exd and the Dpp and Wg signals45,46. In this case, the Dpp and Wg target gene that is involved in the repression of hth/exd seems to be vg, which fulfils a function similar to that of Dll in the leg disc. The involvement of Wg in the trunk/appendage segregation in the wing disc is illustrated by the mutation wg1, which duplicates the thorax at the expense of the wing. In this mutation the early wg function in the wing disc is defective56. Subdividing the appendages

Differential responses to Dpp and Wg determine further genetic diversification during appendage development. Whereas early discs only show abutting Hth and Dll domains, in mature discs the non-hth part contains a region (FIG. 4) that does not express Dll but expresses dac. This gene is activated by lower Dpp and Wg levels than Dll, but is repressed by high signal levels53. The localization of the sources of Wg and Dpp dictate that the central region of the disc will contain Dll activity whereas dac will be activated in a more peripheral ring that overlaps with the Dll domain. This subdivides the appendage part of the leg into three domains of gene expression along the proximodistal axis (FIG. 4). Thus the polarity along the proximodistal axis is a consequence of the localization of Wg and Dpp sources in the disc. The process of genetic diversification along the proximodistal axis is further refined by the activity of Dll and dac target genes that define smaller subdomains. Several genes have

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 2 | FEBRUARY 2001 | 9 3

REVIEWS ence between the more central and the more peripheral regions of the disc. Wg diffuses from the D/V border15,16 and presumably subdivides the wing along the proximodistal axis.

Wing

Specifying appendage identity

pb hth/exd

Antenna

Dll

Scr Antp

I II

Ubx

III Haltere

cad

Analia

vg target genes Ubx

Leg 3 Mouthparts Leg 1

Leg 2

Figure 5 | Specification of appendage identity by the combinatorial contributions of Hox genes and Distal-less and vestigial. In the ventral appendages, the general activity of Distalless (Dll ) is qualified in each segment by the resident Hox gene (proboscipedia (pb), homothorax (hth)/extradenticle (exd), Sex combs reduced (Scr), Antennapedia complex (Antp), Ultrabithorax (Ubx) and caudal (cad )), to determine the identity of the appendage. For the two dorsal appendages — wings and halteres — vestigial (vg) and Ubx seem to specify haltere development, but no Hox gene activity is found in the wing, which may be specified solely by vg.

GENETIC DOMAINS

A general term referring to the region of the body where a particular gene is expressed. They are usually visualized using specific antibody, DNA or RNA probes.

been identified that are associated with individual segments of the leg. These genes (ap, aristaless, Bar, bric a brac, spineless)59–65 are initially expressed in overlapping domains but later establish exclusive domains through transcriptional repression63. Interestingly, all these genes also encode transcription factors, suggesting the existence of several layers of transcriptional regulation during leg development. One outcome of this process of genetic subdivision is the localized activation of Notch ligands Delta and Serrate, which leads to rings of restricted activation of Notch that result in the formation of joints between leg segments66. Other response genes encoding transcription factors such as optomotor-blind (omb) and H15 are induced specifically by either Dpp or Wg and contribute to the identity of the dorsal and ventral regions respectively67,68. The Wg and Dpp pathways antagonize each other68–70 and these antagonistic interactions are probably responsible for the developmental segregation of dorsal and ventral GENETIC DOMAINS in the leg. In the wing disc, diffusion of the Hh-dependent long-range Dpp morphogen from the vicinity of the A/P border patterns the anterior and posterior compartments71,72. As a result, the genes omb and spalt (both of which encode transcription factors) are activated in different domains and presumably affect different sets of target genes. Hh also has a short-range dpp-independent patterning function in the anterior compartment57,58. These functions define distinct genetic domains in the anterior wing. Moreover, the antagonistic function of brinker73, a transcriptional repressor of dpp target genes, also establishes a differ-

Drosophila appendages, then, fall into two categories with broadly similar developmental mechanisms: the ventral ones, exemplified by the leg, and the dorsal ones, exemplified by the wing (FIG. 5). However, superimposed on these are mechanisms to discriminate the identity of the different ventral or dorsal appendages (FIG. 5). Two main components seem to act in combination: a segment-specific property provided by the Hox genes and a general ventral or dorsal property provided by Dpp and Wg response genes such as Dll or vg. The identity of segments along the A/P body axis is established by the Hox genes (reviewed in REF. 21). As parts of thoracic and cephalic segments, the appendages are also specified by the Hox genes. The legs provide the best example74: the identity of the first leg is specified by the Hox gene Sex combs reduced, the second leg is specified by Antp and the third leg by Ubx (FIG. 5). The identity of the antennal segment is specified by hth/exd 75 (BOX 1). Finally, the other appendagelike structures, the analia, are specified by the Hox gene caudal (cad)24. Dll has a critical role in the specification of ventral appendages. It seems to confer a general ‘ventral appendage’ property, which is qualified combinatorially by Hox genes (FIG. 5). Dll activity is necessary for the growth and identity of ventral appendages: legs, antennae, mouthparts and analia2,35,41,76. Moreover, ectopic expression of Dll in dorsal discs induces the formation of distinct ventral appendage structures depending on the pattern of Hox gene expression2. The specification of the identity of the dorsal appendages seems to follow similar rules. vg seems to have a principal role, homologous to that of Dll in the ventral appendages. Its activity is necessary for the formation of wings and halteres and its ectopic expression induces wing or haltere tissue depending on the segment77,78. The fact that it induces wing development in the head, and fore- and midleg segments, but haltere development in the hindleg segment suggests that, like Dll, vg specifies distinct appendage identity in combination with Hox genes. It is less clear how Hox genes contribute to the specification of dorsal appendages. The classical view79–81 is that the development of the haltere is specified by Ubx, whereas wing identity is a default situation that results from no Hox gene activity. Some genetic data support this view. For example, in the homeotic mutation Contrabithorax the wing is transformed into a haltere, and the transformation is accompanied by inappropriate expression of Ubx in the wing cells82. Several Ubx target genes have been identified that are expressed in the wing but not in the haltere83. These genes are probably responsible for some of the differences between wing and haltere. However, some results question the specificity of Ubx in haltere development84. Using new www.nature.com/reviews/molcellbio

| FEBRUARY 2001 | VOLUME 2

REVIEWS

Table 1 | Comparing the development of different appendages Antennae (compared with leg)

Analia (compared with leg)

Mouthparts (compared Halteres (compared with leg) with wing)

Similarities • Similar disposition of the Hh, • hh, wg and dpp expression patterns resemble Dpp and Wg signals76 those in the leg disc103,104 • Dll is activated as a response to • The Dll domain (distal) is not segregated Wg and Dpp signals by cell lineage from the proximal domain, • Legs and antennae can be where Dll expression is repressed transformed one into another by (characterized, in analia, by homeotic mutations75,97–100 brachyenteron expression)24 • Can be homeotically transformed into antennae by ectopically expressing hth75

• Labial disc expresses DII • Disposition of the Hh, in early development76

transform the mouthparts and haltere discs into legs105 • Homeotic transformations between halteres and wings can be induced

Differences • Anteroposterior compartment • Proximal region does not express hth/exd, border is established much later so restriction of Wg and Dpp signalling is than in leg101 not achieved by antagonism from hth/exd22 • Areas of hth/exd and DII expression overlap75; hth/exd and DII pathways seem to co-operate in antennal specification101

methods to target gene expression85 it has been found that the wing-to-haltere transformation can be induced not only by Ubx, but also by the other BX-C genes abdominal-A and Abdominal-B, the normal function of which is to specify abdominal identities86. This indicates a lack of specificity of Hox genes in the determination of appendage identity. This might be due to the fact that the appendages do not have hth/exd function, which is necessary for Hox gene specificity. One possibility is that the differences between wings and halteres are mediated by Dpp and Wg response genes (in addition to vg). Several of these target genes, including the transcription factors spalt, achaete-scute, and serum response factor, are active in the wing but not in the haltere83 and are likely to contribute towards making a haltere different from a wing. Their response in the haltere is blocked by Ubx, which is the BX-C gene that is normally expressed in the haltere, but the other BX-C genes might also have the same ability. In this view the differences between wings and halteres would be due to a partial blockage in the Hh/Dpp/Wg signalling pathway in halteres. A general principle for making an appendage?

A principal feature of the organization of Drosophila appendages is the distinction between the distal part (the appendage proper) that contains full activity of the Hh/Dpp/Wg cascade, and the proximal part that contains hth/exd activity (see TABLE 1 for simalarities and differences in the various appendages). The latter can be considered as an expansion of the trunk. Recent work comparing Drosophila appendages with mouse and chicken limbs 54,87,88 has shown a striking degree of conservation of this configuration: the vertebrate homologues of hth and exd — meis and Pbx1 respectively — are functional in the proximal but not in the distal limb. Even the subcellular regulation of exd by hth in Drosophila is also observed for Pbx1 and meis in mouse limb development. In both Drosophila and chicken, ectopic expression of hth or meis in the distal component blocks limb development 87. Furthermore, there is

Dpp and Wg signals is

• Proboscipedia mutations very similar in the wing

• Some Dpp target genes in the wing are not expressed • No (or less) wg expression in the D/V border in the posterior haltere83

evidence that Hh signalling is required in mouse limbs, for these do not develop properly in Sonic hedgehog (Shh) knockouts89. The mode of action of Shh in chick limb development also resembles the situation in Drosophila: it acts as a short-range activator of bone morphogenetic protein (Bmp2), a vertebrate homologue of dpp, and much of the effect of Shh is mediated by Bmp2 and other Bmps90. Similarly, the role of Dll in appendage formation seems to be highly conserved. Studies done in many animal groups38 have shown that Dll homologues are generally expressed in body outgrowths, be these arthropod appendages, tube feet of sea urchins or mouse and chicken limb buds. These observations indicate that Dll might derive from an ancient homeobox gene involved in the production of body wall outgrowths in the entire animal kingdom38. Thus, the key genes involved in distinguishing trunk from appendage in Drosophila, hth/exd and Dll, have been highly conserved during evolution. The patterning signals Hh, Wg and Dpp also have comparable functions in different animals. Together these observations indicate that there is a universal mechanism to form a limb, shared by all animal groups38. Future goals

The appendages represent specialized forms of development, likely to derive from primitive body wall outgrowths. The formation of these outgrowths requires the identification and developmental isolation of groups of cells in the trunk, and the triggering of their additional proliferation. During this process, Hh signalling seems to function through the activation of specific targets such as Dll or vg. These genes are not only needed for appendage growth, but also to contribute to appendage identity, acting in combination with the Hox genes. Because the Hox genes, Dll and vg encode transcription factors, it is clear that better understanding of appendage development will require the identification of new target genes of both Hox genes and Hh/Dpp/Wg signalling. Moreover, it is expected that there must be molecular interactions

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 2 | FEBRUARY 2001 | 9 5

REVIEWS between the Hox gene and the signalling cascades, as illustrated by the gene spalt, which is a target gene of both Dpp and the Hox gene Ubx83. Another aspect worth mentioning concerns the function of the longrange signals Dpp and Wg. These signals are present both in the adult trunk and in the appendages, but very little is known about their function in the trunk, where, for example, no target gene has yet been identified. Another issue that remains to be explored is the molecular basis of the antagonistic roles of hth/exd and the Hh, Dpp and Wg pathways in growth control. The observation in Drosophila and vertebrates that forcing ectopic hth/exd expression in the domains of Hh, Dpp and Wg blocks appendage development45,46,87 suggests that the proliferative response to the signals is suppressed by hth/exd. How this is achieved remains to be elucidated. Studies in the mechanisms of appendage development will be of interest, not only because of the general conservation of the process, but also because of their importance for understanding the general mechanisms controlling growth and proliferation in higher animals. Mutations of human homologues of exd and hth (Pbx1 and meis) give rise to leukaemias91–93 in humans and

Garcia-Bellido, A., Ripoll, P. & Morata, G. Developmental compartmentalization in the dorsal mesothoracic disc of Drosophila. Dev. Biol. 48, 132–147 (1976). 2. Gorfinkiel, N., Morata, G. & Guerrero, I. The homeobox gene Distal-less induces ventral appendage development in Drosophila. Genes Dev. 11, 2259–2271 (1997). 3. Garcia-Bellido, A., Ripoll, P. & Morata, G. Developmental compartmentalization of the wing disk of Drosophila. Nature New Biol. 245, 251–253 (1973). First demonstration of the existence of compartments in Drosophila. 4. Steiner, E. Establishment of compartments in the developing imaginal leg discs of Drosophila. Wilhelm Roux Arch. Dev. Biol. 180, 31–46 (1976). 5. Lawrence, P. A. & Morata, G. The early development of mesothoracic compartments in Drosophila. An analysis of cell lineage and fate mapping and an assessment of methods. Dev. Biol. 56, 40–51 (1977). 6. Morata, G. & Lawrence, P. A. Control of compartment development by the engrailed gene in Drosophila. Nature 255, 614–617 (1975). Provides evidence for the role of engrailed in maintaining the A/P compartment border in the wing disc. 7. Garcia-Bellido, A. in Cell Patterning, Ciba Found. Symp. Vol. 29 (ed. S. Brenner) 161–182 (Associated Scientific Publishers, New York, 1975). 8. Lawrence, P. A. & Struhl, G. Morphogens, compartments, and pattern: lessons from Drosophila? Cell 85, 951–961 (1996). A lucid review on patterning mechanisms in Drosophila relating the classical ideas of compartments and homeotic genes with the function of the Hh, Wg and Dpp morphogens. 9. Strigini, M. & Cohen, S. Formation of morphogen gradients in the Drosophila wing. Semin. Cell Dev. Biol. 10, 335–344 (1999). 10. Rodriguez, I. & Basler, K. Control of compartmental affinity boundaries by hedgehog. Nature 389, 614–618 (1997). 11. Blair, S. S. & Ralston, A. Smoothened-mediated Hedgehog signalling is required for the maintenance of the anterior–posterior lineage restriction in the developing wing of Drosophila. Development 124, 4053–4063 (1997). References 10 and 11 provide evidence that the role of engrailed in keeping anterior and posterior cells separate is mediated by the signalling gene hedgehog. 12. Dahman, C. & Basler, K. Opposing transcriptional outputs of Hedgehog signaling and engrailed control compartmental cell sorting at the Drosophila A/P

mutant forms of genes of the Hh signalling pathway94,95 are involved in tumorigenesis. Fully understanding appendage development in Drosophila therefore has implications that go beyond developmental and evolutionary biology, to biomedicine.

Links DATABASE LINKS engrailed | hedgehog | wingless |

boundary. Cell 100, 411–422 (2000). 13. Diaz-Benjumea, F. J. & Cohen, S. M. Interaction between dorsal and ventral cells in the imaginal disc directs wing development in Drosophila. Cell 75, 741–752 (1993). 14. Blair, S. S. Mechanisms of compartment formation: evidence that non-proliferating cells do not play a critical role in defining the D/V lineage restriction in the developing wing of Drosophila. Development 119, 339–351 (1993). References 13 and 14 provide evidence for the role of the gene apterous in establishing the D/V compartment border in the wing disc and in acting as a selector gene. 15. Panin, V. M., Papayannoupolos, V., Wilson, R. & Irvine, K. D. Fringe modulates Notch-ligand interactions. Nature 387, 908–912 (1997). 16. Fleming, R. J., Gu, Y. & Hukriede, N. A. Serrate-mediated activation of Notch is specifically blocked by the product of the gene fringe in the dorsal compartment of the Drosophila wing imaginal disc. Development 124, 2973–2981 (1997). 17. Williams, J. A., Bell, J. B. & Carroll, S. B. Control of Drosophila wing and haltere development by the nuclear vestigial gene product. Genes Dev. 5, 2481–2495 (1991). 18. Zecca, M., Basler, K. & Struhl, G. Direct and long range action of a Wingless morphogen gradient. Cell 87, 833–844 (1996). 19. Neumann, C. & Cohen, S. M. A hierarchy of crossregulation involving wingless, vestigial and cut organizes the dorsal/ventral axis of the Drosophila wing. Development 122, 3477–3485 (1996). 20. Rauskolb, C., Correia, T. & Irvine, K. D. Fringe-dependent separation of dorsal and ventral cells in the Drosophila wing. Nature 401, 476–480 (1999). 21. Micchelli, C. A. & Blair, S. S. Dorsoventral lineage restriction in wing imaginal discs requires Notch. Nature 401, 473–476 (1999). 22. Dahmann, C. & Basler, K. Compartment boundaries: at the edge of development. Trends Genet. 15, 320–326 (1999). 23. Lawrence, P. A. & Morata, G. Homeobox genes: their function in Drosophila segmentation and pattern formation. Cell 78, 181–189 (1994). 24. Moreno, E. & Morata, G. caudal is the Hox gene that specifies the most posterior Drosophila segment. Nature 400, 873–877 (1999). 25. McGinnis, W., Levine, M. S., Hafen, E., Kuroiwa, A. & Gehring, W. J. A conserved DNA sequence in homeotic genes of the Drosophila Antennapedia and bithorax complexes. Nature 308, 428–433 (1984). 26. Scott, M. P. & Weiner, A. J. Structural relationships among

27.

28. 29.

30.

31.

32.

33.

34.

35.

36.

37.

38.

genes that control development: sequence homology between Antennapedia, Ultrabithorax and fushi tarazu loci of Drosophila. Proc. Natl Acad. Sci. USA 81, 4115–4119 (1984). References 25 and 26 report the discovery of the homeobox, arguing for a common origin of homeotic genes. They also provided a molecular marker of homeotic genes that eventually led to their discovery in other animal species. Burglin, T. R. in Biodiversity and Evolution (eds Arai, R., Kato, M. & Doi, Y.) 291–336 (The National Science Museum Foundation, Tokyo, 1995). Adams, M. D. et al. The genome sequence of Drosophila melanogaster. Science 287, 2185–2195 (2000). Duboule, D. (ed.) Guidebook to the Homeobox Genes (Sambrook & Tooze Publications, Oxford Univ. Press, Oxford, 1994). Rauskolb, C., Peifer, M. & Wieschaus, E. extradenticle, a regulator of homeotic gene activity, is a homolog of the homeobox-containing human proto-oncogene pbx1. Cell 74, 1–20 (1993). Rieckhof, G., Casares, F., Ryoo, H. D., Abu-Shaar, M. & Mann, R. S. Nuclear translocation of Extradenticle requires homothorax, which encodes an Extradenticle-related homeodomain protein. Cell 91, 171–183 (1997). Mann, R. & Abu-Shaar, M. Nuclear import of the homeodomain protein Extradenticle in response to Decapentaplegic and Wingless signalling. Nature 383, 630–633 (1996). Aspland, S. & White, R. A. H. Nucleocytoplasmic localisation of extradenticle protein is spatially regulated throughout development of Drosophila. Development 124, 741–747 (1997). Abu-Shaar, M. & Mann, R. S. Generation of multiple antagonistic domains along the proximodistal axis during Drosophila leg development. Development 125, 3821–3830 (1998). Cohen, B., Simcox, A. & Cohen, S. M. Allocation of thoracic imaginal primordia in the Drosophila embryo. Development 117, 597–608 (1993). Cohen, S. M., Bronner, G., Kuttner, F., Jurgens, G. & Jackle, H. Distal-less encodes a homeodomain protein required for limb development in Drosophila. Nature 338, 432–434 (1989). Goto, S. & Hayashi, S. Specification of embryonic limb primordium by graded activity of Decapentaplegic. Development 124, 125–132 (1997). Panganiban, G. et al. The origin and evolution of animal appendages. Proc. Natl Acad. Sci. USA 94, 5162–5166 (1997).

www.nature.com/reviews/molcellbio

| FEBRUARY 2001 | VOLUME 2

REVIEWS

39. Vachon, G., Cohen, B., Pfeifle, C., McGuffin, M. E., Botas, J. & Cohen, S. M. Homeotic genes of the bithorax complex repress limb development in the abdomen of the Drosophila embryo through the target gene Distal-less. Cell 71, 437–450 (1992). 40. Lewis, E. B. A gene complex controlling segmentation in Drosophila. Nature 276, 565–570 (1978). 41. Whiteley, M., Noguchi, P. D., Sensabaugh, S. M., Odenwald, W. F. & Kassis, J. A. The Drosophila gene escargot encodes a zinc finger motif found in snail-related genes. Mech. Dev. 36, 117–127 (1992). 42. Cohen, S. M. & Jurgens, G. Proximal-distal pattern formation in Drosophila: cell autonomous requirement for Distal-less gene activity in limb development. EMBO J. 8, 2045–2055 (1989). 43. Fuse, N., Hirose, S. & Hayashi, S. Determination of wing cell fate by the escargot and snail genes in Drosophila. Development 122, 1059–1067 (1996). 44. Boulay, J. L., Dennefeld, C. & Alberga, A. The Drosophila developmental gene snail encodes a protein with nucleic acid binding fingers. Nature 330, 395–398 (1987). 45. Casares, F. & Mann, R. S. A dual role for homothorax in inhibiting wing blade development and specifying proximal wing identities in Drosophila. Development 127, 1499–1508 (2000). 46. Azpiazu, N. & Morata, G. Function and regulation of homothorax in the wing imaginal disc of Drosophila. Development 127, 2685–2693 (2000). 47. Tabata, T. & Kornberg, T. Hedgehog is a signalling protein with a key role in patterning Drosophila imaginal discs. Cell 76, 89–102 (1994). 48. Basler, K. & Struhl, G. Compartment boundaries and the control of Drosophila limb pattern by hedgehog protein. Nature 368, 208–214 (1994). A classic study demonstrating the patterning role of the Hh/Wg/Dpp pathway in appendage development. 49. Campbell, G., Weaver, T. & Tomlinson, A. Axis specification in the developing Drosophila appendage: the role of wingless, decapentaplegic, and the homeobox gene aristaless. Cell 74, 1113–1123 (1993). 50. Diaz-Benjumea, F. J., Cohen, B. & Cohen, S. M. Cell interactions between compartments establish the proximal-distal axis of Drosophila legs. Nature 372, 175–179 (1994). 51. Gonzalez-Crespo, S. & Morata, G. Genetic evidence for the subdivision of the arthropod limb into coxopodite and telopodite. Development 122, 3921–3930 (1996). 52. Mardon, G., Solomon, N. M. & Rubin, G. M. dachshund encodes a nuclear protein required for normal eye and leg development in Drosophila. Development 120, 3473–3486 (1994). 53. Lecuit, T. & Cohen, S. M. Proximal-distal axis formation in the Drosophila leg. Nature 388, 139–145 (1997). Wg and Dpp activate different target genes depending on the local concentration of both morphogens along the proximodistal axis. 54. Gonzalez-Crespo, S., Torres, M., Martinez-A, C., Mann, R. S. & Morata, G. Antagonism between extradenticle function and Hedgehog signalling in the developing limb. Nature 394, 196–200 (1998). Provides evidence both in Drosophila and the mouse limb that the formation of the appendage requires the suppression of extradenticle function. Suppression is achieved by preventing the nuclear transport of the Exd protein. 55. Wu, J. & Cohen, S. M. Proximodistal axis formation in the Drosophila leg: subdivision into proximal and distal domains by Homothorax and Distal-less. Development 126, 109–117 (1999). 56. Ng, M., Diaz-Benjumea, F., Vincent, J. P., Wu, J. & Cohen, S. M. Specification of the wing by localised expression of wingless protein. Nature 381, 316–318 (1996). 57. Mullor, J. L., Calleja, M., Capdevila, J. & Guerrero, I. Hedgehog activity, independent of Decapentaplegic, participates in wing disc patterning. Development 124, 1227–1237 (1997). 58. Strigini, M. & Cohen, S. M. A Hedhehog activity gradient contributes to AP axial patterning of the Drosophila wing. Development 124, 4697–4705 (1997). 59. Cohen, B., McGuffin, M. E., Pfeifle, C., Segal, D. & Cohen, S. M. apterous, a gene required for imaginal disc development in Drosophila encodes a member of the LIM family of developmental regulatory proteins. Genes Dev. 6, 715–729 (1992). 60. Scheitz, K., Spielman, S. & Noll, M. Molecular genetics of

61.

62.

63.

64.

65.

66.

67.

68.

69.

70.

71.

72.

73.

74.

75. 76.

77.

78.

79. 80. 81.

82.

83.

aristaless, a prd-type homeobox involved in the morphogenesis of proximal and distal pattern elements in a subset of appendages in Drosophila. Genes Dev. 7, 114–129 (1993). Campbell, G., Weaver, T. & Tomlinson, A. Axis specification in the developing Drosophila appendage: the role of wingless, decapentaplegic, and the homeobox gene aristaless. Cell 74, 1113–1123 (1993). Campbell, G. & Tomlinson, A. The roles of the homeobox genes aristaless and Distal-less in patterning the legs and wings of Drosophila. Development 125, 4483–4493 (1998). Kojima, T., Sato, M. & Saigo, K. Formation and specification of distal leg segments in Drosophila by dual Bar homeobox genes, BarH1 and BarH2. Development 127, 769–778 (2000). Godt, D., Couderc, J. L., Cramton, S. E. & Laski, F. A. Pattern formation of limbs of Drosophila: bric a brac in both a gradient and a wave-like pattern and is required for specification and proper segmentation of the tarsus. Development 119, 799–812 (1993). Duncan, D. M., Burgess, E. A. & Duncan, I. Control of distal antennal identity and tarsal development in Drosophila by spineless-aristapedia, a homolog of the mammalian dioxin receptor. Genes Dev. 12, 1290–1303 (1998). de Celis, J. F., Tyler, D. M., de Celis, J. & Bray, S. J. Notch signalling mediates segmentation of the Drosophila leg. Development 125, 4617–4626 (1998). Grimm, S. & Pflugfelder, G. O. Control of the gene optomotor-blind in Drosophila wing development by decapentaplegic and wingless. Science 271, 1601–1604 (1996). Brook, W. J. & Cohen, S. M. Antagonistic interactions between Wingless and Decapentaplegic responsible for dorsal-ventral patterning in the Drosophila leg. Science 273, 1373–1377 (1996). Jiang, J. & Struhl, G. Complementary and mutually exclusive activities of Decapentaplegic and Wingless organise axial patterning during Drosophila leg development. Cell 86, 401–409 (1996). Penton, A. & Hoffmann, F. M. Decapentaplegic restricts the domain of wingless during Drosophila limb patterning. Nature 382, 162–165 (1996). Nellen, D., Bruke, R., Struhl, G. & Basler, K. Direct and long-range action of a DPP morphogen gradient. Cell 85, 357–368 (1996). Lecuit, T. et al. Two distinct mechanisms for long-range patterning by Decapentaplegic in the Drosophila wing. Nature 381, 387–393 (1996). References 71 and 72 provide direct evidence that Dpp acts at a distance from its origin and in a dosedependent manner. Jazwinska, A., Kirov, N., Wieschaus, E., Roth, S. & Rushlow, C. The Drosophila gene brinker reveals a novel mechanism of Dpp target activation. Cell 96, 563–573 (1999). Struhl, G. Genes controlling segmental specification in the Drosophila thorax. Proc. Natl Acad. Sci. USA 79, 7380–7384 (1982). A classic paper showing that Hox genes act as a combinatorial code. Casares, F. & Mann, R. S. Control of antennal versus leg development in Drosophila. Nature 392, 723–726 (1998). Cohen, S. M. in The Development of Drosophila melanogaster Vol. 2 (eds Bate, M. & Martinez-Arias, A.) 747–842 (Cold Spring Harbor Laboratory Press, New York, 1993). Kim, J. et al. Integration of positional signals and regulation of wing formation and identity by Drosophila vestigial gene. Nature 382, 133–138 (1996). Maves, L. & Schubiger, G. A molecular basis for transdetermination in Drosophila imaginal discs: interactions between wingless and decapentaplegic signaling. Development 125, 115–124 (1998). Lewis, E. B. Genes and developmental pathways. Am. Zool. 3, 33–56 (1963). Lewis, E. B. in The Role of Chromosomes in Development (ed. Locke, M.) 231–252 (Academic, New York, 1964). Morata, G. & Garcia-Bellido, A. Developmental analysis of some mutants of the bithorax system of the Drosophila. Wilhelm Roux Arch. Dev. Biol. 179, 125–143 (1976). White, R. A. H. & Akam, M. E. Contrabithorax mutations cause inappropriate expression of Ultrabithorax products in Drosophila. Nature 318, 567–569 (1985). Weatherbee, S. D., Halder, G., Kim, J., Hudson, A. &

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

84.

85.

86.

87.

88.

89.

90.

91.

92.

93.

94.

95.

96.

97.

98.

99. 100.

101.

102.

103.

104.

105.

Carroll, S. Ultrabithorax regulates genes at several levels of the wing-patterning hierarchy to shape the development of the Drosophila haltere. Genes Dev. 12, 1474–1482 (1998). Casares, F., Calleja, M. & Sanchez-Herrero, E. Functional similarity in appendage specification by the Ultrabithorax and abdominal-A Drosophila Hox genes. EMBO J. 15, 3934–3942 (1996). A striking result showing that the three BX-C genes Ubx, abd-A and Abd-B can induce haltere development in the wing. It argues for a rethinking of BX-C gene function in the appendages. Brand, A. & Perrimon, N. Targeted gene expression as a means of altering cell fates and generating dominant phenotypes. Development 118, 401–415 (1993). A very useful method to express a cloned gene in specific tissues or body regions. Sanchez-Herrero, E., Vernos, I., Marco, R. & Morata, G. Genetic organization of Drosophila bithorax complex. Nature 313, 108–113 (1985). Mercader, N. et al. Conserved regulation of the proximodistal limb axis development by Meis/Hth. Nature 402, 425–429 (1999). Capdevila, J., Tsukui, T., Rodriguez-Esteban, C., Zappavigna, V. & Izpizua-Belmonte, J. C. Mol. Cell 4, 839–849 (1999). Chiang, C. et al. Cyclopia and defective axial patterning in mice lacking Sonic Hedgehog gene function. Nature 383, 407–413 (1996). Drosospolou, G. et al. A model for anteroposterior patterning of the vertebrate limb based on the sequential long- and short-range Shh signalling and Bmp signalling. Development 127, 1337–1348 (2000). Kamps, M. P., Murre, C., Sun, X. & Baltimore, D. A new homeobox gene contributes to the DNA binding domain of the t(1;19) translocation protein in pre-B ALL. Cell 60, 547–555 (1990). Nourse, J. et al. Chromosomal translocation t(1;19) results of the synthesis of a homeobox fusion mRNA that codes for a potential chimeric transcription factor. Cell 60, 535–545 (1990). Moskow, J., Bulrich, F., Juebner, K., Daar, I. & Buchberg, A. Meis1, a PBX1-related homeobox gene involved in myeloid leukemia in BXH-2 mice. Mol. Cell. Biol. 15, 5434–5443 (1995). Hahn, H. et al. Mutations of the human homolog of Drosophila patched in the nevoid basal cell carcinoma syndrome. Cell 85, 841–851 (1996). Jonhson, R. L. et al. Human homolog of patched, a candidate gene for the basal cell nevus syndrome. Science 272, 1668–1672 (1996). Mann, R. S. & Chan, S. -K. Extra specificity from extradenticle: the partnership between HOX and exd/pbx homeodomain proteins. Trends Genet. 12, 258–262 (1996). Gonzalez-Crespo, S. & Morata, G. Control of Drosophila adult pattern by extradenticle. Development 121, 2117–2125 (1995). Rauskolb, C., Smith, K., Peifer, M. & Wieschaus, E. extradenticle determines segmental identities throughout development. Development 121, 3663–3671 (1995). Struhl, G. A homeotic mutation transforming leg to antenna in Drosophila. Nature 292, 635–638 (1981). Schneuwly, S., Klemenz, R. & Gehring, W. J. Redesigning the body plan of Drosophila by ectopic expression of the homeotic gene Antennapedia. Nature 325, 816–828 (1987). Morata, G. & Lawrence, P. A. Anterior and posterior compartments in the head of Drosophila. Nature 274, 473–474 (1978). Dong, P. D., Chu, J. & Panganiban, G. Co-expression of the homeobox genes Distal-less and homothorax determines Drosophila antennal identity. Development 127, 209–216 (2000). Chen, E. & Baker, B. S. Compartmental organization of the Drosophila genital imaginal discs. Development 124, 205–218 (1996). Sanchez, L., Casares, F., Gorkinkiel, N. & Guerrero, I. The genital disc of Drosophila melanogaster II. Role of the genes hedgehog, decapentaplegic and wingless. Dev. Genes Evol. 207, 229–241 (1997). Lindley, D. & Zimm, G. The Genome of Drosophila melanogaster (Academic, San Diego, California, 1992).

Acknowledgements I thank E. Sanchez-Herrero, J. F. de Celis, M. Calleja and D. Duboule for discussions and their comments on the manuscript.

VOLUME 2 | FEBRUARY 2001 | 9 7

REVIEWS

SNARE-MEDIATED MEMBRANE FUSION Yu A. Chen* and Richard H. Scheller‡ SNARE proteins have been proposed to mediate all intracellular membrane fusion events. There are over 30 SNARE family members in mammalian cells and each is found in a distinct subcellular compartment. It is likely that SNAREs encode aspects of membrane transport specificity but the mechanism by which this specificity is achieved remains controversial. Functional studies have provided exciting insights into how SNARE proteins interact with each other to generate the driving force needed to fuse lipid bilayers. PRESYNAPTIC

Pertaining to the neuron that transmits impulses to a synapse. SYNAPTIC CLEFT

The extracellular space, typically ~20 nm across, that separates the outer membrane of the presynaptic nerve ending from the postsynaptic membrane of the receiving cell in a synapse. POSTSYNAPTIC

Pertaining to the neuron or the muscle cell that is on the efferent side of a synapse, which transduces signals away from the synapse.

*Renovis Inc., 747 Fifty Second Street, Oakland, California 94609, USA. ‡Department of Molecular and Cellular Physiology, Howard Hughes Medical Institute, Stanford University, Stanford, California 94305, USA. e-mails: scheller@cmgm.stanford.edu; yuchen@concentric.net. Correspondence to R.H.S.

In eukaryotic cells, molecules need to be delivered to their correct intracellular destinations without compromising the structural integrity of cellular compartments. To achieve this, transport vesicles bud from an intracellular donor organelle and then target, dock and fuse with an acceptor organelle. Membrane fusion is also involved in organelle inheritance during mitosis and in cell growth or division, which require membrane addition. In the nervous system, membrane fusion is an essential step in chemical synaptic transmission because neurotransmitter-filled PRESYNAPTIC vesicles fuse in a calcium-dependent manner with the plasma membrane to release their content into the SYNAPTIC CLEFT. Furthermore, the fusion of POSTSYNAPTIC-membranereceptor-containing vesicles with the plasma membrane implicates the membrane fusion machinery in longterm modulation of synaptic strength, which is important for memory and learning1. A vesicle fusion event involves many coordinated steps. Before fusion, a vesicle is transported to its specific target membrane and docked or tethered there2. It then goes through several ‘priming’ events to prepare it for release3. A fusion trigger — Ca2+ in many trafficking events4–7 — then directs fusion to proceed to completion. To achieve this, the cellular fusion machinery must overcome the repulsive ionic forces and dissipate the hydration between the two lipid bilayers. This review focuses on the SNARE family of proteins, as they have been implicated as the conserved core protein machinery for all intracellular membrane fusion events. It summarizes the significant progress made

during the past three years in understanding the mechanism of SNARE function. The SNARE superfamily

SNARE (soluble NSF attachment protein receptor where NSF stands for N-ethyl-maleimide-sensitive fusion protein) proteins have been implicated as central in most, if not all, intracellular membrane trafficking events studied so far. The synaptic proteins syntaxin (STX1)8, SNAP-25 (25 kDa synaptosome-associated protein)9 and VAMP10 (vesicle-associated membrane protein, also called synaptobrevin11; for example see VAMP1) were the first SNAREs to be discovered. Yeast proteins that are essential for secretory function, including some SNAREs, were independently discovered in genetic screens12. Both syntaxin and VAMP are anchored to the membrane by a carboxy-terminal transmembrane domain, whereas SNAP-25 is peripherally attached to the membrane by PALMITOYLATION of four cysteine residues in the central region of the protein. SNAREs were originally divided into v-SNAREs and t-SNAREs according to their vesicle or target membrane localization13. However, to avoid ambiguity in the case of homotypic membrane fusion, SNAREs have been reclassified as R-SNAREs (argininecontaining SNAREs) or Q-SNAREs (glutamine-containing SNAREs), based on the identity of a highly conserved residue14. The hallmark of all SNARE proteins is that they contain conserved heptad repeat sequences in their membrane-proximal regions that form coiled-coil structures. More than a hundred other SNARE proteins from www.nature.com/reviews/molcellbio

| FEBRUARY 2001 | VOLUME 2

REVIEWS

STX3 VAMP2 DCV SNAP-23

VAMP1 VAMP2

VAMP1 CCV VAMP2 Early endosome

Golgi

STX5 VAMP4 SNAP-29 Membrin GOS28

STX13 IC Late STX7 endosome STX8 VAMP7 VAMP8

Specificity of membrane trafficking

CCP

STX7 STX8 VAMP3 VAMP8

Fig.1 online), which might indicate that they have selective functional involvement in specific intracellular trafficking steps.

Plasma membrane

STX1 SNAP-25 STX2 SNAP-23 STX3 VAMP5 STX4

STX6 STX10 TGN STX11 STX16 VTI1

STX5 STX11 SEC22B YKT6 BET1 STX18 STX5 SEC22B BET1 Membrin

Nucleus

STX17

RER

STX7 STX8 Lysosome

SER

Figure 1 | Subcellular localization of mammalian SNAREs. The mammalian SNAREs that have been studied so far localize to distinct subcellular compartments in the secretory pathway. (Red, syntaxin family; blue, VAMP family; green, SNAP-25 family; black, others. CCP, clathrin-coated pit; CCV, clathrin-coated vesicles; DCV, dense core vesicles; IC, intermediate compartment; RER, rough endoplasmic reticulum; SER, smooth endoplasmic reticulum; SNAP-25, 25 kDa synaptosome-associated protein; TGN, trans-Golgi network; V, vesicles; VAMP, vesicle-associated membrane protein.)

PALMITOYLATION

Covalent attachment of a palmitate (16-carbon saturated fatty acid) to a cysteine residue through a thioester bond. PC12 CELLS

A clonal line of rat adrenal pheochromocytoma cells which respond to nerve growth factor and can synthesize, store and secrete catecholamines, much like sympathetic neurons. PC12 cells contain small, clear synaptic-like vesicles and larger dense core granules. CRACKED PC12 CELL SYSTEM

Exocytosis assay in which PC12 cells are mechanically permeabilized by a ball homogenizer, and secretion of [3H] noradrenaline from dense core granules is reconstituted and measured.

diverse organisms have been discovered15. On the basis of sequence homology and domain structure, the known mammalian SNAREs have been categorized as members of the syntaxin, VAMP or SNAP-25 families. Most of them are found in specific cellular compartments15,16 (FIG. 1 and hyperlinked version of

It was initially believed that specific interactions of vSNAREs and t-SNAREs confer specificity on intracellular membrane trafficking (BOX 1). Intuitively, however, the specificity of membrane trafficking is most probably defined at the vesicle targeting and tethering stages, during which vesicles are captured and tethered by long extended proteins from the target membrane2. Small GTPases of the Rab family have been proposed to be important in the early stage of vesicle targeting and tethering, and many of them have been found to localize to distinct cellular compartments (see the review by Zerial and McBride on page 107 in this issue). Therefore, it is likely that Rab proteins are important in determining vesicular transport specificity. Most of the recent studies indicate that SNAREs might mediate membrane fusion and not docking. Is it possible, then, that the SNARE-mediated fusion specificity is superimposed on the Rab-mediated docking specificity to make the system even more reliable? A large number of SNARE homologues localize to specific membrane compartments throughout the secretory pathway, which would be wasteful if they all had the same function of fusing two attached membranes. Although some SNAREs can function in several trafficking steps and substitute for other SNAREs17, it is not clear whether this is generally true. In the CRACKED PC12 CELL SYSTEM, in which fusion between DENSE CORE GRANULES and the plasma membrane can be measured, soluble cognate SNAREs rescued or competed more successfully than non-cognate SNAREs, showing a high degree of SNARE specificity16. In recent liposome fusion experiments, both topological specificity (two t-SNAREs on one membrane and one v-SNARE on the other membrane)18 and pairing specificity19 were observed. However, a certain degree of promiscuity was also observed, as any R-SNARE could fuse with plasmamembrane Q-SNAREs19. So, it seems that the intrinsic physio-chemical properties of the SNARE proteins, as well as upstream targeting and tethering factors, encode aspects of specificity in intracellular membrane transport.

Box 1 | The SNARE hypothesis The SNARE (soluble NSF attachment protein receptor, where NSF stands for N-ethyl-maleimide-sensitive fusion protein) hypothesis was proposed in 1993, before most of the current knowledge became available, as the first working model to explain vesicle docking and fusion in molecular terms13. It postulated that each type of transport vesicle has a distinct v-SNARE that pairs up with a unique cognate t-SNARE at the appropriate target membrane, and that this specific interaction docks the vesicles at the correct membrane, with the subsequent dissociation of the SNARE complex by the ATPase activity of NSF driving membrane fusion. Although the biochemical activity of α-SNAP and NSF — dissociating the SNARE complex — has not been disputed, the specific roles of α-SNAP and NSF in this process have since been revised94. The current view is that NSF acts as a chaperone to reactivate SNAREs after one round of fusion, instead of directly driving fusion (FIG. 4). The role of SNAREs in docking, proposed by the SNARE hypothesis, was also challenged by the finding that SNARE-cleaving neurotoxins do not affect vesicle docking at the synapse95, and that SNARE-deficient flies have an increased, not decreased, number of morphologically docked vesicles96,97. Current evidence indicates that small GTPases of the Rab protein family might be crucial for docking or tethering vesicles (see the review by Zerial and McBride on page 107 in this issue), whereas SNARE pairing is involved at a later step of membrane fusion. Moreover, it is the assembly, not disassembly, of the core complex that probably drives lipid fusion (FIG. 4).

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 2 | FEBRUARY 2001 | 9 9

REVIEWS

ports the hypothesis that formation of the SNARE complex fuses two membranes by bringing them into close apposition rather than simply docking them (FIG. 2a). As might be expected for a complex that is so stable, ATP is needed to dissociate it into monomeric components. Disassembly is carried out by two proteins, the ATPase NSF and an adaptor protein, α-SNAP, (soluble NSF attachment protein), which were initially discovered as essential factors in a GOLGI TRANSPORT ASSAY28,29. Several structural studies of α-SNAP, NSF and the SNARE–SNAP–NSF complex25,30–35 indicate a possible model in which the ATP-dependent enzymatic activity of NSF might impart a rotational shear to dissociate the core complex.

VAMP SNAP-25 Syntaxin I

SNARE domains outside the helical bundle

...

Q R

Figure 2 | SNARE proteins form a four-helical bundle complex that drives membrane fusion. a | VAMP (blue) on the vesicle interacts with syntaxin (red) and SNAP-25 (green) on the plasma membrane to form a four-helix bundle that zips up concomitant with bilayer fusion. b | The backbone of the SNARE complex is shown on the left52, with the central ionic layer (red) and 15 hydrophobic layers (black) that mediate the core interactions highlighted. Top-down views of side-chain interactions are shown on the right, with the four SNARE helices shown as ribbons. The ball-and-stick structures represent the indicated amino acids; the dotted lines represent hydrogen bonds or salt bridges that stabilize interactions between SNAREs. QSNAREs and R-SNAREs are characterized by a glutamine (Q) or arginine (R) residue, respectively, in the central layer of the SNARE complex. (SNARE; soluble NSF attachment protein receptor, where NSF stands for N-ethyl-maleimide-sensitive fusion protein; SNAP-25, 25 kDa synaptosome-associated protein; VAMP, vesicle-associated membrane protein.) DENSE CORE GRANULES

Large diameter (80–200 nm) secretory vesicles that have high electron density under electronmicroscopy. They usually contain neuropeptides or catecholamines. CLOSTRIDIAL NEUROTOXINS

Bacterial toxins that potently block neurotransmitter release through their metalloproteolytic activity directed specifically towards SNARE proteins. Includes botulinum neurotoxins and tetanus toxin. GOLGI TRANSPORT ASSAY

In vitro reconstitution assay consisting of isolated Golgi stacks, Mg-ATP and cytosol, where transport-coupled glycosylation is monitored.

100

SNARE core complex

Biochemical studies have shown that the soluble coiled-coil-forming domains of recombinant syntaxin, SNAP-25 and VAMP form an extremely stable complex20,21. This ‘core complex’ is resistant to SDS denaturation22, protease digestion20–22 and CLOSTRIDIAL 22 o NEUROTOXIN cleavage , and is heat stable up to ~90 C (REF. 23). The affinity within the ternary complex is markedly higher than the pairwise binary affinities between the three proteins22,24. However, the exact dissociation constant of the ternary interaction has not yet been determined. The crystal structure of the neuronal SNARE core complex (BOX 2) shows that one coil of syntaxin and VAMP, and two coils of SNAP-25 intertwine to form a four-stranded coiled-coil structure (FIGS 2, 3). This confirms several elegant structural studies that predict a parallel arrangement (with all amino termini at one end of the bundle) of the core complex25–27, which strongly sup-

SNARE domains that are not part of the core complex include the amino-terminal domain of syntaxin, the central palmitoylated region of SNAP-25, the amino-terminal proline-rich region of VAMP, and the transmembrane domains of syntaxin and VAMP (FIG. 3). The long amino-terminal domain of syntaxin forms a three-helix bundle36,37 that competes with the VAMP and SNAP-25 coils for binding of its own carboxy-terminal coil (forming the ‘closed’ conformation of syntaxin)38,39. So, as expected, it decreases the rate of ternary SNARE complex formation in solution40 and the rate of fusion of synthetic liposomes that carry SNARE proteins41. The amino-terminal domain of syntaxin, along with the coil domain, is also required for the interaction between syntaxin and nSec1 (REFS. 24, 39), which is also called Munc-18 (REF. 42). The chaperone protein n-Sec1 binds to the closed conformation of syntaxin and, after a conformational change, probably opens it up to facilitate SNARE complex formation43,44 (FIGS 3, 4). So, the amino terminus of syntaxin is probably important for regulation in vivo. The SNAP-25 loop region between the two coil domains is dispensable for SNAP-25 fusogenic activity16,41,45. However, it is probably crucial for ensuring rapid synaptic transmission by generating a high local concentration of the two required SNAP-25 coils by covalently linking them and attaching them to the plasma membrane. In the case of non-synaptic fusion events, in which speed might be less important, the two SNAP-25-like coils are sometimes contributed by separate proteins46. The function of the proline-rich amino-terminal ~24 amino acids of VAMP (FIG. 3) is not clear. This domain is only about 50% homologous between VAMP1 and VAMP2, but the proline-rich character is maintained47. Although this domain is crucial for inhibition of EXOCYTOSIS by synthetic VAMP peptides in Aplysia californica48, a VAMP coil without the prolinerich domain inhibits exocytosis efficiently in cracked PC12 cells16. Moreover, Fab FRAGMENTS of antibodies directed against the proline-rich region of VAMP2 did not significantly affect exocytosis in adrenal CHROMAFFIN 49 CELLS , indicating that this domain of VAMP is probably not directly involved in membrane fusion. But it is not yet known whether it could be important for interacting with regulators of VAMP function in vivo. www.nature.com/reviews/molcellbio

| FEBRUARY 2001 | VOLUME 2

REVIEWS

EXOCYTOSIS

The discharge by a cell of intracellular materials into the extracellular space through fusion of vesicles (containing these materials) with the plasma membrane. Fab FRAGMENT

Antigen-binding fragment of an immunoglobulin molecule. It is used when multimerization of antibodies caused by their Fc domains is not desirable. CHROMAFFIN CELLS

They arise from the same precursors as sympathetic neurons, and can synthesize, store and secrete catacholamines. They are found in all vertebrates, at various bodily locations but especially in the medulla of the adrenal gland. POLYISOPRENOID

Synthetic molecule consisting of varying numbers of branched five-carbon-atom moieties. BOTULINUM NEUROTOXIN E

Clostridial neurotoxin that cleaves SNAP-25 carboxyterminal coil.

Box 2 | Crystal structure of the core complex Crystallization of the SNARE core complex revealed a four-helix bundle structure52 (FIGS 2, 3). Like other coiled-coil structures, the residues residing at ‘a’ or ‘d’ positions on a heptad helical wheel contribute to the hydrophobic core interactions that are important in stabilizing the structure. These residues are the most conserved residues in the SNARE family14. Interestingly, the Gln and Arg residues mentioned above (Arg from VAMP and three glutamines from syntaxin and the SNAP-25 amino- and carboxy-terminal coils) form a central ionic interaction layer (the zero layer) in the otherwise hydrophobic core of the SNARE complex52 (FIG. 2b). Intrigued by the extreme conservation of these Gln and Arg residues through evolution, many researchers have looked into the exact role of these residues. In yeast, Gln to Arg substitutions were found to result in drastically reduced secretion98,99, whereas an Arg to Gln mutation did not cause any abnormality by itself but rendered the core complex more sensitive to additional mutations98. In PC12 cells, mutating the Gln residue in the SNAP-25 carboxy-terminal coil to alanine (Q174A) only caused a slight decrease in exocytosis45. In adrenal chromaffin cells, overexpression of Q174L SNAP-25 selectively affected the sustained phase of exocytosis and not the exocytic burst100, indicating that this ionic layer might be involved in facilitating SNARE complex disassembly or their initial contacts. It is likely that the 3Q:1R ratio contributes to the correct SNARE binding specificity, as a Q→R mutation can be rescued by a mirror R→Q mutation in the opposite helix in the SNARE complex98,99. This unusual layer might also enforce the correct register during SNARE pairing.

The transmembrane domains of syntaxin and VAMP have been shown to contribute to SNARE binding in vitro21,50,51. It was hypothesized that the transmembrane domains might form α-helices that continue into the cytoplasmic coiled-coil bundle52. However, the introduction of a helix-breaking proline residue between the cytoplasmic and transmembrane domains of VAMP did not significantly affect its ability to fuse liposomes53. Furthermore, both syntaxin and VAMP transmembrane domains can be substituted with bilayer-spanning POLYISOPRENOID synthetic anchors without affecting liposome fusion54. So, at least in the artificial liposome system, the interaction between the transmembrane domains of syntaxin and VAMP does not seem to be crucial for fusion. These precise domain organizations of the three neuronal SNAREs are not always preserved in other members of the SNARE family. However, as these non-coiled-coil domains of SNAREs are more conSNAP-25 N

Coil

served between species than between different SNARE family members, it is likely that they are important for specific regulation of the particular fusion reaction that they are involved in. Do SNAREs mediate membrane fusion?

The ‘zipper’ model of SNARE function postulates that the SNARE core complex ‘zips’ from the membrane-distal amino termini to the membrane-proximal carboxyl termini, and the formation of the stable SNARE complex overcomes the energy barrier to drive fusion of the lipid bilayers25,26. This hypothesis has gained substantial experimental support during the past few years. In Drosophila melanogaster, a temperature-sensitive mutation in syntaxin that abolishes formation of the SNARE complex rapidly blocks exocytosis of docked vesicles55. In the cracked PC12 cell system56, SNAP-25 could be manipulated by inactivating endogenous SNAP-25 with BOTULINUM NEUROTOXIN E, and then adding back a soluble

VAMP CCCC

Coil

Syntaxin Coil

Coil

n-Sec1 Syntaxin

Core complex

n-Sec1–syntaxin complex

Figure 3 | SNARE domain structures and the interaction between syntaxin and its chaperone protein n-Sec1. The amino-terminal domain of syntaxin forms a three-helix bundle (red) that binds to its carboxy-terminal coil domain (purple), forming the closed conformation (right), which is bound and stabilized by n-sec1 (middle)39. A conformational change then occurs to allow dissociation of n-sec1 and the opening up of syntaxin, facilitating core complex formation. The coil domains of syntaxin, SNAP-25 and VAMP form the four-helix bundle core complex (left)52. In addition to the coil domain, VAMP harbours a proline-rich amino-terminal domain (PP) and SNAP-25 harbours a central domain that contains four palmitoylated cysteine residues (CCCC). (SNAP-25, 25 kDa synaptosome-associated protein; TM, transmembrane domain; VAMP, vesicle-associated membrane protein.)

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 2 | FEBRUARY 2001 | 1 0 1

REVIEWS

VAMP

(Rab) Ca2+

Nucleation n-Sec1

n-Sec1 Zipping n-Sec1 SNAP-25 Syntaxin Membrane fusion α-SNAP

n-Sec1

NSF ADP+Pi

ATP

NSF α-SNAP

NSF

α-SNAP

Figure 4 | Molecular model of vesicle exocytosis. Syntaxin is bound to n-Sec1 before formation of the core complex. Rab proteins might facilitate the dissociation of n-Sec1 from syntaxin, allowing subsequent binding (nucleation) between the three neuronal SNAREs, syntaxin, SNAP-25 and VAMP (for simplicity, only one coil is drawn for SNAP-25). Ca2+ triggers the full zipping of the coiled-coil complex, which results in membrane fusion and release of vesicle contents. After the fusion event, recruitment of α-SNAP and NSF from the cytoplasm and subsequent hydrolysis of ATP by NSF causes dissociation of the SNARE complex. Syntaxin, VAMP and SNAP-25 are then free for recycling and another round of exocytosis. (NSF; N-ethyl-maleimide-sensitive fusion protein; SNAP-25, 25 kDa synaptosome-associated protein; SNARE, soluble NSF attachment protein receptor, VAMP, vesicle-associated membrane protein.) EXOCYTIC BURST

Defined by Neher and colleagues as the initial burst of release occurring within a few hundred milliseconds after the stimulus (in the chromaffin cell system), which is probably due to exocytosis of secretory granules that are in a releaseready state. It can be further resolved into two kinetically distinct components. FLUORESCENCE RESONANCE ENERGY TRANSFER

Process of energy transfer between two fluorophores. Can be used to determine the distance between two attachment positions within a macromolecule or between two molecules. YEAST VACUOLAR FUSION SYSTEM

In vitro fusion assay that measures the homotypic fusion of vacuoles isolated from the yeast Saccharomyces cerevisiae using a colorimetric alkaline phosphatase assay. SEA URCHIN EGG FUSION SYSTEM

In vitro fusion assay that measures the homotypic fusion of cortical vesicles isolated from sea urchin eggs upon addition of calcium, by measuring turbidity (A405).

102

SNAP-25 coil to rescue toxin-inhibited exocytosis after toxin washout45. In such a rescue assay, mutating the hydrophobic residues along the carboxy-terminal coil of SNAP-25 most severely affected the rescue, confirming that the formation of the core complex is crucial for SNARE function16,45. Furthermore, the irreversible assembly of the SNARE core complex occurred only after the arrival of Ca2+ and could not be experimentally uncoupled from the membrane fusion process45. Consistent with this result, electrophysiological kinetic analysis of exocytosis in adrenal chromaffin cells revealed that a SNARE assembly-inhibiting antibody and clostridial neurotoxins reduce the initial fast component of the EXOCYTIC BURST, indicating that even the most ready-to-be-released vesicles have not bypassed the SNARE assembly step49,57. These results obtained from physiologically relevant systems are consistent with those from the in vitro synthetic liposome fusion system, in which v- and t-SNAREs are separately reconstituted into synthetic liposomes and fusion of the two liposome populations is measured by either FLUORES58 CENCE RESONANCE ENERGY TRANSFER between lipids or content mixing59. This system showed that SNAREs are necessary and sufficient to mediate lipid and content mixing58,59. So, both in vitro and in vivo evidence indicate that SNAREs might be involved in a late, if not the final, step of membrane fusion. However, there have also been reports to the contrary. In the YEAST VACUOLAR FUSION SYSTEM, it was reported that SNARE complexes could be dissociated and prevented from reassembly with a blocking antibody, and that this disruption of SNARE

complexes did not affect content mixing of the vacuoles60. The conclusion that SNAREs do not mediate membrane fusion was also drawn from studies using the SEA URCHIN EGG FUSION SYSTEM, in which Ca2+ ions were found to dissociate SNARE complexes with no detriment to the fusion process61,62. If SNAREs are not involved in membrane fusion, which proteins actually mediate the bilayer fusion reaction? Two of the downstream factors that have been identified in the yeast vacuole system are calmodulin5 and protein phosphatase 1 (PP1)63. In mammalian systems, calmodulin has been reported to regulate the exocytic machinery64–67 but there has been little indication that mammalian PP1 is crucial for membrane transport. In fact, it seems unlikely that dephosphorylation could be the final Ca2+-triggered event in synaptic vesicle fusion, given that neuronal exocytosis occurs within a millisecond68. The most likely event that could occur within such a short time, other than lipid rearrangement, is probably a protein conformational change. The zipping of the SNARE core complex would be more compatible with the timescale of fusion that is required by neurons. Although it is possible that the molecular machinery mediating bilayer fusion differs fundamentally between organisms (such as yeast and mammals) and/or between different transport steps (such as presynaptic exocytosis and homotypic vacuole fusion), it is more likely that the principle is conserved. We favour the hypothesis that SNARE proteins do mediate yeast vacuole fusion, and that dephosphorylation by PP1 is simultaneously required to facilitate SNARE-catalysed lipid fusion in this system. www.nature.com/reviews/molcellbio

| FEBRUARY 2001 | VOLUME 2

REVIEWS

Figure 5 | Model of SNARE-mediated lipid fusion. a | The two membranes are in the vicinity of each other but the SNAREs are not yet in contact. b | SNARE complexes start zipping from the amino-terminal end, which draws the two membranes further towards each other. c | Zipping proceeds, causing increased curvature and lateral tension of the membranes, exposing the bilayer interior. Spontaneous hemifusion occurs as the separation is sufficiently reduced. d | The highly unfavourable void space at the membrane junction in (c) causes the establishment of contacts between the distal membrane leaflets. e | The lateral tension in the transbilayer contact area induces membrane breakdown, yielding a fusion pore. f | The fusion pore expands and the membrane relaxes. (SNARE, soluble NSF attachment protein receptor, where NSF stands for N-ethyl-maleimide-sensitive fusion protein.)

Do SNAREs assemble before final zipping? TETANUS TOXIN

Clostridial neurotoxin that cleaves VAMP. MEMBRANE CAPACITANCE MEASUREMENTS

Patch-clamp technique that allows indirect measurements of single exocytic events. The technique measures the increase in the capacitance (and therefore surface) of the plasma membrane that results from fusion of exocytic vesicles with the plasma membrane.

When do SNARE complexes form (BOX 3)? If zipping up the SNAREs into a four-helix bundle drives the membrane fusion reaction then the SNARE complex should only exist as a coiled-coil bundle (similar to the crystal structure of an in vitro complex) during or after fusion. However, increasing evidence indicates that before the fusion signal arrives, a reversible or partially assembled trans-SNARE complex (BOX 3) might exist. In crayfish neuromuscular junctions, it was found that the inhibition of synaptic transmission by TETANUS TOXIN, but not botulinum toxin B, requires prior nerve activity69. As the amino-terminal portion of VAMP is required for its proteolysis by tetanus toxin, but not by botulinum toxin

Box 3 | Formation of cis- and trans-SNARE complexes In recent literature, there has been some confusion over the expression ‘SNARE complex formation’. Traditionally, SNAREs have been studied in vitro or in detergent-containing solutions, conditions under which only fully assembled complexes exist. However, in vivo, when the v- and t-SNARE proteins are anchored in membrane lipids, they can be in cis (on the same membrane) or trans (on opposing membranes) conformations. A SNARE complex in the cis conformation probably resembles the fully assembled in vitro complex, whereas SNAREs in a trans complex might only loosely or partially interact because of the resistance posed by the membranes, and its structure and properties are probably very different to the in vitro complex. For example, a trans complex is unlikely to be SDS resistant, clostridial neurotoxin resistant57 or thermostable, and it might not require α-SNAP and NSF for dissociation101, unlike its cis counterpart. However, this distinction was not made experimentally until recently, so the term ‘SNARE complex formation’ might have been used ambiguously in many of the earlier functional studies. Some authors used it to describe pre-fusion partial assembly of trans SNAREs, whereas others used it to describe the irreversible zipping up of the complex leading to ciscomplex formation concurrent with membrane fusion. Some of the controversy in the field is therefore only semantic.

B70,71, this indicates that the amino-terminal portion of VAMP might be shielded in a protein complex before the arrival of the fusion trigger, implying the existence of half-zipped SNARE complexes. In the adrenal chromaffin cell system, kinetic analysis of exocytosis using high time resolution MEMBRANE CAPACITANCE MEASUREMENTS revealed that SNAREs exist in a dynamic equilibrium between a loose and a tight form49. The two forms can be distinguished by their different sensitivities to BOTU57 LINUM NEUROTOXIN A and antibodies that block SNARE complex formation49. In the artificial liposome system, it was shown that prior incubation of v- and t-SNAREcontaining liposomes resulted in liposome docking and an accelerated rate of fusion40,58. All of the above results indicate that inter-SNARE interactions might occur before the actual membrane fusion event. These interactions could be critical for establishing the readiness of the release machinery for fast fusion. What prevents this partial complex from zipping fully? As Ca2+ is known to be the final trigger for many trafficking steps4–7, one possibility is that the SNARE complex is prevented from fully assembling by a Ca2+sensitive clamp. Synaptotagmin I, a synaptic protein that binds Ca2+, SNAREs and phospholipids has been implicated as the Ca2+ sensor in neuronal exocytosis72. It has been proposed that synaptotagmin wraps around the membrane-proximal base of a preassembled transSNARE complex and triggers fast exocytosis through its Ca2+-dependent interactions with SNAREs and/or membranes73,74. It is uncertain, however, how this interaction would prevent full SNARE assembly. It has also been suggested that either the interaction between synaptotagmin and the carboxyl terminus of SNAP-25 (REFS 57,75) or the Ca2+-induced oligomerization of synaptotagmin76 could be important for excitation–secretion coupling. More experimental data are needed before we can build a coherent model of synaptotagmin and SNARE function in synapses. Mechanics of SNARE-mediated lipid fusion

It is energetically expensive to fuse two membranes in an aqueous environment, because the electrostatic repulsive and hydration forces between the two membranes has to be overcome77. In the current fusion model15,78,79 (FIG. 5), once the distance between the two bilayers is sufficiently reduced, HEMIFUSION occurs (FIG. 5c,d), followed by distal leaflet membrane breakdown, resulting in the opening of a ‘fusion pore’ (FIG. 5e). Finally, the fusion pore expands, causing full content mixing and membrane relaxation (FIG. 5f). The first aqueous connection between the lumen of the vesicle and the extracellular space (or the lumen of another membrane compartment) — the fusion pore — was initially shown to exist by both FREEZE–FRACTURE ELECTRON MICROSCOPY80 and PATCH81 CLAMP electrophysiological recordings . The pore was once hypothesized to be initially proteinaceous, consisting of GAP-JUNCTION-like proteins that span two bilayers82. However, evidence has since indicated that the pore is likely to be purely lipidic, with proteins acting as an external scaffold15,78,79, as was initially suggested by Fernandez and colleagues83. In such a model, directed

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 2 | FEBRUARY 2001 | 1 0 3

REVIEWS

BOTULINUM NEUROTOXIN A

Clostridial neurotoxin that cleaves the SNAP-25 carboxyterminal coil. HEMIFUSION

Transient membrane fusion intermediate in which only the two proximal leaflets of the bilayer mix. FREEZE–FRACTURE ELECTRON MICROSCOPY

A technique in which membrane samples are deep frozen and then fractured with the blade of a knife to reveal the internal structure of the membrane. PATCH CLAMP

Technique whereby a very small electrode tip is sealed onto a patch of cell membrane, thereby making it possible to record the flow of current through individual ion channels or pores within the patch. GAP JUNCTION

Communicating junction (permeant to molecules up to 1 kDa) between adjacent cells, which is composed of 12 connexin protein subunits, six of which form a connexon or hemichannel contributed by each of the coupled cells.

104

movement of membranes by proteins under conditions of high membrane tension and curvature is essential; it is this scaffolding that creates the force that allows hemifusion and lipid fusion pore opening and expansion to occur83. SNARE proteins are excellent candidates for doing this. Their unusual ability to form a trans complex might direct the two membranes towards each other and create marked curvature and tension in the membranes, thereby stabilizing the transition state (FIG. 5A–C). In the liposome fusion system, insertion of a linker between the transmembrane domain and the coilforming domains of the SNAREs decreased fusion efficiency with increasing linker length53, indicating that there might be a stringent functional requirement for a certain length of this connection. Furthermore, replacing the membrane anchors of syntaxin and VAMP with short phospholipids that do not span the bilayer prevented lipid mixing54. SNAREs might therefore exert a force through the linker to the membrane anchors by forming a core complex, generating inward and lateral movement in both membranes. As it is the zipping of a trans complex in a parallel fashion that generates this force, it is no surprise that the high in vitro stability of the core complex is functionally relevant. Mutation of the conserved hydrophobic residues that are important for core complex formation results in decreased thermostability of core complexes, which parallels a decreased ability of SNAREs to function in exocytosis in PC12 cells45. In addition to their transmembrane domains, SNAREs might be equipped to affect membranes in several ways. For example, there are basic amino acids at the membrane-proximal end of the core complex, which are well positioned to affect negatively charged membrane surfaces52. Furthermore, two tryptophan residues and a tyrosine residue at the carboxyterminal end of VAMP’s coil domain have been suggested to be involved in potential SNARE–lipid interactions during fusion52. Further structure–function analyses are needed to substantiate the functional importance of these residues. This mechanism of SNARE-mediated fusion has some similarity to viral-protein-mediated membrane fusion. In both cases, coiled-coil helical bundles are the main structural component of the fusion protein, and a pronounced conformational change of the protein promotes fusion15,84. The viral fusion is proposed to use a ‘jack-knife’ mechanism, in which bending of a helix bundle brings the two membrane anchors and the associated bilayers together85, whereas SNAREs are proposed to use the zipper mechanism described above to achieve the same effect. How many SNARE complexes are needed for one fusion event? Intuitively, a ring of complexes seems to be the most optimal to apply uniform force on the membrane lipids to create a fusion pore. However, so far there has been no experimental evidence to indicate that several SNARE complexes might be required to open one fusion pore. Although the linker region of SNAP-25 between the two coiled-coil domains causes multimerization of core complexes in vitro 21, its dele-

tion does not greatly affect membrane fusion in either synthetic liposomes or cracked PC12 cells16,40,45, but fast kinetic analysis of the fusion reaction was not possible in either system. It remains possible that a ring of SNARE complexes is generated by some other means, such as homomultimerization mediated by transmembrane domains51 or interactions with other regulators that can oligomerize, such as synaptotagmin76. Development of single molecule analyses might be necessary to advance our understanding in this respect. Although SNAREs alone are sufficient to fuse synthetic liposomes of certain lipid compositions, it has been suggested that, at least in some cases, SNARE complex formation might induce a hemifusion intermediate state, after which SNAREs become dispensable15. If this was the case then there would be a second catalyst to trigger the fusion pore opening and expansion, causing the distal lipid layer and content mixing. This second catalyst could be synaptotagmin15 or conceivably an asyet-unknown channel-like molecule that can form fusion pores. Or is it possible that Ca2+, the final trigger for many membrane transport steps, could actually be the second catalyst? Divalent cations, such as Ca2+, can cause significant changes in membrane tension77,86. The concentration of Ca2+ that is sensed at the fusion site has been estimated to range from 25 µM at presynaptic terminals87 to about one to two orders of magnitude lower in endocrine exocytosis57,64 and intracellular membrane fusion6. Whether this wide range of Ca2+ concentrations can trigger similar molecular events at distinct membrane fusion steps remains unknown. Perspectives

A progressively more coherent molecular model of SNARE-mediated membrane fusion is emerging from recent mechanistic studies. However, many remaining mysteries need to be solved before SNARE proteins can be used to create artificial fusion systems for drug delivery or to perturb secretion of hormones or other signalling molecules in vivo. In addition to further understanding how SNARE proteins act on membrane lipids to cause fusion, we need to comprehend some of the key regulatory mechanisms of cellular fusion, including Ca2+ regulation and Rab-GTPase-mediated regulation. Do Ca2+ ions act directly on membrane lipids, or activate Ca2+-sensing proteins such as synaptotagmin and/or calmodulin? How does activated synaptotagmin or calmodulin then trigger membrane fusion? Which factors are involved in the Rab-GTP-initiated signalling pathway that leads to formation of the SNARE complex? How is the binding of syntaxin to its chaperone, n-Sec1, regulated? Furthermore, many newly identified proteins need to be better integrated into the SNAREonly picture presented above. For example, dozens of proteins, including complexin88, tomosyn89, Hrs-2 (REF. 90), snapin91, syntaphilin92 and Munc-13 (REF. 93), were identified as potential factors in vesicle exocytosis; however, their precise functions in membrane transport are still largely unknown. Specific disruption of these proteins in functional assays and model organisms will perhaps be required to explain their function. Although www.nature.com/reviews/molcellbio

| FEBRUARY 2001 | VOLUME 2

REVIEWS functional assays facilitate mechanistic understanding, it is often difficult to disrupt a given molecule specifically. Knockout studies can be highly informative, but they leave us with the daunting task of extracting mechanistic information from a phenotype. Another difficulty facing the field is the fact that the molecules of interest often occupy transient states. However, it is to our advantage that membrane fusion is a conserved feature of all eukaryotic cells and so can be studied in diverse

2. 3.

10.

11.

12.

13.

14.

15. 16.

17. 18.

19.

20.

21.

Lledo, P. M., Zhang, X., Südhof, T. C., Malenka, R. C. & Nicoll, R. A. Postsynaptic membrane fusion and long-term potentiation. Science 279, 399–403 (1998). Pfeffer, S. R. Transport-vesicle targeting: tethers before SNAREs. Nature Cell Biol. 1, 17–22 (1999). Klenchin, V. A. & Martin, T. F. Priming in exocytosis: attaining fusion-competence after vesicle docking. Biochimie 82, 399–407 (2000). Heidelberger, R., Heinemann, C., Neher, E. & Matthews, G. Calcium dependence of the rate of exocytosis in a synaptic terminal. Nature 371, 513–515 (1994). Peters, C. & Mayer, A. Ca2+/calmodulin signals the completion of docking and triggers a late step of vacuole fusion. Nature 396, 575–580 (1998). Beckers, C. J. & Balch, W. E. Calcium and GTP: essential components in vesicular trafficking between the endoplasmic reticulum and Golgi apparatus. J. Cell Biol. 108, 1245–1256 (1989). Colombo, M. I., Beron, W. & Stahl, P. D. Calmodulin regulates endosome fusion. J. Biol. Chem. 272, 7707–7712 (1997). Bennett, M. K., Calakos, N. & Scheller, R. H. Syntaxin: a synaptic protein implicated in docking of synaptic vesicles at presynaptic active zones. Science 257, 255–259 (1992). Oyler, G. A. et al. The identification of a novel synaptosomal-associated protein, SNAP-25, differentially expressed by neuronal subpopulations. J. Cell Biol. 109, 3039–3052 (1989). Trimble, W. S., Cowan, D. M. & Scheller, R. H. VAMP-1: a synaptic vesicle-associated integral membrane protein. Proc. Natl Acad. Sci. USA 85, 4538–4542 (1988). Baumert, M., Maycox, P. R., Navone, F., De Camilli, P. & Jahn, R. Synaptobrevin: an integral membrane protein of 18,000 Daltons present in small synaptic vesicles of rat brain. EMBO J. 8, 379–384 (1989). Novick, P., Field, C. & Schekman, R. Identification of 23 complementation groups required for post-translational events in the yeast secretory pathway. Cell 21, 205–215 (1980). Söllner, T., Bennett, M. K., Whiteheart, S. W., Scheller, R. H. & Rothman, J. E. A protein assembly–disassembly pathway in vitro that may correspond to sequential steps of synaptic vesicle docking, activation, and fusion. Cell 75, 409–418 (1993). Fasshauer, D., Sutton, R. B., Brünger, A. T. & Jahn, R. Conserved structural features of the synaptic fusion complex: SNARE proteins reclassified as Q- and RSNAREs. Proc. Natl Acad. Sci. USA 95, 15781–15786 (1998). Jahn, R. & Südhof, T. C. Membrane fusion and exocytosis. Annu. Rev. Biochem. 68, 863–911 (1999). Scales, S. J. et al. SNAREs contribute to the specificity of membrane fusion. Neuron 26, 457–464 (2000). This is the first study in which many cognate and non-cognate SNARE coils were tested for their ability to function in a specific cellular fusion step. Götte, M. & von Mollard, G. F. A new beat for the SNARE drum. Trends Cell Biol. 8, 215–218 (1998). Parlati, F. et al. Topological restriction of SNAREdependent membrane fusion. Nature 407, 194–198 (2000). McNew, J. A. et al. Compartmental specificity of cellular membrane fusion encoded in SNARE proteins. Nature 407, 153–159 (2000). This is a thorough study of SNARE fusion specificity in the synthetic liposome fusion model. Fasshauer, D., Eliason, W. K., Brünger, A. T. & Jahn, R. Identification of a minimal core of the synaptic SNARE complex sufficient for reversible assembly and disassembly. Biochemistry 37, 10354–10362 (1998). Poirier, M. A. et al. Protease resistance of syntaxin–SNAP25–VAMP complexes. Implications for assembly and structure. J. Biol. Chem. 273, 11370–11377 (1998).

systems with different techniques. In the future, with all that is learned from different systems, we will be able to grasp not only what is conserved, but also what is unique to each transport step. Links DATABASE LINKS NSF | STX1 | SNAP-25 | VAMP1 | α-

22. Hayashi, T. et al. Synaptic vesicle membrane fusion complex: action of clostridial neurotoxins on assembly. EMBO J. 13, 5051–5061 (1994). 23. Yang, B. et al. SNARE interactions are not selective. Implications for membrane fusion specificity. J. Biol. Chem. 274, 5649–5653 (1999). 24. Pevsner, J. et al. Specificity and regulation of a synaptic vesicle docking complex. Neuron 13, 353–361 (1994). 25. Hanson, P. I., Roth, R., Morisaki, H., Jahn, R. & Heuser, J. E. Structure and conformational changes in NSF and its membrane receptor complexes visualized by quickfreeze/deep-etch electron microscopy. Cell 90, 523–535 (1997). 26. Lin, R. C. & Scheller, R. H. Structural organization of the synaptic exocytosis core complex. Neuron 19, 1087–1094 (1997). 27. Poirier, M. A. et al. The synaptic SNARE complex is a parallel four-stranded helical bundle. Nature Struct. Biol. 5, 765–769 (1998). 28. Malhotra, V., Orci, L., Glick, B. S., Block, M. R. & Rothman, J. E. Role of an N-ethylmaleimide-sensitive transport component in promoting fusion of transport vesicles with cisternae of the Golgi stack. Cell 54, 221–227 (1988). 29. Clary, D. O., Griff, I. C. & Rothman, J. E. SNAPs, a family of NSF attachment proteins involved in intracellular membrane fusion in animals and yeast. Cell 61, 709–721 (1990). 30. Lenzen, C. U., Steinmann, D., Whiteheart, S. W. & Weis, W. I. Crystal structure of the hexamerization domain of Nethylmaleimide-sensitive fusion protein. Cell 94, 525–536 (1998). 31. Yu, R. C., Hanson, P. I., Jahn, R. & Brünger, A. T. Structure of the ATP-dependent oligomerization domain of Nethylmaleimide sensitive factor complexed with ATP. Nature Struct. Biol. 5, 803–811 (1998). 32. Hohl, T. M. et al. Arrangement of subunits in 20S particles consisting of NSF, SNAPs, and SNARE complexes. Mol. Cell 2, 539–548 (1998). 33. May, A. P., Misura, K. M., Whiteheart, S. W. & Weis, W. I. Crystal structure of the amino-terminal domain of Nethylmaleimide-sensitive fusion protein. Nature Cell Biol. 1, 175–182 (1999). 34. Yu, R. C., Jahn, R. & Brünger, A. T. NSF N-terminal domain crystal structure: models of NSF function. Mol. Cell 4, 97–107 (1999). 35. Rice, L. M. & Brünger, A. T. Crystal structure of the vesicular transport protein Sec17: implications for SNAP function in SNARE complex disassembly. Mol. Cell 4, 85–95 (1999). 36. Fernandez, I. et al. Three-dimensional structure of an evolutionarily conserved N-terminal domain of syntaxin 1A. Cell 94, 841–849 (1998). 37. Lerman, J. C., Robblee, J., Fairman, R. & Hughson, F. M. Structural analysis of the neuronal SNARE protein syntaxin-1A. Biochemistry 39, 8470–8479 (2000). 38. Calakos, N., Bennett, M. K., Peterson, K. E. & Scheller, R. H. Protein–protein interactions contributing to the specificity of intracellular vesicular trafficking. Science 263, 1146–1149 (1994). 39. Misura, K. M. S., Scheller, R. H. & Weis, W. I. Threedimensional structure of the neuronal-Sec1–syntaxin 1a complex. Nature 404, 355–362 (2000). The structure revealed that a pronounced conformational change of syntaxin occurs before fusion. 40. Nicholson, K. L. et al. Regulation of SNARE complex assembly by an N-terminal domain of the t-SNARE Sso1p. Nature Struct. Biol. 5, 793–802 (1998). 41. Parlati, F. et al. Rapid and efficient fusion of phospholipid vesicles by the α-helical core of a SNARE complex in the absence of an N-terminal regulatory domain. Proc. Natl Acad. Sci. USA 96, 12565–12570 (1999). 42. Hata, Y., Slaughter, C. A. & Südhof, T. C. Synaptic vesicle

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

43.

44.

45.

46. 47.

48.

49.

50.

51.

52.

53.

54.

55.

56.

57.

fusion complex contains unc-18 homologue bound to syntaxin. Nature 366, 347–351 (1993). Yang, B., Steegmaier, M., Gonzalez, L. C. Jr & Scheller, R. H. nSec1 binds a closed conformation of syntaxin1A. J. Cell Biol. 148, 247–252 (2000). Dulubova, I. et al. A conformational switch in syntaxin during exocytosis: role of munc18. EMBO J. 18, 4372–4382 (1999). Chen, Y. A., Scales, S. J., Patel, S. M., Doung, Y. -C. & Scheller, R. H. SNARE complex formation is triggered by Ca2+ and drives membrane fusion. Cell 97, 165–174 (1999). This was the first systematic structure–function analysis of a SNARE protein. It provided the first piece of direct functional evidence in a physiological system that SNAREs drive membrane fusion. Fukuda, R. et al. Functional architecture of an intracellular membrane t-SNARE. Nature 407, 198–202 (2000). Elferink, L. A., Trimble, W. S. & Scheller, R. H. Two vesicleassociated membrane protein genes are differentially expressed in the rat central nervous system. J. Biol. Chem. 264, 11061–11064 (1989). Cornille, F., Deloye, F., Fournie-Zaluski, M. C., Roques, B. P. & Poulain, B. Inhibition of neurotransmitter release by synthetic proline-rich peptides shows that the N-terminal domain of vesicle-associated membrane protein/synaptobrevin is critical for neuro-exocytosis. J. Biol. Chem. 270, 16826–16832 (1995). Xu, T. et al. Inhibition of SNARE complex assembly differentially affects kinetic components of exocytosis. Cell 99, 713–722 (1999). The authors proposed a detailed kinetic model of SNARE-mediated fusion reaction based on quantitative electrophysiological analyses. Margittai, M., Otto, H. & Jahn, R. A stable interaction between syntaxin 1a and synaptobrevin 2 mediated by their transmembrane domains. FEBS Lett. 446, 40–44 (1999). Laage, R., Rohde, J., Brosig, B. & Langosch, D. A conserved membrane-spanning amino acid motif drives homomeric and supports heteromeric assembly of presynaptic SNARE proteins. J. Biol. Chem. 275, 17481–17487 (2000). Sutton, R. B., Fasshauer, D., Jahn, R. & Brünger, A. T. Crystal structure of a SNARE complex involved in synaptic exocytosis at 2.4 Å resolution. Nature 395, 347–353 (1998). Provided the first high-resolution crystal structure of a SNARE complex, which was instrumental in elucidating the function of SNARE proteins. McNew, J. A., Weber, T., Engelman, D. M., Söllner, T. H. & Rothman, J. E. The length of the flexible SNAREpin juxtamembrane region is a critical determinant of SNAREdependent fusion. Mol. Cell 4, 415–421 (1999). McNew, J. A. et al. Close is not enough: SNAREdependent membrane fusion requires an active mechanism that transduces force to membrane anchors. J. Cell Biol. 150, 105–117 (2000). Taking advantage of the liposome fusion system (first described in reference 58), the authors delved deep into the mechanisms by which the SNAREs fuse two lipid bilayers. Littleton, J. T. et al. Temperature-sensitive paralytic mutations demonstrate that synaptic exocytosis requires SNARE complex assembly and disassembly. Neuron 21, 401–413 (1998). Hay, J. C. & Martin, T. F. Resolution of regulated secretion into sequential MgATP-dependent and calcium-dependent stages mediated by distinct cytosolic proteins. J. Cell Biol. 119, 139–151 (1992). Xu, T., Binz, T., Niemann, H. & Neher, E. Multiple kinetic components of exocytosis distinguished by neurotoxin sensitivity. Nature Neurosci. 1, 192–200 (1998).

VOLUME 2 | FEBRUARY 2001 | 1 0 5

REVIEWS 58. Weber, T. et al. SNAREpins: minimal machinery for membrane fusion. Cell 92, 759–772 (1998). 59. Nickel, W. et al. Content mixing and membrane integrity during membrane fusion driven by pairing of isolated vSNAREs and t-SNAREs. Proc. Natl Acad. Sci. USA 96, 12571–12576 (1999). 60. Ungermann, C., Sato, K. & Wickner, W. Defining the functions of trans-SNARE pairs. Nature 396, 543–548 (1998). The controversy concerning SNARE function was heavily fuelled by this report, which concluded that SNAREs are not directly involved in the membrane fusion process. 61. Tahara, M. et al. Calcium can disrupt the SNARE protein complex on sea urchin egg secretory vesicles without irreversibly blocking fusion. J. Biol. Chem. 273, 33667–33673 (1998). 62. Coorssen, J. R., Blank, P. S., Tahara, M. & Zimmerberg, J. Biochemical and functional studies of cortical vesicle fusion: the SNARE complex and Ca2+ sensitivity. J. Cell Biol. 143, 1845–1857 (1998). 63. Peters, C. et al. Control of the terminal step of intracellular membrane fusion by protein phosphatase 1. Science 285, 1084–1087 (1999). 64. Chen, Y. A., Duvvuri, V., Schulman, H. & Scheller, R. H. Calmodulin and protein kinase C increase Ca2+-stimulated secretion by modulating membrane-attached exocytic machinery. J. Biol. Chem. 274, 26469–26476 (1999). 65. Chamberlain, L. H., Roth, D., Morgan, A. & Burgoyne, R. D. Distinct effects of α-SNAP, 14-3-3 proteins, and calmodulin on priming and triggering of regulated exocytosis. J. Cell Biol. 130, 1063–1070 (1995). 66. Quetglas, S., Leveque, C., Miquelis, R., Sato, K. & Seagar, M. Ca2+-dependent regulation of synaptic SNARE complex assembly via a calmodulin- and phospholipid-binding domain of synaptobrevin. Proc. Natl Acad. Sci. USA 97, 9695–9700 (2000). 67. Coppola, T. et al. Disruption of Rab3–calmodulin interaction, but not other effector interactions, prevents Rab3 inhibition of exocytosis. EMBO J. 18, 5885–5891 (1999). 68. Mennerick, S. & Matthews, G. Ultrafast exocytosis elicited by calcium current in synaptic terminals of retinal bipolar neurons. Neuron 17, 1241–1249 (1996). 69. Hua, S. -Y. & Charlton, M. P. Activity-dependent changes in partial VAMP complexes during neurotransmitter release. Nature Neurosci. 2, 1078–1083 (1999). This paper presents compelling evidence that partially assembled SNARE complexes exist. 70. Pellizzari, R. et al. Structural determinants of the specificity for synaptic vesicle-associated membrane protein/synaptobrevin of tetanus and botulinum type B and G neurotoxins. J. Biol. Chem. 271, 20353–20358

106

(1996). 71. Foran, P., Shone, C. C. & Dolly, J. O. Differences in the protease activities of tetanus and botulinum B toxins revealed by the cleavage of vesicle-associated membrane protein and various sized fragments. Biochemistry 33, 15365–15374 (1994). 72. Geppert, M. & Südhof, T. C. RAB3 and synaptotagmin: the yin and yang of synaptic membrane fusion. Annu. Rev. Neurosci. 21, 75–95 (1998). 73. Davis, A. F. et al. Kinetics of synaptotagmin responses to Ca2+ and assembly with the core SNARE complex onto membranes. Neuron 24, 363–376 (1999) 74. Sutton, R. B., Ernst, J. A. & Brünger, A. T. Crystal structure of the cytosolic C2A-C2B domains of synaptotagmin III. Implications for Ca2+-independent SNARE complex interaction. J. Cell Biol. 147, 589–598 (1999). 75. Gerona, R. R., Larsen, E. C., Kowalchyk, J. A. & Martin, T. F. The C terminus of SNAP25 is essential for Ca2+dependent binding of synaptotagmin to SNARE complexes. J. Biol. Chem. 275, 6328–6336 (2000). 76. Desai, R. C. et al. The C2b domain of synaptotagmin is a Ca2+-sensing module essential for exocytosis. J. Cell Biol. 150, 1125–1136 (2000). 77. Zimmerberg, J., Vogel, S. S. & Chernomordik, L. V. Mechanisms of membrane fusion. Annu. Rev. Biophys. Biomol. Struct. 22, 433–466 (1993). 78. Monck, J. R. & Fernandez, J. M. The fusion pore and mechanisms of biological membrane fusion. Curr. Opin. Cell Biol. 8, 524–533 (1996). 79. Lee, J. & Lentz, B. R. Evolution of lipidic structures during model membrane fusion and the relation of this process to cell membrane fusion. Biochemistry 36, 6251–6259 (1997). 80. Chandler, D. E. & Heuser, J. E. Arrest of membrane fusion events in mast cells by quick-freezing. J. Cell Biol. 86, 666–674 (1980). 81. Breckenridge, L. J. & Almers, W. Currents through the fusion pore that forms during exocytosis of a secretory vesicle. Nature 328, 814–817 (1987). 82. Almers, W. & Tse, F. W. Transmitter release from synapses: does a preassembled fusion pore initiate exocytosis? Neuron 4, 813–818 (1990). 83. Monck, J. R. & Fernandez, J. M. The exocytotic fusion pore and neurotransmitter release. Neuron 12, 707–716 (1994). 84. Skehel, J. J. & Wiley, D. C. Coiled coils in both intracellular vesicle and viral membrane fusion. Cell 95, 871–874 (1998). 85. Hughson, F. M. Enveloped viruses: a common mode of membrane fusion? Curr. Biol. 7, 565–569 (1997). 86. Ohki, S. Effects of divalent cations, temperature, osmotic pressure gradient, and vesicle curvature on phosphatidylserine vesicle fusion. J. Membr. Biol. 77,

265–275 (1984). 87. Schneggenburger, R. & Neher, E. Intracellular calcium dependence of transmitter release rates at a fast central synapse. Nature 406, 889–892 (2000). 88. McMahon, H. T., Missler, M., Li, C. & Südhof, T. C. Complexins: cytosolic proteins that regulate SNAP receptor function. Cell 83, 111–119 (1995). 89. Fujita, Y. et al. Tomosyn: a syntaxin-1-binding protein that forms a novel complex in the neurotransmitter release process. Neuron 20, 905–915 (1998). 90. Bean, A. J., Seifert, R., Chen, Y. A., Sacks, R. & Scheller, R. H. Hrs-2 is an ATPase implicated in calcium-regulated secretion. Nature 385, 826–829 (1997). 91. Ilardi, J. M., Mochida, S. & Sheng, Z. H. Snapin: a SNAREassociated protein implicated in synaptic transmission. Nature Neurosci. 2, 119–124 (1999). 92. Lao, G. et al. Syntaphilin: a syntaxin-1 clamp that controls SNARE assembly. Neuron 25, 191–201 (2000). 93. Betz, A. et al. Munc13-1 is a presynaptic phorbol ester receptor that enhances neurotransmitter release. Neuron 21, 123–136 (1998). 94. Mayer, A., Wickner, W. & Haas, A. Sec18p (NSF)-driven release of Sec17p (α-SNAP) can precede docking and fusion of yeast vacuoles. Cell 85, 83–94 (1996). This is the key paper that established the current view that α-SNAP and NSF function after fusion. 95. Hunt, J. M. et al. A post-docking role for synaptobrevin in synaptic vesicle fusion. Neuron 12, 1269–1279 (1994). 96. Schulze, K. L., Broadie, K., Perin, M. S. & Bellen, H. J. Genetic and electrophysiological studies of Drosophila syntaxin-1A demonstrate its role in nonneuronal secretion and neurotransmission. Cell 80, 311–320 (1995). 97. Broadie, K. et al. Syntaxin and synaptobrevin function downstream of vesicle docking in Drosophila. Neuron 15, 663–673 (1995). 98. Ossig, R. et al. Exocytosis requires asymmetry in the central layer of the SNARE complex. EMBO J. 19, 6000–6010 (2000). 99. Katz, L. & Brennwald, P. Testing the 3Q:1R ‘rule’: mutational analysis of the ionic ‘zero’ layer in the yeast exocytic SNARE complex reveals no requirement for arginine. Mol. Biol. Cell 11, 3849–3858 (2000). 100. Wei, S. et al. Exocytotic mechanism studied by truncated and zero layer mutants of the C-terminus of SNAP-25. EMBO J. 19, 1279–1289 (2000). 101. Weber, T. et al. SNAREpins are functionally resistant to disruption by NSF and αSNAP. J. Cell Biol. 149, 1063–1072 (2000).

Acknowledgments We thank S. Scales for critically reading the manuscript and L. Gonzalez, S. Scales, B. Yang and R. Lin for the artwork in FIGS 1, 2 and 4.

www.nature.com/reviews/molcellbio

| FEBRUARY 2001 | VOLUME 2

REVIEWS

RAB PROTEINS AS MEMBRANE ORGANIZERS Marino Zerial and Heidi McBride* Cellular organelles in the exocytic and endocytic pathways have a distinctive spatial distribution and communicate through an elaborate system of vesiculo-tubular transport. Rab proteins and their effectors coordinate consecutive stages of transport, such as vesicle formation, vesicle and organelle motility, and tethering of vesicles to their target compartment. These molecules are highly compartmentalized in organelle membranes, making them excellent candidates for determining transport specificity and organelle identity. COGNATE SNARES

SNAREs on opposite membranes that are destined to form trans-SNARE complexes to mediate fusion. NSF

Molecular chaperone involved in recycling SNAREs after one round of fusion. EFFECTOR

A protein or protein complex that binds the GTPase directly and in a GTP-dependent manner and is required for the downstream function determined by that GTPase.

Max-Planck-Institute of Molecular Cell Biology and Genetics, c/o EMBL, Meyerhofstrasse 1, 69117 Heidelberg, Germany. *Current address: The University of Ottawa Heart Institute, 40 Ruskin Street, Ottawa, Ontario, Canada, K1Y 4W7. Correspondence to M.Z. e-mail: Zerial@EMBLHeidelberg.de

How do transport vesicles or tubular structures generated in one compartment encounter their target organelle and engage in membrane fusion? A typical transport reaction can be viewed as a four-step process that consists of formation of a vesicle or tubular intermediate, movement of the vesicle towards its target compartment, tethering/docking with the acceptor membrane and, ultimately, fusion of the lipid bilayers. The specificity of membrane tethering and fusion is critical to preserve organelle identity and the proper flow of cargo within the cell. The first specific event is the Rab-mediated tethering of an incoming vesicle to the correct target organelle. Following this, the specific topological pairing of COGNATE SNARES (soluble NSF attachment protein receptor, where NSF stands for N-ethylmaleimide-sensitive fusion protein) between the two bilayers (SNARE pins) ensures precision in the fusion event1â&#x20AC;&#x201C;4 (see the review by Chen and Scheller on page 98 of this issue). SNAREs are enriched in certain organelles, which helps to identify the correct target and to limit nonspecific fusion events. However, during vesicular transport, SNAREs inevitably spread throughout many cellular compartments. Any given organelle will contain SNARE complexes that must remain inactive until they return to their specific place of function. An additional layer of regulation is therefore inevitable to ensure that trans-SNARE complexes fuse membranes only at the appropriate time and in the correct place. In this review, we discuss how Rab GTPases and their EFFECTORS fit the criteria for a regulatory system that provides the complementary specificity to SNARE

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

complexes during membrane tethering and fusion. The discovery of molecular interactions between Rab effectors and components of the SNARE machinery provides a new understanding of how Rab proteins directly regulate SNARE function. We begin by summarizing the role of Rab proteins and the identification of some of their effectors. We then describe recent evidence indicating that Rab effectors are not randomly distributed on the organelle membrane but are clustered in distinct functional domains. The model emerging from these observations is that, rather than being mere regulators of SNARE protein complexes, Rab GTPases and their effectors are primary determinants of compartmental specificity in the organelles of eukaryotic cells. Heterogeneity of Rab effectors

Rab proteins constitute the largest family of monomeric small GTPases. Eleven Rab (Yptp/Sec4p) proteins are expressed in the yeast Saccharomyces cerevisiae but there might be as many as 63 family members in humans (J. Schultz and P. Bork, personal communication) as estimated from expressed-sequence tags (ESTs) and the sequenced human genome. This increased complexity throughout evolution reflects a greater need for cell organization and intracellular transport in the different cell types of multicellular organisms. Numerous studies have established that Rab proteins are distributed to distinct intracellular compartments and regulate transport between organelles (reviewed in REF. 5, see FIG. 1 and TABLES 1,2) (see extra online material). The regulatory principle of Rab proteins, as for other GTPases, lies in

ÂŠ 2001 Macmillan Magazines Ltd

VOLUME 2 | FEBRUARY 2001 | 1 0 7

REVIEWS

Plasma membrane

Rab21 (apical EC)

CCV Rab5a,b,c

TGN

Rab10 Rab12 Rab13 Rab30 Rab33b Rab36?

Rab6a,b

CCP

Rab3a,b,c,d (SV) Rab8 Rab26 (SG) Rab37

Rab13 (tight junction EC)

Rab11 Early endosome

Golgi

Rab4 Rab15 Rab18 Rab20 Rab22

Rab11 Rab25 Rab17 (EC)

Rab1 Rab2 IC MTOC

Recycling endosome Rab7 Late endosome

Nucleus

Rab9 Rab24

Rab27 (M,T) Lysosome

Endoplasmic reticulum

Figure 1 | Map of intracellular localization of Rab proteins. Summarizes the intracellular localization of Rab proteins in mammalian cells. Some proteins are cell- (for example, Rab3a in neurons) or tissue-specific (for example, Rab17 in epithelia) or show cell-type-specific localization (for example, Rab13 in tight junctions). (CCV, clathrin-coated vesicle; CCP, clathrin-coated pit; EC, epithelial cells; IC, ER–Golgi intermediate compartment; M, melanosomes; MTOC, microtubule-organizing centre; SG, secretory granules; SV, synaptic vesicles; T, T-cell granules; TGN, trans-Golgi nextwork.)

EEA1

The antigen involved in a human autoimmune disease. COPII VESICLES

Coated vesicles involved in transport from the endoplasmic reticulum to the Golgi.

108

their ability to function as molecular switches that oscillate between GTP- and GDP-bound conformations. The GTP-bound form is considered the ‘active’ form. However, with respect to the physiology of the regulated process, the most important feature is the ability of GTPases to cycle regularly between GTP- and GDPbound states. This cycle imposes temporal and spatial regulation to membrane transport, with the Rab proteins acting like timers whose clocks are set depending on the (intrinsic and catalysed) rates of nucleotide exchange and hydrolysis. Their on/off regulatory function is restricted to the membrane compartments where they are localized. Each transport step requires that the activated Rab proteins bind to soluble factors that act as ‘effectors’ to transduce the signal of the Rab GTPase in the transport mechanism. Many established or putative effector proteins and regulators have been identified and characterized (TABLES 1,2). Given the structural conservation of Rab GTPases and SNAREs, one might expect that Rab effectors could also be grouped in a family of structurally conserved proteins. This does not seem to be the case.

| FEBRUARY 2001 | VOLUME 2

The structural heterogeneity shown by Rab effectors implies that these are highly specialized molecules whose activities are exclusively tailored for individual organelles and transport systems. However, some Rab effectors do share structural features. For example, p115/Uso1p, Rabaptin-5 and early endsosome antigen 1 (EEA1), all contain predicted coiled-coil regions, and Rab3-interacting molecule (Rim1), EEA1 and Rabenosyn-5 contain zinc-fingers. A more comprehensive analysis of Rab effectors, taking into account their structural and functional properties, will be necessary to categorize these molecules. Regulation of intracellular transport

It is well established that Rab proteins function in the tethering/docking of vesicles to their target compartment, leading to membrane fusion. However, Rab proteins have also been implicated in vesicle budding and, more recently, in the interaction of vesicles with cytoskeletal elements. The finding that Rab proteins have several functions suggests that all steps of vesicle transport could be coordinated by the same regulatory machinery. Membrane tethering. Over the past few years, several studies have shown that membrane tethering is a conserved mechanism that depends on Rab effectors, rather than on SNARE complexes. Depending on the system, the recruitment of Rab and tethering effector proteins can be either symmetrical or asymmetrical between donor and acceptor membranes. In the yeast secretory pathway, tethering of endoplasmic reticulum (ER)derived vesicles to the Golgi complex depends on the membrane recruitment of Uso1p by Ypt1p but not on SNAREs6. Whereas no direct interaction between Uso1p and the Rab protein could be detected in the latter study, the mammalian homologue of Uso1p, p115 was recently shown to bind directly to Rab1 (REF. 7). However, there are mechanistic differences between mammalian and yeast ER-to-Golgi transport. Whereas Rab1 recruits p115 onto COPII (coat protein 2) VESICLES already at the budding step, in yeast the requirement for Ypt1p might be only at the level of the target membrane8. Beside these proteins, a multi-protein complex called TRAPP (transport protein particle) also targets ER-derived vesicles to the Golgi apparatus9. The TRAPP complex accelerates nucleotide exchange on Ypt1p, probably in the Golgi10 where this GTPase is essential8. The precise function of the complex and its tethering role in relation with Ypt1p activation awaits further analysis. Delivery of post-Golgi vesicles to the plasma membrane in yeast depends on Sec4p and the tethering factor that interacts with this Rab protein is also a multi-protein complex — the exocyst. The exocyst was identified by Novick and co-workers as a complex of seven proteins that are specifically required for exocytosis11. An equivalent complex exists in mammals12. One of its subunits, Sec3p, marks the sites of exocytosis on the plasma membrane in yeast13. The exocyst mediates vesicle targeting and, through the Sec15p subunit, interacts specifically and directly with Sec4p in a GTP-dependent manner14. These data raise several interesting questions. How

www.nature.com/reviews/molcellbio

REVIEWS

Table 1 | Rab proteins and their effectors Rab

Rab function

Direct effector

Effector function

Rab specificity

Effector partners

Partner features

Rab1

• ER–Golgi transport

p115

• Tethering • Sequestering SNAREs into budded vesicles

Rab1–GTP

Giantin GM130

• Tethering of COPIcoated vesicles to Golgi

PRA1

• Rab receptor (proposed)

Rab1 Rab3 Rab4b Rab5a Rab5c

VAMP2

• v-SNARE involved in bilayer fusion

• Rab3a: synaptic vesicle Rabphilin-3 and chromaffin granule secretion • Rab3b, c, d: regulated secretion

• Potentiates fusion

Rab3–GTP

α-actinin • Crosslinks actin filaments Rabaptin-5 into bundles • Stimulated by Rabphilin-3 interactions • Also binds Rabaptin-5, an effector of Rab5 and Rab4

RIM1 RIM2 Calmodulin

• Membrane fusion

Rab3–GTP

RIM–BP1

• Confers calcium sensitivity to protein interactions

Rab3

Many

Rab3

• Contains fibronectin type III repeats and SH3 domains • Multiple functions

Rab4

• Localized to early/recycling Rabaptin-5, • Activates Rab5 through endosomes Rabaptin-5β complex with Rabex-5 • Role in sorting/recycling in early Rabaptin-4 • Implicated in protein endosomes sorting and recycling

Rab4–GTP Rab5–GTP

Rabex-5

Rab5

• Ligand sequestration at plasma Rabaptin-5 • Stabilizes Rabex-5 membrane Rabaptin-5β recruitment • CCV–EE and EE–EE fusion EEA1 • Tethering, core fusion • Endosome motility component p150 • Class III PI(3)K regulatory subunit p110β • Class I PI(3)K catalytic subunit Rabenosyn-5 • Required for CCV–EE and EE–EE fusion

Rab5–GTP Rab4–GTP Rab5–GTP

Rab5–GTP Rab4–GTP

Rabex-5 • Nucleotide exchange factor Rabphilin-3 Syntaxin13 • t-SNAREs essential for Syntaxin6 bilayer fusion hVps34 • Class III PI(3)K catalytic subunit p85-α • Class I PI(3)K regulatory subunit hVps45 • Regulates SNARE complex formation or disassembly

Rab6

• Retrograde Golgi–ER and intra-Golgi transport

Rabkinesin-6 • Vesicle motility • Cytokinesis

Rab6–GTP

Microtubules

Rab8

• TGN–plasma membrane traffic (basolateral in epithelial cells)

Rab8IP

• Stress-activated protein kinase

Rab8–GTP

Rab9

• Late endosome to Golgi

p40

• Stimulates fusion

Rab9–GTP

Rab11

• Recycling through perinuclear recycling endosomes • Plasma membrane–Golgi traffic

Rab11BP

• Unclear

Rab11–GTP

Rab13

• Involved in the formation of the tight junction

δ-PDE

• Extracts Rab13 from membrane

Rab13

Rab5–GTP Rab5–GTP

mSec13

• Nucleotide exchange factor

• Coat component of COPII vesicles

Rab33b • Intra-Golgi transport

Rab33b-BP • Probably regulates motility of Rab33b–GTP Rab33 vesicles

Ypt1p

Uso1p

• Tethering of ER-derived vesicles

Recruitment regulated by Ypt1p

Yp1p–Yif1p complex

• Ypt GTPase binding to the Yip1p–Yif1p complex essential for vesicle docking and fusion

Ypt1p, Ypt31p Sec4p

Sec15p

• Tethering through interaction Sec4p–GTP of vesicular Sec15p and Sec10p with target complex in the bud

Exocyst

• Marks the site for docking or fusion

Ypt51p/ • Golgi–endosome and plasma Vps21p membrane–endosome transport

Vac1p

• TGN–Golgi transport

Ypt51p–GTP

Vps45p Pep12p

• Regulates SNARE complex formation or disassembly • Pep12p is a t-SNARE

Ypt7p

Vam2p– Vam6p

• Tethering and nucleotide exchange activity

Ypt7p–GTP

HOPS complex Vam3p SNARE complex δ-adaptin

• Links SNAREs with Ypt activation • Marks the site for tethering/fusion • AP3-vesicle formation at the Golgi

Sec4p

• ER–Golgi

• Delivery of TGN-derived vesicles into the bud

• Vacuole fusion

• Budding of AP3 vesicles?

• Integral membrane protein • Possible receptor and GDI-releasing factor

(CCV, clathrin-coated vesicle; EE, early endosome; ER, endoplasmic reticulum; PDE, phosphodiesterase; PI(3)K, phosphoinositol-3-OH kinase; SH3, Src homology region 3 domain;TGN, trans-Golgi network; Vamp, vesicle-associated membrane protein.)

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 2 | FEBRUARY 2001 | 1 0 9

REVIEWS

Table 2 | Regulatory proteins Rab/Ypt

GAP

GEF

Rab1

Mss4

Rab3

Rab3-GAP

Mss4, possibly Rabin 3

Rab5

Tuberous sclerosis 2 (?) Rabex5 RN-Tre

Rab6

GAPcenA

Rab8

Mss4

Rab10

Mss4

Ypt1p

TRAPP* Dss4p

Ypt51p/ Vps21p

Gyp1p, Gyp3p

Vps9p

Ypt6p

Gyp2p Gyp3p, Gyp4p, Gyp6p

Ric1p–Rigp1p

Ypt7p

Gyp4p, Gyp7p

HOPS‡

Sec4p

Gyp1p, Gyp2p/Mdr1p, Gyp3p, Gyp4p/Msb4p

Sec2p Dss4p

*TRAPP complex: Trs20p, Trs23p, Trs31p, Trs33p, Trs65p, Trs85p, Trs120p, Trs130p, Bet3p, Bet5p. ‡HOPS complex: Vps11p, Vps16p, Vps18p, Vps33p, Vps39p, Vps41p.

CCP

Area of the plasma membrane where receptors and the clathrin machinery are concentrated, preparing to form a vesicle.

110

and where is assembly of this complex regulated? What is the role of each subunit and that of Sec4p in this process? What determines localization of Sec3p? In the early ENDOCYTIC PATHWAY, Rab5 regulates clathrin-coated-vesicle-mediated transport from the plasma membrane to the early endosomes as well as homotypic early endosome fusion15,16. EEA1 is the Rab5 effector that mediates tethering/docking of early endosomes17. EEA1 is a largely coiled-coil protein that contains two zinc-fingers and two Rab5-binding domains at the amino and carboxyl termini18. Considering that Rab5 is also required on both donor and acceptor membranes for fusion to occur19,20, EEA1 could bridge two membranes that bear Rab5 (see below). A symmetrical requirement for a Rab protein in vesicle docking and fusion has also been determined for Ypt7p in yeast vacuole fusion21,22. Ypt7p binds to and regulates the membrane localization of a multi-protein complex (HOPS, which stands for homotypic fusion and vacuole protein sorting; also referred to as Class C Vps protein complex23) which includes the Vam2p/Vps41p and Vam6p/Vps39p proteins24,25. As these proteins have been implicated in vacuolar membrane docking26, HOPS seems to be the effector that mediates Ypt7p-dependent tethering. Consistent with the symmetrical requirement for Ypt7p, the HOPS complex is also needed on each vacuole partner undergoing homotypic fusion26. Furthermore, HOPS stimulates nucleotide exchange on Ypt7p27. So, as shown earlier for the Rabaptin-5–Rabex-5 complex of Rab5 (REF. 28), HOPS couples nucleotide exchange on a Rab protein to effector recruitment and function. Vesicle budding. More elusive is the role of Rab proteins in budding, for which there is contradictory evidence

| FEBRUARY 2001 | VOLUME 2

depending on the experimental system. For example, in vivo studies have indicated a possible role for Rab1 in budding of vesicles from the ER29 and for Rab9 from endosomes directed to the trans-Golgi network (TGN)30. More recently, however, Rab1 has been shown not to be essential for COPII vesicle formation in vitro, although it commits the vesicles to targeting and fusion7. In yeast, Ypt1p (Rab1) is also not needed for COPII vesicle formation but is exclusively required on target Golgi membranes8,31. A function in vesicle formation has also been attributed to Rab5. Rab5, which modulates the halflife of clathrin-coated pits (CCV) on the plasma membrane in vivo16, is required for vesicle formation in vitro32. Consistent with this finding, overexpression of RN-Tre, a newly identified Rab5 GAP (GTPase-activating protein), downregulates Rab5 and inhibits receptor internalization33. One component of the Ypt7p (Rab7)-tethering complex HOPS, Vam2p/Vps41p24, has also been implicated in the budding of vesicles from the Golgi34, although Ypt7p itself has not yet been directly involved in Golgi budding events26. Thus, depending on the specific transport event, Rab proteins might directly or indirectly influence vesicle budding. They could regulate the concentration and/or assembly of coat components or help to incorporate cargo molecules selectively into the nascent vesicles. Alternatively, their presence in an active state might function as a checkpoint that ensures the delivery of a vesicle to its appropriate target compartment. Vesicle motility. The task of Rab proteins is not restricted to membrane budding and fusion. More recently, these GTPases have been shown to determine the distribution of cellular compartments by regulating the movement of vesicles and organelles along cytoskeletal filaments. A role for Rab6 in microtubule-dependent transport has been inferred from the discovery that this GTPase interacts with a kinesin-like protein, Rabkinesin-6 (REF. 35), which is important for cytokinesis36. Rab5 regulates both the attachment of early endosomes to, and the motility along, microtubules37. There are also functional connections between Rab proteins and motors of the actin cytoskeleton. Genetic interactions have been uncovered in yeast between Sec4p and the myosin heavy chain Myo2p, indicating a possible mechanism whereby vesicles are propelled by motor proteins along polarized actin cables towards the site of exocytosis38,39. Another potential link between Rab and myosin proteins can be deduced from studies of a human disease, Griscelli syndrome. This is a rare, autosomal recessive disorder that is characterized by defective pigmentation of the skin and hair due to an aberrant accumulation of melanosomes in melanocytes. Mutations in two human genes have been associated with the disease, one in the MYO5A gene and the second in the RAB27A gene40. Interestingly, the same proteins are lost in the mouse mutants dilute (myosin-VA) and ashen (Rab27a), which are defective in pigment granule transport 41,42. As Rab27a and myosin-VA

www.nature.com/reviews/molcellbio

REVIEWS function in the same pathway of melanosome transport in melanocytes, it will be interesting to see whether Rab27a and myosin-VA can directly interact. Multiplicity of Rab5 effectors

The identification of interacting molecules has revealed an extraordinary complexity of the machinery downstream of Rab5. Using an affinity-chromatography procedure, it was possible to identify >20 polypeptides from bovine brain cytosol that interact directly or indirectly with the GTP-bound form of Rab5 (REF. 17). An important principle emerging from this analysis is that Rab5 effectors function in a cooperative fashion (BOX 1). The first Rab5 effector identified and found to be essential for early endosome fusion was Rabaptin-5 (REF. 43). Rabaptin-5 forms a complex with another protein, Rabex-5, which catalyses nucleotide exchange on Rab5

(REF. 28). Upon activation of Rab5 by Rabex-5, the Rabaptin-5–Rabex-5 complex induces its own membrane recruitment through Rabaptin-5 (Lippe et al. manuscript in preparation). This positive-feedback loop counteracts GTP hydrolysis44 and is thought to create a microenvironment that is enriched in active Rab5 on the membrane where other Rab5 effectors are recruited28. How does this local clustering of activated Rab5 proteins regulate the tethering machinery? The carboxy-terminal end of the tethering factor EEA1 contains two structural elements that are essential for targeting to the early endosome membrane45. One is the FYVE finger, a zinc-finger that specifically binds to phosphatidylinositol-3-phosphate, PtdIns(3)P46,47, and a Rab5-binding site located immediately upstream of the FYVE finger18. The discovery that phosphatidylinositol-3-OH kinases (PI(3)Ks) are Rab5 effectors48 is

Box 1 | Rab5 effectors cluster in a Rab5 domain

Rabenosyn-5

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

EEA1

The complex network of Rab5 regulators and effectors involves positive Rabaptin-5 T feedback loops and, Rabex-5 according to the model T presented in the figure, is Rabaptin-5 PI(3)K T T Rabex-5 designed to generate a local D T T T T T amplification of active VPS45 T Rab5 and the clustered T T D T T T D recruitment of Rab5 T effectors on the early RNTre endosome membrane. PtdIns(3)P Cytosol Rab5-GTP (T in the figure) is unstable on the Lumen early endosome where it undergoes continuous cycles of GTP hydrolysis (Rab-GDP is shown as D in the figure), catalysed by RN-Tre33 and nucleotide exchange44. The first feedback loop is due to the Rabaptin-5–Rabex-5 complex that activates Rab5 through the nucleotide-exchange activity of Rabex-5 and gets recruited on the early endosome membrane through Rabaptin-5. In this case, the product of the reaction (Rab5-GTP) recruits the enzyme. A second feedback loop is due to the cooperativity between effectors. Active Rab5 interacts with the hVPS34-p150 phosphoinositol-3-OH kinase (PI(3)K), thus coupling phosphatidylinositol-3-phosphate (PtdIns(3)P) production to Rab5 localization. The concomitant presence of Rab5 and PtdIns(3)P allows the recruitment of the Rab5 FYVE effectors early endosome antigen 1 (EEA1) and Rabenosyn5. It is also known that EEA1, Rabaptin-5 and Rabex-5 form high-molecular-weight oligomers on the early endosome membrane. Because the oligomers also incorporate the nucleotide-exchange factor for Rab5, Rabex-5, the effectors themselves feedback on the recruitment and clustering of Rab5 within a limited area of the early endosome. But what determines the specific targeting of Rab5 to the early endosome to initiate a Rab domain? Candidates for Rab receptors and GDI-releasing factors have been found97–101 but, eventually, their localization also needs to be explained. Similarly, what determines the targeting of hVPS34/p150? Important factors for the generation of a Rab domain might be the cooperativity and the self-organization properties of its components described above. We have seen in fact that, through positive feedback loops, the localization of one component depends on the recruitment of the other. None of the individual components (Rab5 or PI(3)K) is sufficient to form a domain but it is the combinatorial use of all components that creates the specificity of that particular membrane environment. For example, EEA1 is absent from the plasma membrane and clathrin-coated vesicles20,102, consistent with the idea that hVPS34 produces PtdIns(3)P on early endosomes48,55. On the plasma membrane, therefore, Rab5 alone is not sufficient to recruit EEA1 and the other FYVE effectors, arguing that clusters of Rab5/hVPS34/PtdIns(3)P/ EEA1/Rabenosyn-5 are present only on a subcompartment of the early endosome57. In addition to these interactions, generation of a Rab domain will most probably require associations between Rab5 effectors and additional membrane proteins. So the Rab5 machinery can be viewed as a typical modular system103, in which specific biochemical interactions between Rab5 effectors and regulators as well as other endosomal proteins create spatial segregation. By regulating the assembly of a specific membrane domain, these molecules contribute to the compartmental specificity, robustness and dynamic properties of the early endosome.

VOLUME 2 | FEBRUARY 2001 | 1 1 1

REVIEWS another example of cooperativity between effector molecules. PI(3)Ks function in various cellular processes49,50. Distinct types of PI(3)K have unique functions in signal transduction and mitogenesis, cytoskeletal organization and membrane transport51. Rab5 interacts directly with two types of PI(3)Ks. The first is p85α/p110β, a type I kinase that mainly phosphorylates PtdIns(4)P and phosphatidylinositol-4,5-bisphosphate (PtdIns(4,5)P2), generating PtdIns(3,4)P2 and phosphatidylinositol-3,4,5-trisphosphate (PtdIns(3,4,5)P3) (REF. 49). The function of p85α/p110β with respect to Rab5 is not known, but, given that type I PI(3)Ks are established components of the signaltransduction machinery, the interaction between p85α/p110β and Rab5 might reveal new properties of Rab5 in response to intracellular signalling. The second PI(3)K is hVPS34/p150, the mammalian homologue of yeast Vps34p/Vps15p52,53. This kinase preferentially phosphorylates phosphatidylinositol to PtdIns(3)P, which is necessary for the recruitment of FYVE finger proteins on the early endosome48,51,54. Rab5 is therefore coupled to the generation of PtdIns(3)P by the recruitment of hVPS34/p150 (BOX. 1), a result that is consistent with the enrichment of PtdIns(3)P on early endosomes55. This mechanism is not limited to the membrane recruitment of EEA1. Rabenosyn-5 is another FYVE finger Rab5 effector that, similar to EEA1, is recruited in a Rab5- and PtdIns(3)P-dependent fashion to early endosomes, where it functions in docking and fusion56. This mechanism could also indirectly regulate the recruitment of FYVE finger proteins other than Rab5 effectors to the early endosome. Purified clathrin-coated vesicles (CCV) fail to recruit EEA1, despite the presence of Rab5 (REF. 20,57). EEA1 (and probably Rabenosyn-5) can be exclusively attached to early endosomes and this asymmetrical recruitment between transport vesicles and their target organelle directly correlates with the asymmetrical distribution of hVPS34 (REF. 48). Binding of EEA1 to activated Rab5 through its amino-terminal site18 is therefore not sufficient for membrane recruitment. However, a patch of EEA1 molecules58 could provide several (low affinity) binding sites that are sufficient to tether an incoming CCV to the early endosome, thus providing directionality to the transport process. So far, less than half of the Rab5-interacting proteins have been identified and several other molecules still need to be integrated in the Rab5 scheme. In addition, if other Rab GTPases interact with effector proteins with a complexity similar to that of Rab5, it means that the number of total Rab effectors is likely to increase considerably with the functional characterization of other Rab proteins. CCV

Coated vesicles involved in the endocytosis of receptors at the plasma membrane. LIPID RAFTS

Lipids including cholesterol and sphingomyelin aggregated laterally to form membrane microdomains.

112

Rab domains

The diversity of biochemical reactions that are regulated by Rab proteins raises the question of how these processes are coordinated. We propose that Rab5 and its effectors are not randomly recruited and distributed on the early endosome membrane but are spatially segregated in a defined membrane domain or Rab domain.

| FEBRUARY 2001 | VOLUME 2

There is increasing evidence that membrane-bound molecules are not randomly distributed in the membrane bilayer but are enriched in membrane domains of varying lipid composition59. However, the membrane arrangement that is regulated by Rab5 is very different from that of other membrane domains such as LIPID RAFTS, which primarily depend on the intercalation of sphingolipids with cholesterol60. First, protein–lipid interactions are a central factor in the generation of the Rab5 domain. The localized synthesis of PtdIns(3)P not only allows the specific recruitment of FYVE effectors in conjunction with Rab5, but also contributes to their clustering. The dynamic properties of a Rab5 domain probably include a spatial and temporal control over PtdIns(3)P synthesis and turnover. As membrane flows through the early endosome, phosphoinositides other than PtdIns(3)P (for example, PtdIns(3,4)P2 and PtdIns(3,4,5)P3)48,61, could be either converted to PtdIns(3)P or excluded from the Rab5 domain. Conversely, the PtdIns(3)P that is generated in the Rab5 domain and not retained by Rab5 effectors could be used in other endosomal sub-compartments, undergo further modifications or be degraded55,62–64. PtdIns (3)P is also enriched in the internal vesicles of multivesicular endosomes55, suggesting that this phospholipid could also be generated at later stages of the endocytic pathway to function in combination with other Rab proteins. A second important factor in the formation of a Rab5 domain is the effector cooperativity. The local generation of lipids and recruitment of individual effectors is not sufficient to maintain a membrane domain. In the absence of lateral interactions between the proteins within the domains, these lipid–protein complexes would rapidly diffuse throughout the plane of the membrane, filling the whole endocytic pathway. One possible mechanism to avoid this is protein oligomerization. On the membrane, EEA1, Rabaptin-5 and Rabex-5 form dynamic oligomeric complexes with NSF58. However, oligomerization still might not be sufficient and a scaffold, such as the actin framework or one that is analogous to the spectrin/ankyrin system65, or to the Golgi stacking machinery66, might be needed to stabilize the local membrane composition of Rab effectors and arrange them in a precise architectural layout. Cytoskeletal tracks could also be connected with this scaffold to increase the efficiency of vesicle delivery to a Rab domain. Last, the generation and maintenance of the Rab5 domain depends on energy. The hydrolysis of GTP regulates the kinetics and limits the extent of effector recruitment. Through PtdIns(3)P production, PI(3)K uses ATP to recruit and, more importantly, to cluster these molecules (BOX 1). The ATPase activity of NSF regulates the dynamics of the hetero-oligomers58. The integration between GTPase and ATPase cycles therefore ensures a dynamic state between assembly and disassembly of oligomeric complexes of proteins and lipids and, consequently, confers a specific control on the size of that membrane domain. Visualization of Rab domains

Morphological studies have lent support to the proposal

www.nature.com/reviews/molcellbio

REVIEWS

Rab5

Rab5 Rab4

Rab 11

Rab 4

Nucleus

Figure 2 | Model of Rab domains on endosomes. Studies of Rab5, Rab4 and Rab11 tagged with green fluorescent protein (GFP) have shown that these Rab proteins are compartmentalized within the membrane of early endosomes68. Cargo flows sequentially through these domains as indicated by the arrows. The Rab domains also have a specific distribution and different pharmacological properties68. We propose that, similarly to the Rab5 effectors, Rab4 and Rab11 effectors are clustered in defined areas of the endosome membranes that are linked to each other through bifunctional Rab effectors.

GREEN FLUORESCENT PROTEIN

Autofluorescent protein originally identified in the jellyfish Aequorea Victoria. RECYCLING ENDOSOME

About 90% of endocytosed receptors are recycled to the plasma membrane. At least part of this traffic occurs through recycling endosomes. TRANSFERRIN

Protein involved in ferric ion uptake into the cell. The pathway followed by transferrin bound to its receptor defines the recycling pathway. APICAL JUNCTIONAL COMPLEX

Desmosomes, adherens junctions and tight junctions make up the apical junctional complex. NEURITE

Process extended by a nerve cell that can give rise to an axon or a dendrite. CIS-SNARE COMPLEX

SNARE pairing occurring within the same membrane.

that Rab proteins and/or Rab effectors are clustered in defined membrane domains. Upon expression of the activated Rab5Q79L mutant of Rab5, Rab5 and EEA1 concentrate in brightly fluorescent spots on the enlarged endosomal membrane58,67. Rab5, and probably its effectors, particularly accumulate at the interface between fusing vesicles and persist in a discrete spot for minutes after endosome fusion67. Immunofluorescence and video microscopy studies conducted on cells that express Rab5, Rab4 and Rab11 tagged with GREEN FLUORESCENT PROTEIN (GFP) have also revealed compartmentalization of these three GTPases within the endosomal membrane68. Three main populations of endosome could be distinguished: one that contains primarily Rab5, a second that contains both Rab5 and Rab4, and a third one that harbours Rab4 and Rab11. This distribution is consistent with biochemical and ultrastructural studies57,69,70. The former two would correspond to early/sorting endosomes and the latter to perinuclear 71 RECYCLING ENDOSOMES . TRANSFERRIN internalized as endocytic tracer was found first to enter the Rab5 domain and then to move sequentially through Rab4- and Rab11-positive structures. Given the function of these Rab proteins16,72,73, the Rab5 domain would be the gateway into the early endosome, whereas the Rab4 and Rab11 domains would contain the machinery that is necessary for sorting and recycling of transferrin to the cell surface. Endosomes can therefore be viewed as a

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

mosaic of Rab4, Rab5 and Rab11 (as well as other Rab) domains that dynamically interact but keep a relatively stable distribution over time (FIG. 2). These Rab domains could be connected through tethering molecules without intermixing. Molecules such as Rabaptin-5, which interacts with Rab5 and Rab4 through two distinct regions (REF. 74), might functionally and structurally link the Rab5 and Rab4 domains. Other molecules should then link other domains together (that is, Rab4 with Rab11). When combined with other types of lipid microdomains that have been shown to exist on endosomes75,76, the multiplicity of membrane platforms could reach a high level of complexity. This membrane compartmentalization is not limited to the Rab5 system, but has been observed for other Rab effectors that are involved in vesicle tethering. Docking sites could be inferred from the fluorescent morphology of discrete punctate sites around the yeast vacuole that contain at least two proteins required for vacuolar protein sorting and morphology, Vam2/Vps41p and Vam6/Vps39p77. Striking fluorescent images of patching were also seen in budding yeast for Sec3p, a component of the exocyst, which is restricted to the bud site13. The exocyst localizes to APICAL JUNCTION78 AL COMPLEXES in polarized mammalian cells , the apical pole of pancreatic acinar cells79, and the tips of growing NEURITES, growth cones and synapse-assembly sites in developing neurons80. Highly dynamic transport carriers bearing Rab6 and containing specific cargo molecules were observed to translocate from the Golgi to the ER81, supporting the idea that this retrograde transport pathway involves a subdomain of the Golgi complex. Thus, patches of Rab effectors have been observed within both endocytic and biosynthetic organelles, indicating that these molecules might generally be restricted spatially on the membrane of organelles. Rabs link to SNAREs and motor proteins

A crucial factor in vesicular transport is the coordination between Rab-dependent membrane tethering, docking and SNARE-dependent membrane fusion. Upon successful tethering and priming, the SNAREs engage in trans-interactions between vesicle- and target-SNAREs. This interaction leads to closely apposed membranes, resulting in fusion. We propose that the selective incorporation of CIS-SNARE COMPLEXES within a Rab domain is a prerequisite for the cognate SNAREs to pair in trans upon tethering/docking (BOX 2). Molecular interactions with Rab effectors within the Rab domain might result in the selective enrichment of cis-SNARE complexes at their site of function. Several recent studies have reported direct molecular links between Rab effector proteins and components of the SNARE machinery (BOX 2). Sec1p and related proteins, Vps33p and Vps45p, regulate SNARE pairing by sequestering syntaxin molecules82,83. Interactions between Rab effectors and Sec1 family members have been observed both in mammalian cells56 and in yeast25,84â&#x20AC;&#x201C;86. These interactions might function to retrieve the inhibitors from the SNAREs and might result in conformational changes that allow

ÂŠ 2001 Macmillan Magazines Ltd

VOLUME 2 | FEBRUARY 2001 | 1 1 3

REVIEWS

Box 2 | Molecular interactions showing cooperativity between Rab and SNARE machineries a | The Rab effectors Vac1p,

a Sec1 homologues bind Rab effectors

Rab effector removes negative regulation Rabenosyn-5 and the HOPS complex can bind Sec1 homologues directly25,56,84–86. The functional importance of Se Rab c1 this interaction can be seen as effector two distinct, but not mutually exclusive, possibilities. First SNAREs Rab (top), the Rab effector might domain remove Sec1 from the primed Rab effector delivers Sec1 to stabilize trans-SNARE pairs cis-SNARE complex to free the SNAREs for interactions in trans. Second (bottom), the Rab effector/Sec1 complex might Se c1 stabilize trans-SNARE pairs 104 after docking . b | In two independent systems, the hydrolysis of ATP by Nethyl-maleimide-sensitive fusion protein (NSF) and αb Priming machinery and Rab effectors SNAP could regulate Rab NSF within the Rab domain primes SNAREs and disassembles effector complexes effectors–SNARE interactions and oligomeric assembly of Rab effectors . The mechanistic ATP consequences of these reactions Rab NSF effectors α-SNAP are not clear but they might ADP mediate the cyclical Pi SNAREs assembly/disassembly of protein complexes during NSF Rab domain membrane transport. In the early endosome, EEA1, Rabaptin-5 and Rabex-5 exist c Rab effectors directly bind SNAREs in oligomeric complexes with Effectors selectively incorporate cognate SNAREs into Rab domains NSF58. The presence of NSF SNAREs within the oligomers could Rab function either to locally prime effector SNAREs or to regulate interactions among Rab Rab domain effectors or between effectors and SNAREs, ultimately leading Assembly of several trans pairs into a pore structure to trans-SNARE pairing. Sec18p (NSF homologue) activity might be required at several stages during vacuole fusion in yeast24–26, in which hydrolysis of ATP by Sec18p releases the Ypt7p (Rab7 homologue) HOPS effector complex from the Vam3p SNARE complex. HOPS can then bind and activate27 Ypt7p leading to vacuole tethering (b, right). In addition, the Ypt7p effector complex can also bind to primed Vam3p, which might aid assembly of trans-SNARE complexes87. Both data from mammalian and yeast systems therefore support the idea that vesicle tethering and SNARE priming are spatially and temporally coupled. c | Multiple direct links have been established between Rab effectors and SNAREs. Complexes formed between EEA1 and two t-SNAREs, syntaxin13 and syntaxin6 (REFS 58,105) within the Rab5 domain might mediate fusion between endosomes, or between trans-Golgi-network-derived vesicles and endosomes, respectively. The Ypt7p effector proteins have also been found in complexes with the SNARE Vam3p complex24,25,87. Through directly binding to syntaxin-5 complexes, the Rab1 effector p115 might ensure their incorporation into COPII vesicles7. The concentration of Rab effectors within a restricted membrane area would increase the apparent affinity and enhance the rate of Rab effectors–SNARE and SNARE–SNARE complex formation. In this way, only the correct SNAREs would be selectively incorporated within the Rab domain , at the target site for fusion (top). These interactions might also contribute to the steady-state localization of SNAREs, despite their regular cycle between organelles. In addition, effector–SNARE interactions could be functionally required for fusion58,87, possibly for the architectural arrangement and bridging of trans-SNARE pairs (bottom).

114

| FEBRUARY 2001 | VOLUME 2

www.nature.com/reviews/molcellbio

REVIEWS

SEC18P (Sec18p)

Saccharomyces cerevisiae homologue of NSF. HAEMAGGLUTININ

Spike protein of the influenza virus. HA is the bestunderstood fusion protein.

SNAREs to be primed or stabilized in a trans-paired conformation (BOX 2). A second set of interactions has been observed between the Rab effectors and the SNARE priming machinery. For example, Vac1p could be precipitated together with Pep12p and SEC18P in the presence of a sec18-1 mutant84. The Rab5 effector proteins EEA1 and Rabaptin-5–Rabex-5 complex form oligomers on the endosome whose assembly is dependent on the ATPase activity of NSF58. Interactions between the Ypt7p effector HOPS and SNAREs also depend on the ATPase activity of Sec18p (REF. 24). Finally, in both systems there are specific links to the SNARE proteins themselves. EEA1 binds directly to syntaxin 13 and this interaction is functionally required for endosome fusion58. The Ypt7p effector proteins have also been found in complexes with the SNARE Vam3p complex24,25,87. The compartmentalization of Rab effectors and the SNARE machinery within specifically assembled domains explains how productive, fusion-competent SNARE pairing would occur only within the membrane environment that has been selected for vesicle tethering, despite the presence of a cycling population of SNAREs at any given time and despite promiscuous pairing. In this context, it is interesting to consider the mechanistic links between the tethering machinery and the activation of SNAREs. The transition from membrane tethering to fusion implies that a relatively long distance must be bridged after activation of the SNARE machinery within the Rab domain. Using electron microscopy, tethered vesicles (50–100 nm in diameter) have been observed ~75–150 nm away from the target88. This is consistent with the length of the rod-like tethering factors. Ultrastructural analysis has shown that EEA1, for example, is associated with filamentous material that extends from the cytoplasmic surface of the endosome57. The current hypothesis is that the tethered vesicle swings through the highly viscous cytosol, attached by its ‘string’, until it collides with the surface of the target. Once the vesicle lands, the activated SNAREs engage for close docking, which closes the gap between the two bilayers. This would suggest that the tethering factors can communicate across the ~100-nm distance the fact that a vesicle has been ‘trapped’ to signal the priming of SNAREs. An alternative possibility is that tethering factors might undergo a large conformational change upon binding a vesicle at the free end, which would reduce the distance gap and result in the vesicle being pulled into close proximity to its target. There are many proteins that undergo extensive conformational changes that are important for their function. For example, HAEMAGGLUTININ (HA) of the influenza virus drastically changes its shape upon pH reduction89, the light-chain-(calmodulin)-binding domain of myosin swings to generate movement in response to conformational changes in the motor domain90,91, and the flagellar protein spasmin extends from a tightly coiled spring to a long rod (a distance difference of microns) to propel the Vorticella velogiines protozoa92. Interestingly, calmodulin is required for endosome fusion93 and EEA1 contains a putative calmodulin-binding IQ motif,

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

which might potentially function in such a conformational change. Functional coordination is not restricted to vesicle tethering and fusion but is also important for vesicle motility. Strikingly, similarly to endosome membrane docking and fusion, Rab5-dependent endosome movement along microtubules depends on hVPS34 PI(3)K. So, Rab5 functionally links regulation of membrane transport, motility and intracellular distribution of early endosomes. The higher the levels of Rab5–GTP, the higher the recruitment of Rab5 effectors, including hVPS34, and the higher the movement, tethering and fusion activity of endosomes. This implies that a microtubule motor or a molecule that regulates its recruitment and/or activity might be a PtdIns(3)P-binding protein. Rab5 effectors implicated in microtubuledependent early endosome motility have not yet been identified, although it is possible that a minus-ended kinesin might be involved37. Rabkinesin-6 is a Rab6 effector35 and a second kinesin-related protein, Rab33bBP, binds to Rab33b94. How these interactions result in vesicle motility remains to be determined. Perspectives for membrane biology

We are beginning to understand the fundamental principles of membrane transport from one intracellular compartment to another. However, a more complex task will be to define an organelle in molecular terms, determine its boundaries, explain how its transport machinery is specifically localized and what controls its size and intracellular distribution. Understanding the mechanisms of membrane compartmentalization and organelle biogenesis will therefore be an important challenge for the future. We have argued that the compartmentalization of Rab proteins and their effectors contributes to the formation of membrane domains and hence to the structural and functional properties of organelles. It will be important to find out whether this mode of function is limited to the early endosomal system or can be generalized to other Rab proteins of different organelles. Moreover, the idea of Rab domains needs to be integrated with other functional modules. For example, functional interactions of Rab proteins with Rho and ARF GTPases and their effectors95,96 could link vesicle formation and targeting with the role of the cytoskeleton in organelle structure and function. However, beside protein machineries, biological membranes contain different kinds of lipid microenvironments59,74,75. It will be interesting to see how the physico-chemical properties of these membrane platforms participate in the membrane compartmentalization of organelles and act on the location and function of the transport machinery. Morphological techniques will become increasingly important to assess the distribution and, most importantly, the dynamics of membrane compartments. In addition to studies in vivo, to elucidate the mechanisms that underlie the compartmentalization of biological membranes it will be necessary to develop biophysical methods (for example, artificial membrane systems) that recapitulate the clustering of molecules and their

VOLUME 2 | FEBRUARY 2001 | 1 1 5

REVIEWS

TRANSCYTOSIS PATHWAYS

Transport of macromolecules across a cell, consisting of endocytosis of a macromolecule at one side of a monolayer and exocytosis at the other side. APICAL

Plasma membrane surface of an epithelial cell that faces the lumen. BASOLATERAL

Plasma membrane surface of an epithelial cell that adjoins underlying tissue.

1. 2. 3.

4. 5. 6.

10.

11.

12.

13.

116

behaviour in vitro . These systems should also help us understand the principles that govern the size of cellular organelles and their sub-compartments. Why are endosomes scattered in comparison with the Golgi complex? What determines the geometry of these compartments? It will be interesting to see, in relation to the regulation of homotypic endosome fusion by Rab5 (REF. 44), whether the timer function of Rab GTPases could have a general function in regulating organelle dynamics. The disclosure of new protein sequences from the different genome projects provides an extraordinary number of novel tools with which to explore the organization of biological membranes. We will also learn a great deal from the generation and expansion of protein families throughout evolution. How has the basic transport machinery of yeast evolved from this simple eukaryote to multicellular organisms? Physiological processes typical of tissues and organs, such as neurotransmitter release, insulin-regulated traffic, recycling and transcytosis, require cell-type specific modifications of the intracellular pathways for the transport of particular cargo molecules. How is transport regulated in polarized epithelial cells and neurons? Different Rab domains containing different sets of Rab effectors could govern

Rothman, J. E. Mechanisms of intracellular protein transport. Nature 372, 55–63 (1994). Weber, T. et al. SNAREpins: minimal machinery for membrane fusion. Cell 92, 759–772 (1998). McNew, J. A. et al. Compartmental specificity of cellular membrane fusion encoded in SNARE proteins. Nature 407, 153–159 (2000). Using the reconstituted liposome fusion assay, Rothman and colleagues systematically tested liposomes containing each of the yeast v-SNAREs for fusion with three potential t-SNARE complexes. The cognate SNARE pairs showed a high degree of selectivity, an important tenant of the SNARE hypothesis. Parlati, F. et al. Topological restriction of SNARE-dependent membrane fusion. Nature 407, 194–198 (2000). Novick, P. & Zerial, M. The diversity of Rab proteins in vesicle transport. Curr. Opin. Cell Biol. 9, 496–504 (1997). Cao, X., Ballew, N. & Barlowe, C. Initial docking of ERderived vesicles requires Uso1p and Ypt1p but is independent of SNARE proteins. EMBO J. 17, 2156–2165 (1998). Allan, B. B., Moyer, B. D. & Balch, W. E. Rab1 recruitment of p115 into a cis-SNARE complex: programming budding COPII vesicles for fusion. Science 289, 444–448 (2000). Shows that the previously identified tethering factor, p115, is a Rab1 effector and direct ly interacts with the SNARE machinery. The functional importance of an interaction between a Rab effector and the cisSNARE complex during vesicle budding highlights the multiple roles of Rab effectors. Cao, X. & Barlowe, C. Asymmetric requirements for a Rab GTPase and SNARE proteins in fusion of COPII vesicles with acceptor membranes. J. Cell Biol. 149, 55–66 (2000). Sacher, M. et al. TRAPP, a highly conserved novel complex on the cis-Golgi that mediates vesicle docking and fusion. EMBO J. 17, 2494–2503 (1998). Wang, W., Sacher, M. & Ferro-Novick, S. TRAPP stimulates guanine nucleotide exchange on Ypt1p. J. Cell Biol. 151, 289–296 (2000). TerBush, D. R., Maurice, T., Roth, D. & Novick, P. The exocyst is a multiprotein complex required for exocytosis in Saccharomyces cerevisiae. EMBO J. 15, 6483–6494 (1996). Kee, Y. et al. Subunit structure of the mammalian exocyst complex. Proc. Natl Acad. Sci. USA 94, 14438–14443 (1997). Finger, F. P., Hughes, T. E. & Novick, P. Sec3p is a spatial landmark for polarized secretion in budding yeast. Cell 92, 559–571 (1998). Exocytosis in yeast occurs in a polarized fashion. This study shows that Sec3p localizes to the site of

| FEBRUARY 2001 | VOLUME 2

14.

15.

16.

17.

18.

19.

20.

21.

22.

23.

protein sorting and transport in the secretory, endocytic/recycling and TRANSCYTOTIC PATHWAYS of polarized cells. Efforts in this direction are expected to provide clues into, for example, what makes APICAL and BASOLATERAL early endosomes functionally distinct in polarized epithelial cells. How is their spatial distribution controlled? It will be fascinating to see how the basic molecular principles that account for membrane compartmentalization and organelle polarity can contribute, at larger scale, to an integrated understanding of cell polarity and complex tissues in multicellular organisms.

polarized exocytosis and is required for the targeting of secretory vesicles to the bud. This is in line with the concept that Rab effectors specify where the vesicles should tether on their target compartment, allowing trans-SNARE complex formation to mediate fusion. Guo, W., Roth, D., Walch-Solimena, C. & Novick, P. The exocyst is an effector for Sec4p, targeting secretory vesicles to sites of exocytosis. EMBO J. 18, 1071–1080 (1999). This work established that Sec15p, a subunit of the exocyst, is a Sec4p effector protein. It supports the idea that Rab proteins regulate the function of multimeric protein complexes important in the initial recognition of the docking site for an incoming vesicle. Gorvel, J. -P., Chavrier, P., Zerial, M. & Gruenberg, J. Rab5 controls early endosome fusion in vitro. Cell 64, 915–925 (1991). Bucci, C. et al. The small GTPase rab5 functions as a regulatory factor in the early endocytic pathway. Cell 70, 715–728 (1992). Chistoforidis, S., McBride, H. M., Burgoyne, R., D. & Zerial, M. The Rab5 effector EEA1 is a core component of endosome docking. Nature 397, 621–625 (1999). The identification of over 20 proteins that bind specifically to activated Rab5 opened up the idea that a much more complex protein machinery could be downstream of a Rab protein and have a wider regulatory role than previously imagined. Second, this paper shows that EEA1 alone can tether early endosomes and allow the SNARE machinery to mediate fusion. Simonsen, A. et al. EEA1 links phosphatidylinositol 3-kinase function to Rab5 regulation of endosome fusion. Nature 394, 494–498 (1998). Barbieri, M. A. et al. Evidence for a symmetrical requirement for Rab5-GTP in in vitro endosome–endosome fusion. J. Biol. Chem. 273, 25850–25855 (1998). Rubino, M., Miaczynska, M., Lippe, R. & Zerial, M. Selective membrane recruitment of EEA1 suggests a role in directional transport of clathrin-coated vesicles to early endosomes. J. Biol. Chem. 275, 3745–3748 (2000). Haas, A., Scheglmann, D., Lazar, T., Gallwitz, D. & Wickner, W. The GTPase Ypt7p of Saccharomyces cerevisiae is required on both partner vacuoles for the homotypic fusion step of vacuole inheritance. EMBO J. 14, 5258–5270 (1995). Mayer, A. & Wickner, W. Docking of yeast vacuoles is catalyzed by the Ras-like GTPase Ypt7p after symmetric priming by Sec18p (NSF). J. Cell Biol. 136, 307–317 (1997). Rieder, S. E. & Emr, S. D. A novel RING finger protein complex essential for a late step in protein transport to the

yeast vacuole. Mol. Biol. Cell 8, 2307–2327 (1997). 24. Price, A., Seals, D., Wickner, W. & Ungermann, C. The docking stage of yeast vacuole fusion requires the transfer of proteins from a cis-SNARE complex to a Rab/Ypt protein. J. Cell Biol. 148, 1231–1238 (2000). These experiments have shown that the HOPS complex is an effector of Ypt7p in vacuole tethering. Furthermore, this work, together with References 57 and 86 provided the first demonstrations of the dynamics of the interactions between Rab effector and SNARE machineries. 25. Seals, D. F., Eitzen, G., Margolis, N., Wickner, W. T. & Price, A. A Ypt/Rab effector complex containing the Sec1 homolog Vps33p is required for homotypic vacuole fusion. Proc. Natl Acad. Sci. USA 97, 9402–9407 (2000). 26. Price, A., Wickner, W. & Ungermann, C. Proteins needed for vesicle budding from the Golgi complex are also required for the docking step of homotypic vacuole fusion. J. Cell Biol. 148, 1223–1230 (2000). 27. Wurmser, A. E., Sato, T. K. & Emr, S. D. New component of the vacuolar class C-Vps complex couples nucleotide exchange on the Ypt7 GTPase to SNARE-dependent docking and fusion. J. Cell Biol. 151, 551–562 (2000). 28. Horiuchi, H. et al. A novel Rab5 GDP/GTP exchange factor complexed to Rabaptin-5 links nucleotide exchange to effector recruitment and function. Cell 90, 1149–1159 (1997). 29. Nuoffer, C., Davidson, H. W., Matteson, J., Meinkoth, J. & Balch, W. E. A GDP-bound form of rab1 inhibits protein export from the endoplasmic reticulum and transport between Golgi compartments. J. Cell Biol. 125, 225–237 (1994). 30. Riederer, M. A., Soldati, T., Shapiro, A. D., Lin, J. & Pfeffer, S. Lysosome biogenesis requires Rab9 function and receptor recycling from endosomes to the trans Golgi network. J. Cell Biol. 125, 573–582 (1994). 31. Barlowe, C. et al. COPII: a membrane coat formed by Sec proteins that drive vesicle budding from the endoplasmic reticulum. Cell 77, 895–907 (1994). 32. McLauchlan, H. et al. A novel role for Rab5-GDI in ligand sequestration into clathrin-coated pits. Curr. Biol. 8, 34–45 (1998). 33. Lanzetti, L. et al. The Eps8 protein coordinates EGF receptor signalling through Rac and trafficking through Rab5. Nature 408, 374–377 (2000). 34. Rehling, P., Darsow, T., Katzmann, D. J. & Emr, S. D. Formation of AP-3 transport intermediates requires Vps41 function. Nature Cell Biol. 1, 346–353 (1999). 35. Echard, A. et al. Interaction of a Golgi-associated kinesinlike protein with Rab6. Science 279, 580–585 (1998). A yeast two-hybrid search for effectors identified a kinesin as a Rab6-interacting protein. In line with the

www.nature.com/reviews/molcellbio

REVIEWS

36.

37.

38.

39.

40.

41.

42.

43.

44. 45.

46.

47.

48.

49.

50.

51.

52.

53.

54.

55.

56.

57.

concept that Rab proteins can have different functions, this result suggests that a Rab protein might be functionally connected with the microtubule-dependent motility of vesicles and organelles. Hill, E., Clarke, M. & Barr, F. A. The Rab6-binding kinesin, Rab6-KIFL, is required for cytokinesis. EMBO J. 19, 5711–5719 (2000). Nielsen, E., Severin, F., Backer, J. M., Hyman, A. A. & Zerial, M. Rab5 regulates motility of early endosomes on microtubules. Nature Cell Biol. 1, 376–382 (1999). A role of Rab5 in the attachment and motility of early endosomes along microtubules was uncovered using in vitro assays. These data together with the findings of Reference 35, strongly support an additional role of Rab proteins in organelle movement along the cytoskeleton. Pruyne, D. W., Schott, D. H. & Bretscher, A. Tropomyosincontaining actin cables direct the Myo2p-dependent polarized delivery of secretory vesicles in budding yeast. J. Cell Biol. 143, 1931–1945 (1998). Schott, D., Ho, J., Pruyne, D. & Bretscher, A. The COOHterminal domain of Myo2p, a yeast myosin V, has a direct role in secretory vesicle targeting. J. Cell Biol. 147, 791–808 (1999). Menasche, G. et al. Mutations in RAB27A cause Griscelli syndrome associated with haemophagocytic syndrome. Nature Genet. 25, 173–176 (2000). Mercer, J. A., Seperack, P. K., Strobel, M. C., Copeland, N. G. & Jenkins, N. A. Novel myosin heavy chain encoded by murine dilute coat colour locus. Nature 349, 709–713 (1991); erratum 352, 547 (1992). Wilson, S. M. et al. A mutation in Rab27a causes the vesicle transport defects observed in ashen mice. Proc. Natl Acad. Sci. USA 97, 7933–7938 (2000). Stenmark, H., Vitale, G., Ullrich, O. & Zerial, M. Rabaptin-5 is a direct effector of the small GTPase Rab5 in endocytic membrane fusion. Cell 83, 423–432 (1995). Rybin, V. et al. GTPase activity of Rab5 acts as a timer for endocytic membrane fusion. Nature 383, 266–269 (1996). Stenmark, H., Aasland, R., Toh, B. H. & D’Arrigo, A. Endosomal localization of the autoantigen EEA1 is mediated by a zinc-binding FYVE finger. J. Biol. Chem. 271, 24048–24054 (1996). Stenmark, H. & Aasland, R. FYVE-finger proteins — effectors of an inositol lipid. J. Cell Sci. 112, 4175–4183 (1999). Lawe, D. C., Patki, V., Heller-Harrison, R., Lambright, D. & Corvera, S. The FYVE domain of early endosome antigen 1 is required for both phosphatidylinositol 3-phosphate and Rab5 binding. Critical role of this dual interaction for endosomal localization. J. Biol. Chem. 275, 3699–3705 (2000). Christoforidis, S. et al. Phosphatidylinositol-3-OH kinases are Rab5 effectors. Nature Cell Biol. 1, 249–252 (1999). The identification of two distinct lipid kinases as effectors of a Rab protein linked the Rab field to signal transduction and strengthened the idea that Rabs regulate the assembly of specialized lipid and protein microdomains. Vanhaesebroeck, B., Leevers, S. J., Panayotou, G. & Waterfield, M. D. Phosphoinositide 3-kinases: a conserved family of signal transducers. Trends Biochem. Sci. 22, 267–272 (1997). Rameh, L. E. & Cantley, L. C. The role of phosphoinositide 3-kinase lipid products in cell function. J. Biol. Chem. 274, 8347–8350 (1999). Siddhanta, U., McIlroy, J., Shah, A., Zhang, Y. & Backer, J. M. Distinct roles for the p110α and hVPS34 phosphatidylinositol 3′-kinases in vesicular trafficking, regulation of the actin cytoskeleton, and mitogenesis. J. Cell Biol. 143, 1647–1659 (1998). Schu, P. V. et al. Phosphatidylinositol 3-kinase encoded by yeast VPS34 gene essential for protein sorting. Science 260, 88–91 (1993). First demonstration of a functional connection between PI(3)K and membrane transport. Volinia, S. et al. A human phosphatidylinositol 3-kinase complex related to the yeast Vps34p-Vps15p protein sorting system. EMBO J. 14, 3339–3348 (1995). Burd, C. G. & Emr, S. D. Phosphatidylinositol(3)-phosphate signaling mediated by specific binding to RING FYVE domains. Mol. Cell 2, 157–162 (1998). Gillooly, D. J. et al. Localization of phosphatidylinositol 3phosphate in yeast and mammalian cells. EMBO J. 19, 4577–4588 (2000). Nielsen, E. et al. Rabenosyn-5, a novel Rab5 effector, is complexed with hVPS45 and recruited to endosomes through a FYVE finger domain. J. Cell Biol. 151, 601–612 (2000). Wilson, J. M. et al. EEA1, a tethering protein of the early sorting endosome, shows a polarized distribution in

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

58.

59.

60. 61.

62.

63.

64.

65.

66.

67.

68.

69.

70.

71. 72.

73.

74.

75.

76.

77.

78.

79.

80.

81.

82.

hippocampal neurons, epithelial cells, and fibroblasts. Mol. Biol. Cell 11, 2657–2671 (2000). References 57, 68–70 support the idea that Rab proteins and their effectors are compartmentalized in early endosomes and show a non-random distribution. McBride, H. M. et al. Oligomeric complexes link Rab5 effectors with NSF and drive membrane fusion via interactions between EEA1 and syntaxin 13. Cell 98, 377–386 (1999). Mukherjiee, S. & Maxfield, F. R. Role of membrane organization and membrane domains in endocytic lipid trafficking. Traffic 1, 203–211 (2000). Simons, K. & Ikonen, E. Functional rafts in cell membranes. Nature 387, 569–572 (1997). Domin, J., Gaidarov, I., Smith, M. E., Keen, J. H. & Waterfield, M. D. The class II phosphoinositide 3-kinase PI3K-C2α is concentrated in the trans-Golgi network and present in clathrin-coated vesicles. J. Biol. Chem. 275, 11943–11950 (2000). Wurmser, A. E. & Emr, S. D. Phosphoinositide signaling and turnover: PtdIns(3)P, a regulator of membrane traffic, is transported to the vacuole and degraded by a process that requires lumenal vacuolar hydrolase activities. EMBO J. 17, 4930–4942 (1998). Gary, J. D., Wurmser, A. E., Bonangelino, C. J., Weisman, L. S. & Emr, S. D. Fab1p is essential for PtdIns(3)P 5-kinase activity and the maintenance of vacuolar size and membrane homeostasis. J. Cell Biol. 143, 65–79 (1998). Blondeau, F. et al. Myotubularin, a phosphatase deficient in myotubular myopathy, acts on phosphatidylinositol 3kinase and phosphatidylinositol 3-phosphate pathway. Hum. Mol. Genet. 9, 2223–2229 (2000). De Matteis, M. A. & Morrow, J. S. The role of ankyrin and spectrin in membrane transport and domain formation. Curr. Opin. Cell Biol. 10, 542–549 (1998). Shorter, J. et al. GRASP55, a second mammalian GRASP protein involved in the stacking of Golgi cisternae in a cellfree system. EMBO J. 18, 4949–4960 (1999). Roberts, R. L. et al. Endosome fusion in living cells overexpressing GFP-rab5. J. Cell Sci. 112, 3667–3675 (1999). Sonnichsen, B., De Renzis, S., Nielsen, E., Rietdorf, J. & Zerial, M. Distinct membrane domains on endosomes in the recycling pathway visualized by multicolor imaging of Rab4, Rab5, and Rab11. J. Cell Biol. 149, 901–914 (2000). Sheff, D. R., Daro, E. A., Hull, M. & Mellman, I. The receptor recycling pathway contains two distinct populations of early endosomes with different sorting functions. J. Cell Biol. 145, 123–139 (1999). Trischler, M., Stoorvogel, W. & Ullrich, O. Biochemical analysis of distinct Rab5- and Rab11-positive endosomes along the transferrin pathway. J. Cell Sci. 112, 4773–4783 (1999). Gruenberg, J. & Maxfield, F. R. Membrane transport in the endocytic pathway. Curr. Opin. Cell Biol. 7, 552–563 (1995). van der Sluijs, P. et al. The small GTP-binding protein rab4 controls an early sorting event on the endocytic pathway. Cell 70, 729–740 (1992). Ullrich, O., Reinsch, S., Urbé, S., Zerial, M. & Parton, R. G. Rab11 regulates recycling through the pericentriolar recycling endosome. J. Cell Biol. 135, 913–924 (1996). Vitale, G. et al. Distinct Rab-binding domains mediate the interaction of Rabaptin-5 with GTP-bound Rab4 and Rab5. EMBO J. 17, 1941–1951 (1998). Simons, K. & Gruenberg, J. Jamming the endosomal system: lipid rafts and lysosomal storage diseases. Trends Cell Biol. 10, 459–462 (2000). Burd, C. G., Babst, M. & Emr, S. D. Novel pathways, membrane coats and PI kinase regulation in yeast lysosomal trafficking. Semin. Cell Dev. Biol. 9, 527–533 (1998). Nakamura, N., Hirata, A., Ohsumi, Y. & Wada, Y. Vam2/Vps41p and Vam6/Vps39p are components of a protein complex on the vacuolar membranes and involved in the vacuolar assembly in the yeast Saccharomyces cerevisiae. J. Biol. Chem. 272, 11344–11349 (1997). Grindstaff, K. K. et al. Sec6/8 complex is recruited to cell–cell contacts and specifies transport vesicle delivery to the basal-lateral membrane in epithelial cells. Cell 93, 731–740 (1998). Shin, D. M., Zhao, X. S., Zeng, W., Mozhayeva, M. & Muallem, S. The mammalian Sec6/8 complex interacts with Ca2+ signaling complexes and regulates their activity. J. Cell Biol. 150, 1101–1112 (2000). Hazuka, C. D. et al. The sec6/8 complex is located at neurite outgrowth and axonal synapse-assembly domains. J. Neurosci. 19, 1324–1334 (1999). White, J. et al. Rab6 coordinates a novel Golgi to ER retrograde transport pathway in live cells. J. Cell Biol. 147, 743–760 (1999). Pevsner, J. et al. Specificity and regulation of a synaptic

vesicle docking complex. Neuron 13, 353–361 (1994). 83. Yang, B., Steegmaier, M., Gonzalez, L. C, Jr & Scheller, R. H. nSec1 binds a closed conformation of syntaxin1A. J. Cell Biol. 148, 247–252 (2000). 84. Burd, C. G., Peterson, M., Cowles, C. R. & Emr, S. D. A novel Sec18p/NSF-dependent complex required for Golgi-to-endosome transport in yeast. Mol. Biol. Cell 8, 1089–1104 (1997). 85. Peterson, M. R., Burd, C. G. & Emr, S. D. Vac1p coordinates Rab and phosphatidylinositol 3-kinase signaling in Vps45p-dependent vesicle docking/fusion at the endosome. Curr. Biol. 9, 159–162 (1999). 86. Tall, G. G., Hama, H., DeWald, D. B. & Horazdovsky, B. F. The phosphatidylinositol 3-phosphate binding protein Vac1p interacts with a Rab GTPase and a Sec1p homologue to facilitate vesicle-mediated vacuolar protein sorting. Mol. Biol. Cell 10, 1873–1889 (1999). 87. Sato, T. K., Rehling, P., Peterson, M. R. & Emr, S. D. Class C Vps protein complex regulates vacuolar SNARE pairing and is required for vesicle docking/fusion. Mol. Cell 6, 661–671 (2000). 88. Orci, L., Perrelet, A. & Rothman, J. E. Vesicles on strings: morphological evidence for processive transport within the Golgi stack. Proc. Natl Acad. Sci. USA 95, 2279–2283 (1998). 89. Wiley, D. C. & Skehel, J. J. The structure and function of the hemagglutinin membrane glycoprotein of influenza virus. Annu. Rev. Biochem. 56, 365–394 (1987). 90. Uyeda, T. Q., Abramson, P. D. & Spudich, J. A. The neck region of the myosin motor domain acts as a lever arm to generate movement. Proc. Natl Acad. Sci. USA 93, 4459–4464 (1996). 91. Howard, J. & Spudich, J. A. Is the lever arm of myosin a molecular elastic element? Proc. Natl Acad. Sci. USA 93, 4462–4464 (1996). 92. Mahadevan, L. & Matsudaira, P. Motility powered by supramolecular springs and ratchets. Science 288, 95–100 (2000). 93. Colombo, M. I., Beron, W. & Stahl, P. D. Calmodulin regulates endosome fusion. J. Biol. Chem. 272, 7707–7712 (1997). 94. Koda, T., Zheng, J. Y. & Ishibe, M. Rab33B regulates intraGolgi transport and utilizes a novel kinesin as an effector. 39th Annual Meeting of the American Society for Cell Biology Abstract no. 1244, 214a (Washington, D. C., 1999). 95. Imamura, H. et al. Rho and Rab small G proteins coordinately reorganize stress fibers and focal adhesions in MDCK cells. Mol. Biol. Cell 9, 2561–2575 (1998). 96. Chavrier, P. & Goud, B. The role of ARF and Rab GTPases in membrane transport. Curr. Opin. Cell Biol. 11, 466–475 (1999). 97. Dirac-Svejstrup, A. B., Sumizawa, T. & Pfeffer, S. R. Identification of a GDI displacement factor that releases endosomal Rab GTPases from Rab-GDI. EMBO J. 16, 465–472 (1997). 98. Ayad, N., Hull, M. & Mellman, I. Mitotic phosphorylation of rab4 prevents binding to a specific receptor on endosome membranes. EMBO J. 16, 4497–4507 (1997). 99. Bucci, C., Chiariello, M., Lattero, D., Maiorano, M. & Bruni, C. B. Interaction cloning and characterization of the cDNA encoding the human prenylated rab acceptor (PRA1). Biochem. Biophys. Res. Commun. 258, 657–662 (1999). 100. Matern, H. et al. A novel Golgi membrane protein is part of a GTPase-binding protein complex involved in vesicle targeting. EMBO J. 19, 4485–4492 (2000). 101. Hoffenberg, S. et al. A novel membrane-anchored Rab5 interacting protein required for homotypic endosome fusion. J. Biol. Chem. 275, 24661–24669 (2000). 102. Mu, F. T. et al. EEA1, an early endosome-associated protein. EEA1 is a conserved α-helical peripheral membrane protein flanked by cysteine ‘fingers’ and contains a calmodulin-binding IQ motif. J. Biol. Chem. 270, 13503–13511 (1995). 103. Hartwell, L. H., Hopfield, J. J., Leibler, S. & Murray, A. W. From molecular to modular cell biology. Nature 402, C47–C52 (1999). 104. Carr, C. M., Grote, E., Munson, M., Hughson, F. M. & Novick, P. J. Sec1p binds to SNARE complexes and concentrates at sites of secretion. J. Cell Biol. 146, 333–344 (1999). 105. Simonsen, A., Gaullier, J. M., D’Arrigo, A. & Stenmark, H. The Rab5 effector EEA1 interacts directly with syntaxin–6. J. Biol. Chem. 274, 28857–28860 (1999).

Acknowledgments We thank members of the Zerial lab, R. Lippe, M. Miaczynska and S. De Renzis, as well as our colleagues K. Simons, J. Gruenberg, G. Griffiths, J. Howard for their helpful comments and critical reading of the manuscript. We are grateful to I. Kaestner for superb secretarial help.

VOLUME 2 | FEBRUARY 2001 | 1 1 7

REVIEWS

Marino Zerial did his biology studies in Triest (Italy) where he worked on lysosomal storage disorders. He conducted postdoctoral work in Paris with Giorgio Bernardi (1983; Institut Jacques Monod) and in Heidelberg at the EMBL with Henrik Garoff (1985). From 1989, he was an EMBL staff scientist with Kai Simons. Then, in 1991, he became a group leader and, in 1998, a Director in the new Max Planck Institute MPI-CGB in Dresden. He is interested in the mechanisms of endocytosis and cell polarity. Heidi McBride began her career in 1991 as a graduate student working in the field of mitochondrial biogenesis in Gordon Shore’s laboratory at McGill University in Montreal, Quebec. In 1996, she moved to the EMBL in Heidelberg to complete her training as a cell biologist working with Marino Zerial. In 2000, Heidi returned home as Assistant Professor at the University of Ottawa Heart Insititute in Canada.

RAB1 http://www.ncbi.nlm.nih.gov/LocusLink/LocRpt.cgi?l=5861 RAB3 http://www.ncbi.nlm.nih.gov/LocusLink/list.cgi?Q=rab3a%20or%20rab3b%20or%20r ab3d%20not%20interacting&ORG=Hs

α-actinin http://www.ncbi.nlm.nih.gov/LocusLink/list.cgi?Q=actn*&ORG=Hs Rabex 5 http://www.ncbi.nlm.nih.gov/LocusLink/LocRpt.cgi?l=27342 syntaxin 13 http://www.ncbi.nlm.nih.gov/LocusLink/LocRpt.cgi?l=23673 syntaxin 6 http://www.ncbi.nlm.nih.gov/LocusLink/LocRpt.cgi?l=10228 hVps34 http://www.ncbi.nlm.nih.gov/LocusLink/LocRpt.cgi?l=5289

p150 http:www.ncbi.nlm.nih.gov/Locuslink/locRpt.cgi?1=30849 p85α http://www.ncbi.nlm.nih.gov/LocusLink/LocRpt.cgi?l=5290 rabenosyn-5 http://www.ncbi.nlm.nih.gov/LocusLink/LocRpt.cgi?l=64145 rabkinesin-6 http://www.ncbi.nlm.nih.gov/LocusLink/LocRpt.cgi?l=19348 RAB8IP http://www.ncbi.nlm.nih.gov/LocusLink/LocRpt.cgi?l=5871 p40 http://www.ncbi.nlm.nih.gov/LocusLink/LocRpt.cgi?l=10244 Rab11bp http://www.ncbi.nlm.nih.gov/LocusLink/LocRpt.cgi?l=54521

RAB4 http://www.ncbi.nlm.nih.gov/LocusLink/list.cgi?Q=rab4%20or%20rab4b&ORG=Hs

VPS45 http://www.ncbi.nlm.nih.gov/LocusLink/LocRpt.cgi?l=11311

RAB5 http://www.ncbi.nlm.nih.gov/LocusLink/list.cgi?Q=rab5a%20or%20rab5b%20or%20r ab5c&ORG=Hs

Sec13r http://www.ncbi.nlm.nih.gov/LocusLink/LocRpt.cgi?l=20332

RAB6 http://www.ncbi.nlm.nih.gov/LocusLink/list.cgi?Q=rab6%20or%20rab6b%20not%20i nteracting%20not%20similar%20not%20activating&ORG=Hs RAB8 http://www.ncbi.nlm.nih.gov/LocusLink/LocRpt.cgi?l=4218 RAB9 http://www.ncbi.nlm.nih.gov/LocusLink/LocRpt.cgi?l=9367 RAB11 http://www.ncbi.nlm.nih.gov/LocusLink/list.cgi?Q=rab11a%20or%20rab11b&ORG= Hs RAB13 http://www.ncbi.nlm.nih.gov/LocusLink/LocRpt.cgi?l=5872 Rab33b http://www.ncbi.nlm.nih.gov/LocusLink/LocRpt.cgi?l=19338 p115 http://www.ncbi.nlm.nih.gov/LocusLink/LocRpt.cgi?l=8615 PRA1 http://www.ncbi.nlm.nih.gov/LocusLink/LocRpt.cgi?l=10567 rabphilin3 http://www.ncbi.nlm.nih.gov/LocusLink/LocRpt.cgi?l=19894 calmodulin http://www.ncbi.nlm.nih.gov/LocusLink/LocRpt.cgi?l=801 rabaptin 5 http://www.ncbi.nlm.nih.gov/LocusLink/LocRpt.cgi?l=9135 EEA1 http://www.ncbi.nlm.nih.gov/LocusLink/LocRpt.cgi?l=8411 p110β http://www.ncbi.nlm.nih.gov/LocusLink/LocRpt.cgi?l=5291 VAMP2 http://www.ncbi.nlm.nih.gov/LocusLink/LocRpt.cgi?l=6844

MSS4 http://www.ncbi.nlm.nih.gov/LocusLink/LocRpt.cgi?l=5877 RAB3GAP http://www.ncbi.nlm.nih.gov/LocusLink/LocRpt.cgi?l=22930 Tuberous sclerosis 2 http://www.ncbi.nlm.nih.gov/LocusLink/LocRpt.cgi?l=7249 EPS8 http://www.ncbi.nlm.nih.gov/LocusLink/LocRpt.cgi?l=2059 δ-PDE http://www.ncbi.nlm.nih.gov/LocusLink/LocRpt.cgi?l=5147 YEAST GENES Sec4 http://genome-www4.stanford.edu/cgi-bin/SGD/locus.pl?locus=sec4 Exocyst http://genome-www4.stanford.edu/cgi-bin/SGD/GO/go.pl?goid=145 Sec15 http://genome-www4.stanford.edu/cgi-bin/SGD/locus.pl?locus=sec15 Vps45 http://genome-www4.stanford.edu/cgi-bin/SGD/locus.pl?locus=vps45 Pep12p http://genome-www4.stanford.edu/cgi-bin/SGD/locus.pl?locus=pep12 Vam3p http://genome-www4.stanford.edu/cgi-bin/SGD/locus.pl?locus=vam3 Vac1p http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?cmd=Retrieve&db=Nucleotide&list_uids=173156&dopt= GenBank Vps39p http://genome-www4.stanford.edu/cgi-bin/SGD/locus.pl?locus=vps39 Vps41p http://genome-www4.stanford.edu/cgi-bin/SGD/locus.pl?locus=vps41 Yip1p

118

| FEBRUARY 2001 | VOLUME 2

www.nature.com/reviews/molcellbio

REVIEWS http://genome-www4.stanford.edu/cgi-bin/SGD/locus.pl?locus=yip1 Yif1 http://genome-www4.stanford.edu/cgi-bin/SGD/locus.pl?locus=yif1 Ypt51p http://genome-www4.stanford.edu/cgi-bin/SGD/locus.pl?locus=ypt51 Ypt6p http://genome-www4.stanford.edu/cgi-bin/SGD/locus.pl?locus=ypt6 Ypt7p http://genome-www4.stanford.edu/cgi-bin/SGD/locus.pl?locus=ypt7

Ypt1 http://genome-www4.stanford.edu/cgi-bin/SGD/locus.pl?locus=ypt1

Vps33p http://genome-www4.stanford.edu/cgi-bin/SGD/locus.pl?locus=vps33 Vps39p http://genome-www4.stanford.edu/cgi-bin/SGD/locus.pl?locus=vps39 Vps41p http://genome-www4.stanford.edu/cgi-bin/SGD/locus.pl?locus=vps41

RAB13 http://www.ncbi.nlm.nih.gov/LocusLink/LocRpt.cgi?l=5872 Rab21 http://www.ncbi.nlm.nih.gov/LocusLink/LocRpt.cgi?l=19333

Dss4p http://genome-www4.stanford.edu/cgi-bin/SGD/locus.pl?locus=dss4

Rab6a,b http://www.ncbi.nlm.nih.gov/LocusLink/list.cgi?Q=rab6%20or%20rab6b%20not %20interacting%20not%20similar%20not%20activating&ORG=Hs

Gyp1 http://genome-www4.stanford.edu/cgi-bin/SGD/locus.pl?locus=gyp1

Rab10 http://www.ncbi.nlm.nih.gov/LocusLink/LocRpt.cgi?l=10890

Gyp2 http://genome-www4.stanford.edu/cgi-bin/SGD/locus.pl?locus=gyp2

Rab12 http://www.ncbi.nlm.nih.gov/LocusLink/LocRpt.cgi?l=19328

Gyp3 http://genome-www4.stanford.edu/cgi-bin/SGD/locus.pl?locus=gyp3

Rab30 http://www.ncbi.nlm.nih.gov/LocusLink/LocRpt.cgi?l=27314

Gyp6 http://genome-www4.stanford.edu/cgi-bin/SGD/locus.pl?locus=gyp6

Rab33b http://www.ncbi.nlm.nih.gov/LocusLink/LocRpt.cgi?l=19338

Vps9p http://genome-www4.stanford.edu/cgi-bin/SGD/locus.pl?locus=vps9

Rab36 http://www.ncbi.nlm.nih.gov/LocusLink/LocRpt.cgi?l=9609

Sec2p http://genome-www4.stanford.edu/cgi-bin/SGD/locus.pl?locus=sec2

RAB6a,b http://www.ncbi.nlm.nih.gov/LocusLink/list.cgi?Q=rab6%20or%20rab6b%20not %20interacting%20not%20similar%20not%20activating&ORG=Hs

Gyp7p http://genome-www4.stanford.edu/cgi-bin/SGD/locus.pl?locus=gyp7 TRAPP intermediate page Bet3 http://genome-www4.stanford.edu/cgi-bin/SGD/locus.pl?locus=bet3

RAB1 http://www.ncbi.nlm.nih.gov/LocusLink/LocRpt.cgi?l=5861 Rab2 http://www.ncbi.nlm.nih.gov/LocusLink/LocRpt.cgi?l=5862

Bet5 http://genome-www4.stanford.edu/cgi-bin/SGD/locus.pl?locus=bet5

RAB3 http://www.ncbi.nlm.nih.gov/LocusLink/list.cgi?Q=rab3a%20or%20rab3b%20or %20rab3d%20not%20interacting&ORG=Hs

Trs20 http://genome-www4.stanford.edu/cgi-bin/SGD/locus.pl?locus=trs20

RAB8 http://www.ncbi.nlm.nih.gov/LocusLink/LocRpt.cgi?l=4218

Trs23 http://genome-www4.stanford.edu/cgi-bin/SGD/locus.pl?locus=trs23

Rab26 http://www.ncbi.nlm.nih.gov/LocusLink/LocRpt.cgi?l=25837 Rab37 http://www.ncbi.nlm.nih.gov/LocusLink/LocRpt.cgi?l=58222 RAB11 http://www.ncbi.nlm.nih.gov/LocusLink/list.cgi?Q=rab11a%20or%20rab11b&OR G=Hs Rab17 http://www.ncbi.nlm.nih.gov/LocusLink/LocRpt.cgi?l=19329 RAB9 http://www.ncbi.nlm.nih.gov/LocusLink/LocRpt.cgi?l=9367 Rab24 http://www.ncbi.nlm.nih.gov/LocusLink/LocRpt.cgi?l=19336 RAB5 http://www.ncbi.nlm.nih.gov/LocusLink/list.cgi?Q=rab5a%20or%20rab5b%20or %20rab5c&ORG=Hs RAB4 http://www.ncbi.nlm.nih.gov/LocusLink/list.cgi?Q=rab4%20or%20rab4b&ORG= Hs Rab25 http://www.ncbi.nlm.nih.gov/LocusLink/LocRpt.cgi?l= 53868 Rab18 http://www.ncbi.nlm.nih.gov/LocusLink/LocRpt.cgi?l=22931 Rab20 http://www.ncbi.nlm.nih.gov/LocusLink/LocRpt.cgi?l=19332 Rab22 http://www.ncbi.nlm.nih.gov/LocusLink/LocRpt.cgi?l=19334 RAB27A http://www.ncbi.nlm.nih.gov/LocusLink/LocRpt.cgi?l=5873 Rab33b http://www.ncbi.nlm.nih.gov/LocusLink/LocRpt.cgi?l=19338

Trs 31 http://genome-www4.stanford.edu/cgi-bin/SGD/locus.pl?locus=trs31 Trs33 http://genome-www4.stanford.edu/cgi-bin/SGD/locus.pl?locus=trs33 Trs65 http://genome-www4.stanford.edu/cgi-bin/SGD/locus.pl?locus=trs65 Trs85 http://genome-www4.stanford.edu/cgi-bin/SGD/locus.pl?locus=trs85 Trs120 http://genome-www4.stanford.edu/cgi-bin/SGD/locus.pl?locus=trs120 Trs130 http://genome-www4.stanford.edu/cgi-bin/SGD/locus.pl?locus=trs130

HOPS complex intermediate page Vps11p http://genome-www4.stanford.edu/cgi-bin/SGD/locus.pl?locus=vps11 Vps16p http://genome-www4.stanford.edu/cgi-bin/SGD/locus.pl?locus=vps16 Vps18p http://genome-www4.stanford.edu/cgi-bin/SGD/locus.pl?locus=vps18

Zerial Lab home page http://www.mpi-cbg.de/content.php3?lang=en&aktID=zerial University of Ottawa Heart Institute http://www.ottawaheart.ca/hcwelcome.htm

ÂŠ 2001 Macmillan Magazines Ltd

REVIEWS

PRIONS: HEALTH SCARE AND BIOLOGICAL CHALLENGE Adriano Aguzzi, Fabio Montrasio and Pascal S. Kaeser Although human prion diseases are rare, the incidence of ‘new variant’ Creutzfeldt–Jakob disease in the United Kingdom is increasing exponentially. Given that this disease is probably the result of infection with bovine prions, understanding how prions replicate — and how to counteract their action — has become a central issue for public health. What are the links between the bovine and human prion diseases, and how do prions reach and damage the central nervous system? EPIZOOTIC

Refers to a disease that is temporally prevalent and widespread in a population of animals. PRION-ONLY HYPOTHESIS

States that the prion is devoid of informational nucleic acid and consists of protein (or glycoprotein) as its essential pathogenic component.

Institute of Neuropathology, University of Zurich, Schmelzbergstrasse 12, CH8091 Zurich, Switzerland Correspondence to A.A. e-mail: Adriano@pathol.unizh.ch

118

Twenty years after the inferred beginning1,2 of the bovine spongiform encephalopathy (BSE) epidemic, in-depth understanding of the effects of this EPIZOOTIC on human health has become more important than ever. The incidence of BSE in the British ‘national herd’ (the total cattle population in the United Kingdom) reached a peak in 1992 and has declined since2,3. The first histopathological confirmation of BSE was reported in November 1986 for a case that had occurred in April 1985. Contaminated meat-andbone meal (which had been used as a protein supplement in ruminant food) was soon recognized as the main mode of transmission of the disease, and this feeding practice was banned in 1989. Given the average incubation time of the disease — a matter of years — one could argue that the measures introduced were highly effective. However, several mathematical models proposed over the past years predicted that the prevalence of the disease would level off to zero around the turn of the century — a prediction that has, unfortunately, proved untrue. Projections were highest if the agent was assumed to transmit horizontally, and — counterintuitively — lowest if maternal transmission was assumed (FIG. 1). Switzerland has the dubious privilege of being the nation with the largest number of reported BSE cases after the United Kingdom (and, recently, Portugal and Ireland)4. Although the peak of the epidemic hit Switzerland some three years after it hit the United Kingdom, it has leveled off to relatively low but stable levels over the past 24 months. Of even more concern is

the fact that in Portugal, Ireland, France and, most recently, Germany, the number of BSE cases is actually increasing5 (and see link to the World Organization for Animal Health), and that BSE is now being seen in BARB (‘born after the real ban’) cows that were born after 1996 — when a very rigid ‘zero tolerance’ meatand-bone meal ban became effective. The molecular basis of prion diseases

To understand why this is happening, much research is going into the molecular events that underlie replication of the infectious agent — the prion. According to the terminology adopted here, the term ‘prion’ does not have structural implications, other than that a protein is an essential component. This can be defined6 as the agent that causes BSE, scrapie (a prion disease in sheep), Creutzfeldt–Jakob disease (CJD) in humans, and other transmissible spongiform encephalopathies, such as chronic wasting diseases of mule deer and elk, and other less common diseases that affect exotic ungulates such as kudu and nyalas, and captive large cats. This definition has been useful to foster understanding, but it says nothing about the true physical nature of the agent. A different definition that has become popular among yeast geneticists centres on the structural biology of prions. According to this second definition, prions are proteins that can exist in at least two conformations, one of which can induce the conversion of further prion molecules from one conformation into the other. So prion proteins can act as true genetic elements — even though they do not conwww.nature.com/reviews/molcellbio

| FEBRUARY 2001 | VOLUME 2

REVIEWS tain informational nucleic acids — in that they are self-perpetuating and heritable7. Eighteen years after Stanley Prusiner formulated his PRION-ONLY HYPOTHESIS (FIG. 2) — for which he was awarded the Nobel Prize in 1997 — there is still uncertainty as to whether these two definitions coincide in the case of mammalian prions, as it has not yet been 4,000

unequivocally established that the disease-associated protein isolated by Prusiner and termed PRP represents the infectious agent. Another problem is that, although all AMYLOID PROTEINS and their precursors would fit the second definition, these proteins do not seem to be transmissible or infectious in vivo or in cell culture. Although several yeast proteins have been shown to fulfil both of these criteria (BOX 1), no such successes have been reported for mammalian prions. SC

What is a prion?

Cases

3,000

2,000

1,000

Feed ban introduced First histopathological confirmation

0 1986

1988

1990

1992

1994

1996

1998

2000

Year

Figure 1 | Confirmed cases of BSE plotted by month and year of onset. Data valid to end of January 2000 (produced in June 2000). The good news is that the epizootic has been receding since 1992. The not-so-good news is that, despite several predictions, the incidence has not reached zero, and seems to be levelling off asymptotically at a low, but measurable, height. (Source: British Ministry of Agriculture, Fisheries and Food.)

a 'Refolding' model PrPC

Conversion prevented by energy barrier

How the prion damages its host Heterodimer

PrPSc

Homodimer

Amyloid; not essential for replication

b 'Seeding' model PrPC

Equilibrium between both forms

PrPSc

The normal mammalian prion protein is known as PRP . In vitro conversion of this protein can yield a moiety that has many of the physico-chemical properties that are characteristic of PrP Sc, the disease-associated prion protein. These include aggregation into higher-order quasi-crystalline complexes that are birefringent when observed under polarized light (especially when stained with amyloid dyes such as Congo red), formation of fibrils that are identifiable by electron microscopy and partial resistance to proteolytic enzymes, as identified by digestion with proteinase K8–10. The crucial element that is common to the two definitions mentioned above, and that is absolutely required for the classification of a protein as a prion, is transmissibility. None of the experimental procedures reported so far has unambiguously accomplished transformation of the cellular prion protein PrPC into a transmissible agent. Speculations abound as to why this has not been possible: the requirement for additional cellular factors distinct from PrPC, for example, has been invoked on the basis of genetic evidence11, but has never been proved. Universal consensus about the nature of the agent will predictably be reached only once a synthetic reconstitution has been done from non-infectious material.

Seed formation (very slow)

Recruitment Infectious of monomeric seed PrPSc (fast)

Amyloid

Fragmentation into several infectious seeds

Figure 2 | The ‘protein-only’ hypothesis and two popular models for the conformational conversion of PrPC into PrPSc. a | The ‘refolding’ or template-assistance model postulates an interaction between exogenously introduced PrPSc and endogenous PrPC, which is induced to transform itself into further PrPSc. A high-energy barrier might prevent spontaneous conversion of PrPC into PrPSc. b | The ‘seeding’ or nucleation–polymerization model proposes that PrPC and PrPSc are in a reversible thermodynamic equilibrium. Only if several monomeric PrPSc molecules are mounted into a highly ordered seed can further monomeric PrPSc be recruited and eventually aggregate to amyloid. Within such a crystal-like seed, PrPSc becomes stabilized. Fragmentation of PrPSc aggregates increases the number of nuclei, which can recruit further PrPSc and thus results in apparent replication of the agent. In sporadic prion disease, fluctuations in local PrPC concentration might — exceedingly rarely — trigger spontaneous seeding and self-propagating prion replication.

Notwithstanding all the unresolved problems, certain properties of the infectious agent can be studied. Perhaps the most obvious question is this: how do prions damage the central nervous system, which is the only compartment of the body that undergoes histopathologically (FIG. 3) and clinically detectable degeneration in prion diseases? Cellular models of prion disease might be useful to address this question, although prions replicate inefficiently in most established cell lines. Many studies have been done using a synthetic peptide from the central region of the PrPC molecule, which spontaneously assembles into amyloid-like structures. In vitro, this peptide can elicit reactions that resemble those seen in brain cells during the late stages of prion disease. These include activation of microglial cells, stimulation of the production of intermediate filaments by astrocytes, and even the death of neurons, all of which seem to depend on the presence of the normal prion protein in target cells12,13. But all of these studies suffer from a fundamental problem. It is not clear whether the phenomenon observed in conjunction with exposure of cells to this small, amyloidogenic peptide bear much relevance to what is happening in vivo during prion

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 2 | FEBRUARY 2001 | 1 1 9

REVIEWS

PRPSC OR PRP-RES

An ‘abnormal’ form of the mature Prnp gene product found in tissue of transmissable spongiform encephalopathy sufferers, operationally defined as being partly resistant to proteinase K digestion under defined reaction conditions. It is believed to differ from PrPC only (or mainly) conformationally, and is rich in β-sheet structure. Within the framework of the protein-only hypothesis it is often considered to be the transmissible agent or prion. AMYLOID PROTEINS

A term introduced by Rudolf Virchow a century ago to designate proteins that show birefringence under polarized light when stained with Congo red. A more modern concept defines amyloids as proteins that attain their energy minimum in a highly ordered, aggregated state. PRPC OR PRP-SEN

The naturally occurring form of the mature Prnp gene product.

Box 1 | Yeast prion studies Over the past several months, there have been breathtaking advances in the understanding of prion phenomena in yeast, and there is no doubt that at least two yeast proteins fulfil the molecular definition of a protein that can propagate its conformational state to its siblings in an ‘infectious’ manner. The ultimate experiment to show that a given protein is a prion is ‘in vitro conversion’: this term defines a cell-free manipulation by which the non-contagious conformation is transformed into a transmissible agent. Ideally, this manipulation should occur without participation of the pathological, transmissible prion, to exclude the possibility of cross-contamination. These conditions can be met in the case of the yeast prions identified so far, Sup35 (REFS 83,84) and Ure2p (REFS 85,86). The prion state (denominated ψ+) of the Sup35 protein brings the cell into a state of ‘translational infidelity’ that allows ribosomal read-through across stop codons. In the long run this cannot be a good thing for the cell, but a tantalizing conceptual development has just been proposed by Susan Lindquist87 — ψ+ might provide a crucial evolutionary buffer that allows cells to explore, for a limited period of time, a large variety of combinatorial mutational hits in the hope of finding traits that provide selective advantage in a fluctuating environment.

replication — a process that may be very different. Moreover, some of the published data have been challenged recently14. To ask whether cerebral accumulation of PrPSc is enough to damage nerve cells, PrP-deficient mice15 (which are resistant to scrapie16 and do not replicate prions17) have been grafted with brain cells derived from transgenic mice that overexpress PrPC. These mice were subsequently infected with prions18, and the pathology was confined to the regions of the brain that expressed PrPC. In the surrounding, PrP-deficient brain, no pathological changes were detected — even if there was H-E

GFAP

PrP

Control

CJD

Figure 3 | Neuropathological features of transmissible spongiform encephalopathies. Histological and immunohistochemical analysis of frontal cortex samples from the brain of a patient who died of non-cerebral causes (upper row) and of a patient suffering from CJD (lower row). Brain sections were stained with hematoxylin-eosin (H-E, left panels), with antibodies against glial fibrillary acidic protein (GFAP, middle panels) and with antibodies against the prion protein (PrP, right panels). Neuronal loss and prominent spongiosis are visible in the H-E stain. Strong proliferation of reactive astrocytes (gliosis) and perivacuolar prion protein deposits are detectable in the GFAP and PrP immunostains of the CJD brain samples.

120

substantial accumulation of pathological PrPSc (REF. 18). Although interpretation of this experiment requires certain caveats (notably, the possibility that a threshold concentration of PrPSc is needed for neurodegeneration, and that this level is not reached outside the grafted tissue), it is difficult to avoid the conclusion that the neuronal cytotoxicity of PrPSc depends on the expression of cellular PrPC by target cells. Why should that be? Perhaps PrPC acts as a receptor for PrPSc. There is some evidence that this might be the case19. Alternatively, the process of converting PrPC into PrPSc — rather than exposure to the disease-associated prion protein — might be the main deleterious event. This second possibility has been investigated in a series of elegant papers by Lingappa and co-workers. These authors have identified an atypical form of PrPC that undergoes a peculiar biogenesis. Most cellular PrPC is translocated into the lumen of the endoplasmic reticulum (ER) by virtue of its secretory signal peptide. The PrPC is then routed to the cell surface as a glycophosphatidylinositol-linked membrane-associated protein. But a small proportion of PrPC is made as a transmembrane form that later leaves the ER. Lingappa and coworkers have named these forms of PrPC according to their orientation20,21 — CtmPrP and NtmPrP (carboxyand amino-terminal transmembrane prion protein, respectively, FIG. 4). Lingappa and colleagues found that levels of CtmPrP correlate well with the neurodegenerative changes in pathological conditions — in fact, the correlation is much better than that with the accumulation of PrPSc itself22,23. These observations formed the basis for two hypotheses: that CtmPrP might be a marker of prion-induced neurodegeneration; or that the conversion of PrPC into PrPSc might trigger the formation of CtmPrP, which might, in turn, be an effector of neurotoxicity. Strong circumstantial evidence from transgenic mice and from patients with the hereditary prion disease, Gerstmann–Sträussler–Scheinker syndrome, favours the second hypothesis, although there is no information about putative signalling pathways that might be involved. The march of prions …

In most cases of prion infection in both humans and animals, the point of entry is outside the nervous system. In the case of BSE (and also, possibly, of new variant (nv) CJD, which develops owing to infection of humans with the BSE agent), exposure might be oral. By contrast, most cases of CJD brought about by medical procedures have occurred by parenteral administration (for example, intramuscular injection) of prion-contaminated growth hormone and gonadotropins of human cadaveric origin — in the age preceding recombinant DNA technology. But how do prions that are administered to the periphery of the body reach the central nervous system (CNS)? By analogy with viruses that affect the nervous system, there might be two main pathways of neural invasion. Many viruses — for example those that cause rabies and herpes — exploit the anatomical connections provided by peripheral nerves, and reach the CNS www.nature.com/reviews/molcellbio

| FEBRUARY 2001 | VOLUME 2

REVIEWS

PrPC

Mutation

PrPSc inoculation

PrPSc

Stage 1

Transmission Accumulation

Stage 2

secPrP NtmPrP

Nascent PrP Mutation

Stage 3

CtmPrP

Degradation?

Exit to post-endoplasmic reticulum compartment Neurodegeneration

PrPC

CtmPrP

PrPC mediating juxtaposition between transmembrane molecules A and B

CtmPrP disrupting juxtaposition

Signal for cell survival

Signal for cell survival missing

between transmembrane molecules A and B

Cell death

GERMINAL CENTRES

Specialized areas of lymphoid organs that are important for affinity maturation of B cells. PRION STRAINS

Prion strains have different phenotypes — for example, incubation times, distribution of lesions in the brain, relative abundance of mono-, di- and unglycosylated PrPSc, and electrophoretic mobility of the protease-resistant part of PrPSc. But they can all be propagated in the same inbred mouse strain, indicating that, within the framework of the proteinonly hypothesis, the same polypeptide chain can mediate different strain phenotypes.

through axonal transport. Human immunodeficiency virus (HIV), however, uses a different mechanism: it reaches cerebral microglial cells using a ‘Trojan horse’ mechanism that involves infection of macrophages. What about prions? Available evidence indicates that both pathways are involved: prions can colonize the immune system as well as lymphocytes24 and follicular dendritic cells25 (which are located in the GERMIC NAL CENTRES and express considerable amounts of PrP ) (FIG. 5). Blättler and colleagues have shown 18 that extracerebral prion protein is required for neural invasion: prion knockout (Prnp) mice that harbour a PrPC-expressing graft in their brains consistently develop spongiform encephalopathy (restricted to the graft) upon intracerebral inoculation with the infectious agent26. However, they do not develop such encephalopathies upon intra-ocular, intraperitoneal or even intravenous administration of the agent27. In the case of intra-ocular inoculation, neural invasion is impaired even when a specific transgenic manipulation prevents all antibodies against PrPC from being generated28. In other words, it is the absence of PrPC — rather than an immune response against prions — that prevents the spread of the infectious agent within the body of a PrPC-deficient mouse29. … from spleen to brain

In which cellular compartment must PrPC be expressed to support neural invasion? Reconstitution of the haematopoietic and lymphopoietic system with stem cells derived from wild-type or transgenic mice that overexpress PrPC is not enough to restore neural invasion30. These results imply that the crucial compartment is sessile, and that it cannot be transferred by

Figure 4 | Three-stage model of prion pathogenesis, and possible role of CtmPrP in cell death. a | Stage 1 represents the formation and accumulation of PrPSc, initiated by either inoculation or spontaneous conversion of a mutated PrPC to PrPSc. Stage 2 symbolizes the events involved in generating Ctm PrP, either by an unknown process that involves PrPSc (characterized by dashed lines) or by certain mutations within PrP. Two distinct forms of PrP can be made at the endoplasmic reticulum (ER): one that is fully translocated (secPrP) and one that is a transmembrane form. Digestion with proteases of the transmembrane form results in two fragments. One is derived from the carboxyl terminus and is glycosylated (designated C-transmembrane PrP; CtmPrP). Its carboxyl terminus is in the ER lumen and the amino terminus is accessible to proteases in the cytosol. The other fragment is derived from the amino terminus and is unglycosylated (termed N-transmembrane PrP; NtmPrP), and has the opposite conformation. Stage 3 depicts the hypothetical events involved in CtmPrP-mediated neurodegeneration, possibly involving the exit of CtmPrP to a post-ER compartment (adapted from REF. 29). b | Possible role of CtmPrP in cell death. Full-length PrP might act as a co-receptor on the cell surface, mediating the juxtaposition of two cell-surface transmembrane molecules A and B. This generates a signal for cell survival in the cytosol. Failure of CtmPrP to bind B could induce cell death by not facilitating the association of A to B. This mechanism also could explain effects of expression of an amino-terminally truncated PrP89, as well as the Doppel gene product90 (Figure adapted from REF. 88).

adoptive bone-marrow reconstitution27. There are at least two likely candidates for this compartment — the peripheral nerves31 and the follicular dendritic cells. However, titration experiments indicate that adoptive bone-marrow transfer reconstitutes the ability of the spleen to accumulate (and, perhaps, to replicate) prions of the Rocky mountain laboratory (RML) STRAIN after intraperitoneal inoculation27. This unexpected result might indicate that haematopoietic cells — perhaps lymphocytes — can replicate prions, or that they transport the infectious agent from the site of inoculation to the spleen. Our laboratory has repeated these experiments and confirmed their unambiguous reproducibility (P.S.K., Michael A. Klein, Petra Schwarz and A.A., unpublished observations). But Brown and colleagues reported32 that, using a different prion strain called ME7, they could not detect any accumulation of prions in the spleens of PrP C-knockout mice that had been reconstituted with PrPC-positive haematopoietic cells and killed at unspecified intervals through the incubation period. Assuming that the experimental design of the Blättler study using RML prions and the Brown experiments using ME7 is indeed comparable, the discrepancies between the outcomes might point to the possibility that different prion strains show a varying propensity to colonize specific components of the immune system. There might be precedents for this: BSE prions are hardly detectable in lymphoid organs of cows (with the possible exception of gutassociated lymphoid tissue over a transient period of time), whereas nvCJD prions extensively colonize human lymphoid organs. By identifying the molecular determinants of such differences in organ tropism,

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 2 | FEBRUARY 2001 | 1 2 1

REVIEWS

PrP-expressing neurons Autonomic nervous system?

Central nervous system · Spinal cord · Brain

Peripheral nervous system M cells? Macrophages? B cells? Dendritic cells?

Prions

PrP-expressing neurons Autonomic nervous system?

Blood stream?

Periphery · Intestine · Peritoneal cavity

Peripheral nervous system M cells? Macrophages? B cells? Dendritic cells?

B cells PrP-expressing FDCs Complement system?

Lymphoreticular system · Peyer's patches · Spleen · Lymph nodes

Figure 5 | Model of prion neuroinvasion in mice. After peripheral inoculation, prions colonize the lymphoreticular system. Possible elements that might contribute to the transport into the lymphoreticular system are M cells, macrophages, B cells and dendritic cells. Required elements of secondary lymphoid organs for prion replication are functional follicular dendritic cells (FDCs) and B cells. In the progress of the disease, prions find access to the central nervous system, probably through elements of the autonomic nervous system. Alternatively, direct invasion of the central nervous system by nerves might occur after peripheral infection. Central nervous system infection through the blood cannot be excluded, but is, in our opinion, unlikely.

we might learn more about the basic mechanism of prion pathogenesis. Anatomy of prion neuroinvasion

What are the cellular requirements for invasion of prions into lymphoid tissues? This question has been addressed by screening mouse strains with spontaneous and engineered deficiencies in various compartments of the immune system, and one clear-cut result has emerged: any genetic defect that impairs the terminal differentiation of B-cell precursors into antibody-secreting cells completely blocks the colonization of lymphoid organs by prions. Such defects also block the development of disease in the CNS upon peripheral inoculation33. This phenomenon is obviously due to a ?

? TNFα3

TNFR1 TNFα3

LTα3

B cell LTα1β2

Radioresistant stromal cell

LTα3

TNFR1

LTα1β2

Radioresistant stromal cell LTβR

LTβR

Unknown factors

Mature FDC FDC precusor

122

block of neuroinvasion, because B-cell-deficient mice show the same susceptibility to disease as wild-type mice when inoculated intracerebrally33. This dependence of neuroinvasion on B cells is absolute — we are not aware of any experimental design, mouse strain or prion strain in which the spread of prions from the periphery to the CNS is not impaired by ablation of B cells. The necessity for B cells does not imply that they are sufficient for neuroinvasion. However, attempts to identify other necessary compartments have yielded ambiguous results. One candidate is the follicular dendritic cell (FDC), which is the main site for accumulation of PrPSc in lymphoid organs25 and functions as an antibody-dependent antigen trap. But again, experimental evidence has not been completely conclusive. All published data indicate that, in intraperitoneally inoculated mice, prions can accumulate only in spleens with properly formed germinal centres and immunohistochemically identifiable FDCs. Nobody has recovered prions from the spleens of intraperitoneally inoculated mice deficient in tumour necrosis factor receptor 1 (TNFR1) (REF. 33), TNF-α32, or lymphotoxin-β34, none of which contain identifiable FDCs in their spleens. Moreover, administration of soluble lymphotoxin-β receptor efficiently prevents the build-up of a splenic prion burden in wild-type mice35. This result is also valid for the ME7 prion strain36, despite its many alleged differences from the RML strain. The molecular mechanisms by which FDCs capture prions are, of course, of immediate interest. Capture of conventional antigens by FDCs occurs through Fcγ receptors and complement receptors: there is reason to believe that the same systems are operational in prion capture when limiting amounts of infectivity are introduced37, but further molecules are likely to be involved in this process. On the other hand, neuroinvasion — the development of brain disease after peripheral challenge — is completely unaffected in TNFR1 (REF. 33) and lymphotoxin-β34 knockout mice, and cannot even be repressed fully by treatment with an antibody against the lym-

Figure 6 | Models of the signalling pathways required for the establishment and maintenance of functional follicular dendritic cells. For the formation of follicular dendritic cell (FDC) networks in the follicles of secondary lymphoid organs, both the tumour necrosis factor (TNF) and the lymphotoxin (LT)-β signalling pathways are necessary. However, maintenance of mature FDCs seems to depend solely on the continuous activation of the lymphotoxin-β pathway. Soluble and membrane-bound forms of TNFα3 homotrimers and LTα3 homotrimers, tethered to B cells, signal through the TNFR1, whereas the LTα1β2 heterotrimer signals through the LTβR. Follicular B cells provide the ligands, whereas expression of both TNFR1 (TNFRp55) and LTβR is required on radioresistant stromal cells. Two alternative FDC-maturation models are conceivable: FDC precursor cells are stromal radioresistant cells that differentiate into mature FDCs by activation of both TNFR1 and LTβR signal pathways; or radioresistant stromal cells, which differ from FDC precursors, are activated by TNFR1 and LTβR signalling pathways and produce molecules that stimulate the maturation of FDC precursors. (Figure adapted from REF. 91.) www.nature.com/reviews/molcellbio

| FEBRUARY 2001 | VOLUME 2

REVIEWS

Table 1 | Incidence of nvCJD since 1985* nvCJD‡

Total||

Year

Sporadic CJD

1985

1986

1987

1988

1989

1990

1991

1992

1993

1994

1995

1996

1997

1998

1999

2000

25¶

* In the United Kingdom. ‡ Including probable nvCJD cases awaiting autopsy results, or that are still alive. || Including hereditary and iatrogenic cases. ¶ Counted as of 3 January 2001.

photoxin-β receptor35,36. Therefore, although the lack of signalling from lymphotoxin-β to FDCs probably accounts for some of the protection from peripheral prion inoculation that is observed in B-cell-deficient mice, B cells probably have an additional role in prion neuroinvasion. This function is clearly independent of PrP expression38, and it must also be distinct from lymphotoxin-β/TNF signalling to FDCs (FIG. 6). The arm of neuroinvasion that takes prions from germinal centres to the central nervous system has been less intensely studied. Peripheral nerves, probably belonging to the autonomic nervous system, are likely to be important39–41, and lymphocytes might be involved in the migration of prions from FDCs to peripheral nerves. From bench to clinic

IATROGENIC

Caused by medical treatment. THALAMUS

A conglomerate of neuronal groups in the diencephalon.

How have all of these molecular studies helped us to understand transmissible spongiform encephalopathies? For example, as prions can be detected in lymphoreticular tissues (such as spleen, lymph nodes, tonsils and appendix) of patients with nvCJD, is there a risk of IATROGENIC transmission through exposure to blood or tissues from people with preclinical nvCJD, or from exposure to contaminated surgical instruments? Epidemiological surveys over the past two decades have not implicated blood transfusions or administration of blood products as risk factors for prion diseases. However, a small increase in relative risk for the ‘classical’, sporadically occurring disease (sCJD) is associated with surgery of all kinds42, and it might indeed indicate unrecognized iatrogenic transmission. In the case of nvCJD the situation might not be as simple — for one thing, we do not know as much about the epidemiology and iatrogenic transmissibility of this

new disease as we do about sCJD. New variant CJD was first described in 1996 (REF. 43), and has claimed almost 100 lives in the United Kingdom and in France so far44 (TABLE 1). Several striking characteristics of nvCJD set it apart from sCJD, which was described 80 years ago45,46 (TABLE 2). For one thing, sCJD typically affects elderly people, whereas nvCJD has mainly hit very young people, with a range of between 12 and 52 years of age. The reason for this age distribution remains unclear47. Also, the clinical courses of the two diseases are different. Whereas sCJD is typically a rapidly progressing illness, leading to severe dementia and then death within months or even weeks, nvCJD tends to develop over a more protracted period. Also, the initial symptoms of nvCJD are usually personality changes and psychosis, rather than dementia48. Even under the microscope, the two diseases are very different. One feature is common to sCJD and nvCJD: widespread vacuolation of the cortical neuropil (the meshwork of axons and dendrites) which, in its most extreme manifestation, makes the brain resemble a sponge under low-magnification microscopy — hence the designation ‘spongiform encephalopathy’. Another hallmark of nvCJD is the prominent accumulation of small spherical protein deposits termed plaques in the brains of affected people. Although some plaques can be seen in a minority of patients with sCJD, the plaques of nvCJD have a specific morphology that includes a characteristic rim of microvacuolated tissue49. Further peculiarities of nvCJD include severe destruction of neurons in the THALAMUS . The latter can be recognized by noninvasive neural imaging methods, such as the so-called pulvinar sign50 (a hyperintense area in the posterior thalamus that can be visualized by magnetic resonance imaging). In nvCJD there is generalized colonization of lymphoid organs by the infectious agent, and immunochemical methods (typically western blots) have shown deposition of disease-associated PrPSc in germinal centres of lymph nodes, tonsils and spleen51–53. The new disease has so far exclusively struck patients whose prion gene is homozygous for methionine at position 129. In our experience, this group represents only one-third of the population54 and around twothirds of patients with sCJD. For completeness, we Table 2 | Diagnostic criteria for nvCJD I

• Progressive neuropsychiatric disorder • Duration of illness > 6 months • Routine investigations do not suggest an alternative diagnosis

• No history of potential iatrogenic exposure II

• • • • •

III

• Electroencephalogram does not show

Early psychiatric symptoms Persistent painful sensory symptoms Ataxia Myoclonus or chorea or dystonia Dementia

the typical appearance of sporadic CJD (or no electroencephalogram done) • Bilateral pulvinar high signal on MRI scan IV

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

• Positive tonsil biopsy

VOLUME 2 | FEBRUARY 2001 | 1 2 3

REVIEWS

CORTICAL CEREBRAL RIBBON

The grey matter located underneath the surface of the brain. FLORID PLAQUES

Deposits of prion protein surrounded by a rim of vacuolated brain tissue.

should also mention that CJD is not a totally new phenomenon in very young people55. Over the past 20 years, CJD has been recorded in almost 100 children and teenagers. But in most of these cases, this was a result of documented iatrogenic exposure to the infectious agent. Typically, it came from growth hormone or pituitary gonadotropins from corpses, which were given to children to treat pituitary dwarfism and other conditions in the era before recombinant DNA technology. BSE: a human prion disease?

What is the evidence that the agent causing nvCJD might be identical with that of BSE when transmitted to humans? So far, none of the arguments is conclusive, but each — and particularly when they are all considered together — is tantalizing. Much effort has gone into characterizing the ‘strain properties’ of the agent that affects cows and humans. Because the molecular substrate that underlies the nature of prion strains (which are heritable phenotypic traits that can be reproduced upon serial passage through experimental animals) is not known, strain typing of prions has to rely on surrogate markers. Two such markers have been particularly useful. One is the distribution of neuronal vacuoles in the brains of affected animals: for example, whereas some strains target the CORTICAL CEREBRAL RIBBON, others mainly affect the midbrain56. The BSE prion strain attacks the dorsal medulla and the superior colliculus (a part of the optical pathway) virulently and consistently57. Worryingly, BSE prions (extracted from the brains of affected cows) and nvCJD prions derived from the brains of British patients produce the same lesional patterns when transmitted to panels of susceptible mice49,57,58. The second marker for strain typing of prions comes from analysing the biochemical properties of disease-associated PrP recovered from the brains of cattle and humans. Different steric conformations of PrP (which, according to the most popular hypothesis, account for the phenotypic strain properties) expose different sites to the action of proteolytic enzymes. These sites, in turn, can be identified by the different molecular weights of the resulting fragments. When used in conjunction with the ratio of diglycosylated to monoglycosylated PrP (the glycotype ratio) — another parameter that seems to correlate with strain properties — these traits were again found to be indistinguishable between BSE and human nvCJD prions59,60. A third line of argument that relates BSE and human nvCJD concerns epidemiology of the disease. So far, the total number of definite or probable cases of nvCJD is 82 in the United Kingdom, one in Ireland and two in France61. Assuming that the quality of the epidemiological surveillance is similar in these countries and in the rest of Europe (which has not reported cases of nvCJD), the unavoidable conclusion is that the incidence of nvCJD correlates with the prevalence of BSE. One of the most powerful arguments is the study of pathogenesis in primates. In a classical experiment, Lasmézas and colleagues62 inoculated brain extracts from BSE-affected cows into cynomolgus macaques.

124

After about three years, all inoculated primates (two adults and one infant) developed spongiform encephalopathy. The histopathological appearance of the disease was identical to that of nvCJD and included characteristic FLORID PLAQUES, which have been recognized in every case of nvCJD. Characteristically, nvCJD plaques are surrounded by a rim of microvacuolated brain tissue — a feature that they share with the deposits in scrapie. But this feature is not seen in classical CJD, nor in any other human spongiform encephalopathy. However, impressive as all of these arguments might seem, each is phenomenological rather than causal. Distribution of histopathological lesions, as well as morphology of plaque deposits, is downstream of the molecular events that are responsible for prion strain specificity. Measurements of the ‘glycotype ratio’ might be more directly related to the essence of strains, but there is still no way to tell whether they might simply be surrogate markers. It would be desirable to measure the conformation of disease-associated PrP more directly. Some inroads have been made with a method that exploits the relative affinities of antibodies against PrPC (REFS 63,64), but, to our knowledge, this possibility is restricted to differentiation of mouse and hamster PrPs, and has not yet been applied to investigating BSE and nvCJD. So how many nvCJD victims will there be in the future? Terrible as the disease has been for patients, we have not yet seen a large-scale epidemic. And, although many mathematical models have been generated65,66, the number of cases is still too small to predict future developments with any certainty. The number of cases diagnosed in the 12 months that preceded the writing of this article has risen to 36 (from 14 in the 12 months before) — certainly a cause for concern, if not for alarm. Another question relates to the possibility of chronic subclinical disease or a permanent ‘carrier’ status in cows as well as in humans. Evidence that such a carrier status might be produced by the passage of the infectious agent across species was first reported by Race and Chesebro67,68, and has recently been confirmed69 — at least for the passage between hamsters and mice. Immune deficiency can also lead to a similar situation in which prions replicate silently in the body, even when there is no species barrier70. So the problem of animal transmissible spongiform encephalopathies could be more widespread than is assumed, and might call for drastic measures in farming. Moreover, people carrying the infectious agent might transmit it horizontally71, and the risks associated with this possibility can be met only if we know more about how the agent is transmitted and how prions reach the brain from peripheral sites. Preventive measures

What can be done in the meantime to prevent the spread of the disease? As discussed above, nvCJD seems to be much more ‘lymphoinvasive’ than sCJD. In particular, nvCJD prions can be detected easily in lymphatic organs such as tonsils and the appendix51,52,72; this is also the case for scrapie73–75 but not for sCJD prions. Although prion infectivity of circulating lymphowww.nature.com/reviews/molcellbio

| FEBRUARY 2001 | VOLUME 2

REVIEWS cytes seems to be at least two orders of magnitude lower than that detected in splenic lymphocytes76, the possibility that circulating lymphocytes might be in equilibrium with their splenic siblings call for cautionary measures when dealing with blood products. But what should these be? Leukodepletion — a filtering process that aims to reduce the number of leukocytes in transfused blood units — has been advocated, but there is still no certainty about its efficacy. In addition, even if blood prion infectivity is initially contained in lymphocytes in vivo, lysis of cells might lead to contamination of blood units with infectious ‘microparticles’77, which might be difficult to remove by any method (short of ultracentrifugation). However, many of the virusremoval steps involved in the manufacture of stable blood products have some positive effects on prion removal, so the possibility of such contamination can be regarded as a worst-case scenario. A final consideration applies to secondary prophylaxis. Given the large amount of infectious BSE material that has entered the human food chain, it is possible that many people harbour preclinical nvCJD. Unfortunately, the distribution of preclinical disease in the United Kingdom and other countries is totally obscure. The only available information is a retrospective immunohistochemical analysis of British appen-

10.

11.

12.

13.

14.

Cohen, C. H. & Valleron, A. J. When did bovine spongiform encephalopathy (BSE) start? Implications on the prediction of a new variant of Creutzfeldt-Jakob disease (nvCJD) epidemic. Int. J. Epidemiol. 28, 526–531 (1999). Anderson, R. M. et al. Transmission dynamics and epidemiology of BSE in British cattle. Nature 382, 779–788 (1996). Ferguson, N. M., Donnelly, C. A., Woolhouse, M. E. & Anderson, R. M. Estimation of the basic reproduction number of BSE: the intensity of transmission in British cattle. Proc. R Soc. Lond. B 266, 23–32 (1999). Doherr, M. G., Heim, D., Vandevelde, M. & Fatzer, R. Modelling the expected numbers of preclinical and clinical cases of bovine spongiform encephalopathy in Switzerland. Vet. Rec. 145, 155–160 (1999). Donnelly, C. A., Santos, R., Ramos, M., Galo, A. & Simas, J. P. BSE in Portugal: anticipating the decline of an epidemic. J. Epidemiol. Biostat. 4, 277–283 (1999). Aguzzi, A. & Weissmann, C. Prion research: the next frontiers. Nature 389, 795–798 (1997). Most of the points that were unresolved when this review article was written continue to constitute the next frontiers. Serio, T. R. et al. Nucleated conformational conversion and the replication of conformational information by a prion determinant. Science 289, 1317–1321 (2000). Kocisko, D. A. et al. Cell-free formation of proteaseresistant prion protein. Nature 370, 471-474 (1994). A milestone discovery in mammalian prion biochemistry: the first demonstration that PrPC can be made protease-resistant by PrPSc in vitro. Bessen, R. A. et al. Non-genetic propagation of strainspecific properties of scrapie prion protein Nature 375, 698–700 (1995). Raymond, G. J. et al. Molecular assessment of the potential transmissibilities of BSE and scrapie to humans. Nature 388, 285–288 (1997). Telling, G. C. et al. Prion propagation in mice expressing human and chimeric PrP transgenes implicates the interaction of cellular PrP with another protein. Cell 83, 79–90 (1995). Peyrin, J. M. et al. Microglial cells respond to amyloidogenic PrP peptide by the production of inflammatory cytokines. Neuroreport 10, 723–729 (1999). Brown, D. R., Schmidt, B. & Kretzschmar, H. A. Role of microglia and host protein in neurotoxicity of a prion protein fragment. Nature 380, 345–347 (1996). Kunz, B., Sandmeier, E. & Christen, P. Neurotoxicity of

15.

16.

17.

18.

19.

20.

21.

22.

23.

24.

25.

26.

27.

dices and tonsils78 — a well-meant study, but of limited sensitivity. Most available assays for PrPSc are breathtakingly insensitive, although the identification of proteins that selectively bind PrPSc (but not PrPC) might herald some developments in this field79. Once subclinical carriers are identified, it will be imperative to develop strategies that will help to control spread of the agent and prevent the outbreak of symptoms in these people. Indeed, some promising approaches have been identified80,81. Possible targets for interfering with neural invasion are rate-limiting processes that control prion replication in the infected person. In light of the knowledge discussed above, treatments that target the neuroimmune interface of prion replication and neural invasion82 continue to be a promising area for research. Links DATABASE LINKS Creutzfeldt-Jakob disease | PRPC | Prp |

prion peptide 106–126 not confirmed. FEBS Lett. 458, 65–68 (1999). Büeler, H. R. et al. Normal development and behaviour of mice lacking the neuronal cell-surface PrP protein. Nature 356, 577–582 (1992). Büeler, H. R. et al. Mice devoid of PrP are resistant to scrapie. Cell 73, 1339–1347 (1993). A ‘citation classic’, whose title says it all. Sailer, A., Büeler, H., Fischer, M., Aguzzi, A. & Weissmann, C. No propagation of prions in mice devoid of PrP. Cell 77, 967–968 (1994). Brandner, S. et al. Normal host prion protein necessary for scrapie-induced neurotoxicity. Nature 379, 339–343 (1996). The normal prion protein is not only necessary for prion propagation, but also for induction of neurotoxicity. Horiuchi, M. & Caughey, B. Specific binding of normal prion protein to the scrapie form via a localized domain initiates its conversion to the protease-resistant state. EMBO J. 18, 3193–3203 (1999). Lopez, C. D., Yost, C. S., Prusiner, S. B., Myers, R. M. & Lingappa, V. R. Unusual topogenic sequence directs prion protein biogenesis. Science 248, 226–229 (1990). Yost, C. S., Lopez, C. D., Prusiner, S. B., Myers, R. M. & Lingappa, V. R. Non-hydrophobic extracytoplasmic determinant of stop transfer in the prion protein. Nature 343, 669–672 (1990). Hegde, R. S. et al. A transmembrane form of the prion protein in neurodegenerative disease. Science 279, 827–834 (1998). Hegde, R. S. et al. Transmissible and genetic prion diseases share a common pathway of neurodegeneration. Nature 402, 822–826 (1999). Cashman, N. R. et al. Cellular isoform of the scrapie agent protein participates in lymphocyte activation. Cell 61, 185–192 (1990). Kitamoto, T., Muramoto, T., Mohri, S., Doh ura, K. & Tateishi, J. Abnormal isoform of prion protein accumulates in follicular dendritic cells in mice with Creutzfeldt-Jakob disease. J. Virol. 65, 6292–6295 (1991). This paper showed that FDCs accumulate diseaseassociated prion protein, and kick-started the whole field of prion neural immunology. Brandner, S., Isenmann, S., Kuhne, G. & Aguzzi, A. Identification of the end stage of scrapie using infected neural grafts. Brain Pathol. 8, 19–27 (1998). Blättler, T. et al. PrP-expressing tissue required for transfer

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

28.

29. 30.

31.

32.

33. 34.

35.

36.

37. 38.

39.

40.

41.

42.

43.

of scrapie infectivity from spleen to brain. Nature 389, 69–73 (1997). Brandner, S. et al. Normal host prion protein (PrPC) is required for scrapie spread within the central nervous system. Proc. Natl Acad. Sci. USA 93, 13148–13151 (1996). Aguzzi, A. Neuro-immune connection in spread of prions in the body? Lancet 349, 742–743 (1997). Fischer, M. et al. Prion protein (PrP) with amino-proximal deletions restoring susceptibility of PrP knockout mice to scrapie. EMBO J. 15, 1255–1264 (1996). Race, R., Oldstone, M. & Chesebro, B. Entry versus blockade of brain infection following oral or intraperitoneal scrapie administration: role of prion protein expression in peripheral nerves and spleen. J. Virol. 74, 828–833 (2000). Brown, K. L. et al. Scrapie replication in lymphoid tissues depends on prion protein- expressing follicular dendritic cells. Nature Med. 5, 1308–1312 (1999). Klein, M. A. et al. A crucial role for B cells in neuroinvasive scrapie. Nature 390, 687–690 (1997). Manuelidis, L. et al. Follicular dendritic cells and dissemination of Creutzfeldt-Jakob disease. J. Virol. 74, 8614–8622 (2000). Montrasio, F. et al. Impaired prion replication in spleens of mice lacking functional follicular dendritic cells. Science 288, 1257–1259 (2000). Mabbott, N. A., Mackay, F., Minns, F. & Bruce, M. E. Temporary inactivation of follicular dendritic cells delays neuroinvasion of scrapie. Nature Med. 6, 719–720 (2000). Klein, M. A. et al. A role for complement in early prion pathogenesis. (submitted). Klein, M. A. et al. PrP expression in B lymphocytes is not required for prion neuroinvasion. Nature Med. 4, 1429–1433 (1998). Baldauf, E., Beekes, M. & Diringer, H. Evidence for an alternative direct route of access for the scrapie agent to the brain bypassing the spinal cord. J. Gen. Virol. 78, 1187–1197 (1997). Glatzel, M. et al. Adenoviral and adeno-associated viral transfer of genes to the peripheral nervous system. Proc. Natl Acad. Sci. USA 97, 442–447 (2000). Glatzel, M. & Aguzzi, A. PrPC expression in the peripheral nervous system is a determinant of prion neuroinvasion. J. Gen. Virol. 81, 2813–2821 (2000). Collins, S. et al. Surgical treatment and risk of sporadic Creutzfeldt-Jakob disease: a case-control study. Lancet 353, 693–697 (1999). Will, R. et al. A new variant of Creutzfeldt-Jakob disease in

VOLUME 2 | FEBRUARY 2001 | 1 2 5

REVIEWS

the UK. Lancet 347, 921–925 (1996). 44. Government report on the epidemiology of CJD. http://www.doh.gov.uk/cjd/stats/sept00. htm. (Department of Health, United Kingdom, 2000). 45. Creutzfeldt, H. G. Über eine eigenartige herdförmige Erkrankung des Zentralnervensystems. Z. ges. Neurol. Psychiatr. 57, 1–19 (1920). 46. Jakob, A. Über eigenartige Erkrankungen des Zentralnervensystems mit bemerkenswertem anatomischem Befunde. (Spastische PseudoskleroseEncephalomyelopathie mit disseminierten Degenerationsherden). Z. ges. Neurol. Psychiatr. 64, 147–228 (1921). 47. Politopoulou, G., Seebach, J. D., Schmugge, M., Schwarz, H. P. & Aguzzi, A. Age-related expression of the cellular prion protein in human peripheral blood leukocytes. Haematologica 85, 580–587 (2000). 48. Zeidler, M. et al. New variant Creutzfeldt-Jakob disease: psychiatric features. Lancet 350, 908–910 (1997). 49. Aguzzi, A. Between cows and monkeys. Nature 381, 734 (1996). 50. Zeidler, M. et al. The pulvinar sign on magnetic resonance imaging in variant Creutzfeldt-Jakob disease. Lancet 355, 1412–1418 (2000). 51. Hill, A. F., Zeidler, M., Ironside, J. & Collinge, J. Diagnosis of new variant Creutzfeldt-Jakob disease by tonsil biopsy. Lancet 349, 99–100 (1997). A short communication with far-reaching consequences for public health. 52. Hill, A. F. et al. Investigation of variant Creutzfeldt-Jakob disease and other human prion diseases with tonsil biopsy samples. Lancet 353, 183–189 (1999). 53. Collinge, J. Variant Creutzfeldt-Jakob disease. Lancet 354, 317–323 (1999). 54. Palmer, M. S., Dryden, A. J., Hughes, J. T. & Collinge, J. Homozygous prion protein genotype predisposes to sporadic Creutzfeldt-Jakob disease. Nature 352, 340–342 (1991). 55. Parchi, P. et al. Classification of sporadic CreutzfeldtJakob disease based on molecular and phenotypic analysis of 300 subjects. Ann. Neurol. 46, 224–233 (1999). 56. Fraser, H. & Dickinson, A. G. Distribution of experimentally induced scrapie lesions in the brain. Nature 216, 1310–1311 (1967). 57. Bruce, M. E. et al. Transmissions to mice indicate that ‘new variant’ CJD is caused by the BSE agent. Nature 389, 498–501 (1997). 58. Aguzzi, A. & Weissmann, C. Spongiform encephalopathies: a suspicious signature. Nature 383, 666–667 (1996). 59. Collinge, J., Sidle, K. C., Meads, J., Ironside, J. & Hill, A. F. Molecular analysis of prion strain variation and the aetiology of ‘new variant’ CJD. Nature 383, 685–690 (1996). 60. Hill, A. F. et al. The same prion strain causes vCJD and BSE. Nature 389, 448–450 (1997).

126

61. Chazot, G. et al. New variant of Creutzfeldt-Jakob disease in a 26-year-old French man. Lancet 347, 1181 (1996). 62. Lasmézas, C. I. et al. BSE Transmission to Macaques. Nature 381, 743–744 (1996). 63. Safar, J. et al. Eight prion strains have PrPSc molecules with different conformations. Nature Med. 4, 1157–1165 (1998). 64. Aguzzi, A. Protein conformation dictates prion strain. Nature Med. 4, 1125–1126 (1998). 65. Ghani, A. C., Donnelly, C. A., Ferguson, N. M. & Anderson, R. M. Assessment of the prevalence of vCJD through testing tonsils and appendices for abnormal prion protein. Proc. R Soc. Lond. B 267, 23–29 (2000). 66. Ghani, A. C., Ferguson, N. M., Donnelly, C. A., Hagenaars, T. J. & Anderson, R. M. Estimation of the number of people incubating variant CJD. Lancet 352, 1353–1354 (1998). 67. Race, R. & Chesebro, B. Scrapie infectivity found in resistant species. Nature 392, 770 (1998). 68. Aguzzi, A. & Weissmann, C. Spongiform encephalopathies. The prion’s perplexing persistence. Nature 392, 763–764 (1998). 69. Hill, A. F. et al. Species-barrier-independent prion replication in apparently resistant species. Proc. Natl Acad. Sci. USA 97, 10248–10253 (2000). 70. Frigg, R., Klein, M. A., Hegyi, I., Zinkernagel, R. M. & Aguzzi, A. Scrapie pathogenesis in subclinically infected Bcell-deficient mice. J. Virol. 73, 9584–9588 (1999). 71. Aguzzi, A. Prion diseases, blood and the immune system: concerns and reality. Haematologica 85, 3–10 (2000). 72. Hilton, D. A., Fathers, E., Edwards, P., Ironside, J. W. & Zajicek, J. Prion immunoreactivity in appendix before clinical onset of variant Creutzfeldt-Jakob disease. Lancet 352, 703–704 (1998). 73. Schreuder, B. E., van Keulen, L. J., Vromans, M. E., Langeveld, J. P. & Smits, M. A. Tonsillar biopsy and PrPSc detection in the preclinical diagnosis of scrapie. Vet. Rec. 142, 564–568 (1998). 74. Schreuder, B. E., van Keulen, L. J., Smits, M. A., Langeveld, J. P. & Stegeman, J. A. Control of scrapie eventually possible? Vet. Q. 19, 105–113 (1997). 75. Vankeulen, L. J. M. et al. Immunohistochemical detection of prion protein in lymphoid tissues of sheep with natural scrapie. J. Clin. Microbiol. 34, 1228–1231 (1996). 76. Raeber, A. J. et al. PrP-dependent association of prions with splenic but not circulating lymphocytes of scrapieinfected mice. EMBO J. 18, 2702–2706 (1999). 77. Gidon-Jeangirard, C. et al. Annexin V delays apoptosis while exerting an external constraint preventing the release of CD4+ and PrPC+ membrane particles in a human T lymphocyte model. J. Immunol. 162, 5712–5718 (1999). 78. Ironside, J. W. et al. Retrospective study of prion-protein accumulation in tonsil and appendix tissues. Lancet 355, 1693–1694 (2000). 79. Fischer, M. B., Roeckl, C., Parizek, P., Schwarz, H. P. & Aguzzi, A. Binding of disease-associated prion protein to

80.

81.

82.

83.

84.

85.

86.

87.

88. 89.

90.

91.

plasminogen. Nature 408, 479–483 (2000). Plasminogen seems to be the first endogenous protein that discriminates between normal and pathological prion proteins. Priola, S. A., Raines, A. & Caughey, W. S. Porphyrin and phthalocyanine antiscrapie compounds. Science 287, 1503–1506 (2000). Soto, C. et al. Reversion of prion protein conformational changes by synthetic β-sheet breaker peptides. Lancet 355, 192–197 (2000). Aguzzi, A. & Collinge, J. Post-exposure prophylaxis after accidental prion inoculation. Lancet 350, 1519–1520 (1997). Paushkin, S. V., Kushnirov, V. V., Smirnov, V. N. & TerAvanesyan, M. D. In vitro propagation of the prion-like state of yeast Sup35 protein. Science 277, 381–383 (1997). Sparrer, H. E., Santoso, A., Szoka, F. C., Jr & Weissman, J. S. Evidence for the prion hypothesis: induction of the yeast [PSI+] factor by in vitro- converted Sup35 protein. Science 289, 595–599 (2000). In this paper, yeast geneticists have achieved what continues to be a Holy Grail for mammalian prion science: de novo generation of recombinant, infectious prions. Taylor, K. L., Cheng, N., Williams, R. W., Steven, A. C. & Wickner, R. B. Prion domain initiation of amyloid formation in vitro from native Ure2p. Science 283, 1339–1343 (1999). Wickner, R. B. [URE3] as an altered URE2 protein: evidence for a prion analog in Saccharomyces cerevisiae Science 264, 566–569 (1994). The first suggestion that prions exist in yeast. True, H. L. & Lindquist, S. L. A yeast prion provides a mechanism for genetic variation and phenotypic diversity. Nature 407, 477–483 (2000). Hope, J. Spongiform encephalopathies. Breech–birth prions. Nature 402, 737–739 (1999). Shmerling, D. et al. Expression of amino-terminally truncated PrP in the mouse leading to ataxia and specific cerebellar lesions. Cell 93, 203–214 (1998). Moore, R. C. et al. Ataxia in prion protein (PrP)-deficient mice is associated with upregulation of the novel PrP–like protein doppel. J. Mol. Biol. 292, 797–817 (1999). Endras, R. et al. Mature follicular dendritic cell networks depend on expression of lymphotoxin β receptor by radioresistant stromal cells and of lymphotoxin β and tumor necrosis factor by B cells. J. Exp. Med. 189, 159–167 (1999).

Acknowledgments We thank B. Chesebro, V. Lingappa, and B. Caughey for critical reading. The work of our laboratory is supported by the Kanton of Zürich, the Bundesämter für Gesundheit, Veterinärwesen, Bildung und Wissenschaft, by grants of the Swiss Nationalfonds, and by the companies Baxter, Abbott, Migros and Coop.

www.nature.com/reviews/molcellbio

| FEBRUARY 2001 | VOLUME 2

REVIEWS

UNTANGLING THE ErbB SIGNALLING NETWORK Yosef Yarden* and Mark X. Sliwkowski‡ When epidermal growth factor and its relatives bind the ErbB family of receptors, they trigger a rich network of signalling pathways, culminating in responses ranging from cell division to death, motility to adhesion. The network is often dysregulated in cancer and lends credence to the mantra that molecular understanding yields clinical benefit: over 25,000 women with breast cancer have now been treated with trastuzumab (Herceptin®), a recombinant antibody designed to block the receptor ErbB2. Likewise, small-molecule enzyme inhibitors and monoclonal antibodies to ErbB1 are in advanced phases of clinical testing. What can this pathway teach us about translating basic science into clinical use? MESENCHYME

Immature connective tissue that consists of cells embedded in extracellular matrix. NEUREGULINS

EGF-like ligands whose primary receptor is ErbB3 and/or ErbB4. Four types of neuregulin are known. STROMA

Supporting connective tissue in which a glandular or other epithelium is embedded.

*Department of Biological Regulation, The Weizmann Institute of Science, Rehovot 76100, Israel. ‡Department of Molecular Oncology, Genentech Inc., South San Francisco, California 94080, USA. e-mails: yosef.yarden@ weizmann.ac.il; marks@gene.com. Correspondence to Y.Y.

ErbBs are typical receptor tyrosine kinases that were implicated in cancer in the early 1980s when the avian erythroblastosis tumour virus was found to encode an aberrant form of the human epidermal growth factor (EGF) receptor (also known as ErbB1, HER or EGFR). Since then, the ErbB family has grown to four, and we are beginning to appreciate that the normal function of ErbBs and their ligands is to mediate cell–cell interactions in organogenesis and adulthood (reviewed in REF. 1). In the epithelium, the basolateral location of ErbBs enables them to mediate signals between the MES2 ENCHYME and the epithelium for cell growth . The mesenchyme serves as a storehouse for many ligands including NEUREGULINS (NRGs), which bind ErbB3 and ErbB4. ErbB2 (also known as HER2) is a more potent oncoprotein than the other ErbBs, but no known ligand binds it with high affinity. It was first discovered as a rodent carcinogen-induced oncogene that encodes a variant of ErbB2 with a mutation that makes its tyrosine kinase constitutively active. ErbB2 is a shared coreceptor for several STROMAL ligands. Blocking the action of ErbB2 might thus inhibit a myriad of mitogenic pathways affecting ErbB-expressing tumour cells3. Although several strategies are being developed, Herceptin® — a HUMANIZED MONOCLONAL ANTIBODY to ErbB2 — has been the first to reach widespread clinical use, in particular for the treatment of metastatic breast cancer4,5.

A layered signalling network

The components of the ErbB signalling pathway are evolutionarily ancient (BOX 1), and at first glance resemble a simple growth factor signalling pathway: ligand binding to a monomeric receptor tyrosine kinase activates the cytoplasmic catalytic function by promoting receptor dimerization and self-phosphorylation on tyrosine residues. The latter serve as docking sites for various ADAPTOR PROTEINS or enzymes, which simultaneously initiate many signalling cascades to produce a physiological outcome (FIG. 1). In higher eukaryotes, the simple linear pathway has evolved into a richly interactive, multilayered network, in which combinatorial expression and activation of components permits context-specific biological responses throughout development and adulthood. The input layer. This comprises the ligands (EGF family of growth factors) and their receptors — the ErbBs (FIG. 1). All high-affinity ErbB ligands have an EGF-LIKE DOMAIN and three disulphide-bonded intramolecular loops. This receptor-binding domain is usually part of a large transmembrane precursor containing other structural motifs such as IMMUNOGLOBULIN-LIKE DOMAINS, heparin-binding sites and glycosylated linkers. Expression and processing of the precursor are highly regulated. For example, transformation by active Ras, or exposure to steroid hormones6 leads to increased expres-

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 2 | FEBRUARY 2001 | 1 2 7

REVIEWS

Box 1 | Evolution of the ErbB signalling network Both the nematode Caenorhabditis elegans and the fruitfly Drosophila melanogaster have primordial linear versions of the ErbB signalling pathway. In higher organisms, this has evolved into a complex network, probably because an interconnected layered structure can confer selective gains in terms of adaptation, tolerance to mutations and signal diversification91. The main functional features of the ErbB module were defined in invertebrates: ErbB regulates the fate of diverse cell lineages in different developmental stages through short-range paracrine interactions. C. elegans and Drosophila each contain a single ErbB homologue; however, the only EGF-like ligand of C. elegans, called Lin-3, is replaced by four ligands in Drosophila. Vulva development is a well-characterized function of the Lin-3 signalling pathway: the six vulva precursor cells (VPCs) respond to an inductive signal from a gonadal anchor cell, which is thought to secrete Lin-3. Lin-3 binds the juxtaposed receptor on one of the VPCs and instructs it to undergo several cell cycles and develop concomitantly a more differentiated phenotype. The Lin-3 pathway functions in other inductive morphogenic events; loss-of-function mutations in the receptor result not only in a vulvaless phenotype, but also in sterility, abnormal male tail development and death92. The Drosophila EGF receptor (DER) is used repeatedly in several stages of development, including oogenesis, embryogenesis, and wing and eye development. Likewise, differentiation of the DER-expressing tendon cell is regulated by the myotubederived NRG-like ligand, Vein93. Gurken, a homologue of transforming growth factor-α (TGF-α), functions primarily in the oocyte. Activation of another ligand, Spitz, which is anchored to the cell surface, requires proteolytic cleavage94. By contrast, Argos, a secreted DER ligand, is unique in that it negatively acts on receptor signalling95.

HUMANIZED MONOCLONAL ANTIBODY

An antibody, usually from a rodent, engineered to contain mainly human sequences. This process reduces the immune response to the antibody in humans. ADAPTOR PROTEINS

Proteins that augment cellular responses by recruiting other proteins to a complex. They usually contain several protein–protein interaction domains. EGF-LIKE DOMAIN

A motif with ~50 amino acids, including six cysteine residues and a mainly β-sheet structure, found in all ErbB-binding growth factors and in extracellular matrix proteins. IMMUNOGLOBULIN-LIKE DOMAIN

A protein domain composed of two β-pleated sheets held together by a disulphide bond. METALLOPROTEINASES

Proteinases that have a metal ion at their active sites.

128

sion of several ErbB ligands, and cleavage of ligand precursors by a METALLOPROTEINASE can be stimulated by activation of other receptors, such as G-protein-coupled receptors7 (FIG. 2). An important issue relates to the multiplicity and possible redundancy of ErbB ligands. This issue is particularly relevant to the many NRGs and their splice variants. Studies in cultured cells and initial attempts to address this issue in animals suggest that ErbB ligands have non-overlapping functions. For example, ligands such as EGF and NRG4, which bind to ErbB1 and ErbB4, respectively, have narrow specificity, whereas others such as epiregulin, NRG1β and betacellulin bind to two distinct primary receptors8. Overexpression of ErbB2, which biases heterodimer formation, can broaden ligand specificity (FIG. 1, dotted lines), and ligands that are better at recruiting this co-receptor can reduce the binding of less effective ligands. In addition, splice variants of NRGs and various ligand–receptor complexes also differ in their ability to recruit a partner receptor9–11, which affects their potency and kinetics of signalling. The four ErbBs share an overall structure of two cysteine-rich regions in their extracellular region, and a kinase domain flanked by a carboxy-terminal tail with tyrosine autophosphorylation sites. With few exceptions (for example, haematopoietic cells), ErbB proteins are expressed in cells of MESODERMAL and ECTODERMAL origins. Examination of the intracellular and extracellular domains of the ErbBs provides a satisfying explanation as to why a horizontal network of interactions is crucial to the ErbB signalling pathway: ErbB3 is devoid of intrinsic kinase activity12, whereas ErbB2 seems to have no direct ligand13. Therefore, in isolation neither ErbB2 nor ErbB3 can support linear signalling (FIG. 3). Most inter-receptor interactions are mediated by ligands, and

ErbB2-containing heterodimers are formed preferentially14,15. Nevertheless, overexpression of a specific receptor can bias dimer formation, especially in the case of ErbB2, whose homodimers can spontaneously form in ErbB2overexpressing cells. Many cancers of epithelial origin have an amplification of the ErbB2 gene, which pushes the equilibrium towards ErbB2 homodimer and heterodimer formation. By contrast, ErbB4, whose expression pattern is relatively limited, has several isoforms that differ in their juxtamembrane and carboxyl termini, resulting in differences in the recruitment of phosphatidylinositol-3-OH kinase (PI(3)K)16, which activates cell-survival pathways. Signal-processing layers. The specificity and potency of intracellular signals are determined by positive and negative effectors of ErbB proteins, as well as by the identity of the ligand, oligomer composition and specific structural determinants of the receptors. The main determinant, however, is the vast array of phosphotyrosine-binding proteins that associate with the tail of each ErbB molecule after engagement into dimeric complexes (FIG. 1). Which sites are autophosphorylated, and hence which signalling proteins are engaged, are determined by the identity of the ligand as well as by the heterodimer partner17. The Ras- and Shc-activated mitogen-activated protein kinase (MAPK) pathway is an invariable target of all ErbB ligands, and the PI(3)K-activated AKT PATHWAY and p70S6K/p85S6K pathway are downstream of most active ErbB dimers. The potency and kinetics of PI(3)K activation differ, however, probably because PI(3)K couples directly with ErbB3 and ErbB4, but indirectly with ErbB1 and ErbB2 (REF. 18). Simultaneous activation of linear cascades, such as the MAPK pathway, the STRESS-ACTIVATED PROTEIN KINASE cascade, protein kinase C (PKC) and the Akt pathway translates in the nucleus into distinct transcriptional programmes. These involve not only the proto-oncogenes fos, jun and myc, but also a family of zinc-fingercontaining transcription factors that includes Sp1 and Egr1, as well as Ets family members such as GA-binding protein (GABP)19. Despite sharing some pathways, each receptor is coupled with a distinct set of signalling proteins. For example, unlike ErbB1, the kinase-defective ErbB3 cannot interact with the adaptor protein and UBIQUITIN LIGASE c-Cbl, the adaptor protein Grb2, the second-messenger-generating enzyme phospholipase Cγ or the Ras-specific GTPase-activating protein (GAP)20, but it can associate with the adaptors Shc and Grb7 (FIG. 1).In addition to combinatorial interactions, an important determinant of signalling outcome is variation in the kinetics of specific pathways. The principal process that turns off signalling by the ErbB network is ligand-mediated receptor endocytosis, and the kinetics of this process also depend heavily on receptor composition (BOX 2). The output layer. The output of the ErbB network ranges from cell division and migration (both associated with tumorigenesis) to adhesion, differentiation and www.nature.com/reviews/molcellbio

| FEBRUARY 2001 | VOLUME 2

REVIEWS

a LPA, thrombin, ET, etc.

TGF-α (1)

EGF (1)

β-cellulin (1)

Epiregulin (1,4)

HB-EGF (1,4)

Amphiregulin (1)

NRG1 (3,4) α β

NRG2 (4) α β

NRG3 (4)

NRG4 (4)

Cytokines

Ligands

Input layer 1 1

Receptor dimers

b Src Cbl

PLCγ

PI(3)K

Shp2

Grb2

GAP Ras–GTP

Signal-processing layer

RAF MEK MAPK

Akt PKC

Bad

S6K

Sp1

Myc

Jun Fos

Elk

Jak

Shc

Ras–GDP

Nck

Vav

Crk

Grb7

Adaptors and enzymes

Rac

Sos PAK JNKK JNK

Egr1

Cascades

Abl

Transcription factors

Stat

c Output layer

Apoptosis

Migration

Growth

Adhesion

Differentiation

Figure 1 | The ErbB signalling network. a | Ligands and the ten dimeric receptor combinations comprise the input layer. Numbers in each ligand block indicate the respective high-affinity ErbB receptors8. For simplicity, specificities of receptor binding are shown only for epidermal growth factor (EGF) and neuregulin 4 (NRG4). ErbB2 binds no ligand with high affinity, and ErbB3 homodimers are catalytically inactive (crossed kinase domains). Trans-regulation by G-protein-coupled receptors (such as those for lysophosphatidic acid (LPA), thrombin and endothelin (ET)), and cytokine receptors is shown by wide arrows. b | Signalling to the adaptor/enzyme layer is shown only for two receptor dimers: the weakly mitogenic ErbB1 homodimer, and the relatively potent ErbB2–ErbB3 heterodimer. Only some of the pathways and transcription factors are represented in this layer. c | How they are translated to specific types of output is poorly understood at present. (Abl, a proto-oncogenic tyrosine kinase whose targets are poorly understood; Akt, a serine/threonine kinase that phosphorylates the anti-apoptotic protein Bad and the ribosomal S6 kinase (S6K); GAP, GTPase activating protein; HB-EGF, heparin-binding EGF; Jak, janus kinase; PKC, protein kinase C; PLCγ, phospholipase Cγ; Shp2, Src homology domain-2-containing protein tyrosine phosphatase 2; Stat, signal transducer and activator of transcription; RAF–MEK–MAPK and PAK–JNKK–JNK, two cascades of serine/threonine kinases that regulate the activity of a number of transcription factors.)

MESODERM

The middle germ layer of the developing embryo. It gives rise to the musculoskeletal, vascular and urinogenital systems, and to connective tissue (including that of the dermis). ECTODERM

The outermost germ layer of the developing embryo. It gives rise to the epidermis and the nerves. AKT PATHWAY

Akt (or protein kinase B) is a serine/threonine protein kinase activated by the phosphatidylinositol-3-OH kinase pathway that activates survival responses.

apoptosis (FIG. 1). Output depends on cellular context, as well as the specific ligand and ErbB dimer. This has been best shown in terms of mitogenic and transforming responses: homodimeric receptor combinations are less mitogenic and transforming than the corresponding heterodimeric combinations, and ErbB2-containing heterodimers are the most potent complexes21–23 (FIG. 3). Perhaps the best example of the ability of the ErbB module to tune mitogenic signalling is provided by the ErbB2–ErbB3 heterodimer: although neither ErbB2 nor ErbB3 alone can be activated by ligand, the heterodimer is the most transforming24,25 and mitogenic21 receptor complex. The ErbB2–ErbB3 heterodimer also increases cell motility on stimulation with a ligand26; but the other NRG receptor, ErbB4, which exists in several isoforms, has been associated with processes varying from cellular chemotaxis27 to proliferation and differentiation28.

A network of networks?

The ErbB network might integrate not only its own inputs but also heterologous signals, including hormones, neurotransmitters, lymphokines and stress inducers29 (FIG. 1). Many of these trans-regulatory interactions are mediated by protein kinases that directly phosphorylate ErbBs, thereby affecting their kinase activity or endocytic transport29. The most extensively studied mechanism involves activation of G-proteincoupled receptors (GPCRs) by agonists such as lysophosphatidic acid (LPA), carbachol (which specifically activates muscarinic acetylcholine receptors) or thrombin (FIG. 2). Experiments done with mutants and inhibitors of ErbBs imply that the mitogenic activity of some GPCR agonists requires transactivation of ErbB proteins. These agents increase tyrosine phosphorylation of ErbB1 and ErbB2, either by increasing their intrinsic

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 2 | FEBRUARY 2001 | 1 2 9

REVIEWS

Inactive ErbB

LPA, thrombin, MMP ET, etc +

α β γ

Steroid hormone

HB-EGF

G protein

+ + Ca2+

Pyk2

Src

Ras

MAPK Steroid hormone receptor Transcription

Figure 2 | Crosstalk between the ErbB network and other signalling pathways. G-protein-coupled receptors (GPCRs) such as those for lysophosphatidic acid (LPA), thrombin and endothelin (ET) can have positive effects on ErbB signalling through two mechanisms. First, through a poorly defined mechanism, they can activate matrix metalloproteinases (MMPs), which cleave membrane-tethered ErbB ligands (such as heparinbinding EGF-like factor, HB-EGF), thereby freeing them to bind to ErbBs. Second, GPCRs indirectly activate Src (perhaps via Pyk2), which phosphorylates the intracellular domains of ErbBs on tyrosine residues. Steroid hormones can have a positive effect on ErbB signalling by activating the transcription of genes encoding ErbB ligands. Finally, ErbB activation can activate a positive feedback loop through the Ras–MAPK (mitogen-activated protein kinase) pathway, which also activates transcription of ErbB ligand genes.

ErbB ligand gene

STRESS-ACTIVATED PROTEIN KINASES

Members of the mitogenactivated protein kinase (MAPK) family that respond to stress. They include the Jun amino-terminal kinases (JNKs) and the p38 MAPKs.

kinase activity30 or by inhibiting an associated phosphatase activity. Signalling events downstream of ErbB1 are subsequently triggered, and this might account for the mitogenic potential of the heterologous agonists. Apparently, a cascade of tyrosine kinases links GPCRs such as the LPA receptor or the β-adrenergic receptor to ErbB1 and subsequently to MAPK. The cascade culminates in the stimulation of Src family kinases31, which are recruited by either the calcium-regulated tyrosine kinase Pyk2 (REF. 32) or a GPCR-coupled kinase and an adaptor protein (for example, arrestin33). Another kinase that phosphorylates ErbB1 is the cytokine-regulated tyrosine kinase Jak2: on stimulation of adipocytes by growth hormone, Jak2 phosphorylates ErbB1, thus allowing MAPK activation even by a kinase-defective mutant of ErbB1 (REFS 34, 35). Yet another cytokine, interleukin-6, elevates tyrosine phosphorylation of ErbB2 by increasing its intrinsic catalytic activity36. By contrast, factors that activate PKC, such as certain growth factors and hormones (for example, PDGF, LPA and EGF by itself), increase threonine and serine phosphorylation of ErbB1 and ErbB2, which decrease tyrosine phosphorylation and ligand binding affinity through a mechanism involving accelerated recycling of internalized receptors (BOX 2). These interconnections to other signalling modules help to integrate and coordinate cellular responses to extracellular stimuli.

testinal tract. These processes are probably regulated by growth factors from the local mesenchyme. Mice lacking expression of transforming growth factor-α (TGFα) have abnormal skin, hair and eye development40,41 but, in contrast with ErbB1-deficient mice, which undergo massive apoptosis in cortical and thalamic NRG ErbB3

NRG ErbB4

Enzymes that catalyse the last stage of ubiquitylation, in which the small protein ubiquitin is transferred from a ubiquitin-conjugating enzyme (UBC or E2) to its target protein. They are also known as E3 enzymes. GAPS

Proteins that inactivate small GTP-binding proteins, such as Ras family members, by increasing their rate of GTP hydrolysis.

130

The ErbB network is a key developmental signalling pathway throughout evolution. Its functions in worm and fly development are now well understood (BOX 1), but recent research using knockout and transgenic mice is beginning to clarify the functions of individual ErbBs and specific ligands in mammalian development. ErbB1 and its ligands. Inactivation of ErbB1 impairs epithelial development in many organs, including those involved in tooth growth and eye opening37–39. Likewise, transgenic and in vitro studies implicate ErbB1 in promoting proliferation and differentiation of the epithelial component of skin, lung, pancreas and the gastroin-

ErbB4

Weak signalling

EGF

No signalling

ErbB2 overexpression

ErbB1

Weak signalling

EGF and NRG

· Slow ligand dissociation · Relaxed ligand specificity · Slow endocytosis · Rapid recycling · Prolonged firing

Integrating developmental cues UBIQUITIN LIGASES

ErbB3

ErbB1/3/4

ErbB2

· Increased cell proliferation · Increased cell migration · Resistance to apoptosis

Potent signalling

Figure 3 | Signalling by ErbB homodimers in comparison with ErbB2-containing heterodimers. Receptors are shown as two lobes connected by a transmembrane stretch. Binding of a ligand (EGF-like or NRG) to the extracellular lobe of ErbB1, ErbB3 (note inactive kinase, marked by a cross) or ErbB4 induces homodimer formation. When ErbB2 is overexpressed, heterodimers form preferentially. Unlike homodimers, which are either inactive (ErbB3 homodimers) or signal only weakly, ErbB2-containing heterodimers have attributes that prolong and enhance downstream signalling (green box) and their outputs (yellow box). Apparently, homodimers of ErbB2 are weaker signalling complexes than heterodimers containing ErbB2. (EGF, epidermal growth factor; NRG, neuregulin.)

www.nature.com/reviews/molcellbio

| FEBRUARY 2001 | VOLUME 2

REVIEWS

Box 2 | Turning off the ErbB response On ligand binding, ErbB1 molecules cluster over clathrin-coated regions of the plasma membrane, which invaginate to form endocytic vesicles. These mature to early and late endosomes, while gradually decreasing their internal pH and accumulating hydrolytic enzymes that lead to receptor degradation. Importantly, the other three ErbB proteins are endocytosis impaired and are more often recycled back to the cell surface21,96. Sorting to degradation is determined by the composition of the dimer: ErbB1 homodimers are targeted primarily to the lysosome; ErbB3 molecules are constitutively recycled97; and heterodimerization with ErbB2 decreases the rate of endocytosis and increases recycling of its partners98,99. Receptor internalization is determined by cytoplasmic motifs100, but sorting in the early endosome seems to depend on the differential dissociation of ligand–ErbB complexes at mildly acidic pH. Complex dissociation leads to recycling, whereas continuous activation of tyrosine phosphorylation in the endosome leads to recruitment of c-Cbl, a ubiquitin ligase that preferentially binds to ErbB1 homodimers101 and directs them to lysosomal degradation by tagging with polyubiquitin tracts102.

brain regions38, mice homozygous for a disrupted TGFα gene show no brain abnormalities. So, the limited penetrance of TGF-α mutations and the confinement of the phenotype to the skin and eye suggest that each ErbB ligand has a distinct functional role and tissue specificity, analogous to the different roles played by each of the Drosophila EGF receptor ligands in insect development (BOX 2). Neuregulins and their receptors. Like ErbB1 and its ligands involved in mesenchyme–epithelium interactions, the NRGs and their receptors are involved in the interaction between nerves and their target cells (for example, muscle, GLIA and SCHWANN CELLS), and are essential for cardiac and neural development. Mice defective in ErbB4, ErbB2 and NRG-1 die at embryonic day 10.5 from similar heart defects1. Endocardium-derived

Table 1 | Expression of ErbBs and their ligands in cancer Molecule

Nature of dysregulation

Type of cancer

Notes

References

Overexpression

Prostate

Expressed by stroma in early, androgen-dependent prostate cancer and by tumours in advanced, androgen-independent cancer

Overexpression

Pancreatic

Correlates with tumour size and decreased patient survival; may be due to overexpression of Ki-Ras, which also drives expression of HB-EGF and NRG1

Ligands TGF-α

108

Overexpression

Lung, ovary, colon

Correlates with poor prognosis when co-expressed with ErbB1

Overexpression

Mammary adenocarcinomas

Necessary, but not sufficient for tumorigenesis in animal models

109

Overexpression

Head and neck, breast, bladder, prostate, kidney, nonsmall-cell lung cancer

Significant indicator for recurrence in operable breast tumours; associated with shorter diseasefree and overall survival in advanced breast cancer; may serve as a prognostic marker for bladder, prostate, and non-small-cell lung cancers

110,111

Overexpression

Glioma

Amplification occurs in 40% of gliomas; overexpression correlates with higher grade and reduced survival

Mutation

Glioma, lung, ovary, breast

Deletion of part of the extracellular domain yields a constitutively active receptor

ErbB2

Overexpression

Breast, lung pancreas, colon oesophagus, endometrium, cervix

Overexpressed owing to gene amplification in 15–30% of invasive invasive ductal breast cancers Overexpression correlates with tumour size, spread of the tumour to lymph nodes, high grade, high percentage of S-phase cells, aneuploidy and lack of steroid hormone receptors

ErbB3

Expression

Breast, colon gastric, prostate, other carcinomas

Co-expression of ErbB2 with ErbB1 or ErbB3 in breast cancer improves predicting power

64,65

Overexpression

Oral squamous Overexpression correlates with lymph node cell cancer involvement and patient survival

112

Reduced expression

Breast, prostate Correlates with a differentiated phenotype

Expression

Childhood medulloblastoma

NRG1

Receptors ErbB1

GLIA

Supporting cells of the nervous system, including oligodendrocytes and astrocytes in the central nervous system, and Schwann cells in the peripheral nervous system. Glia surround neurons, providing mechanical and physical support, and electrical insulation between neurons. SCHWANN CELLS

Cells that produce myelin and ensheath axons in the peripheral nervous system.

ErbB4

Co-expression with ErbB2 has a prognostic value

(TGF-α, transforming growth factor-α; NRG1, neuregulin-1; HB-EGF, heparin-binding epidermal growth factor.)

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 2 | FEBRUARY 2001 | 1 3 1

REVIEWS NRG1 stimulates an ErbB2–ErbB4 heterodimer on adjacent myocytes to initiate formation of the TRABECULAE. Surprisingly, the immunoglobulin domain and the cytoplasmic part of NRG1 — regions that are not involved in receptor binding — are essential for proper heart development42,43. ErbB3-deficient mice survive to embryonic day 13.5 and suffer from defective cardiac formation44,45. The alternative NRG-promoted heterodimer, ErbB2–ErbB3, is involved in different morphogenic events: mice lacking ErbB2, ErbB3 or NRG1 have a severely underdeveloped SYMPATHETIC GANGLION chain. This is probably caused by defective migration of neural progenitors from the NEURAL CREST44. The Schwann cell lineage is also controlled by the ErbB2–ErbB3 heterodimer. In vitro studies showed that NRG1 biases differentiation of neural crest progenitors towards a glial fate, and ErbB3-deficient mice showed partial lack of Schwann cells along peripheral and sensory neurons45,46. The ability of NRGs to control transcription of several ion channels underlies involvement of ErbBs in the neuromuscular junction47. NRGs elevate

transcription of all subunits of the postsynaptic nicotinic acetylcholine receptor, but a nerve-derived splice variant seems to bias replacement of the γ-subunit with the ε-chain, which increases single-channel conductance. A similar subunit switch might occur at central synapses; NRG1β can markedly increase expression of the NR2C subunit of the N-methyl-D-aspartate receptor in slices of cerebellum48. The cancer connection

The potent cell proliferation signals generated by the ErbB network are used by cancer cells to fix oncogenic mutations by CLONAL EXPANSION. In addition, many types of oncogenic viruses exploit the ErbB network by manipulating its components (BOX 3). Human cancers use several mechanisms to activate the network at different layers. In many different cancer cell types, the ErbB pathway becomes hyperactivated by a range of mechanisms, including overproduction of ligands, overproduction of receptors, or constitutive activation of receptors (TABLE 1). It is extremely useful to know whether a

Box 3 | How do viruses harness the ErbB network?

TRABECULAE

Finger-like projections of cardiac muscle cells that form ridges in the ventricular wall. SYMPATHETIC GANGLIA

Clusters of sympathetic neurons in which a glandular or other epithelium is embedded. NEURAL CREST

A group of embryonic cells that separate from the embryonic neural plate and migrate, giving rise to the spinal and autonomic ganglia, peripheral glia, chromaffin cells, melanocytes and some haematopoietic cells. CLONAL EXPANSION

Growth of a population of cells from a single precursor cell.

132

Receptor Growth factor Several transforming and v-ErbB non-transforming viruses (AEV) constitutively elevate ErbB signalling by expressing an active component or by interfering with signalling Recycling P P P P v-Cbl shut-off. The hepatitis B virus (HBV), which is associated with hepatocellular v-Ras v-Crk carcinoma, upregulates Endosome v-Src transcription from the ErbB1 promoter103. Likewise, v-Raf c-Cbl v-Akt expression is deregulated by LMP1, a protein encoded by v-Jun the Epstein–Barr virus v-Fos (EBV), which is associated Poxviruses with several malignancies, H+ E5(HPV) Lysosome ATPincluding nasopharyngeal ase 104 carcinoma . Most members of the largest group of DNA viruses, poxviruses, encode EGF-like ligands, whose EBV,HBV expression at sites of infection significantly Receptor gene increases pathogenicity105. Retroviruses RNA tumour viruses present the most divergent strategy to Growth factor gene harness ErbB signalling: the avian erythroblastosis virus (AEV) encodes a truncated form of ErbB1 lacking most of the ectodomain and carrying many intracellular mutations. The oncoprotein v-ErbB forms ligand-independent covalent dimers at the cell surface106. Active mutants of various ErbB target proteins, including small GTP-binding proteins (v-Ras), adaptors (v-Crk), protein kinases (v-Src, v-Akt, vRaf) and transcription factors (v-Jun, v-Fos), are encoded by oncogenes of different strains of retroviruses. In addition, the mouse Cas NS-1 retrovirus, which induces pre-B cell lymphomas and myeloid leukaemia, encodes a dominant active form of c-Cbl, a ubiquitin ligase that targets ErbB proteins to lysosomal degradation102. This interferes with receptor ubiquitylation and degradation, similar to the effect of E5, a product of the human papilloma virus (HPV) that inhibits ErbB1 degradation through inhibition of an endosomal proton-ATPase107. Both E5 and vCbl increase the rate of receptor recycling back to the cell surface.

www.nature.com/reviews/molcellbio

| FEBRUARY 2001 | VOLUME 2

REVIEWS

Anti-ErbB antibody (e.g. Herceptin) –

Unfolded, inactive ErbB dimer

ErbB dimer

Ligand

– Tyrosine kinase inhibitors (e.g. tyrphostins)

– –

Phosphoproteins and downstream signalling events

Hsp90 inhibitors (e.g. geldanamycin)

Hsp90

– scFvs Immature ErbB

ER/Golgi

Ribozymes

Translation –

Transcription

Triplex-forming oligos, antisense oligos

ErbB gene

Figure 4 | Therapeutic strategies for blocking the ErbB signalling network. Anti-ErbB antibodies (such as Herceptin®, which binds ErbB2) block ligand binding and stimulate receptor internalization. Tyrosine kinase inhibitors such as tyrphostins block downstream signalling of the receptor–ligand complex, and Hsp90 inhibitors (for example, geldanamycin) prevent stabilization of ErbBs at the membrane. The active conformation of ErbB2 is maintained through interactions with a chaperone (Hsp90), and therefore chaperone antagonists inactivate the oncoprotein. It might also be possible to prevent ErbBs from reaching the cell surface, by blocking their transcription with triplex-forming oligonucleotides, their translation with antisense oligonucleotides or ribozymes, or their trafficking to the cell surface with intracellular single-chain Fv fragments of antibodies (scFvs). (ER, endoplasmic reticulum.) CARCINOMA

A malignant tumour of epithelial origin. PROGNOSIS

The likely outcome or course of a disease. ANDROGEN-DEPENDENT PROSTATE CANCER

An early form of prostate cancer that is responsive to androgens and anti-androgen therapy. AUTOCRINE

Activation of cellular receptors by ligands produced by the same cell. GENE AMPLIFICATION

A differential increase in a specific portion of the genome. Amplification is associated with neoplastic transformation and acquisition of drug resistance.

particular tumour has an overactive ErbB pathway because of mutation, overexpression or amplification of a component of the ErbB pathway, as it can tell us what the patient’s chance of survival is and with what drug they should be treated (FIG. 4). Ligands. The relationship between ErbB ligand expression and tumorigenicity is complex: growth factors can be induced secondarily by a primary oncogene; either the stroma or the tumour can act as a ligand source; or the ligand can be expressed but unprocessed or sequestered in an inactive form49. Of all the ErbB ligands, the relevance of TGF-α to human cancer is best characterized. TGF-α and ErbB1 are co-expressed in several types of CARCINOMAS50, and expression of TGF-α, particularly in lung, ovary and colon tumours co-expressing ErbB1, correlates with poor PROGNOSIS (reviewed in REF. 51). In prostate cancer, the pattern of expression of TGF-α seems to change as the disease progresses52. In early, ANDROGEN-DEPENDENT PROSTATE CANCER, TGF-α is expressed primarily in the

tumour stroma, which suggests paracrine signalling. In advanced, androgen-independent disease, TGF-α is expressed by the tumour cells themselves, indicating AUTOCRINE signalling. Less information is available on other ligands (TABLE 1). ErbB1. Both overexpression and structural alterations of ErbB1 are frequent in human malignancies. However, in vitro studies suggest that overexpression of the normal receptor leads to transformation only in the presence of a ligand. Accordingly, expression of EGF-like ligands often accompanies ErbB1 overexpression in primary tumours. Overexpression of ErbB1 is a very frequent genetic alteration in brain tumours; amplification of the gene occurs in 40% of gliomas53. Overexpression is associated with higher grade, higher proliferation and reduced survival. In a significant proportion of tumours, GENE AMPLIFICATION is accompanied by rearrangements. The most common mutation (type III) deletes part of the extracellular domain35, yielding a constitutively active receptor. Recent studies identified an identical alteration in carcinomas of the lung, ovary and breast, suggesting broader implications to human cancer54. ErbB2. Several types of cancers overexpress ErbB2 (reviewed in REF. 56). The association of ErbB2 expression with cancer is best studied in breast cancer, where protein is overexpressed owing to gene amplification in 15–30% of invasive DUCTAL BREAST CANCERS55. Overexpression correlates with tumour size, spread of the tumour to lymph nodes, high grade, high percentage of S-phase cells, ANEUPLOIDY and lack of steroid hormone receptors, implying that ErbB2 confers a strong proliferative advantage to tumour cells56,57. Paradoxically, a higher degree of ErbB2 overexpression is reported in early forms of breast cancer relative to more advanced invasive carcinomas, suggesting that alterations in ErbB2 alone are insufficient for breast tumour progression from a relatively benign to a more malignant phenotype56. The identification of ErbB2 amplifications by FLUORESCENCE IN SITU HYBRIDIZATION (FISH; FIG. 5 ) has now been approved by the US Food and Drug Administration to pinpoint patients at high risk for recurrence and disease-related death with node-negative invasive breast cancer56,58. Efforts are also being made to correlate ErbB2 status with predictive value — in other words, do patients with ErbB2 amplifications benefit from particular types of therapy? Again FISH technology can identify patients who might benefit from more aggressive therapy59. Several studies have shown that ErbB2 overexpression is associated with resistance to anti-oestrogen therapy60. Most ErbB2-overexpressing tumours do not express the oestrogen and progesterone receptors, indicating inverse relationships between the steroid hormone axis and the ErbB network. Clinically, this crosstalk might be critical: patients treated with an anti-oestrogen drug were found to have a worse outcome if their tumours overexpressed ErbB2

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 2 | FEBRUARY 2001 | 1 3 3

REVIEWS a

Figure 5 | Molecular diagnosis of breast cancer. a | Immunohistochemistry and b | fluorescence in situ hybridization (FISH) analysis of ErbB2 in human breast cancer. Immunohistochemistry was performed using HercepTest and FISH using a PathVysion ErbB2 DNA probe kit. The ErbB2 gene is seen as red fluorescence and the chromosome-17 centromeric α-satellite probe as green fluorescence. (Image courtesy of D. Eberhard, E. Huntzicker and B. Wright, Genentech, Inc.)

(REF. 61). On the one hand, in vitro studies indicate that overexpression of ErbB2 or NRG confers resistance to anti-oestrogens and renders cancer cells independent of oestrogen62. On the other hand, oestrogen suppresses transcription from the ErbB2 promoter, and specifically inhibits growth of ErbB2-overexpressing mammary cells63. Taken together, the molecular and clinical observations imply that the steroid and ErbB pathways are alternative, but functionally linked pathways that enhance cell proliferation (FIG. 2).

DUCTAL BREAST CANCER

Cancer arising from the lining of the milk ducts, as opposed to the lobules of the breast (lobular breast cancer). ANEUPLOIDY

An abnormal number of chromosomes caused by their inaccurate segregation during cell division. FLUORESCENCE IN SITU HYBRIDIZATION

Visualizing a genetic marker on a chromosome by using a fluorescently labelled polynucleotide probe that hybridizes to a gene on a chromosome during metaphase. FARNESYLTRANSFERASE INHIBITORS

Inhibitors that block the activity of Ras by preventing the addition of a farnesyl group that targets it to the plasma membrane. TYRPHOSTINS

A type of tyrosine kinase inhibitor.

134

Neuregulin receptors. The catalytically inactive member of the ErbB family, ErbB3, is expressed in several cancers, but there is no evidence for gene amplification and overexpression is limited. However, a large recent study found that co-expression of ErbB2 with ErbB1 or ErbB3 in oral squamous-cell carcinoma was significant and it critically improved the predicting power64, consistent with the non-autonomous role of ErbB3. Similarly, analysis of prostate cancer suggests the existence of a paracrine loop involving NRG1 and the ErbB2–ErbB3 heterodimer65. Some studies observed lower expression of ErbB4 in breast and prostate tumours relative to normal tissues, and an association with a relatively differentiated histological phenotype66. By contrast with epithelial tumours, childhood medulloblastomas often express ErbB4, whose co-expression with ErbB2 has a prognostic value67, in line with the importance of receptor heterodimerization. The network as a target for cancer therapy

The central role of the ErbB network in the development of solid tumours, its availability to extracellular manipulation, and detailed understanding of the underlying biochemistry have made the ErbB network an attractive target for pharmacological intervention (FIG. 4). Most efforts have concentrated on ErbB2 and ErbB1 owing to their increased expression in certain tumour cells relative to normal cells. Immunological strategies. One approach — a humanized antibody to ErbB2 called Herceptin® — has been

approved for clinical use, both alone and in combination with chemotherapeutic agents. In addition to downregulating surface ErbB2, Herceptin induces the cyclin-dependent kinase inhibitor p27Kip1 and the Rbrelated protein p130, which reduce the number of cells in S phase68. The recruitment and activation of immune effector cells to the ErbB2-overexpressing tumour might also contribute to Herceptin’s mechanism of action69. Alternative approaches to the use of naked monoclonal antibodies to ErbBs include making antibodies toxic to cancer cells by linking them to radionuclides, toxins or prodrugs. Active immunization with portions of ErbB2 is another promising approach70. Monoclonal antibodies directed to a mutant form of ErbB1 (EGFRvIII) found in gliomas and carcinomas inhibit brain tumours in a manner dependent on the Fc receptor71. Comparison of two tumour-inhibitory monoclonal antibodies to ErbB1 revealed that only one depends on immune mechanisms; the other acts primarily by altering receptor functions. The chimeric version of this antibody, C225, competes with ligand binding to ErbB1 and arrests cultured cells at G1 because of an elevation in p27KIP1 (REF. 72). This therapeutic antibody is now in late-stage clinical testing in patients with colorectal or head and neck cancers. Low molecular weight inhibitors. The discovery of naturally occurring compounds capable of inhibiting the ErbB network (for example, herbimycin, genistein and emodin) led to the synthesis of analogues specific to the nucleotide-binding sites of ErbB proteins or their putative chaperones, the 90-kDa heat-shock proteins (Hsp90)73. The chaperone might escort ErbB proteins from the endoplasmic reticulum to the plasma membrane, where it might stabilize the active conformation of the kinase. The crystal structures of related kinases were used to enhance selectivity of synthetic tyrosine kinase inhibitors to ErbBs73. Both reversible and irreversible inhibitors74 capable of discriminating between ErbBs and other kinases have been developed. When applied in vitro and in animal models, the compounds variably inhibited cell growth with some specificity for ErbB1- and ErbB2-expressing cells. At least five of these compounds are now being tested in human clinical studies. Because some studies indicated that Ras and Src are essential for transformation by ErbB proteins, FARNESYL TRANSFERASE INHIBITORS, Srcspecific TYRPHOSTINS, MAPK inhibitors and Akt inhibitors might also be therapetically effective in containing activated ErbB pathways75. Gene therapy. Strategies aimed at blocking transcription, translation or maturation of ErbB transcripts or proteins are candidates for gene therapy. Early studies have shown that the adenovirus type 5 early region 1A (E1A) gene product can block ErbB2 overexpression and suppress the tumorigenic potential of ErbB2-overexpressing ovarian cancer cells76. This method is now being tested in a phase I trial with ovarian cancer patients. Intracellular single chain antibodies (scFvs) directed to either ErbB1 or ErbB2 can effectively inhibit www.nature.com/reviews/molcellbio

| FEBRUARY 2001 | VOLUME 2

REVIEWS receptor transfer from the endoplasmic reticulum to the plasma membrane, and thereby reduce signalling77. A human protocol for the treatment of ErbB2-positive ovarian cancer with scFvs has been developed following demonstration of selectivity and phenotypic effects in vitro78. Triplex-forming oligonucleotides that bind to a purine-rich sequence in the ErbB2 promoter are potent and specific inhibitors of ErbB2 transcription in an in vitro assay79. Antisense oligonucleotides80, various dominant-negative mutants of ErbBs81 and specific ribozymes82 show specificity and efficacy in blocking receptor expression in cultured cells, and therefore might also prove useful as therapeutic lead compounds. Perspectives

Successful treatments have been or are being developed to target aberrant ErbB receptor signalling in cancer; however, the potential for exploiting this pathway is still in its infancy. Antagonizing ErbB signalling might be a useful strategy for treating proliferative diseases other than cancer. One such opportunity might be coronary atherosclerosis. The migration of vascular smooth muscle cells in the arterial intima contributes to this cardiovascular disorder, particularly restenosis. Activation of the thrombin receptor is required for smooth muscle cell migration and proliferation, and activation of this G-protein-coupled receptor depends on transactivation by ErbB1 in response to heparin-binding EGF. Blockade of ErbB1 activation might therefore aid in the treatment of this disorder83. Another opportunity for intervention by targeted ErbB therapy might be psoriasis84. In normal skin, ErbB1 expression is restricted to the basal layer whereas in psoriatic skin, ErbB1 and one of its ligands, amphiregulin, are highly expressed throughout the entire epidermal layer85. Inhibition of ErbB1 activation might help

Burden, S. & Yarden, Y. Neuregulins and their receptors: a versatile signalling module in organogenesis and oncogenesis. Neuron 18, 847–855 (1997). Borg, J. -P. et al. ERBIN: a basolateral PDZ protein that interacts with the mammalian ERBB2/HER2 receptor. Nature Cell Biol. 2, 407–414 (2000). Monilola, A. O., Neve, R. M., Lane, H. A. & Hynes, N. E. The ErbB signaling network: receptor heterodimerization in development and cancer. EMBO J. 19, 3159–3167 (2000). Baselga, J. et al. Phase II study of weekly intravenous recombinant humanized anti-p185HER2 monoclonal antibody in patients with HER2/neu-overexpressing metastatic breast cancer. J. Clin. Oncol. 14, 737–744 (1996). First full report of clinical tests of an anti-ErbB2 antibody as a single agent. Patients with metastatic breast cancer were intravenously treated with the recombinant drug. Toxicity was minimal and objective response was observed in several organs of a small group of patients. Cobleigh, M. A. et al. Multinational study of the efficacy and safety of humanized anti-HER2 monoclonal antibody in women who have HER2-overexpressing metastatic breast cancer that has progressed after chemotherapy for metastatic disease. J. Clin. Oncol. 17, 2639–2648 (1999). Dickson, R. B. & Lippman, M. E. Estrogenic regulation of growth and polypeptide growth factor secretion in human breast carcinoma. Endocrin. Rev. 8, 29–43 (1987). Prenzel, N. et al. EGF receptor transactivation by Gprotein-coupled receptors requires metalloproteinase cleavage of proHB-EGF. Nature 402, 884–888 (1999). Jones, J. T., Akita, R. W. & Sliwkowski, M. X. Binding

9. 10.

11.

12.

13.

14.

15.

control the spread or recurrence of psoriatic lesions. In contrast to inhibiting ErbB signalling, potential also exists for activating the pathway in clinically meaningful ways. For example, ErbB ligands might promote wound healing86. ErbB signalling is also involved in fetal lung development, and appropriate activation of these pathways might benefit premature infants87. Neuregulins, which are also known as glial growth factors, are potent mitogens for Schwann cells88. Activation of Schwann cells with NRG might help resolve peripheral nerve injuries or neuropathies89. In summary, the ErbB field has made significant strides since Stanley Cohen’s initial observation that EGF induces precocious eyelid opening in neonatal mice90. Although many of the individual molecules involved in ErbB signalling have been characterized, a full understanding of how the network functions in homeostasis — or malfunctions in a number of diseases — requires further definition. Regardless, the interface between basic and translational science has been established, and exploiting the ErbB pathway will probably yield other meaningful advances in the very near future. Links DATABASE LINKS ErbB1 | NRGs | ErbB3 | EGF | epiregulin | NRG1β | betacellulin | PI(3)K | Shc | p70S6K | PKC | Akt | Grb7 | fos | jun | myc | zinc finger | Sp1 | Egr1 | GABP | Grb2 | phospholipase Cγ | Src | Pyk2 | arrestin | Jak2 | interleukin-6 | TGF-α | Drosophila EGF receptor | NRG1 | Herceptin | p27Kip1 | Rb | p130 | C225 | Hsp90 | amphiregulin | Vein | Gurken | Spitz | Argos FURTHER INFORMATION The tumour gene database ENCYCLOPEDIA OF LIFE SCIENCES C. elegans vulval induction | Drosophila embryo: dorsal–ventral specification

specificities and affinities of EGF domains for ErbB receptors. FEBS Lett. 447, 227–231 (1999). Relative binding affinities of the EGF domains of 11 ErbB ligands were measured on 6 ErbB receptor combinations using soluble receptors. This format allowed precise determination of the effect of heterodimerization on ligand affinity and specificity. Tzahar, E. et al. Bivalence of EGF-like ligands drives the ErbB signalling network. EMBO J. 16, 4938–4950 (1997). Landgraf, R. & Eisenberg, D. Heregulin reverses the oligomerization of HER3. Biochemistry 39, 8503–8511 (2000). Ferguson, K. M., Darling, P. J., Mohan, M. J., Macatee, T. L. & Lemmon, M. A. Extracellular domains drive homo- but not heterodimerization of erbB receptors. EMBO J. 19, 4632–4643 (2000). Guy, P. M., Platko, J. V., Cantley, L. C., Cerione, R. A. & Carraway, K. L. Insect cell-expressed p180ErbB3 possesses an impaired tyrosine kinase activity. Proc. Natl Acad. Sci. USA 91, 8132–8136 (1994). Klapper, L. N. et al. The ErbB2/HER2 oncoprotein of human carcinomas may function solely as a shared coreceptor for multiple stroma-derived growth factors. Proc. Natl Acad. Sci. USA 96, 4995–5000 (1999). Tzahar, E. et al. A hierarchical network of interreceptor interactions determines signal transduction by Neu differentiation factor/neuregulin and epidermal growth factor. Mol. Cell. Biol. 16, 5276–5287 (1996). Graus Porta, D., Beerli, R. R., Daly, J. M. & Hynes, N. E. ErbB2, the preferred heterodimerization partner of all ErbB receptors, is a mediator of lateral signalling. EMBO J. 16, 1647–1655 (1997).

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

16. Elenius, K. et al. Characterization of a naturally occurring ErbB4 isoform that does not bind or activate phosphatidyl inositol 3-kinase. Oncogene 18, 2607–2615 (1999). 17. Olayioye, M. A. et al. ErbB1 and ErbB2 acquire distinct signalling properties dependent upon their dimerization partner. Mol. Cell. Biol. 18, 5042–5051 (1998). 18. Soltoff, S. P. & Cantley, L. C. p120cbl is a cytosolic adapter protein that associates with phosphoinositide 3-kinase in response to epidermal growth factor in PC12 and other cells. J. Biol. Chem. 271, 563–567 (1996). 19. Schaeffer, L., Duclert, N., Huchet Dymanus, M. & Changeux, J. P. Implication of a multisubunit Ets-related transcription factor in synaptic expression of the nicotinic acetylcholine receptor. EMBO J. 17, 3078–3090 (1998). 20. Fedi, P., Pierce, J., Di Fiore, P. P. & Kraus, M. H. Efficient coupling with phosphatidylinositol 3-kinase, but not phospholipase Cγ or GTPase-activating protein, distinguishes ErbB3 signalling from that of other ErbB/EGFR family members. Mol. Cell. Biol. 14, 492–500 (1994). 21. Pinkas-Kramarski, R. et al. Diversification of Neu differentiation factor and epidermal growth factor signalling by combinatorial receptor interactions. EMBO J. 15, 2452–2467 (1996). References 21 and 22 describe analyses of signal transduction by individual ErbB proteins and their combinations expressed in isolation in myeloid cells. This allows the pan-ErbB stimulatory effect of ErbB2 to be demonstrated. 22. Riese, D. J. II, van Raaij, T. M., Plowman, G. D., Andrews, G. C. & Stern, D. F. The cellular response to neuregulins is governed by complex interactions of the ErbB receptor

VOLUME 2 | FEBRUARY 2001 | 1 3 5

REVIEWS family. Mol. Cell. Biol. 15, 5770–5776 (1995). 23. Kokai, Y. et al. Synergistic interaction of p185c-neu and the EGF receptor leads to transformation of rodent fibroblasts. Cell 58, 287–292 (1989). 24. Alimandi, M. et al. Cooperative signalling of ErbB3 and ErbB2 in neoplastic transformation of human mammary carcinoma cells. Oncogene 15, 1813–1821 (1995). 25. Wallasch, C. et al. Heregulin-dependent regulation of HER2/neu oncogenic signalling by heterodimerization with HER3. EMBO J. 14, 4267–4275 (1995). 26. Chausovsky, A. et al. Molecular requirements for the effect of neuregulin on cell spreading, motility and colony organization. Oncogene 19, 878–888 (2000). 27. Kainulainen, V. et al. A natural ErbB4 isoform that does not activate phosphoinositide 3-kinase mediates proliferation but not survival or chemotaxis. J. Biol. Chem. 275, 8641–8649 (2000). 28. Vaskovsky, A., Lupowitz, Z., Erlich, S. & Pinkas-Kramarski, R. ErbB4 activation promotes neurite outgrowth in PC12 cells. J. Neurochem. 74, 979–987 (2000). 29. Carpenter, G. Employment of the epidermal growth factor receptor in growth factor-independent signalling pathways. J. Cell Biol. 146, 697–702 (1999). 30. Daub, H., Wallasch, C., Lankenau, A., Herrlich, A. & Ullrich, A. Signal characteristics of G protein-transactivated EGF receptor. EMBO J. 16, 7032–7044 (1997). Elucidation in cultured cells of crosstalk between Gprotein-coupled receptors and ErbB signalling. Ectopic expression of Gq- or Gi-coupled receptors revealed the essential function of ErbB1 in downstream signalling of these G proteins to mitogen-activated protein kinases. 31. Luttrell, L. M., Della Rocca, G. J., van Biesen, T., Luttrell, D. K. & Lefkowitz, R. J. Gβγ subunits mediate Srcdependent phosphorylation of the epidermal growth factor receptor. A scaffold for G protein-coupled receptormediated Ras activation. J. Biol. Chem. 272, 4637–4644 (1997). 32. Dikic, I., Tokiwa, G., Lev, S., Courtneidge, S. A. & Schlessinger, J. A role for Pyk2 and Src in linking Gprotein-coupled receptors with MAP kinase activation. Nature 383, 547–550 (1996). 33. Luttrell, L. M. et al. β-Arrestin-dependent formation of β2 adrenergic receptor–Src protein kinase complexes. Science 283, 655–661 (1999). 34. Yamauchi, T. et al. Tyrosine phosphorylation of the EGF receptor by the kinase Jak2 is induced by growth hormone. Nature 390, 91–96 (1997). 35. Wong, A. J. et al. Structural alterations of the epidermal growth factor receptor gene in human gliomas. Proc. Natl Acad. Sci. USA 89, 2965–2969 (1992). 36. Qiu, Y., Ravi, L. & Kung, H. J. Requirement of ErbB2 for signalling by interleukin-6 in prostate carcinoma cells. Nature 393, 83–85 (1998). 37. Miettinen, P. et al. Epithelial immaturity and multiorgan faliure in mice lacking epidermal growth factor receptor. Nature 376, 337–341 (1995). 38. Sibilia, M., Steinbach, J. P., Stingl, L., Aguzzi, A. & Wagner, E. F. A strain-independent postnatal neurodegeneration in mice lacking the EGF receptor. EMBO J. 17, 719–731 (1998). 39. Threadgill, D. W. et al. Targeted disruption of mouse EGF receptor: effect of genetic background on mutant phenotype. Science 269, 230–234 (1995). 40. Mann, G. et al. Mice with null mutations of the TGFα gene have abnormal skin architecture, wavy hair, and curly whiskers and often develop corneal inflammation. Cell 73, 249–261 (1993). 41. Luetteke, N. C. et al. TGFα deficiency results in hair follicles and eye abnormalities in targeted and Waved-1 mice. Cell 73, 263–278 (1993). 42. Liu, X. et al. Domain-specific gene disruption reveals critical regulation of neuregulin signalling by its cytoplasmic tail. Proc. Natl Acad. Sci. USA 95, 13024–13029 (1998). 43. Kramer, R. et al. Neuregulins with an Ig-like domain are essential for mouse myocardial and neuronal development. Proc. Natl Acad. Sci. USA 93, 4833–4838 (1996). 44. Britsch, S. et al. The ErbB2 and ErbB3 receptors and their ligand, neuregulin-1, are essential for development of the sympathetic nervous system. Genes Dev. 12, 1825–1836 (1998). By targeting ErbB2, ErbB3 and NRG1, the authors revealed that the ternary complex has an essential role in the developing sympathetic nervous system. Apparently, signalling by this complex drives migration of sympathetic cells from the neural crest. 45. Erickson, S. L. et al. ErbB3 is required for normal cerebellar and cardiac development: a comparison with ErbB2-and heregulin-deficient mice. Development 124, 4999–5011

136

(1997). 46. Riethmacher, D. et al. Severe neuropathies in mice with targeted mutations in the ErbB3 receptor. Nature 389, 725–730 (1997). 47. Fischbach, G. D. & Rosen, K. M. ARIA: a neuromuscular junction neuregulin. Annu. Rev. Neurosci. 20, 429–458 (1997). 48. Ozaki, M., Sasner, M., Yano, R., Lu, H. S. & Buonanno, A. Neuregulin-β induces expression of an NMDA-receptor subunit. Nature 390, 691–694 (1997). Synaptogenesis in the central nervous system involves marked changes in the composition of Nmethyl-D-aspartate receptors. This in vitro study implies that neuregulins regulate the composition of the neurotransmitter receptor in maturing synapses in the brain, in a manner analogous to the neuromuscular junction. 49. Dong, J. et al. Metalloprotease-mediated ligand release regulates autocrine signalling through the epidermal growth factor receptor. Proc. Natl Acad. Sci. USA 96, 6235–6240 (1999). 50. Modjtahedi, H. & Dean, C. The receptor for EGF and its ligands: Expression, prognostic value and target for therapy in cancer. Int. J. Oncol. 4, 277–296 (1994). 51. Salomon, D. S., Brandt, R., Ciardiello, F. & Normanno, N. Epidermal growth factor-related peptides and their receptors in human malignancies. Crit. Rev. Oncol. Hematol. 19, 183–232 (1995). 52. Scher, H. I. et al. Changing pattern of expression of the epidermal growth factor receptor and transforming growth factor-α in the progression of prostatic neoplasms. Clin. Cancer Res. 1, 545–550 (1995). 53. Wikstrand, C. J., Reist, C. J., Archer, G. E., Zalutsky, M. R. & Bigner, D. D. The class III variant of the epidermal growth factor receptor (EGFRvIII): characterization and utilization as an immunotherapeutic target. J. Neurovirol. 4, 148–158 (1998). 54. Moscatello, D. K. et al. Frequent expression of a mutant epidermal growth factor receptor in multiple human tumours. Cancer Res. 55, 5536–5539 (1995). 55. Slamon, D. J. et al. Human breast cancer: correlation of relapse and survival with amplification of the HER-2/neu oncogene. Science 235, 177–182 (1987). First demonstration of the prognostic value of ErbB2 amplification in breast cancer. Gene amplification was correlated with several disease parameters, and was a significant predictor of patient survival and time to relapse. 56. Ross, J. S. & Fletcher, J. A. The HER-2/neu oncogene in breast cancer: prognostic factor, predictive factor, and target for therapy. Stem Cells 16, 413–428 (1998). 57. Paik, S. & Liu, E. T. HER2 as a predictor of therapeutic response in breast cancer. Breast Dis. 11, 91–102 (2000). 58. Press, M. F. et al. HER-2/neu gene amplification characterized by fluorescence in situ hybridization: poor prognosis in node-negative breast carcinomas. J. Clin. Oncol. 15, 2894–2904 (1997). 59. Muss, H. B. c-erbB-2 expression and response to adjuvant therapy in women with node-positive early breast cancer. N. Engl. J. Med. 330, 1260–1266 (1994). First evidence that ErbB2 can predict therapeutic response in breast cancer. Using tissue blocks obtained from patients treated with adjuvant chemotherapy, the authors found that patients randomly assigned to a high-dose regimen of adjuvant chemotherapy had significantly longer disease-free and overall survival if their tumours overexpressed c-ErbB2. 60. Borg, A. et al. ERBB2 amplification is associated with tamoxifen resistance in steroid-receptor positive breast cancer. Cancer Lett. 81, 137–144 (1994). 61. Carlomagno, C. et al. c-erb B2 overexpression decreases the benefit of adjuvant tamoxifen in early-stage breast cancer without axillary lymph node metastases. J. Clin. Oncol. 14, 2702–2708 (1996). 62. Pietras, R. J. HER-2 tyrosine kinase pathway targets estrogen receptor and promotes hormone-independent growth in human breast cancer cells. Oncogene 10, 2435–2446 (1995). 63. Giani, C. et al. Increased expression of c-ErbB2 in hormone-dependent breast cancer cells inhibits cell growth and induces differentiation. Oncogene 17, 425–432 (1998). 64. Xia, W. et al. Combination of EGFR, HER-2/neu, and HER3 is a stronger predictor for the outcome of oral squamous cell carcinoma than any individual family members. Clin. Cancer Res. 5, 4164–4174 (1999). 65. Lyne, J. C. et al. Tissue expression of neu differentiation factor/heregulin and its receptor complex in prostate cancer and its biologic effects on prostate cancer cells in

vitro. Cancer J. Sci. Am. 3, 21–30 (1997). 66. Kew, T. Y. et al. c-ErbB4 protein expression in human breast cancer. Br. J. Cancer 82, 1163–1170 (2000). 67. Gilbertson, R. J., Perry, R. H., Kelly, P. J., Pearson, A. D. & Lunec, J. Prognostic significance of HER2 and HER4 coexpression in childhood medulloblastoma. Cancer Res. 57, 3272–3280 (1997). 68. Sliwkowski, M. X. et al. Nonclinical studies addressing the mechanism of action of trastuzumab (Herceptin). Semin. Oncol. 26, 60–70 (1999). 69. Clynes, R. A., Towers, T. L., Presta, L. G. & Ravetch, J. V. Inhibitory Fc receptors modulate in vivo cytotoxicity against tumor targets. Nature Med. 6, 443–446 (2000). 70. Disis, M. L. & Cheever, M. A. HER-2/neu protein: a target for antigen-specific immunotherapy of human cancer. Adv. Cancer Res. 71, 343–371 (1997). 71. Sampson, J. H. et al. Unarmed, tumour-specific monoclonal antibody effectively treats brain tumours. Proc. Natl Acad. Sci. USA 97, 7503–7508 (2000). 72. Wu, X. et al. Involvement of p27KIP1 in G1 arrest mediated by an anti-epidermal growth factor receptor monoclonal antibody. Oncogene 12, 1397–1403 (1996). 73. Levitzki, A. & Gazit, A. Tyrosine kinase inhibition: an approach to drug development. Science 267, 1782–1788 (1995). 74. Fry, D. W. et al. Specific, irreversible inactivation of the epidermal growth factor receptor and erbB2, by a new class of tyrosine kinase inhibitor. Proc. Natl Acad. Sci. USA 95, 12022–12027 (1998). 75. Brugge, J. S. New intracellular targets for drug design, Science 260, 918–919 (1993). 76. Chang, J. Y. et al. The tumour suppression activity of E1A in HER-2/neu-overexpressing breast cancer. Oncogene 14, 561–568 (1997). 77. Beerli, R. R., Wels, W. & Hynes, N. E. Intracellular expression of single chain antibodies reverts ErbB2 transformation. J. Biol. Chem. 269, 23931–23936 (1994). 78. Alvarez, R. D. & Curiel, D. T. A phase I study of recombinant adenovirus vector-mediated delivery of an anti-ErbB2 single chain (sFv) antibody gene for previously treated ovarian and extraovarian cancer patients. Hum. Gene Ther. 8, 229–242 (1997). 79. Ebbinghaus, S. W. et al. Triplex formation inhibits HER2/neu transcription in vitro. J. Clin. Invest. 92, 2433–2439 (1993). 80. Vaughn, J. P. et al. Antisense DNA downregulation of the ERBB2 oncogene measured by a flow cytometric assay. Proc. Natl Acad. Sci. USA 92, 8338–8342 (1995). 81. Hsieh, S. S. et al. ERBB-2 expression is rate-limiting for epidermal growth factor-mediated stimulation of ovarian cancer cell proliferation. Int. J. Cancer 86, 644–651 (2000). 82. Qian, X. et al. Kinase-deficient neu proteins suppress epidermal growth factor receptor function and abolish cell transformation. Oncogene 9, 1507–1514 (2000). 83. Kalmes, A., Vesti, B., Daum, G., Abraham, J. A. & Clowes, A. W. Heparin blockade of thrombin-induced smooth muscle cell migration involves inhibition of epidermal grwoth factor (EGF) receptor transactivation by heparinbinding EGF-like growth factor. Circulation Res. 87, 92–98 (2000). 84. Ben-Bassat, H. & Klein, B. Y. Inhibitors of tyrosine kinases in the treatment of psoriasis. Curr. Pharm. Des. 6, 933–942 (2000). 85. Nakata, A. et al. Localization of heparin-binding epidermal growth factor-like growth factor in human coronary arteries. Possible roles of HB-EGF in the formation of coronary atherosclerosis. Circulation 94, 2778–2786 (1996). 86. Wells, A. EGF receptor. Int. J. Biochem. Cell Biol. 31, 637–643 (1999). 87. Patel, N. V. et al. Neuregulin-1 and human epidermal growth factor receptors 2 and 3 play a role in human lung development in vitro. Am. J. Respir. Cell Mol. Biol. 22, 432–440 (2000). 88. Levi, A. D. et al. The influence of heregulins on human Schwann cell proliferation. J. Neurosci. 15, 1329–1340 (1995). 89. Levi, A. D. et al. The role of cultured Schwann cell grafts in the repair of gaps within the peripheral nervous system of primates. Exp. Neurol. 143, 25–36 (1997). 90. Cohen, S. & Carpenter, G. Human epidermal growth factor: isolation and chemical and biological properties. Proc. Natl Acad. Sci. USA 72, 1317–1321 (1975). 91. Bray, D. & Lay, S. Computer simulated evolution of a network of cell-signalling molecules. Biophys. J. 66, 972–977 (1994). 92. Aroian, R. V. & Sternberg, P. W. Multiple functions of let-23, a Caenorhabditis elegans receptor tyrosine kinase gene required for vulval induction. Genetics 128, 251–267 (1991).

www.nature.com/reviews/molcellbio

| FEBRUARY 2001 | VOLUME 2

REVIEWS 93. Volk, T. Signalling out Drosophila tendon cells. Trends Genet. 15, 448–453 (1999). 94. Schweitzer, R., Shaharabany, M., Seger, R. & Shilo, B.-Z. Secreted Spitz triggers the DER signalling pathway and is a limiting component in embryonic ventral ectoderm determination. Genes Dev. 10, 1518–1529 (1995). 95. Schweitzer, R. et al. Inhibition of Drosophila EGF receptor activation by the secreted protein Argos. Nature 376, 699–702 (1995). 96. Baulida, J., Kraus, M. H., Alimandi, M., Di Fiore, P. P. & Carpenter, G. All ErbB receptors other than the epidermal growth factor receptor are endocytosis impaired. J. Biol. Chem. 271, 5251–5257 (1996). Comparative analysis of ErbB proteins that shows that there are substantial differences in mechanisms of endocytosis-based attenuation. 97. Waterman, H., Sabanai, I., Geiger, B. & Yarden, Y. Alternative intracellular routing of ErbB receptors may determine signalling potency. J. Biol. Chem. 273, 13819–13827 (1998). 98. Worthylake, R. & Wiley, H. S. Structural aspects of the epidermal growth factor receptor required for transmodulation of ErbB2/neu. J. Biol. Chem. 272, 8594–8601 (1997). 99. Lenferink, A. E. et al. Differential endocytic routing of homo- and hetero-dimeric ErbB tyrosine kinases confers signalling superiority to receptor heterodimers. EMBO J. 17, 3385–3397 (1998). 100. Sorkin, A., Di Fiore, P. P. & Carpenter, G. The carboxyl

101.

102.

103.

104.

105.

106.

terminus of epidermal growth factor receptor/ErbB2 chimera is internalization impaired. Oncogene 8, 3021–3028 (1993). Muthuswamy, S. K., Gilman, M. & Brugge, J. Controlled dimerization of ErbB receptors provide evidence for differential signalling by homo- and heterodimers. Mol. Cell. Biol. 19, 6845–6857 (1999). Levkowitz, G. et al. c-Cbl/Sli-1 regulates endocytic sorting and ubiquitination of the epidermal growth factor receptor. Genes Dev. 12, 3663–3674 (1998). First report of the endocytic-sorting function of cCbl, a ubiquitin ligase that is a major substrate of ErbB1. Unlike sorting by c-Cbl to the late endosome, the oncogenic viral form, v-Cbl, shunts internalized receptors to the recycling pathway. Menzo, S. et al. Transactivation of epidermal growth factor receptor gene by the hepatitis B virus X-gene product. Virology 196, 878–882 (1993). Miller, W. E., Earp, H. S. & Raab-Traub, N. The EpsteinBarr virus latent membrane protein 1 induces expression of the epidermal growth factor receptor. J. Virol. 69, 4390–4398 (1995). Opgenorth, A., Nation, N., Graham, K. & McFadden, G. Transforming growth factor-α, Shope fibroma virus factor, and vaccinia growth factor can replace myxoma growth factor in the induction of myxomatosis in rabbits. Virology 192, 701–709 (1993). Adelsman, M. A., Huntley, B. K. & Maihle, N. J. Ligandindependent dimerization of oncogenic v-erbB products

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

107.

108.

109.

110.

111. 112.

involves covalent interactions. J. Virol. 70, 2533–2544 (1996). Straight, S. W., Herman, B. & McCance, D. J. The E5 oncoprotein of human papillomavirus type 16 inhibits the acidification of endosomes in human keratinocytes. J. Virol. 69, 3185–3192 (1995). Yamanaka, Y. et al. Co-expression of epidermal growth factor receptor and ligands in human pancreatic cancer is associated with enhanced tumour aggressiveness. Anticancer Res. 13, 565–569 (1993). Krane, I. M. & Leder, P. NDF/heregulin induces persistence of terminal end buds and adenocarcinomas in the mammary glands of transgenic mice. Oncogene 12, 1781–1788 (1996). Gorgoulis, V. et al. Expression of EGF, TGF-α and EGFR in squamous cell lung carcinomas. Anticancer Res. 12, 1183–1187 (1992). Irish, J. C. & Bernstein, A. Oncogenes in head and neck cancer. Laryngoscope 103, 42–52 (1993). Shintani, S., Funayama, T., Yoshihama, Y., Alcalde, R. E. & Matsumura, T. Prognostic significance of ERBB3 overexpression in oral squamous cell carcinoma. Cancer Lett. 95, 79–83 (1995).

Acknowledgements Y.Y. acknowledges support by the Israel Science Fund, the US Army Medical Research and Material Command and the M.D. Institute for Cancer Research.

VOLUME 2 | FEBRUARY 2001 | 1 3 7

REVIEWS

FILAMINS AS INTEGRATORS OF CELL MECHANICS AND SIGNALLING Thomas P. Stossel*, John Condeelis‡, Lynn Cooley§, John H. Hartwig*, Angelika Noegel||, Michael Schleicher¶ and Sandor S. Shapiro# Filamins are large actin-binding proteins that stabilize delicate three-dimensional actin webs and link them to cellular membranes. They integrate cellular architectural and signalling functions and are essential for fetal development and cell locomotion. Here, we describe the history, structure and function of this group of proteins.

PHAGOCYTOSIS

Actin-dependent process, by which cells engulf external particulate material by extension and fusion of pseudopods. OSMOTIC FLUID FLOW

The movement of fluid across semi-permeable membranes from lesser to greater solute concentrations.

*Brigham and Women’s Hospital, Boston, Massachusetts, 02115 USA. ‡Albert Einstein College of Medicine, New York, New York 10461, USA. §Yale Medical School, New Haven, Connecticut 06520, USA. ||University of Cologne, Cologne D-50931, Germany. ¶LudwigMaximilians University, Munich D-80336, Germany. #Jefferson Medical College, Philadelphia, Pennsylvania 19107, USA. Correspondence to T. P. S. e-mail: tstossel@rics.bwh. harvard.edu

138

Dynamic reorganization of the actin cytoskeleton transforms cell shapes and generates the forces necessary for cell locomotion, cell division, PHAGOCYTOSIS and a host of other cellular processes. Structures based on actin filaments range from parallel bundles to gel networks. The bundles provide tensile strength, serve as cables for strong contractile activity and as tracks for organelle transport. Three-dimensional orthogonal networks provide elasticity, accommodate internal diffusion of water, solutes and small organelles, resist weak OSMOTIC and HYDROSTATIC FLUID FLOWS and serve as scaffolds for weak contractions. They are barriers to spontaneous movement of large organelles and they localize signal transduction and other intracellular reactions. Actin-binding proteins orchestrate the engineering of the actin cytoskeleton in response to signalling cascades. These proteins have diverse functions — regulating the addition and loss of actin subunits to and from linear actin polymers, organizing actin polymers into and out of three-dimensional scaffolding and attaching actin filaments to cellular membranes. This review describes the history, structure and functions of one group of these proteins — the filamins.

large amounts of purified muscle actin to precipitate2. Antibodies recognizing this protein decorated actinrich stress fibres in chicken FIBROBLASTS. This appearance led to the naming of the antigen as ‘filamin’3,4. Such anti-filamin antibodies identified filamin-like proteins in diverse vertebrate cells, including muscle cells5,6. Independently, a search for actin crosslinking factors in cytoplasmic extracts of Dictyostelium discoideum led to the purification of a PARALOGUE of the filamins7,8, later also found in Entamoeba histolytica, the pathogen responsible for human amoebic dysentery9. The complementary DNA of this protein was the first filamin family member to be cloned10.Two filamin isoforms in Drosophila melanogaster are separate products of the cheerio gene, which was identified by studies of sterile fly mutants unable to construct normal OVARIAN 11,12 RING CANALS ; a second as yet unnamed gene is under investigation (D. Brown, personal communication). The fact that the discovery of different members of the filamin family arose through different approaches has led to a lack of consistency in the nomenclature. To address this problem, we propose here a uniform naming system (BOX 1; TABLE 1).

Discovery of the filamin protein family

Structure of the filamins

The purification of the first filamin was an accidental by-product of an attempt to isolate a Ca2+-sensitive myosin from rabbit MACROPHAGES1. Small quantities of this protein, then called ‘actin-binding protein’, caused

As filamins are extended ‘filamentous’ dimers when viewed by electron microscopy (FIG. 1),‘filamin’ (abbreviated FLN) is a reasonable name for this class of molecule, although filamins do not necessarily produce the www.nature.com/reviews/molcellbio

| FEBRUARY 2001 | VOLUME 2

REVIEWS

HYDROSTATIC FLUID FLOW

The movement of fluid under mechanical pressure in the direction of least resistance. MACROPHAGE

Any cell of the mononuclear phagocyte system that is characterized by its ability to phagocytose foreign particulate and colloidal material. FIBROBLAST

Box 1 | Proposed nomenclature for the filamin family The prefixes hs (Homo sapiens), gg (Gallus gallus), dm (Drosophila melanogaster), dd (Dictyostelium discoideum) denote the systematic name of the filamin-expressing species. The alphabetical listing of human filamins in TABLE 1 reflects the order in which complete sequence data entered the human genome database and the convention of using letters rather than numbers to describe them. An additional designation, h+ or h–, describes whether the proteins contain hinge 1. For example, a human FLNb that contains hinge 1 would be called hsFLNbh+. The Drosophila filamins are not orthologues of the human proteins, leading to the use of numbers rather than letters to specify the two recognized Drosophila filamin genes. A suffix designating the number of filamin repeats discriminates the two products of the Drosophila FLN1 gene.

Common cell type found in connective tissue in many parts of the body, which secretes an extracellular matrix rich in collagen and other macromolecules.

Table 1 | Proposed naming system

PARALOGUE

Gene products on opposite branches of a duplicated gene family. Orthologues are on the same branch (for example, FLNa, FLNb and FLNc). Paralogues and orthologues are homologues.

Proposed name

Previous names

Number of Expression repeats

Other information

hsFLNa

ABP, ABP280, FLN1, non-muscle FLN, αFLN

Most abundant and widely expressed variant in human tissues16,74,75

Contains hinge 2

hsFLNb

βFLN, FH1, FLN3

Broad distribution76–80, less abundant than hsFLNa

Contains hinge 2

ggFLNb

FLN

Retinal epithelium

>90% identity with hsFLNb81

hsFLNc

γFLN, ABP-L, FLN2

Predominant in muscle

~80 amino-acid insert in repeat 20 confers specific localization in developing muscle cells75,80,82–84 Contains hinge 2

ddFLN

ABP120, gelation factor

OVARIAN RING CANAL

In Drosophila, a single oocyte develops in egg chambers containing 15 nurse cells connected by intercellular bridges — the ring canals.

Also in E. histolytica

dmFLN1-20 Filamin-240, Filamin 1

Ovarian ring canals, follicle cell membranes

Amino terminus is 70% identical to hsFLN. Contains 2 hinges11,12

dmFLN1-9

Ovarian muscle sheath

No actin-binding domain Expressed from a different promoter than dmFLN1-20

Filamin 90

dmFLN2

b N

N β-sheet repeat 1

Actin-binding domain (α-actinin domain)

Rod domain 1

Hinge 1 Calpain cleavage sites

Rod domain 2

Hinge 2 C C Dimerization domain

200 nm

Figure 1 | The structure of human FLNa. a | The amino-terminal actin-binding domain contains sequence motifs found in many actin-filament-binding proteins15. The rest of the protein is made of 24 repeats of ~96 amino acids each. These probably fold into antiparallel βpleated sheet domains that overlap so as to generate a rod. Dimerization occurs through the twenty-fourth repeat. All vertebrate filamins have a stretch of 35 amino acids between repeats 23 and 24 designated ‘hinge 2’. Some filamins also have a second hinge designated ‘hinge 1’ between repeats 15 and 16. The subunits have straight segments corresponding to the βsheet rods interrupted by bends corresponding to sites susceptible to proteolysis by calpain68. b | Electron micrographs of tantalum–tungsten-cast rabbit macrophage FLNa dimers31,69.

Contains 2 hinges

actin filament bundles seen by light microscopy, as the name originally implied. Vertebrate and some Drosophila filamins are dimers with large (240–280 kDa) polypeptide chains that associate at their carboxyl termini11–14. The amino-terminal actin-binding domain of filamin subunits encompasses a stretch of 275 amino acids with sequence motifs first recognized in β-spectrin, dystrophin and α-actinin and found in other actin-filament-binding proteins, such as calponin and utrophin15. The rest of the polypeptide consists of 24 repeated sequences of ~96 amino acids, interrupted by one or two short ‘hinge’ segments. Secondary structure algorithm modelling of the repeats reveals that these segments form antiparallel β-sheet domains that overlap so as to generate a rod16. The minimal requirement for membership of the filamin family is a string of the characteristic repeated β-sheet units. The human filamin amino-acid sequences show about 60–80% overall sequence homology, with the greatest variation towards the carboxy-terminal self-association domains. But some members of the family are much more divergent: amoeba filamins and one Drosophila filamin have much shorter dimerization domains than vertebrate filamins (FIG. 2) and a Drosophila filamin isoform even lacks the amino-terminal actin-binding motif.

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 2 | FEBRUARY 2001 | 1 3 9

REVIEWS Vertebrate filamins

GEL

A liquid becomes a gel when a ‘giant molecule’ occupies the entire liquid.

dmFLN1-20 (Drosophila) 1

dmFLN1-9 (Drosophila)

POINTED END

Defined by arrowhead appearance of myosin head fragments bound to the actin filaments.

ddFLN (Dictyostelium) 6 5 4 3 2 1 1 2 3 4 5 6

LEADING EDGE

The thin margin of a lamellipodium spanning the area of the cell from the plasma membrane to about 1 µm back into the lamellipodium. PLATELETS

The smallest blood cells, which are important in haemostasis and blood coagulation.

Figure 2 | Schematic subunit structures of some filamins. The Dictyostelium filamin is shown in dimer form70,71. The crystal structure of rod repeats 5 and 6 of Dictyostelium filamin reveals that the two monomers overlap only within the sixth repeats, thus forming an antiparallel dimer. The sixth repeats from both subunits fold around each other to effect a very tight association. Rods 1–5 do not pair and allow high flexibility between the actin-binding domains72.

Constructing orthogonal actin networks in vitro

Around the time of the discovery of the first filamins, cell biologists began to examine a phenomenon in which initially soluble extracts of diverse cell types solidified, or ‘gelled’, as actin polymerized, and the GELS seemed to be more solid than equivalent concentrations of pure actin17. As macrophage FLNa so efficiently precipitates filamentous actin, it was a prime candidate to explain this actin gelation in unstirred extracts. Gelation is due to the crosslinking of actin filaments into orthogonal networks, and subsequent studies showed that FLNa is the most potent actin-filamentcrosslinking protein18,19. Indeed, only one FLNa dimer per actin filament is sufficient to induce gelation20. a

As most actin-filament-crosslinking proteins resemble the filamins in having two actin-filamentbinding sites of similar affinity21–23, a greater valence or binding affinity cannot explain the high actin gelling efficiency of FLNa relative to other actin crosslinkers. The answer lies in the geometry of actin-filament branching imposed by different actin-filamentcrosslinking molecules. At low molar ratios compared with actin, filamins produce an optically clear gel23,24. Electron micrographs of actin filaments crosslinked by FLNa25,26 or Dictyostelium filamins 27 in vitro show striking high-angle branching (FIG. 3a–c), and the interbranch distances are inversely proportional to the filamin concentration. At branching points, the POINTED ENDS of actin filaments intersect with the sides of other filaments to form T- or Y-shaped structures (FIG. 3b). The mechanism by which filamin promotes actinfilament branching is incompletely understood. What is clear is that filamin must dimerize28 and must have intact amino-terminal actin-binding domains29,30. The extended end-to-end length of the FLNa dimer in which sequential rigid rod domains are interspersed with flexible hinges has been proposed to make filamin behave like a ‘molecular leaf spring’, lending FLNa the mix of flexibility and stiffness required to tether large actin filaments and hold them in an improbable perpendicular arrangement16,31. The LEADING EDGE of motile cells contains threedimensional orthogonal networks of short filaments overlapping in X-, T- or Y-shaped junctions32. FLNa has been detected at such junctions in rabbit macrophage, human PLATELET and tumour cell cytoskeletons33–36. The inter-branch distances are shorter in platelet than in macrophage actin cytoskeletons, consistent with the higher FLNa content of platelets. One study that used different antibodies to those used in previous labelling experiments failed to localize filamin at filament junctions in the leading edge of a fibroblast cell line37. However, in many other studies, filamin molecules were found to localize uniformly throughout the actin

Figure 3 | Electron micrographs showing crosslinked complexes of actin filaments. a | Short filaments of rabbit skeletal muscle actin polymerized in the presence of rabbit macrophage FLNa and rotary shadowed with platinum25. b | Pieces of cortical actin filaments released from human platelet cytoskeletons by centrifugation73. The actin filaments on glass coverslips were reacted with myosin subfragment 1, leading to the appearance of twisted ropes. The narrow ends of the repeat structures on the cables point to the slow-growing (pointed) ends of actin filaments and reveal multiple filaments intersecting point-to-side with other filaments (arrows). c | Electron micrograph showing an orthogonal network of rabbit skeletal muscle actin polymerized in the presence of rabbit lung macrophage FLNa (molar ratio 25:1), rapidly frozen, and then critical-point dried and rotary shadowed with carbon platinum. d | Identically prepared electron micrograph of actin bundles, following polymerization with rabbit macrophage α-actinin19 at the same molar ratio of actin to crosslinker. The bars indicate 100 nm.

140

www.nature.com/reviews/molcellbio

| FEBRUARY 2001 | VOLUME 2

REVIEWS that has been observed in branched filaments made with filamin26,27. Third, the Arp2/3 complex increases the elasticity of actin filament solutions in vitro, but much higher Arp2/3 to actin molar ratios42 are necessary to produce elastic values similar to those generated by FLNa, at least under conditions examined so far43. Fourth, the linkages between actin filaments formed by the Arp2/3 complex are metastable and dissociate40, whereas interactions between FLNa and actin filaments are robust on a long timescale43,44. And last, although some actin filament–filament junctions in cells have the 70o angle associated with Arp2/3 complex-mediated branching, others do not, indicating that, after the Arp2/3 complex initiates actin-filament branching, filamins might stabilize the branches and weld spots where growing filaments overlap. The greater flexibility of filamin-mediated branching could be important in this three-dimensional interlocking process. Integrators of cell mechanics and signalling

Figure 4 | Electron micrograph of the actin cytoskeleton at the periphery of a human blood platelet. The cell was activated for 5 min, extracted with Triton X-100, fixed with 1% glutaraldehyde and then reacted sequentially with affinity-purified rabbit anti-FLNa peptide antibodies and with anti-rabbit immunoglobulin G complexed to 10 nm gold beads. The specimens were then rapidly frozen, critical-point dried, and cast with tantalum–tungsten35. The replica reveals extensive labelling of the cytoskeleton from the cell edge to deep within the cell interior, especially at filament–filament junctions.

cytoskeletal network, right up to the plasma membrane33–36 (FIG. 4). Do filamin and the Arp2/3 complex cooperate?

MELANOMA

Cancer derived from melanocytes, the cells that synthesize melanin pigments. RHO FAMILY GTPASES

Ras-related GTPases involved in controlling the polymerization of actin. DICTYOSTELIUM SLUGS

Aggregated form of Dictyostelium amoebae.

Filamins can orientate actin filaments perpendicularly as they polymerize spontaneously in vitro, but what happens in cells that tightly regulate actin polymerization? One candidate for working with filamins in establishing cortical actin architecture is the Arp2/3 complex, an assembly of seven polypeptides. The Arp2/3 complex nucleates branching actin filament growth in the barbed-end direction off a pre-existing actin filament at nearly fixed angles of 70o in vitro39–41. The belief has spread that Arp2/3 is the most important actin-filament-crosslinking factor in cortical cytoplasm37. However, there are several arguments that the Arp2/3 complex synergizes with other actin-filament-crosslinking proteins such as filamin. First, and most importantly, the Arp2/3 complex is abundant in filamin-deficient MELANOMA cells (described below), but evidently is unable to stabilize their membranes or support locomotion. Second, polymer branching per se is not sufficient for polymer gelation. Electron micrographs of branched filaments made with Arp2/3 show brush-like structures but not the closing of loops39,41 that is necessary to form three-dimensional gels, and

Filamins bind various other macromolecules in addition to filamentous actin. So far, 20 binding partners have been found, and more are expected. Most of the interactions occur through the carboxy-terminal end of filamins. Although the physiological significance of some of the interactions listed in TABLE 2 is unknown, one obvious theme is that the interwebbing of actin scaffolds with membrane receptors is a way to provide mechanical stability to the cell membrane and to maintain cell–cell and cell–matrix connections. By bringing together receptors such as β-integrins, the submembrane actin network and intracellular signalling components, filamins can facilitate the activation of local cellular processes, in particular those involving actin polymerization (FIG. 5). A good example of the potential of such macromolecular cooperation is the fact that FLNa binds to RHO FAMILY GTPASES and to some of their regulatory cofactors that are strongly implicated in the regulation of actin assembly and myosin activation45. Ral binds FLNa in a GTP-dependent manner, whereas Cdc42, RhoA and Rac1 bind FLNa constitutively46,47. Trio, a guanine nucleotide exchange factor for Rho GTPases, binds FLNa, implying that switching on and off Rho GTPases constitutively bound to filamins might regulate the spatial positioning of actin assembly48. So by interacting directly with the Rho signalling apparatus, filamins can organize actin filaments that elongate under the influence of these GTPases into three-dimensional configurations useful to the cell. A growing body of evidence points towards the involvement of filamins in signal transduction. For example, DICTYOSTELIUM SLUGS that lack filamin have markedly impaired photo- and thermosensory responses, indicating that filamin has an adaptor role in these processes49. In mammalian fibroblasts, FLNa transduces stress signals to the actin cytoskeleton. In cells subjected to shear stress, β1 integrin directly associates with filamin, signalling the stressed cell to stiffen to render it resistant to subsequent strains. Cells lacking FLNa do not show a stiffening response50. This so-called

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 2 | FEBRUARY 2001 | 1 4 1

REVIEWS

CALPAIN

Calcium-dependent cysteine proteases involved in signal transduction in a variety of cellular processes.

mechanoprotection function of filamins might be especially important in muscle tissues repeatedly subjected to powerful contractile forces. Regulation of filamins

Although abundant evidence links filamins to cell signalling pathways, it is unclear how the signalling reactions affect filamin function. Several serine/threonine

protein kinases phosphorylate filamins, including protein kinase A, protein kinase C, Ca2+/calmodulindependent protein kinase II and p90 ribosomal S6 kinase51–55. Phosphorylation of FLNa by protein kinase A increases its resistance to CALPAIN cleavage52 and phosphorylation of chicken gizzard filamin by Ca2+/calmodulin-dependent protein kinase II decreases its actin-binding affinity56. But the physio-

Table 2 | Filamin-binding partners Partners

Method used

GpIb/V/IX (Von Direct binding Willebrand receptor) complex

Binding site on filamin*

Functional significance of the association

Other information

Repeat 17–19 (FLNa,b)

Promotes cell spreading78,85

Filamin binds to the GpIbα subunit

β1A, β1D, β2, Direct binding Carboxy-terminal Mechanoprotection86 β3, β7 integrins Yeast two-hybrid half

Filamin binds to the β-chain

Tissue factor

Direct binding

Repeats 23–24

Co-association in vivo87

FcR1 (CD64)

Direct binding

Ligand-sensitive dissociation88

Furin

Direct binding Repeats 13–14 Yeast two-hybrid

Directs furin from early endosomes to Golgi89 Promotes furin internalization

δ-Sarcoglycan Yeast two-hybrid Repeats 23–24 Direct binding (FLNc)

Possible role in limb girdle muscular dystrophy90

Myotilin

Yeast two-hybrid Repeats 19–21 (FLNc)

Possible role in limb girdle muscular dystrophy84

Caveolin-1

Yeast two-hybrid Carboxy-terminal ? Direct binding half

Co-localization in caveolar structures91

Presenilins

Yeast two-hybrid Carboxyl terminus FLN1 overexpression inhibits Direct binding (FLNa,b, presenilin overexpression Dm FLN1) phenotype in Drosophila76,92,93

Filamin binds to the hydrophilic loop Co-localization in Tlymphocytes, Cos-1 cells and Alzheimer disease plaques

Dopamine D2 receptor

Yeast two-hybrid Repeats 16–19 Direct binding

Participates in response to dopamine94

Filamin binds to third loop

Granzyme B

Yeast two-hybrid Repeat 24 Direct binding

Participates in granzyme Bmediated apoptosis95

Hydrolysis of FLNa by granzyme B

Toll

Yeast two-hybrid Repeats 21–24 (dmFLN1)

FLN1 also binds Tube96

TRAF2

Yeast two-hybrid Repeats 21–24

Involved in SAPK or NF-κB activation by TRAF2 or TNF97

SEK-1

Yeast two-hybrid Repeats 21–24 Direct binding

Involved in response to SAPKactivators, lysophosphatidic acid and TNF46

Androgen receptor

Yeast two-hybrid Carboxyl terminus Involved in translocation of androgen receptor to the nucleus98

Rho, Rac, Cdc42

Direct binding46,47 Repeats 21–24 (GTP-independent)

RalA

Direct binding Repeat 24 (GTP-dependent) Yeast two-hybrid

Promotes the formation of microspikes downstream of Cdc4247

Trio

Yeast two-hybrid Repeats 23–24 Direct binding

Promotes dorsal ruffling48

cvHSP

Yeast two-hybrid99 Repeats 21–24

Kv4.2 Direct binding Carboxyl terminus Promotes current density potassium Yeast two-hybrid mediated by this specific channel (post- Co-localization channel100 synaptic axons)

Filamin binds to GEF domain 1 Filamin binds to amino terminus

*FLNa unless otherwise specified. (GEF, guanine nucleotide exchange factor; cvHSP, cardiovascular heat-shock protein; NF-κB, nuclear factor of kappa light polypeptide gene enhancer in B-cells; SAPK, stress-activated protein kinase; TNF, tumour necrosis factor; TRAF, TNF receptor-associated factor.)

142

www.nature.com/reviews/molcellbio

| FEBRUARY 2001 | VOLUME 2

REVIEWS

GP1bα

Integrins

PtdIns(4,5)P2 PI(5)K Rac Trio Cdc42

Figure 5 | Possible organization of FLNa at the surface of a human blood platelet. FLNa is shown crosslinking actin filaments and engaging transmembrane proteins (Gp1bα and the β-chains of β3 integrin) as well as the small GTPases Cdc42 and Rac and their guanine nucleotide exchange factor, Trio. Activated Rac, which is known to bind and activate phosphatidylinositol-5-OH kinase (PI(5)K), could stimulate biosynthesis of phosphatidylinositol-4,5-bisphosphate (PtdIns(4,5)P2), leading to promotion of actin assembly from monomeric actin.

LAMELLA

Flat, sheet-like protrusions at the edge of the cell. A fanshaped lamella is a prominent feature identifying the leading edge of a cell undergoing locomotion on a flat surface. Actin networks are the principal structures within these lamellae. EPILEPTIC SEIZURE

Commonly known as fits, seizures are the result of uncontrolled bursts of neuronal activation in the brain that usually lead to repetitive motor activity, such as shaking of extremities. Epilepsy is the term applied to recurrent seizures. Anatomical malformations, tumours, scars, infection, inflammation, haemorrhage and metabolic derangements can cause seizures. LATERAL VENTRICLES

Cavernous structures in the middle of the cerebrum that contain cerebrospinal fluid.

logical importance of these phosphorylations requires elucidation. Life without filamin

One look at bushes and trees is enough to convey how branched scaffoldings can produce a tremendous variety of shapes. Such plasticity is essential for modulating cell shape and integrating the complex surface movements required for crawling. The amoeboid forms of Dictyostelium genetically depleted of filamin by homologous recombination have difficulty extending broad surface protrusions and, as a result, have impaired locomotion and phagocytosis57–59. The actin networks in extended LAMELLAE of filamin-null amoebae are denser and less three-dimensional than those of wildtype amoebae. This appearance implies that the deficient cells compensate for the lack of filamin by increasing their content of polymerized actin60. Filamin-deficient Dictyostelium amoebae also have a diminished osmotic resistance to swelling induced by diminution of the osmotic pressure of the extracellular medium61,62. Several tumour cell lines grown from genetically unrelated humans with malignant melanoma express no FLNa protein. The FLNa-deficient cells do not

undergo locomotion in response to factors that elicit migration of FLNa-expressing melanoma cells cultivated from other tumours. Transfection of one of the FLNa-deficient cell lines (called M2) with FLNa cDNA yielded sublines that crawled with velocities proportional to the expression levels of the protein. Expressing levels that were higher than wild type decreased locomotion rates again63. The M2 cells have an unstable surface63,64. Under conditions in which wild-type cultured cells extend one or a few flattened ruffling lamellae, M2 cells (as well as other melanoma cell lines lacking FLNa) protrude and retract multiple spherical aneurysms, described as blebs (see movie online). The blebbing phenomenon results from an inability of the cell cortex that lacks FLNa to withstand internal hydrostatic pressures as efficiently as wild-type or rescued cells. The initiating event that leads to bleb protrusion is a weakening of the actin structure, because the blebs initially contain mostly monomeric, not filamentous, actin. Subsequently, actin polymerization and whatever actin crosslinking is achievable in the absence of FLNa limit bleb enlargement and lead to retraction of the bleb structure. Eventually the FLNa-null cells can stabilize their membranes and flatten, although this stabilization occurs when the cells polymerize cellular actin to a level significantly higher than that of filamin-expressing cells65. Mutations in the FLNa gene that completely block its expression are the cause of human periventricular heterotopia, an X-linked developmental disease66. In this disorder, females suffer from recurrent EPILEPTIC SEIZURES. Their brains have neuronal cell body accumulations along the LATERAL VENTRICLES as a result of failed neuronal migration to the cortex, and these foci are the sources of seizure activity. Heterotopia patients seem to have an unusually high incidence of vascular complications as well. In most affected families, inheritance of the heterotopia trait is lethal to males in utero. This implies that essential embryonic cell migration can only occur in FLNa-expressing cells. Drosophila ovarian ring canals, which normally accommodate the mass movement of cytoplasm from nutritive nurse cells to the growing oocyte, fail to accumulate filamentous actin in cheerio mutants and do not grow to wild-type size. As a consequence, cytoplasmic flow to the oocyte is inadequate and leads to the production of small, nonviable eggs. In weaker alleles of cheerio, ring canals accumulate some, but not enough, filamentous actin. The intercellular bridges in the mutants ultimately degenerate and destabilize the nurse-cell plasma membranes. These data indicate that filamin in Drosophila ovarian germline cells might be essential for the recruitment of F-actin to ring canal rims and its organization at that site, and that the cells are unable to compensate for filamin loss with another actin-binding protein11,12,67. The simplest explanation for many of the structural and functional defects in filamin mutants is that they lack an efficient mechanism to create peripheral actin gel networks by orthogonal crosslinking of actin filaments. For example, the adhesive properties of

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 2 | FEBRUARY 2001 | 1 4 3

REVIEWS FLNa-deficient melanoma cells seem roughly normal, although these cells have a 50% reduction in bulk elasticity compared with FLNa-expressing cells63.

the unusual situation where the number of laboratories active in the field (almost) doesn’t match the number of interesting questions.

Future directions

Many questions remain to be answered about filamin function. How do filamins dimerize and orientate so as to organize actin filaments? How do filamins, the Arp2/3 complex and other actin filament crosslinking proteins work together during cell protrusive activity? In this context, it would be interesting to compare filamin-mediated actin-filament crosslinking with the actin branching caused by the Arp2/3 complex. How do cellular signals regulate filamin function and the numerous interactions between filamins and a variety of cellular partners? At present, filamin research is in

10.

11.

12.

13. 14. 15.

16.

17.

144

Hartwig, J. & Stossel, T. Isolation and properties of actin, myosin, and a new actin-binding protein in rabbit alveolar macrophages. J. Biol. Chem. 250, 5696–5705 (1975). Stossel, T. P. & Hartwig, J. H. Interactions of actin, myosin and an actin-binding protein of rabbit alveolar macrophages. Macrophage myosin Mg2+-adenosine triphosphatase requires a cofactor for activation by actin. J. Biol. Chem. 250, 5706–5712 (1975). Wang, K., Ash, J. F. & Singer, S. J. Filamin, a new highmolecular weight protein found in smooth muscle and nonmuscle cells. Proc. Natl Acad. Sci. USA. 72, 4483–4486 (1975). Shizuta, Y., Shizuta, H., Gallo, M., Davies, P. & Pastan, I. Purification and properties of filamin, an actin-binding protein from chicken gizzard. J. Biol. Chem. 251, 6562–6567 (1976). Wallach, D., Davies, P. J. A. & Pastan, I. Purification of mammalian filamin. Similarity to high molecular weight actin-binding protein in macrophages, platelets, fibroblasts and other tissues. J. Biol. Chem. 253, 3328–3335 (1978). Bechtel, P. J. Identification of a high molecular weight actin-binding protein in skeletal muscle. J. Biol. Chem. 254, 1755–1758 (1979). Condeelis, J., Salisbury, J. & Fujiwara, K. A new protein that gels F actin in the cell cortex of Dictyostelium discoideum. Nature 292, 161–163 (1981). Condeelis, J., Geosits, J. & Vahey, M. Isolation of a new actin binding protein from Dictyostelium. Cell Motility 2, 273–285 (1982). Vargas, M., Sansonetti, P. & Guillén, N. Identification and cellular localization of the actin-binding protein ABP-120 from Entamoeba histolytica. Mol. Microbiol. 22, 849–857 (1996). Noegel, A., Rapp, S., Lottspeich, F., Schleicher, M. & Stewart, M. The Dictyostelium gelation factor shares a putative actin binding site with α-actinins and dystrophin and also has a rod domain containing six 100-residue motifs that appear to have a cross-beta conformation. J. Cell Biol. 109, 607–618 (1989). Li, M. -G. et al. Filamin is required for ring canal assembly and actin organization during Drosophila oogenesis. J. Cell Biol. 146, 1061–1073 (1999). Sokol, N. & Cooley, L. Drosophila filamin encoded by the cheerio locus is a component of ovarian ring canals. Curr. Biol. 9, 1221–1230 (1999). Introduction to Drosophila oocyte development and overview of filamins in Drosophila. Weihing, R. R. The filamins: properties and functions. Can. J. Biochem. Cell Biol. 63, 397–413 (1985). Hartwig, J. in Protein Profile (ed. Sheterline, P.) 711–778 (Academic, London, 1994). Lebart, M. -C. et al. Characterization of the actin binding site on smooth muscle filamin. J. Biol. Chem. 269, 4279–4284 (1994). Gorlin, J. et al. Human endothelial actin-binding protein (ABP, nonmuscle filamin): a molecular leaf spring. J. Cell Biol. 111, 1089–1105 (1990). This paper describes the cloning and structure of FLNa. Taylor, D. & Condeelis, J. Cytoplasmic structure and contractility in amoeboid cells. Int. Rev. Cytol. 56, 57–144 (1979).

Links DATABASE LINKS filamin | cheerio | β-spectrin |

18. Brotschi, E. A., Hartwig, J. H. & Stossel, T. P. The gelation of actin by actin-binding protein. J. Biol. Chem. 253, 8988–8993 (1978). 19. Bennett, J., Zaner, K. & Stossel, T. Isolation and some properties of macrophage α-actinin. Evidence that it is not an actin gelling protein. Biochemistry 23, 5081–5086 (1984). 20. Ito, T., Suzuki, A. & Stossel, T. Regulation of water flow by actin-binding protein-induced actin gelation. Biophys. J. 61, 1301–1305 (1992). 21. Goldmann, W. & Isenberg, G. Analysis of filamin and αactinin binding to actin by the stopped flow method. FEBS Lett. 336, 408–410 (1993). 22. Matsudaira, P. Actin crosslinking proteins at the leading edge. Semin. Cell Biol. 5, 165–174 (1994). 23. Honda, M., Takiguchi, K., Ishikawa, S. & Hotani, H. Morphogenesis of liposomes encapsulating actin depends on the type of actin-crosslinking. J. Mol. Biol. 287, 293–300 (1999). 24. Stossel, T. & Hartwig, J. Interactions of actin, myosin and an actin-binding protein of rabbit pulmonary macrophages. II. Role in cytoplasmic movement and phagocytosis. J. Cell Biol. 68, 602–614 (1976). 25. Hartwig, J. H., Tyler, J. & Stossel, T. P. Actin-binding protein promotes the bipolar and perpendicular branching of actin filaments. J. Cell Biol. 87, 841–848 (1980). 26. Niederman, R., Amrein, P. & Hartwig, J. H. The three dimensional structure of actin filaments in solution and an actin gel made with actin-binding protein. J. Cell Biol. 96, 1400–1413 (1983). 27. Wolosowick, J. & Condeelis, J. Fine structure of gels prepared from an actin-binding protein: comparison to cytoplasmic extracts and cortical cytoplasm in ameboid cells of Dictyostelium discoideum. J. Cell Biochem. 30, 227–243 (1986). 28. Weihing, R. R. Actin-binding and dimerization domains of HeLa cell filamin. Biochemistry 27, 1865–1869 (1988). 29. Bresnick, A., Warren, V. & Condeelis, J. Identification of a short sequence essential for actin binding by Dictyostelium ABP-120. J. Biol. Chem. 265, 9236–9240 (1990). 30. Bresnick, A., Janmey, P. & Condeelis, J. Evidence that a 27-residue sequence is the actin-binding site of ABP-120. J. Biol. Chem. 266, 12989–12993 (1991). 31. Hartwig, J. H. & Stossel, T. P. The structure of actin-binding protein molecules in solution and interacting with actin filaments. J. Mol. Biol. 145, 563–581 (1981). 32. Schliwa, M. & Van Blerkom, J. Structural interactions of cytoskeletal components. J. Cell Biol. 90, 225–235 (1981). 33. Hartwig, J. H. & Shevlin, P. The architecture of actin filaments and the ultrastructural location of actin-binding protein in the periphery of lung macrophages. J. Cell Biol. 103, 1007–1020 (1986). 34. Hartwig, J. H., Chambers, K. A. & Stossel, T. P. Assocation of gelsolin with actin filaments and cell membranes of macrophages and platelets. J. Cell Biol. 108, 467–480 (1989). 35. Hartwig, J. Mechanisms of actin rearrangements mediating platelet activation. J. Cell Biol. 118, 1421–1442 (1992). 36. Bailly, M. et al. Relationship between Arp2/3 complex and the barbed ends of actin filaments at the leading edge of carcinoma cells after epidermal growth factor stimulation.

J. Cell Biol. 145, 331–345 (1999). 37. Svitkina, T. & Borisy, G. Arp2/3 complex and actin depolymerizing factor/cofilin in dendritic organization and treadmilling of actin filament array in lamellipodia. J. Cell Biol. 145, 1009–1026 (1999). 38. Langanger, G. et al. Ultrastructural localization of α-actinin and filamin in cultured cells with the immunogold staining (IGS) method. J. Cell Biol. 99, 1324–1334 (1984). 39. Blanchoin, L. et al. Direct observation of dendritic actin filament networks nucleated by Arp2/3 complex and WASP/Scar proteins. Nature 404, 1007–1011 (2000). 40. Pantaloni, D., Boujemaa, R., Didry, D., Gounon, P. & Carlier, M. -F. The Arp2/3 complex branches filament barbed ends: functional antagonism with capping proteins. Nature Cell Biol. 2, 385–391 (2000). 41. Mullins, R., Heuser, M. & Pollard, T. The interaction of Arp2/3 complex with actin: nucleation, high affinity pointed end capping, and formation of branching networks of filaments. Proc. Natl Acad. Sci. USA 95, 6181–6186 (1998). 42. Mullins, R., Kelleher, J., Xu, J. & Pollard, T. Arp2/3 complex from Acanthamoeba binds profilin and crosslinks actin filaments. Mol. Biol. Cell 9, 841–852 (1998). 43. Janmey, P. A., Hvidt, S., Lamb, J. & Stossel, T. P. Resemblance of actin-binding protein/actin gels to covalently crosslinked networks. Nature 345, 89–92 (1990). This paper describes the rheological background of filamin–actin gels. 44. Janssen, K.-P. et al. Viscoelastic properties of F-actin solutions in the presence of normal and mutated actinbinding proteins. Arch. Biochem. Biophys. 325, 183–189 (1996). 45. Bishop, A. & Hall, A. Rho GTPases and their effector proteins. Biochem. J. 348, 241–255 (2000). 46. Marti, A. et al. ABP-280 binds the SAPK activator SEK-1 and is required for TNFα activation of SAPK in melanoma cells. J. Biol. Chem. 272, 2620–2628 (1997). 47. Ohta, Y., Suzuki, N., Nakamura, S., Hartwig, J. & Stossel, T. The small GTPase RalA targets filamin to induce filopodia. Proc. Natl Acad. Sci. USA 96, 2122–2128 (1999). 48. Bellanger, J. -M. et al. The Rac1- and RhoG-activating domain of the bifunctional guanine nucleotide exchange factor Trio targets filamin to remodel cytoskeletal actin. Nature Cell Biol. 2, 888–892 (2000). 49. Fisher, P. et al. Photosensory and thermosensory responses in Dictyostelium slugs are specifically impaired by absence of the F-actin crosslinking gelation factor (ABP-120). Curr. Biol. 7, 889–892 (1997). 50. Glogauer, M. et al. The role of actin-binding protein 280 in integrin-dependent mechanoprotection. J. Biol. Chem. 273, 1689–1698 (1998). 51. Wallach, D., Davies, P. J. A. & Pastan, I. Cyclic AMPdependent phosphorylation of filamin in mammalian smooth muscle. J. Biol. Chem. 253, 4739–4745 (1978). 52. Chen, M. & Stracher, A. In situ phosphorylation of platelet actin-binding protein by cAMP-dependent protein kinase stabilizes it against proteolysis by calpain. J. Biol. Chem. 264, 14282–14289 (1989). 53. Kawamoto, S. & Hidaka, H. Ca2+-activated phospholipiddependent protein kinase catalyses the phosphorylation of

www.nature.com/reviews/molcellbio

| FEBRUARY 2001 | VOLUME 2

REVIEWS

54.

55.

56.

57.

58.

59.

60.

61.

62.

63.

64. 65.

66.

67.

68.

actin-binding proteins. Biochem. Biophys. Res. Commun. 118, 736–742 (1984). Jay, D., García, E., Lara, J., Medina, M. & de la Cruz Ibarra, M. Determination of a cAMP-dependent protein kinase phosphorylation site in the C-terminal region of human endothelial actin-binding protein. Arch. Biochem. Biophys. 377, 80–84 (2000). Ohta, Y. & Hartwig, J. Phosphorylation of actin-binding protein 280 by growth factors is mediated by p90 ribosomal protein S6 kinase. J. Biol. Chem. 271, 11858–11864 (1996). Ohta, Y. & Hartwig, J. Actin filament crosslinking by chicken gizzard filamin is regulated by phosphorylation in vitro. Biochemistry. 34, 6745–6754 (1995). Cox, D. et al. Targeted disruption of the ABP-120 gene leads to cells with altered motility. J. Cell Biol. 116, 943–955 (1992). Cox, D., Wessels, D., Soll, D., Hartwig, J. & Condeelis, J. Re-expression of ABP-120 rescues cytoskeletal, motility, and phagocytosis defects of ABP-120-Dictyostelium mutants. Mol. Biol. Cell 7, 803–823 (1996). Ponte, E., Rivero, F., Fechheimer, M., Noegel, A. & Bozzaro, S. Severe developmental defects in Dictyostelium null mutants for actin-binding proteins. Mech. Dev. 91, 153–161 (2000). Cox, D., Ridsdale, J., Condeelis, J. & Hartwig, J. Genetic deletion of ABP-120 alters the three-dimensional organization of actin filaments in Dictyostelium pseudopods. J. Cell Biol. 128, 819–835 (1995). Description of filamin-deficient Dictyostelium amoebae. Rivero, F. et al. The role of the cortical cytoskeleton: F-actin crosslinking proteins protect against osmotic stress, ensure cell size, cell shape and motility, and contribute to phagocytosis and development. J. Cell Sci. 109, 2679–2691 (1996). Rivero, F., Furukawa, R., Fechheimer, M. & Noegel, A. Three actin crosslinking proteins, the 34 kDa actinbundling protein, α-actinin and gelation factor (ABP-120), have both unique and redundant roles in the growth and development of Dictyostelium. J. Cell Sci. 112, 2737–2751 (1999). Cunningham, C. et al. Actin-binding protein requirement for cortical stability and efficient locomotion. Science 255, 325–327 (1992). Description of filamin-deficient melanoma cells and their rescue. Dai, J. & Sheetz, M. Membrane tether formation from blebbing cells. Biophys. J. 77, 3363–3370 (1999). Cunningham, C. Actin polymerization and intracellular solvent flow in cell surface blebbing. J. Cell Biol. 129, 1589–1599 (1995). Fox, J. et al. Mutations in filamin 1 prevent migration of cerebral cortical neurons in human periventricular heterotopia. Neuron 21, 1315–1325 (1998). Description of the first human filamin mutation, periventricular heterotopia. Robinson, D., Smith-Leiker, T., Sokol, N., Hudson, A. & Cooley, L. Formation of the Drosophila ovarian ring canal inner rim depends on cheerio. Genetics 145, 1063–1072 (1997). Ezzell, R. M., Kenney, D. M., Egan, S., Stossel, T. P. & Hartwig, J. H. Localization of the domain of actin-binding protein that binds to membrane glycoprotein Ib and actin in human platelets. J. Biol. Chem. 263, 13303–13309 (1988).

69. Tyler, J. M., Anderson, J. M. & Branton, D. Structural comparison of several actin-binding molecules. J. Cell Biol. 85, 489–495 (1980). 70. Fucini, P., Renner, C., Heberhold, C., Noegel, A. & Holak, T. The repeating segments of the F-actin crosslinking gelation factor (ABP-120) have an immunoglobulin-like fold. Nature Struct. Biol. 4, 223–230 (1997). 71. Fucini, P. et al. Molecular architecture of the rod domain of the Dictyostelium gelation factor (ABP-120). J. Mol. Biol. 291, 1017–1023 (1999). 72. McCoy, A., Fucini, P., Noegel, A. & Stewart, M. Structural basis for dimerization of the Dictyostelium gelation factor (ABP120) rod. Nature Struct. Biol. 9, 836–841 (1999). This paper presents the atomic structure of a Dictyostelium filamin self-association domain. 73. Hartwig, J. & DeSisto, M. The cytoskeleton of the resting human blood platelet: structure of the membrane skeleton and its attachment to actin filaments. J. Cell Biol. 112, 407–425 (1991). 74. Gorlin, J. et al. Actin-binding protein (ABP-280) filamin gene (FLN) maps telomeric to the color vision locus (R/CGP) and centromeric to G6PD in Xq28. Genomics 17, 496–498 (1993). 75. Maestrini, E. et al. Mapping of two genes encoding isoforms of the actin-binding protein ABP-280, a dystrophin like protein, to Xq28 and to chromosome 7. Hum. Mol. Genet. 2, 761–766 (1993). 76. Zhang, W., Han, S., McKeel, D., Goate, A. & Wu, J. Interaction of presenilins with the filamin family of actinbinding proteins. J. Neurosci. 18, 914–922 (1998). 77. Takafuta, T., Wu, G., Murphy, G. & Shapiro, S. Human ßfilamin is a new protein that interacts with the cytoplasmic tail of glycoprotein Ibα. J. Biol. Chem. 273, 17531–17538 (1998). 78. Xu, W. -F, Xie, Z. -W, Chung, D. & Davie, E. A novel human actin-binding protein homologue that binds to platelet glycoprotein Ibα. Blood 92, 1268–1276 (1998). 79. Bröcker, F. et al. Assignment of human filamin gene FLNB to human chromosome band 3p14. 3 and identification of YACs containing the complete FLNB transcribed region. Cytogenet. Cell Genet. 85, 267–268 (1999). 80. Chakarova, C. et al. Genomic structure and fine mapping of the two human filamin gene paralogues FLNB and FLNC and comparative analysis of the filamin gene family. Hum. Genet. 107, 597–611 (2000). 81. Barry, C. P., Xie, J., Lemmon, V. & Young, A. P. Molecular characterization of a multipromoter gene encoding a chicken filamin protein. J. Biol. Chem. 268, 25577–25586 (1993). 82. Xie, Z., Wu, W., Davie, E. & Chung, D. Molecular cloning of human ABPL, an actin-binding protein homologue. Biochem. Biophys. Res. Commun. 251, 914–919 (1998). 83. van der Ven, P., Obermann, W., Lemke, B., Gautel, M., Weber, K. & Fürst, D. Characterization of muscle filamin isoforms suggests a possible role of γ-filamin/ABP-L in sarcomeric Z-disc formation. Cell Motil. Cytoskeleton 45, 149–162 (2000). 84. van der Ven, P. et al. Indications for a novel muscular dystrophy pathway: γ-filamin, the muscle-specific filamin isoform, interacts with myotilin. J. Cell Biol. 151, 235–247 (2000). 85. Meyer, S. et al. Identification of the region in actin-binding protein that binds to the cytoplasmic domain of glycoprotein Ib. J. Biol. Chem. 272, 2914–2919 (1997). 86. Calderwood, D., Shattil, S. & Ginsberg, M. Integrins and

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

87.

88.

89.

90.

91.

92.

93.

94.

95.

96.

97.

98.

99.

100.

actin filaments: reciprocal regulation of cell adhesion and signaling. J. Biol. Chem. 275, 22607–22610 (2000). Ott, I., Fischer, E., Miyagi, Y., Mueller, B. & Ruf, W. A role for tissue factor in cell adhesion and migration mediated by interaction with actin-binding protein 280. J. Cell Biol. 140, 1241–1253 (1998). Ohta, Y., Stossel, T. & Hartwig, J. Ligand-sensitive binding of actin-binding protein (ABP) to immunoglobulin G Fc receptor I(FcγR1, CD64). Cell. 67, 275–282 (1991). Liu, G. et al. Cytoskeletal protein ABP-280 directs the intracellular trafficking of furin and modulates proprotein processing in the endocytic pathway. J. Cell Biol. 139, 1719–1733 (1997). Good example of the identification and functional workup of a filamin–partner relationship. Thompson, T. et al. Filamin 2 (FLN2): a muscle-specific sarcoglycan interacting protein. J. Cell Biol. 148, 115–126 (2000). Stahlhut, M. & van Deurs, B. Identification of filamin as a novel ligand for caveolin-1: evidence for the organization of caveolin-1-associated membrane domains by the actin cytoskeleton. Mol. Biol. Cell 11, 325–337 (2000). Schwartzman, A. et al. Endogenous presenilin 1 redistributes to the surface of lamellipodia upon adhesion of Jurkat cells to a collagen matrix. Proc. Natl Acad. Sci. USA 96, 7932–7937 (1999). Guo, Y., Zhang, S., Sokol, N., Cooley, L. & Boulianne, G. Physical and genetic interaction of filamin with presenilin in Drosophila. J. Cell Sci. 113, 3499–3508 (2000). Li, M., Bermak, C., Wang, Z. & Zhou, Q. Modulation of dopamine D2 receptor signaling by actin-binding protein (ABP-280). Mol. Pharmacol. 57, 446–452 (2000). Browne, K., Johnstone, R., Jans, D. & Trapani, J. Filamin (280-kDa actin-binding protein) is a caspase substrate and is also cleaved directly by the cytotoxic T lymphocyte protease granzyme B during apoptosis. J. Biol. Chem. 275, 39262–39266 (2000). Edwards, D., Towb, P. & Wasserman, S. An activitydependent network of interactions links the Rel protein Dorsal with its cytoplasmic regulators. Development 124, 3855–3864 (1997). Leonardi, A., Ellinger-Ziegelbauer, H., Franzoso, G., Brown, K. & Siebenlist, U. Physical and functional interaction of filamin (actin-binding protein-280) and tumor necrosis receptor-associated factor 2. J. Biol. Chem. 275, 271–278 (2000). Ozanne, D., Brady, M., Gaughan, L., Cook, S., Neal, D. & Robson, C. Androgen receptor nuclear translocation is faciliated by the F-actin crosslinking protein filamin. Mol. Endocrinol. 14, 1618–1626 (2000). Krief, S. et al. Identification and characterization of cvHsp. A novel human small stress protein selectively expressed in cardiovascular and insulin-sensitive tissues. J. Biol. Chem. 274, 36592–36600 (1999). Petrecca, K., Miller, D. & Shrier, A. Localization and enhanced current density of the Kv4. 2 potassium channel by interaction with the actin-binding protein filamin. J. Neurosci. 20, 8736–8744 (2000).

Acknowledgements Some of the research reported in this review was supported by the NIH grants and by the Edwin S Webster Foundation (to T.P.S and J.H.H). We appreciate comments on this paper by Kuan Wang, NIAMS, NIH, who with S. John Singer was the first to purify avian gizzard filamin.

VOLUME 2 | FEBRUARY 2001 | 1 4 5

PERSPECTIVES OPINION

Confounded cytosine! Tinkering and the evolution of DNA Anthony Poole, David Penny and Britt-Marie Sjöberg Early in the history of DNA, thymine replaced uracil, thus solving a short-term problem for storing genetic information — mutation of cytosine to uracil through deamination. Any engineer would have replaced cytosine, but evolution is a tinkerer not an engineer. By keeping cytosine and replacing uracil the problem was never eliminated, returning once again with the advent of DNA methylation.

The origin of DNA is a fundamental question in evolution. Early on, DNA replaced RNA, reflecting the superior information-storage capacity of DNA1,2. Modern biochemical pathways provide an insight into this transition, as do RNA and uracil-DNA (U-DNA) viruses2,3, suggesting that the replacement took place in two steps (FIGS 1, 2a): replacement of ribose with deoxyribose, then replacement of uracil (U) with thymine (T)4. The first step was probably very complex, and has recently been reviewed elsewhere2,5. Here we look at the second (U→T) replacement, which is emerging as another example of why evolution is best viewed as a tinkerer, not as an engineer with an eye for ‘good’ design (BOX 1). Central to the story is cytosine (C), which readily deaminates to form U. This turns C•G pairs into U•G mispairs, and is an ongoing process in DNA6,7 (FIG. 2b). Without repair, replication of a U•G mispair would give one U•A pair (which is read as a T•A pair) and one C•G pair. All organisms carry the machinery for repair of C deaminations — uracil-N-glycosylase (UNG), which recognizes and removes any U that it detects, leaving an aba-

sic site. This is patched up by base-excision repair8,9 (FIG. 3), which creates a gap in the DNA opposite G. DNA polymerase then fills the gap with dC, thus repairing the mutation. Occasionally, U (from dUTP) is incorporated opposite A, so both U•G and U•A pairs turn up in DNA. The UNG recognizes and removes U arising from either C deamination or misincorporation, allowing DNA to be faithfully repaired10–12. Before T was a constituent of DNA, it would have been harder to detect C deaminaC

RNA

Uracil-DNA

Ribonucleotide reductase

dUTPase

Thymidylate synthase

dT Modern DNA

Figure 1 | Stepwise evolution of DNA on the basis of what is inferred from modern biochemical pathways. Ribonucleotide reductases catalyse synthesis of deoxyribonucleotides from ribonucleotides, with the exception of dT, which is synthesized from dU by thymidylate synthase (TS). dUTP is acted on by dUTPase to produce dUMP, the substrate for dTMP synthesis by TS. This is then brought up to the triphosphate level (indicated by the last arrow) before incorporation into DNA.

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

tions, because U was a bona fide constituent of early DNA at U•A pairs. It is widely accepted that U→T replacement solved this problem because it allowed any U arising by C deamination to be detected unambiguously13 (FIG. 4). However, replacing U with T would not by itself eliminate mutations arising from C deamination — it simply allows C→U mutations to be recognized, because U would otherwise be absent from DNA9. In the absence of repair (assuming the only role for T was to provide a means of recognizing C→U deaminations), there is no obvious selection pressure for the U→T replacement. François Jacob14 has likened evolution to “tinkering”. In contrast to an engineer, who works by design, obtaining all the necessary materials needed for construction of a prototype and finally testing it before putting it to work, a tinkerer makes use of whatever is at hand. This means that the result, although functional, is often far from perfect. A consequence of this modus operandi is that if something works in the short term it will be used, even if a better alternative is conceivable. New innovations cannot arise only to become useful when a subsequent function evolves, because there is no selection to maintain such innovations before they become useful. Recent progress on the biochemistry of U removal reveals an unexpected diversity of reactions catalysed by members of the uracilDNA glycosylase family (even though they all share a common origin), and allows the U→T conundrum to be resolved. New data15 on a closely related phenomenon — the repair of deaminated 5-methylcytosine (5meC, which deaminates to T, resulting in a T•G mismatch; FIG. 2b) — highlights the usefulness of the tinkering analogy for evolution. The problems solved by replacing U with T resurfaced once again when C methylation became a feature of the genome, with a member of the U-DNA glycosylase family being recruited to repair 5-meC→T deaminations. Driving thymidylate synthase evolution

If we view the U→T replacement in terms of

VOLUME 2 | FEBRUARY 2001 | 1 4 7

PERSPECTIVES

Box 1 | Tinkering versus engineering Is cytosine a component of the genetic material simply because there is no possible alternative to cytosine? If this were so, then cytosine deamination would be unavoidable. Jacob’s point, however, is that in evolution solutions are found using what is readily available14. An engineer can choose to scrap an entire project and start again from scratch. In the current context, if C is no good because of deamination, an engineer can replace it. If G is no good — because there is no alternative to C — then it too can be replaced. Indeed, if DNA therefore turns out to be no good, because there is no alternative to C•G pairs, this can be replaced. The problem of C deamination is recognized by prebiotic chemists36 as causing headaches for solving the origin-of-life problem. There is nevertheless no expectation that nucleic acids (with C) are the only possible medium for biological information storage — they are merely the only medium we are aware of. So there are two positions that can be held: that DNA containing C is the only possible genetic material, so the tinkering analogy does not hold; or that C and DNA came to be used for storage of genetic material, in spite of the tendency for C to deaminate. Our point is that if C were available, and were selected for in evolution, it would not make any difference if it later turned out to have an Achilles’ heel — once it became central to the genetic material, it was effectively impossible to replace it, at any level. A helpful analogy is Francis Crick’s37 concept of a “frozen accident” — a feature on which other features are built becomes so central to the working of the “machine” that it cannot be replaced in evolution, even if there is a conceivably better alternative. All those features that are built on the central feature would also have to be changed, so the central feature is effectively “frozen”. Frozen accidents are therefore a consequence of evolution through tinkering.

tinkering, repair of C→U deaminations probably evolved before — or perhaps concurrent with — thymidylate synthase (which synthesizes dTMP from dUMP for the incorporation of T (from dTTP) into DNA; BOX 2). T could not have arisen in the absence of UNG as this implies forethought, not tinkering. It makes more sense if some form of repair arose first. This could have been by removal of all U, removal and repair only at U•G mispairs or preferential (but imperfect) U•G repair. We will consider each of these possibilities in turn. If repair were by removal of all U, as per modern UNG, U would be removed where it had arisen by C deamination (which would be beneficial), but also where it was correctly paired with A (FIG. 4). Removal of U opposite A would be energetically wasteful, because the base-excision repair pathway would be stuck in a futile cycle. Nevertheless, if the cost (in terms of energy) of removing and reincorporating U is less than the benefit (in terms of improved genetic stability) of repairing C deamination events, then UNG function might have been selected for (FIG. 4). The cost of the futile cycle would have driven selection towards replacement of U by T (that is, driven the evolution of thymidylate synthase), and subsequently favoured evolution of mechanisms (such as dUTPase) for reducing the misincorporation of U into DNA. However, extensive removal of U at U•A pairs was probably a greater cost than the benefit of repairing U•G. On average, 25% of the genome would be U, and the risk of mutation by subsequent misincorporation

148

would be ever present, as futile cycling would be continuous. At the other end of the spectrum is the second possibility — removal and repair only at U•G mispairs, as per the UNG-related enzyme, mismatch-specific uracil DNA-glycosylase (MUG)16,17. This would leave U•A pairs untouched (FIG. 4). However, if U•G mispairs were recognized and replaced unambiguously from the outset, what selection would there have been for U→T replacement? Leaky MUG — a possible solution

The suggestion that either UNG or MUG predate thymidylate synthase does not seem to solve the problem of the U→T transition. The a

Uracil

Thymine

-O P O

O-

-O P O

ORibonucleotide (RNA)

NH2 H3C

N N

N Cytosine

5-methylcytosine

O Deamination

OO OH O=P O-

NH2

CH3

HN N

HN O

ideal solution would be a trade-off between UNG and MUG activities — preferential (but imperfect) U•G repair (FIG. 4). Selection for U replacement could then have been driven by initial ‘leakiness’ of repair (that is, occasional U-excision repair at U•A pairs). Such a ‘leaky’ MUG (FIG. 4), with properties common to both modern enzymes, would have later evolved by duplication and divergence into the two specific enzymes seen in modern repair. Increasing knowledge of the modern enzymes from sequence, structural, mutagenesis and biochemical data provides a powerful way to explore this leaky MUG idea. Extant MUG specifically flips out U binding to G opposite, thus precluding N-glycoside cleavage at U•A pairs. ‘Flipping-out’ of U is common to both MUG and UNG, but MUG recognizes G opposite U, with little affinity for U itself18, whereas UNG has high affinity for U, ignoring the complementary strand and even excising U from single-stranded DNA16. In MUG, interaction with G is mediated largely by two residues (glycine 143 and serine 145), which hydrogen-bond with groups on G (REF. 16) that normally pair with C. Mutating these residues should therefore weaken discrimination between G and A, producing a leaky MUG. More significantly, substrate-affinity studies using catalytically inactive mutants of UNG show that UNG naturally has a tenfold faster association rate for U•G pairs than for U•A pairs19. This results from the greater ease with which U•G pairs can be prised apart, allowing the enzyme to bind U (REF. 19). So, even for a rudimentary enzymatic reaction of U-specific recognition and cleavage, U•G mismatches might have been preferred by the prototype UNG/MUG.

O H O=P O-

Deamination O

ODeoxyribonucleotide (DNA)

O NH

N Uracil

H3C

NH N

Thymine

Figure 2 | DNA structure and the chemistry of cytosine deamination. a | Differences between RNA and DNA. RNA and DNA differ at the 2′ position of the ribose — RNA has a hydroxyl group, whereas in DNA this is reduced to hydrogen. The second difference is at the level of the base — RNA contains uracil, which is equivalent to thymine in DNA. The difference between these two bases is the presence, in thymine, of a methyl group at the 5′ position of the pyrimidine ring. b | Cytosine deamination. Spontaneous deamination of cytosine and 5-methylcytosine is shown. The former deaminates to uracil, whereas the latter deaminates to give thymine. Both deamination events are mutagenic if unrepaired, leading to C•G→T•A changes in one daughter DNA strand after replication.

www.nature.com/reviews/molcellbio

| FEBRUARY 2001 | VOLUME 2

PERSPECTIVES 5′ G A U G T C 3′ 3′ C T G C A G 5′ UNG G A G T C C T G C A G AP endonuclease H

G AO P OG T C C T G C A G dRPase P O Deoxyribose-5-P H G AO PG T C C T G C A G

DNA polymerase H G A O PC G T C C T G C A G

DNA polymerase P G T O H G A C G TO C C T G C A G

DNA polymerase 5′→3′ exonuclease H G A C G TO PC C T G C A G

DNA ligase 5′ G A C G T C 3′ 3′ C T G C A G 5′

Figure 3 | Base-excision repair pathways. The two alternative pathways for base-excision repair of C→U deaminations are shown. All steps downstream of uracil-N-glycosylase (UNG) are part of general base-excision repair, which acts on abasic lesions from a number of sources, including those produced by action of DNA glycosylases. It can be argued that the right-hand pathway pre-dates that on the left, given that deoxyribophosphatase (dRPase) would further reduce the possibility of errors introduced during repair — removal of only the abasic site eliminates the possibility of downstream errors introduced by resynthesis of a short patch by DNA polymerase. (AP endonuclease, apurinic/apyrimidic endonuclease.) (Adapted from REF. 8.)

vided further improvement, turning the dUTP pool into dUMP (FIG. 5), which thymidylate synthase could convert to dTMP. This would have reduced dUTP misincorporation and enhanced dTTP incorporation (FIG. 5). In this model, then, the base-excision repair pathway pre-dated leaky MUG activity (which excises only the base, leaving an abasic site). Abasic lesions arise non-enzymatically in DNA and are a major source of mutation9,20. So base-excision repair probably evolved to repair abasic lesions before its role in excision of U. Why MUG?

The above picture describes the probable nature of the proto-UNG/MUG and how it could have driven U→T replacement, as well as how UNG could subsequently evolve. But what selection pressure gave rise to MUG? Given the much higher rate at which UNG removes U from DNA, MUG is probably not important in U-base-excision repair in vivo 21,22, and the function of discriminate U•G removal in bacteria is unclear. There is evidence, however, that MUG might provide a means of maintaining genetic material during viral infection. Two bacteriophage that infect Bacillus subtilis — PBS1 and PBS2 — are unusual among DNA viruses as their DNA contains U not T (REF. 3). These viruses make a protein that inhibits UNG but not MUG17. So, in an infected bacterium, UNG-mediated repair is

The trouble with T

As eukaryotic genomes became more complex, additional mechanisms of gene regulation developed. One such mechanism is DNA methylation, where a methyl group is added to position 5 on the cytosine ring, forming 5-meC (REF. 15). The ability to regu-

CAGGCCUA GUCCGGAU

Stopping the cycle: eliminating uracil

If the proto-UNG/MUG preferentially excised U at U•G pairs, but occasionally excised U opposite A (leaky MUG; FIG. 4), evolution of thymidylate synthase and consequent U→T replacement would have been advantageous. It would result in improved fidelity — and possibly energy conservation — by eliminating occasional futile removal at U•A pairs. Reduction in base-excision repair activity would have reduced errors resulting from incorrect replacement of adjacent downstream nucleotides during re-synthesis8,9 (FIG. 3). The U•G preference would have become relaxed, with speed of recognition and removal being selected for as genomes increased in size. We therefore propose that the advent of thymidylate synthase permitted the elimination of occasional U excision opposite A. Hence the role of T in DNA was simply to allow the specific repair of U•G mispairs. Other than this function, there is no clear advantage in having T in DNA rather than U. The advent of dUTPase would have pro-

inhibited, allowing the production of phage U-DNA. But any U•G mispairs arising from C→U deamination in the host cannot be repaired. Although U•A pairs are tolerated, lack of repair during infection could leave a bacterium with serious lesions, so the ability to specifically repair U•G mispairs would be advantageous. Both PBS1 and PBS2 are lytic phage (where the host cell is killed upon phage release), although given predicted phage diversity23,24 some lysogenic phage (where the host cell survives phage release) might also have U-DNA genomes. The hypothesis that MUG maintains genome fidelity during lysogenic infection by U-containing DNA phage would be borne out if MUG were found only in prokaryotes infected by U-DNA phage. Whereas hostmediated repair of U•G mispairs would be beneficial for both lytic and lysogenic phage, the selective advantage of MUG to the host would be in genome integrity during lysogenic infection. The existence of MUG in bacteria opens up the exciting possibility that, as per the RNA world, the U-DNA world is alive and well in viruses.

Cytosine deamination CAGGCUUA GUCUGGAU TS

UNG U

CAGGCUTA GTCUGGAT No repair: U→T replacement in absence of selection?

MUG

'Leaky' MUG U

CAGGCUUA GUCUGGAU U U U U U removal: Mutagenic futile cycling at U·A pairs drives U→T

CAGGCCUA GUCCGGAU 100% U·G repair: C deamination cannot drive U→T

CAGGCCUA GUCCGGAU Leaky U·G repair: Occasional action at U·A pairs drives U→T

Figure 4 | ‘Leaky’ mismatch-specific uracil-DNA glycosylase as a driving force for evolution of U→T replacement in DNA. Replacement of dU with dT occurred after the origin of DNA, probably in response to the problem of C deamination to U. This leads to loss of information as C is incorrectly read as U. Four evolutionary hypotheses are shown. The standard explanation for the origin of thymidylate synthase (TS) is that it solves this problem. However, replacing U with T without pre-existing uracil-Nglycosylase (UNG) activity does not eliminate the problem. An alternative explanation is that UNG arose first, allowing repair of deaminated cytosines, but creating an additional problem — extensive futile repair at undamaged sites. Unlike UNG, mismatch-specific uracil-DNA glycosylase (MUG) acts specifically on U•G mismatches, so if this evolved before TS, it is hard to see what selection pressure would drive U→T replacement. We propose that a ‘leaky’ MUG arose in response to C→U deamination. The enzyme would have had a preference for U•G repair, but occasionally act on U•A pairs. Occasional futile repair might have favoured evolution of TS and replacement of U by T.

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 2 | FEBRUARY 2001 | 1 4 9

PERSPECTIVES

Box 2 | The biochemistry of U and T The synthesis of deoxythymidine triphosphate (dTTP) differs from that of other deoxyribonucleotides. Thymidylate synthase produces deoxythymidine monophosphate (dTMP) from deoxyuridine monophosphate (dUMP), rather than producing it directly from the ribonucleoside triphosphate (rTTP in this case) as for other deoxyribonucleotides13 (FIGS 1,5). DNA polymerases incorporate either dUTP or dTTP opposite A (REF. 13), as shown by near thymineless Escherichia coli mutants38, and bacteriophage that contain U-DNA and use the host’s DNA polymerase for replication3. So, given the ease of misincorporation, it is essential to keep U out of DNA (and T out of RNA). The latter is easily achieved, as rT is only ever produced post-transcriptionally39. Misincorporation of dUTP into DNA instead of conversion to dTTP is minimized by dUTPase, which converts dUTP synthesized by ribonucleotide reductase to dUMP, the substrate for thymidylate synthase (FIG. 5). So modern metabolism seems geared up to keep dUTP out of DNA at all cost.

late genes by C methylation would have been beneficial, but it came with a catch: 5-meC deaminates to T at a rate 2–4-fold higher than C deaminates to U (FIG. 2)21, meaning that a new form of mismatch became a problem. In eukaryotes, thymine-DNA glycosylase (TDG), which is evolutionarily related to UNG and MUG, repairs T•G mismatches arising from deamination of 5-meC (REFS 17,25). Somewhat controversially, Yoder and colleagues26 argue that, other than in X-inactivation and imprinting, there is little evidence for DNA methylation in transcriptional regulation. Their hypothesis is that methylation provides genomic defence against the insertion of transposable elements26. If correct, then the high deamination rate of 5-meC could provide a way of inactivating transposable elements through mutation, meaning that 5-meC deamination is actually beneficial, not detrimental. Although there is intense debate as to whether the role of C methylation is in genome defence or gene regulation27–31, both possibilities highlight the validity of the ‘tinkering’ analogy. For instance, if C methylation arose and was selected for in gene regulation, those sites would also become more vulnerable to mutation by deamination, so thymine-DNA glycosylase would be advantageous. Mutations arising by 5-meC deamination might also have provided a means of inactivating transposable elements. The presence of TDG would then represent a trade-off, because accurate 5-meC→T repair limits inactivation of transposable elements. Indeed, methylation cannot be limited solely to transposable elements, as there is no selection to repair these with TDG. Perhaps in mammals, where 5-meC is used for both imprinting and X-inactivation, the need for repair was greater; hence the selection and maintenance of two thymine-DNA glycosylases (mammals have a second unrelated thymine-DNA glycosylase called MBD4 (REF. 25)). One result of the alternative ‘tinkering’ model, whereby 5-meC arose for element inactivation and was later recruited to gene

150

regulation, would have been an increased susceptibility of regulated genes to mutation. Hence repair would have arisen at the expense of element inactivation. Either way, it is possible to see the trade-offs that establishing one brings to the other. Eukaryotic TDG most probably arose by U + P

Deoxyribose-5-P O

UNG/BER 5′ G A U G T C 3′

dCTP dCTP dUTP dTTP deaminase NDP NDP RNR kinase kinase dUTPase rCTP rUTP dCDP dUDP dTDP CMP: Thymidylate UMP kinase kinase TS dTMP dUMP dCMP Deoxycytidine kinase Deoxycytidine dC deaminase dU

Thymidine kinase dT Thymidine phosphorylase T

U Deoxyribose-1-P Uracil pool

Deoxyribose-5-P Deoxyriboaldolase Acetaldehyde + Glyceraldehyde-3-P

Figure 5 | De novo and salvage pathways of pyrimidine deoxyribonucleotide metabolism. U and T are acted on by a common set of enzymes at all steps in the de novo and salvage pathways. C is acted on by a separate set, with the exception of NDP kinase, which acts on all diphosphate nucleotides. The CMP:UMP kinase overlaps with thymidylate kinase activity as both can act on dUMP. However, CMP:UMP kinase also acts on ribonucleotides. Free uracil and deoxyribose-5-phosphate, which are released as a result of the action of uracil-N-glycosylase (UNG) and base-excision repair (BER), are both salvaged in modern metabolism, although deoxyribose-5-phosphate can also be broken down into acetaldehyde and glyceraldehyde-3phosphate, which are then absorbed into central metabolism. (TS, thymidylate synthase) (Adapted from REF. 40.)

duplication of UNG, being selected for as a means of dealing with this additional source of lowered genome stability (although MBD4 is unrelated to either25). Deletion studies reveal that a core region of the human TDG protein that no longer removes T opposite G can remove U opposite G (REF. 17), thus conferring MUG-like activity on TDG. Interestingly, in E. coli, 5-meC deaminations are repaired by a mismatch-specific endonuclease called Vsr, the first step in the very short patch-repair pathway32,33. This is not related to the DNA glycosylases33, suggesting 5-meC and repair of 5-meC→T deaminations arose independently in eukaryotes and bacteria. Several species of insect apparently lack UNG (the genome sequence of Drosophila also shows that UNG is absent34) and biochemical studies indicate that these instead carry an enzyme that is probably closely related to TDG17. The insect homologues preferentially act at U•G mispairs17, suggesting that U•A pairs might even be tolerated in some insects, and that only U•G mispairs are repaired. Interestingly, 5-meC is absent from Drosophila. The machinery for C methylation has apparently been lost35, which is cited as an explanation for why Drosophila is so vulnerable to invasion by transposable elements26. Conclusions

The tendency for C to undergo deamination, and the way in which deaminations are repaired, provides a clear example of evolution as tinkering. Viewing evolution in this way sheds light on the likely nature of the selection pressure that drove the U→T transition. Examination of both UNG and MUG allows us to hypothesize that U excision-repair arose before thymidylate synthase, and that occasional removal of U opposite A by the protoUNG/MUG (or ‘leaky’ MUG) drove the U→T replacement. Overall, we suggest the following steps for the U→T transition: baseexcision repair→leaky MUG→thymidylate synthase→dUTPase. Replacing U with T provided a means by which to fine-tune repair of C→U deaminawww.nature.com/reviews/molcellbio

| FEBRUARY 2001 | VOLUME 2

PERSPECTIVES tions, but the problem of C deamination was never eliminated — it re-emerged in the form of 5-meC deamination. Tinkering also makes sense of the evolution of the 5-meC apparatus, which subsequently drove the recruitment of the U-excision apparatus into T excision because of the ‘unforeseen’ side effect of 5-meC→T deamination. All this could have been avoided simply by eliminating C early in the evolution of the genetic material — but how boring life would be if evolution worked by engineering. Anthony Poole and David Penny are at the Institute of Molecular BioSciences, PO Box 11222, Massey University, Palmerston North, New Zealand. Britt-Marie Sjöberg is in the Department of Molecular Biology, Stockholm University, 106 91 Stockholm, Sweden. e-mails: a.m.poole@massey.ac.nz; d.penny@massey.ac.nz; brittmarie.sjoberg@molbio.su.se

Links ENCYCLOPEDIA OF LIFE SCIENCES Codon

usage in molecular evolution 1. 2.

8. 9. 10.

11.

12. 13. 14. 15. 16.

17.

18.

19.

Gilbert, W. The RNA world. Nature 319, 618 (1986). Poole, A., Penny, D. & Sjöberg, B. -M. Methyl-RNA: an evolutionary bridge between RNA and DNA? Chem. Biol. 7, R207–R216 (2000). Hitzeman, R. A. & Price, A. R. Relationship of Bacillus subtilis DNA polymerase III to bacteriophage PBS2induced DNA polymerase and to the replication of uracilcontaining DNA. J. Virol. 28, 697–709 (1978). Lazcano, A., Guerrero, R., Margulis, L. & Oró, J. The evolutionary transition from RNA to DNA in early cells. J. Mol. Evol. 27, 283–290 (1988). Sutherland, J. D. & Whitfield, J. N. Prebiotic chemistry: a bioorganic perspective. Tetrahedron 53, 11493–11527 (1997). Frederico, L. A., Kunkel, T. A. & Ramsay Shaw, B. A sensitive genetic assay for the detection of cytosine deamination: determination of rate constants and the activation energy. Biochemistry 29, 2532–2537 (1990). Frederico, L. A., Kunkel, T. A. & Ramsay Shaw, B. Cytosine deamination in mismatched base pairs. Biochemistry 32, 6523–6530 (1993). Dianov, G. & Lindahl, T. Reconstitution of the DNA base excision-repair pathway. Curr. Biol. 4, 1069–1076 (1994). Lindahl, T. & Wood, R. D. Quality control by DNA repair. Science 286, 1897–1905 (1999). Mosbaugh, D. W. & Bennett, S. E. Uracil-excision DNA repair. Prog. Nucleic Acid Res. Mol. Biol. 48, 315–370 (1994). Krokan, H. E., Standal, R. & Slupphaug, G. DNA glycosylases in the base excision repair of DNA. Biochem. J. 325, 1–16 (1997). Cunningham, R. P. DNA glycosylases. Mutat. Res. 383, 189–196 (1997). Kornberg, A. & Baker, T. A. DNA Replication 2nd edn Ch. 2 (W. H. Freeman, New York, 1992). Jacob, F. Evolution and tinkering. Science 196, 1161–1166 (1977). Jones, P. A. The DNA methylation paradox. Trends Genet. 15, 34–37 (1999). Barrett, T. E. et al. Crystal structure of a G:T/U mismatchspecific DNA glycosylase: mismatch recognition by complementary-strand interactions. Cell 92, 117–129 (1998). Gallinari, P. & Jiricny, J. A new class of uracil-DNA glycosylases related to human thymine-DNA glycosylase. Nature 383, 735–738 (1996). Barrett, T. E. et al. Crystal structure of a thwarted mismatch glycosylase DNA repair complex. EMBO J. 18, 6599–6609 (1999). Panayotou, G., Brown, T., Barlow, T., Pearl, L. H. & Savva, R. Direct measurement of the substrate preference of uracil-DNA glycosylase. J. Biol. Chem. 273,

45–50 (1998). 20. Lindahl, T. Instability and decay of the primary structure of DNA. Nature 362, 709–715 (1993). 21. Lutsenko, E. & Bhagwat, A. S. Principal causes of hot spots for cytosine to thymine mutations at sites of cytosine methylation in growing cells. A model, its experimental support and implications. Mutat. Res. 437, 11–20 (1999). 22. Lutsenko, E. & Bhagwat A. S. The role of the Escherichia 4 coli MUG protein in the removal of uracil and 3,N Ethenocytosine from DNA. J. Biol. Chem. 274, 31034–31038 (1999). 23. Maniloff, J. & Ackermann, H.-W. Taxonomy of bacterial viruses: establishment of tailed virus genera and the order Caudovirales. Arch. Virol. 143, 2051–2063 (1998). 24. Hendrix, R. W., Smith, M. C. M., Burns, R. N., Ford, M. E. & Hatfull, G. F. Evolutionary relationships among diverse bacteriophages and prophages: all the world’s a phage. Proc. Natl Acad. Sci. USA 96, 2192–2197 (1999). 25. Hendrich, B., Hardeland, U., Ng, H. -H., Jiricny, J. & Bird, A. The thymine glycosylase MBD4 can bind to the product of deamination at methylated CpG sites. Nature 401, 301–304 (1999). 26. Yoder, J. A., Walsh, C. P. & Bestor, T. H. Cytosine methylation and the ecology of intragenomic parasites. Trends Genet. 13, 335–340 (1997). 27. O’Neill, R. J. W., O’Neill, M. J. & Graves, J. A. M. Undermethylation associated with retroelement activation and chromosome remodelling in an interspecific mammalian hybrid. Nature 393, 68–72 (1998). 28. Walsh, C. P., Chaillet, J. R. & Bestor, T. H. Transcription of IAP endogenous retroviruses is constrained by cytosine methylation. Nature Genet. 20, 116–117 (1998). 29. Garrick, D., Fiering, S., Martin, D. I. K. & Whitelaw, E. Repeat induced gene silencing in mammals. Nature Genet. 18, 56–59 (1998).

30. Regev, A., Lamb, M. J. & Jablonka, E. The role of DNA methylation in invertebrates: developmental regulation or genome defence? Mol. Biol. Evol. 15, 880–891 (1998). 31. Simmen, M. W. et al. Nonmethylated transposable elements and methylated genes in a chordate genome. Science 283, 1164–1167 (1999). 32. Lieb, M. & Bhagwat, A. S. Very short patch repair: reducing the cost of cytosine methylation. Mol. Microbiol. 20, 467–473 (1996). 33. Tsutakawa, S. E., Jingami, H. & Morikawa, K. Recognition of a TG mismatch: the crystal structure of very short patch repair endonuclease in complex with a DNA duplex. Cell 99, 615–623 (1999). 34. Adams, M. A. et al. The genome sequence of Drosophila melanogaster. Science 287, 2185–2195 (2000). 35. Tweedie, S. et al. Vestiges of a DNA methylation system in Drosophila melanogaster? Nature Genet. 23, 389–390 (1999). 36. Shapiro, R. Prebiotic cytosine synthesis: a critical analysis and implications for the origin of life. Proc. Natl Acad. Sci. USA 96, 4396–4401 (1999). 37. Crick, F. H. C. The origin of the genetic code. J. Mol. Biol. 38, 367–379 (1968). 38. El-Hajj, H. H., Wang, L. & Weiss, B. Multiple mutant of Escherichia coli synthesizing virtually thymineless DNA during limited growth. J. Bacteriol. 174, 4450–4456 (1992). 39. Garcia, G. A. & Goodenough-Lashua, D. M. in Modification and Editing of RNA (eds Grosjean, H. & Benne, R.) 135–168 (ASM, Washington DC, 1998). 40. El-Hajj, H. H., Zhang, H. & Weiss, B. Lethality of a dut (deoxyuridine triphosphatase) mutation in Escherichia coli. J. Bacteriol. 170, 1069–1075 (1988).

Acknowledgements. This work was supported by the Swedish Natural Science Research Council (to B.M.S) and by the New Zealand Marsden Fund.

OPINION

Retroviral recombination: what drives the switch? Matteo Negroni and Henri Buc The high rate of recombination in retroviruses is due to the frequent template switching that occurs during reverse transcription. Although the mechanism that leads to this switch is still a matter of debate, there is increasing evidence that specific RNA structures are involved. And the implications might go beyond retroviral genetic variability.

Diploidy is great! We, eukaryotes, have developed a specialized system to exchange genetic information between homologous chromosomes at meiosis. By passing only half our genetic complement to the next generation, and leaving a mating partner to provide the other half, genetic information is reshuffled. Diploidy also ensures that if the DNA is damaged on one chromosome, its homologous counterpart can be copied by the cell’s repair machinery. Retroviruses seem to share similar concerns. They have developed a sophisticated

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

Entry

Reverse transcription Translation

Integration

Transcription Provirus

Capsid assembly Budding

Figure 1 | Reverse transcription in the life cycle of retroviruses. The retroviral infectious cycle. Reverse transcription takes place in the cytoplasm of the host cell within the viral structure called the capsid (hexagon). The reverse transcriptase converts the viral RNA into doublestranded DNA, which integrates in the host genome to form what is called a ‘provirus’. Transcription by the cellular RNA polymerase II generates the viral messenger RNAs, as well as the new genomic RNAs that will be packaged into the budding virions. (Adapted from REF. 34.)

VOLUME 2 | FEBRUARY 2001 | 1 5 1

Links ENCYCLOPEDIA OF LIFE SCIENCES Codon

usage in molecular evolution 1. 2.

8. 9. 10.

11.

12. 13. 14. 15. 16.

17.

18.

19.

Acknowledgements. This work was supported by the Swedish Natural Science Research Council (to B.M.S) and by the New Zealand Marsden Fund.

OPINION

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

Entry

Reverse transcription Translation

Integration

Transcription Provirus

Capsid assembly Budding

VOLUME 2 | FEBRUARY 2001 | 1 5 1

PERSPECTIVES 5′

Polymerase active site 3′

Fingers

Thumb

RNase H Palm 3′

5′

system to ensure that two copies of their single-stranded genomic RNA are present in each viral particle1. When a retrovirus enters the host cell, the viral DNA polymerase (reverse transcriptase) converts the genomic RNA into a double-stranded DNA that integrates into the host’s genome2,3 (FIG. 1). During synthesis of the first DNA strand, the reverse transcriptase can switch template from one to the other copy of genomic RNA4,5 — a phenomenon known as copy-choice6. If the two molecules are not identical, this leads to genetic recombination7,8. The two RNAs will be considerably different if they come from viruses that belong to distinct viral strains but have co-infected the same cell, leading to the formation of ‘heterozygous’ virions. The frequent occurrence of such events in vivo is highlighted by the finding that up to 10% of the genomes among HIV-1 subtypes are genetic mosaics9. The term ‘template switching’ implies that the nascent DNA strand is transferred from one RNA (the ‘donor’) to the other (the ‘acceptor’). This process depends on a ribonuclease (RNase) H activity, carried by the reverse transcriptase, which degrades the RNA template once it has been copied10,11 (FIG. 2). So the reverse transcription complex is precarious, and relies on the stability of the RNA/DNA heteroduplex, which spans from the polymerization catalytic site to the RNaseH domain of the reverse transcriptase (FIG. 2). The presence of an acceptor RNA, complementary to the nascent DNA, makes the situation even more favourable for changing the template. Still, something must trigger the switch.

152

Figure 2 | The catalytic subunit of HIV-1 reverse transcriptase. The HIV-1 reverse transcriptase is a heterodimer composed of the p66 catalytic subunit and a proteolytic fragment of p66 named p51. Because the p66 subunit resembles a right hand, its subdomains are called palm (beige), thumb (purple) and fingers (green)35. The RNA and the nascent DNA are drawn as blue and grey strings, respectively. The orange domain is the RNaseH domain, which degrades the RNA once it has been copied. Degradation of the RNA is not necessarily complete all along the template, and short stretches of RNA might temporarily persist, hybridized onto the nascent DNA36. The RNA slides into the reverse transcriptase through a path along which the polymerase and RNaseH active sites are located37. The reverse transcriptase not only grabs the RNA/DNA hybrid, but also contacts the incoming template ahead of the polymerase active site in the reverse transcriptase ‘finger’ domain38. The RNA slides up the finger domain and points towards the reader. (Modified from REF. 35, according to data in REF. 38.)

can potentially be continued on the same RNA template (processive copying) or be transferred onto the other RNA (strand transfer). Any situation that impairs the first process would therefore push the equilibrium towards the second. If there is a discontinuity in the RNA template, the only way to continue synthesis is to transfer the nascent DNA onto the other genomic RNA6. This process is not only an important ‘escape lane’ for the virus (rescue synthesis) but, in a heterozygous virion, it might also lead to genetic recombination6 — the ‘forced copy-choice’ model (BOX 1a). In this case, the nascent DNA is thought to melt from its original template and, guided by the sequence complementarity, to hybridize on the acceptor RNA in the corresponding position. Resumption of DNA synthesis will generate a recombinant DNA. A similar situation can be found on intact RNA templates if the incorporation of one nucleotide is particularly difficult12 (‘pausedriven’ recombination; BOX 1a), as a consequence of the primary sequence of the RNA13, of its secondary structure14, or a combination of these two factors. The reverse transcriptase

Mechanisms of recombination

For each nucleotide incorporated into the growing chain of DNA, reverse transcription

Box 1 | Mechanisms for retroviral recombination a

STOP

3′

5′

5′ 3′

5′

3′ 3′

3′

5′

STOP

5′ 5′

3′

STOP

3′

5′

3′

5′ 3′

STOP

5′ 3′

3′

5′ 5′

5′

3′

3′ 3′

5′ 5′

Donor and acceptor RNAs are shown as red and blue lines, respectively; nascent DNA is shown as a black line. The arrow indicates the direction of synthesis. a | Forced copy-choice and pausedriven recombination. Reverse transcription is stopped on the donor RNA at a given position (‘stop’), either by a nick on the RNA (according to the forced copy-choice model6) or when the incorporation of the following nucleotide becomes particularly difficult12. The reverse transcriptase (yellow ellipse) dissociates from the primer–template complex (second panel), and the nascent DNA progressively hybridizes to the acceptor RNA (third panel). Once the 3′ end of the nascent DNA is annealed to the acceptor template, re-binding of the reverse transcriptase will eventually resume DNA synthesis (bottom panel). b | The ‘interactive hairpin’ model. The palindromic sequences that form the stem of the transactivation response element are shown as boxes. The stem melts as reverse transcription progresses through it, allowing the formation of a new interaction in the hairpin region, this time between donor and acceptor RNAs. During copying of the descending part of the hairpin, the nascent DNA eventually hybridizes to the acceptor RNA and diverts synthesis from the donor RNA. c | Pause-independent model. Strand transfer was proposed to occur in two steps. First, the acceptor RNA docks on the nascent DNA behind the polymerizing enzyme (top panel). The second step is the displacement of the donor RNA by the acceptor template (bottom panel). The structure of the acceptor template is supposed to be crucial at this stage.

www.nature.com/reviews/molcellbio

| FEBRUARY 2001 | VOLUME 2

PERSPECTIVES would stall at this position, leading to an arrest of reverse transcription similar to that shown in BOX 1a. Strand transfer then proceeds through a mechanism analogous to the one described for forced copy-choice. In both cases described above, strand transfer is promoted indirectly — by the difficulty in continuing synthesis. Another possibility is that transferring synthesis on the acceptor RNA becomes, in itself, preferable to processive copying. A prerequisite for the switch would be a close proximity between the acceptor RNA and the catalytic site of a reverse transcriptase engaged in DNA synthesis on the donor RNA. The switch would then be determined by a displacement of the donor RNA by the acceptor template. Kim and co-workers15 proposed a model to account for efficient strand transfer taking place in the descending portion of the stem of the transactivation response element of HIV1. The proximity between donor and acceptor RNAs is thought to be determined by an interaction in a short region of the hairpin structure, ahead of the reverse transcriptase (‘interactive hairpin’ model; BOX 1b). During reverse transcription of this region, the acceptor RNA eventually diverts DNA synthesis, leading to template switching15. These observations, made on short model templates, indicate that pausing of reverse transcription or RNA hairpin structures can promote template switching. However this does not imply that, during reverse transcription of the much longer genomic RNAs, recombination preferentially occurs at pause sites or on RNA hairpins. To address this issue, long RNAs are needed, for which the dynamics of alternative folding are more complex than on short templates. Furthermore, reverse transcriptase and RNA are not the only components of the reverse-transcription complex. In vivo, reverse transcription occurs on a ribonucleoprotein complex. The most abundant component of this is the RNA-binding nucleocapsid protein3,16, which is known to enhance strand transfer15,17–21. The nucleocapsid protein behaves as an ‘RNA chaperone’, promoting the intermolecular annealing of complementary nucleic acids and the transient intramolecular rearrangement of RNA secondary structures22–25. We have addressed these issues by doing a random search for recombination events on long RNA templates either naked or coated with RNA chaperones26. We observed that template switching preferentially occurs at certain positions (‘hotspots’) on the RNA. These hotspots do not correlate with the positions at which reverse transcription pauses. Instead, they are located in strongly structured

extents. We discuss here what, in our opinion, might be the contribution of each mechanism to genetic recombination in vivo.

5′

Polymerase active site

3′ Fingers Thumb

RNase H Palm 3′

5′

Figure 3 | A pre-recombination intermediate? The catalytic subunit of HIV-1 reverse transcriptase is depicted as in FIG. 2. An acceptor RNA (brown) annealed to the nascent DNA behind the RNaseH catalytic site is shown. The transient path followed by the acceptor RNA is arbitrary, and is meant to rise above the plane of the drawing. This picture, a more detailed representation of the top panel in BOX 1c, mimics a hypothetical situation preceding strand invasion of the primer–template hybrid by the acceptor RNA. This step could be favoured by a molecule of nucleocapsid protein bound to the acceptor RNA, as proposed in REF. 26.

regions of the template. The nucleocapsid protein enhances recombination, modulating the distribution of the positions of strand transfer, an effect essentially exerted at the level of the acceptor RNA26. We proposed that strand transfer occurs through two distinct steps: docking of the acceptor RNA onto the nascent DNA behind the reverse transcriptase; and strand invasion near the catalytic site of the enzyme, with replacement of the donor RNA (BOX 1c). We suggested that this step is mediated by a transient interaction with the reverse transcriptase. A prevailing mechanism?

In vivo recombination occurs along the whole genome27, suggesting that triggers for template switching are frequently found within the genomic RNA. As discussed above, retroviral recombination seems to follow several alternative paths in vitro. Because these models are not mutually exclusive, a realistic hypothesis is that different mechanistic traits might promote template switching to various

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

Forced copy-choice. This process closely resembles the obligatory strand transfer on the repeated terminal region R of the genomic RNA (strong-stop strand transfer), with the difference that it can occur on any sequence of the genome. Template switching is caused by a nick on the RNA, and the level of integrity of the viral RNA at the moment of reverse transcription is not known. It is therefore difficult to estimate the frequency of forced copy-choice in vivo. However, strand transfer occurs in vitro at high rates in the absence of manifest breaks on the RNA (about 2–5 x 10–4 per nucleotide26). So although forced copychoice is crucial for rescue synthesis, its contribution to genetic variability might not be predominant, except when viral genomes are considerably damaged. Pause-driven recombination. At first glance, this process resembles forced copy-choice, but the nature of the obstacle opposed to reverse transcription is different. For both models, an asymmetry is required between the donor and the acceptor templates, because the obstacle to reverse transcription must be absent on the acceptor RNA. In forced copy-choice, this asymmetry is obvious, as the probability of the donor and acceptor RNAs being broken at the same position is low. By contrast, in pausedriven recombination, whatever causes reverse transcription to pause on one RNA is intrinsic to the sequence (see above) and will also do the same on its homologous counterpart (the ‘stop’ signal in BOX 1a should be drawn on both templates). So transferring the nascent DNA on the acceptor template will not solve the problem of restoring DNA synthesis, unless this symmetry is broken. How could such a symmetry-breaking event occur? As reverse transcription progresses, the RNaseH activity degrades the donor RNA (FIG. 2). Portions of the template involved in forming secondary structures on the donor could therefore be destroyed, altering the global folding of the molecule28. Hence, although the primary sequences of the donor and acceptor RNAs are the same in terms of the position of transfer, their secondary structures might differ — one having been reverse transcribed and the other not. This could accidentally create a better structure for resumption of reverse transcription on the acceptor RNA. However, we find it difficult to imagine that all these requirements are recurrently fulfilled along the genomic RNA. Rather, we feel that pause-driven

VOLUME 2 | FEBRUARY 2001 | 1 5 3

PERSPECTIVES a

LTR

R U5 PBS 5′ tRNA

U3 R

Non-LTR AAA

3′

5′ Beginning of reverse transcription

LTR

AAA

LTR

Final DNA product

AAA New genomic RNA

3′

AAA

AAA New 'genomic' RNA

b 1

LTR

Non-LTR

5 6 7

5 6

scriptase (FIG. 3). It could compete with the donor RNA for an interaction with the enzyme. A similar situation has already been shown to occur during in vitro experiments that mimic strong-stop strand transfer10. We suggest that a similar interaction takes place here, stabilizes the recombination intermediate, and favours strand exchange at the catalytic site. We reason that the various local secondary structures adopted by the acceptor RNA could be crucial for such interactions, or simply for positioning the acceptor RNA close to the active site, favouring strand displacement. These features could account for most of the site preferences shown by copy-choice in vitro26. Although it is difficult at present to test these hypotheses, a search for a ‘consensus’ structure at recombination hotspots might allow these issues to be addressed in the near future.

50 aa

Figure 4 | Reverse transcriptases from retroelements. a | Left part. In retroviruses and retroelements that will form long terminal repeats (LTRs), reverse transcription begins near the 5′ end using a tRNA hybridized to a region of the viral genome called the primer binding site (PBS) and soon reaches this end of the RNA molecule. The presence of the repeated sequence R (in yellow) at both ends of the RNA allows transfer of the nascent DNA to the 3′ end of the genome (strong-stop strand transfer). The integrated form of the virus (provirus) is flanked by the LTRs. These are generated during reverse transcription, and contain the repeated sequence R and the unique sequences adjacent to it on the viral RNA (U5 in grey, and U3 in green, respectively). Synthesis of the new genomic RNAs will occur from a promoter in the U3 region. (Dark blue, cellular genomic DNA (the poly-A sequence is drawn as three As at the 3′ end of the RNA)). Right part. By contrast, non-LTR elements begin reverse transcription near the 3′ end of their RNA (using various primers39) and do not require strand transfer. The example in the drawing represents replication of the human long interspersed element (LINE). Synthesis of the RNA will occur from an internal promoter located at the 5′ end of the element. b | Sequence alignment of the primary structure of the catalytic core of the reverse transcriptases from various retroelements. There are seven conserved regions (grey) situated in the ‘finger’ and ‘palm’ domains33. The positions of the catalytic residues involved in DNA polymerization are indicated by stars. All reverse transcriptases from LTR retroelements and retroviruses have a compact organization (top part of figure), with short ‘spacer’ sequences among the conserved domains (in yellow). Conversely, reverse transcriptases from non-LTR retroelements have the organization shown in the bottom part of the figure. The bar represents 50 amino acids.

recombination applies to isolated cases and that it is not a prevailing mechanism. This view is supported by the poor correlation between recombination hotspots and the pause sites observed in vitro on long RNAs26. Active displacement. In our opinion, the most likely possibility is that the acceptor template efficiently displaces the donor RNA without requiring strong pauses of reverse transcription. However, this raises a new problem. For forced copy-choice and pausedriven strand transfer, the hybrid formed by the nascent DNA and the donor RNA could melt before the nascent DNA re-hybridizes with the acceptor RNA. But if no pausing of DNA synthesis is required, the reverse transcriptase active site would already be occupied by the donor template annealed to the nascent DNA (FIG. 2).

154

How can the acceptor RNA displace the donor template? We reason that, in accordance with the scheme presented in BOX 1c, a close proximity between donor and acceptor RNAs can be easily achieved on long RNAs through a ‘docking step’ that occurs behind the reverse transcriptase. Because the acceptor template, in contrast to the donor one, is not degraded by the RNaseH activity of the reverse transcriptase, annealing of the growing DNA strand onto the acceptor RNA yields a more extended — and therefore more stable — hybrid. Once docking has occurred on the trailing part of the nascent DNA (BOX 1c), strand exchange becomes an intramolecular process, which is far more efficient than an intermolecular reaction. The acceptor RNA, docked onto the nascent DNA behind the RNaseH active site, constantly ‘hangs around’ the reverse tran-

Implications of retroviral recombination

During a single cycle of replication, recombination reshuffles several nucleotide substitutions into a new combination. So it allows sudden and large sequence perturbations. This is a crucial parameter to take into consideration when trying to understand the evolution of retroviruses. Traditional analyses of phylogenetic data rely on the assumption that the rates of appearance and fixation of point mutations have the dominant role in the divergence of viral subtypes. On this basis, the data cannot be reconciled with another simplifying assumption — that the mutation rate of retroviruses is constant. It was proposed that this rate varies among retroviral lineages and over time. But it has recently been argued29 that these variations are not significant, and that they merely reflect the occurrence of recombination within viral populations. Neglecting recombination might therefore have led to incorrect conclusions on crucial issues about the AIDS pandemic, such as the dating of its origin or the time at which new subtypes emerged29. It will, however, be difficult to take into account the incidence of recombination when trying to establish evolutionary trees. First, because the recombination rates along a model template fluctuate within a fivefold to eightfold range26,27, in principle it would be more appropriate to apply corrections, segment by segment, within the genome considered. Furthermore, the efficiency of recombination also probably depends on the extent of local homology between the two genomic RNAs in the region of transfer. When the homology is completely lost, the www.nature.com/reviews/molcellbio

| FEBRUARY 2001 | VOLUME 2

PERSPECTIVES rate of recombination drops by a factor of 100 to 1,000 (REF. 30) compared with the values found for strictly homologous sequences. Between these two extreme cases, the efficiency of recombination is expected to drop as the local degree of genetic heterogeneity is increased. Therefore, for hypervariable segments of genomes, the effect of recombination could be particularly weak. The evolution of such sequences would be mostly determined by point mutations. Small populations of highly variable viruses are known to incorporate mutations, most of them resulting in a lower fitness31. These mutations, which accumulate in the population under non-competitive conditions, lead to a sharp restriction of viral load when a strong selective pressure is applied, owing to the elimination of these ‘suboptimal’ classes of genome32. Recombination might allow the rapid resurgence of infectious strains in response to drastic environmental changes32. From this standpoint we could speculate that, if recombination occurs more efficiently on certain sequences, it would lead to the preferential emergence of specific subclasses of recombinants. As discussed above, it is possible that the sequence requirements for efficient retroviral recombination will be understood in the future. Such an achievement could have practical consequences for optimizing the generation of retroviral vectors for gene therapy or vaccine applications, because unwanted genomic rearrangements could be considerably reduced. Another implication of the existence of specific sequences that lead to efficient recombination is more evolutionary, suggesting a relationship between structural features of the reverse transcriptase and its ability to switch template. Retroviruses belong to a family of mobile genetic elements called retrotransposons, which propagate by reverse transcription. Some retrotransposons replicate as retroviruses, a strategy that includes an obligatory step of strand transfer on the repeated sequence R (strong-stop strand transfer), whereas others do not require such a step (FIG. 4). An evolutionary analysis of the primary sequences of the reverse transcriptases from these elements shows a good correlation between the modular organization of the core of these enzymes and the presence of an obligatory strand transfer during reverse transcription33 (FIG. 4). It is therefore tempting to imagine that the need to switch template efficiently is a strong constraint, which has led to a given modular organization being maintained in the catalytic core of the

enzyme. If this were the case, genetic recombination and the choice of a particular strategy of reverse transcription in retroviruses could turn out to be strictly correlated. These considerations, although speculative, suggest that a better understanding of retroviral recombination might shed light not only on its effect on retroviral genomic variability, but also on how the ability of reverse transcriptases to switch template branched out from their main task of perpetuating genetic information.

18.

19.

20.

Matteo Negroni and Henri Buc are in the Department of Molecular Biology, Institut Pasteur, 75724 Paris cedex 15, France. e-mail: matteo@pasteur.fr

21.

Links

22.

ENCYCLOPEDIA OF LIFE SCIENCES Retroviral

replication 1. 2.

10.

11.

12.

13.

14.

15.

16.

17.

Rein, A. Retroviral RNA packaging: a review. Arch. Virol. Suppl. 9, 513–522 (1994). Gilboa, E., Mitra, S. W., Goff, S. & Baltimore, D. A detailed model of reverse transcription and tests of crucial aspects. Cell 18, 93–100 (1979). Coffin, J. M., Hughes, S. H. & Varmus, H. E. Retroviruses (Cold Spring Harbor Laboratory Press, Cold Spring Harbor, New York, 1997). Goodrich, D. W. & Duesberg, P. H. Retroviral recombination during reverse transcription. Proc. Natl Acad. Sci. USA 87, 2052–2056 (1990). Hu, W. S. & Temin, H. M. Genetic consequences of packaging two RNA genomes in one retroviral particle: pseudodiploidy and high rate of genetic recombination. Proc. Natl Acad. Sci. USA 87, 1556–1560 (1990). Coffin, J. M. Structure, replication, and recombination of retrovirus genomes: some unifying hypotheses. J. Gen. Virol. 42, 1–26 (1979). Elder, J. H. et al. Biochemical evidence that MCF murine leukemia viruses are envelope (env) gene recombinants. Proc. Natl Acad. Sci. USA 74, 4676–4680 (1977). Blair, D. G. Genetic recombination between avian leukosis and sarcoma viruses. Experimental variables and the frequencies of recombination. Virology 77, 534–544 (1977). Sharp, P. M., Bailes, E., Robertson, D. L., Gao, F. & Hahn, B. H. Origins and evolution of AIDS viruses. Biol. Bull. 196, 338–342 (1999). Peliska, J. A. & Benkovic, S. J. Mechanism of DNA strand transfer reactions catalysed by HIV-1 reverse transcriptase. Science 258, 1112–1118 (1992). DeStefano, J. J., Mallaber, L. M., Rodriguez-Rodriguez, L., Fay, P. J. & Bambara, R. A. Requirements for strand transfer between internal regions of heteropolymer templates by human immunodeficiency virus reverse transcriptase. J. Virol. 66, 6370–6378 (1992). DeStefano, J. J., Bambara, R. A. & Fay, P. J. The mechanism of human immunodeficiency virus reverse transcriptase-catalysed strand transfer from internal regions of heteropolymeric RNA templates. J. Biol. Chem. 269, 161–168 (1994). Wu, W., Blumberg, B. M., Fay, P. J. & Bambara, R. A. Strand transfer mediated by human immunodeficiency virus reverse transcriptase in vitro is promoted by pausing and results in misincorporation. J. Biol. Chem. 270, 325–332 (1995). Suo, Z. & Johnson, K. A. Effect of RNA secondary structure on the kinetics of DNA synthesis catalysed by HIV-1 reverse transcriptase. Biochemistry 36, 12459–12467 (1997). Kim, J. K., Palaniappan, C., Wu, W., Fay, P. J. & Bambara, R. A. Evidence for a unique mechanism of strand transfer from the transactivation response region of HIV-1. J. Biol. Chem. 272, 16769–16777 (1997). Darlix, J. L., Lapadat-Tapolsky, M., de Rocquigny, H. & Roques, B. P. First glimpses at structure-function relationships of the nucleocapsid protein of retroviruses. J. Mol. Biol. 254, 523–537 (1995). Negroni, M. & Buc, H. Recombination during reverse

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

23.

24.

25. 26.

27.

28.

29.

30.

31. 32. 33.

34. 35.

36.

37.

38.

39.

transcription: an evaluation of the role of the nucleocapsid protein. J. Mol. Biol. 286, 15–31 (1999). Guo, J., Henderson, L. E., Bess, J., Kane, B. & Levin, J. G. Human immunodeficiency virus type 1 nucleocapsid protein promotes efficient strand transfer and specific viral DNA synthesis by inhibiting TAR-dependent selfpriming from minus-strand strong-stop DNA. J. Virol. 71, 5178–5188 (1997). Peliska, J. A., Balasubramanian, S., Giedroc, D. P. & Benkovic, S. J. Recombinant HIV-1 nucleocapsid protein accelerates HIV-1 reverse transcriptase catalysed DNA strand transfer reactions and modulates RNase H activity. Biochemistry 33, 13817–13823 (1994). Rodriguez-Rodriguez, L., Tsuchihashi, Z., Fuentes, G. M., Bambara, R. A. & Fay, P. J. Influence of human immunodeficiency virus nucleocapsid protein on synthesis and strand transfer by the reverse transcriptase in vitro. J. Biol. Chem. 270, 15005–15011 (1995). Allain, B., Lapadat-Tapolsky, M., Berlioz, C. & Darlix, J. L. Transactivation of the minus-strand DNA transfer by nucleocapsid protein during reverse transcription of the retroviral genome. EMBO J. 13, 973–981 (1994). Clodi, E., Semrad, K. & Schroeder, R. Assaying RNA chaperone activity in vivo using a novel RNA folding trap. EMBO J. 18, 3776–3782 (1999). Tsuchihashi, Z. & Brown, P. DNA strand exchange and selective DNA annealing promoted by the human immunodeficiency virus type 1 nucleocapsid protein. J. Virol. 68, 5863–5870 (1994). Rein, A., Henderson, L. E. & Levin, J. G. Nucleic-acidchaperone activity of retroviral nucleocapsid proteins: significance for viral replication. Trends Biochem. Sci. 23, 297–301 (1998). You, J. C. & McHenry, C. S. HIV nucleocapsid protein. J. Biol. Chem. 268, 16519–16527 (1993). Negroni, M. & Buc, H. Copy-choice recombination by reverse transcriptases: reshuffling of genetic markers mediated by RNA chaperones. Proc. Natl Acad. Sci. USA 97, 6385–6390 (2000). Jetzt, A. E. et al. High rate of recombination throughout the human immunodeficiency virus type 1 genome. J. Virol. 74, 1234–1240 (2000). Suo, Z. & Johnson, K. A. RNA secondary structure switching during DNA synthesis catalysed by HIV-1 reverse transcriptase. Biochemistry 36, 14778–14785 (1997). Schierup, M. H. & Hein, J. Consequences of recombination on traditional phylogenetic analysis. Genetics 156, 879–891 (2000). Zhang, J. & Temin, H. M. Rate and mechanism of nonhomologous recombination during a single cycle of retroviral replication. Science 259, 234–238 (1993). Chao, L. Fitness of RNA virus decreased by Muller’s ratchet. Nature 348, 454–455 (1990). Muller, H. J. The relation of recombination to mutational advance. Mutat. Res. 1, 2–9 (1964). Xiong, Y. & Eickbush, T. H. Origin and evolution of retroelements based upon their reverse transcriptase sequences. EMBO J. 9, 3353–3362 (1990). Coffin, J. M. in Fields Virology (eds Fields, B. N. et al.) 1437–1500 (Raven Press, New York, 1990). Kohlstaedt, L. A., Wang, J., Friedman, J. M., Rice, P. A. & Steitz, T. A. Crystal structure at 3.5 Å resolution of HIV1 reverse transcriptase complexed with an inhibitor. Science 256, 1783–1790 (1992). Furfine, E. S. & Reardon, J. E. Reverse transcriptase. RNase H from the human immunodeficiency virus. Relationship of the DNA polymerase and RNA hydrolysis activities. J. Biol. Chem. 266, 406–412 (1991). Jacobo-Molina, A. et al. Crystal structure of human immunodeficiency virus type 1 reverse transcriptase complexed with double-stranded DNA at 3.0 Å resolution shows bent DNA. Proc. Natl Acad. Sci. USA 90, 6320–6324 (1993). Huang, H., Chopra, R., Verdine, G. L. & Harrison, S. C. Structure of a covalently trapped catalytic complex of HIV-1 reverse transcriptase: implications for drug resistance. Science 282, 1669–1675 (1998). Levin, H. L. It’s prime time for reverse transcriptase. Cell 88, 5–8 (1997).

Acknowledgements We acknowledge the French Agency for Research on AIDS (ANRS) for financial support. We are also indebted to M. Ricchetti for constant and helpful discussions, and to T. Heidmann for his help in elaborating concepts on the biology of retroelements.

VOLUME 2 | FEBRUARY 2001 | 1 5 5