Nature Reviews - Molecular Cell Biology - October 2000 vol1 December 2000 vol1 by Nature Magazine

toc_r

Page 1 of 2

CONTENTS

December 2000 Vol 1 No 3

159 | In this issue doi:10.1038/35043040

169 | HOW CELLS READ TGFSIGNALS Joan Massagué doi:10.1038/35043051

Highlights PDF

[1587K]

161 | MEMBRANE DYNAMICS Independence day doi:10.1038/35043000

162 | IN THE NEWS Don't shoot the messenger doi:10.1038/35043003

162 | CELL ADHESION How to lead a double life

[1012K]

179 | THE MANY SUBSTRATES AND FUNCTIONS OF ATM Michael B. Kastan & Daesik Lim doi:10.1038/35043058 [356K]

187 | THREE WAYS TO MAKE A VESICLE Tomas Kirchhausen doi:10.1038/35043117 [2973K]

doi:10.1038/35043005

162 | RNA LOCALIZATION She's got a ticket to ride doi:10.1038/35043008

199 | P63 AND P73: P53 MIMICS, MENACES AND MORE Annie Yang & Frank McKeon doi:10.1038/35043127 [1550K]

163 | CIRCADIAN RHYTHMS Working the night shift doi:10.1038/35043010

163 | DEVELOPMENT Divide and rule doi:10.1038/35043013

164 | WEB WATCH Sweet talk doi:10.1038/35043045

164 | P53 REGULATION Tipping the balance doi:10.1038/35043015

208 | ARE DESMOSOMES MORE THAN TETHERS FOR INTERMEDIATE FILAMENTS? Kathleen J. Green & Claire A. Gaudry doi:10.1038/35043032 [1188K]

217 | INTRAMEMBRANE PROTEOLYSIS BY PRESENILINS Harald Steiner & Christian Haass doi:10.1038/35043065 [1194K]

164 | DIFFERENTIATION Molecular alchemy doi:10.1038/35043017

165 | TECHNIQUE A certain FLAIR doi:10.1038/35043020

225 | TIMELINE KREBS AND HIS TRINITY OF CYCLES Hans Kornberg doi:10.1038/35043073 [331K]

165 | IN BRIEF T-CELL SIGNALLING | CELL DIVISION | CELLULAR MICROBIOLOGY | PROTEIN METHYLATION doi:10.1038/35043047

166 | IN BRIEF MEIOSIS | CHROMOSOME

228 | TIMELINE THE ELUSIVE CYTOSTATIC FACTOR IN THE ANIMAL EGG Yoshio Masui doi:10.1038/35043096 [1301K]

http://www.nature.com.ezproxy.library.wisc.edu/nrm/journal/v1/n3/toc_r.html

12/3/2006

toc_r

Page 2 of 2

BIOLOGY | CYTOSKELETON doi:10.1038/35043049

166 | DNA REPAIR A taxing question

233 | OPINION HUMAN CANCER CELL LINES: FACT AND FANTASY John R. W. Masters doi:10.1038/35043102 [311K]

doi:10.1038/35043023

237 | NatureView

167 | PROTEIN METABOLISM Gates of destruction

doi:10.1038/35043105

doi:10.1038/35043026

167 | TRANSLATION Unmasking the effects of CPEB doi:10.1038/35043029

NATURE REVIEWS | MOLECULAR CELL BIOLOGY

http://www.nature.com.ezproxy.library.wisc.edu/nrm/journal/v1/n3/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 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

M E M B R A N E DY N A M I C S

Independence day Virtually every aspect of Golgi biology is surrounded by some controversy, and they all boil down to one basic question: is the Golgi subordinate to the endoplasmic reticulum or is it an independent organelle? Graham Warren and colleagues have now found some ingenious ways to show that the Golgi is an independent organelle. Why should we think that the Golgi depends on the ER in the first place? Under conditions in which ER-to-Golgi transport is blocked, the Golgi stack completely disappears and Golgi enzymes end up in the ER. This means that Golgi enzymes cycle constitutively between the two organelles, and raises the possibility that the Golgi could be a transient structure. If Golgi biogenesis and function depend on the ER, one prediction is that it should be possible to recreate a functional Golgi from the ER. Pelletier et al. report in Nature Cell Biology that this is not the case. They used microsurgery to create Golgifree cell pieces that contain parts of the ER. In these cell pieces, newly synthesized proteins could be transported out of the ER, but they did not make it to the plasma membrane. In a parallel study published in Nature, Seemann et al. tested whether the ER is necessary to form a Golgi structure, and found that it’s not. They first treated cells with the fungal metabolite brefeldin A, which fragments the Golgi stack and sends Golgi enzymes back to the ER. This is

a reversible process, and removing brefeldin A leads to the reformation of a functional Golgi. But while washing out brefeldin A, the authors injected the cells with a dominantnegative construct of a small GTPase, which inhibits exit from the ER. Despite the absence of traffic from the ER to the Golgi, structures were formed that resembled a Golgi, were properly localized in the cell and contained some Golgi matrix proteins. But they were devoid of oligosaccharide-modifying Golgi enzymes, which remained stuck in the ER. So the ER seems to be neither sufficient nor necessary for Golgi structure and function. The picture that emerges is that the Golgi is an independent organelle containing

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

recycling enzymes as well as resident matrix proteins. The resident proteins have all the necessary information for Golgi structure and location, and form a scaffold around which the Golgi enzymes assemble to modify cargo in transit. Will this new view of the Golgi survive the test of time? Raluca Gagescu References and links ORIGINAL RESEARCH PAPER Pelletier, L.,

Jokitalo, E. & Warren, G. The effect of Golgi depletion on exocytic transport. Nature Cell Biol. 2, 840–845 (2000) | Seemann, J., Jokitalo, E., Pypaert, M. & Warren, G. Matrix proteins can generate the higher order architecture of the Golgi apparatus. Nature 407, 1022–1026 (2000) FURTHER READING Pelham, H. R. B. & Rothman, J. E. The debate about transport in the Golgi — two sides of the same coin? Cell 102, 713–719 (2000) WEB SITE Graham Warren’s laboratory

VOLUME 1 | DECEMBER 2000 | 1 6 1

HIGHLIGHTS

IN THE NEWS Don’t shoot the messenger The release in late October of the Phillips report — the 18volume result of a three-year investigation into how Britain’s bovine spongiform encephalopathy (BSE) crisis was handled — sparked a wave of media coverage in the United Kingdom. The report concludes that ministers and civil servants were acting under the genuine belief that the risks of eating British beef were minimal. But there have been attempts by the British press to lay some of the blame on another doorstep — that of the scientists. A feature in The Guardian, for example, preaches that “BSE is the latest crisis to dent public faith in those who should know better”, citing genetically modified foods, fluoride in water and antibiotics in foods as other instances of scientific “scare stories”. In the case of BSE, however, the article claims that scientists are under fire not for “their pure scientific method, or even their conclusions”, but for “the way they allowed themselves and their opinions to be manipulated by civil servants”. Writing in The Independent on Sunday, Geoffrey Lean dubs the crisis “a kind of Stockholm syndrome — where captives come to identify with those who take them hostage — in reverse. Seduced by a hazard they are supposed to be controlling, regulators come to believe that it poses no threat”. He groups scientists among those who fell into this trap, but also speaks out against the “villification and marginalisation of … dissident scientists”, citing Professor Derek Bryce-Smith as an example. He, apparently, warned long ago about the dangers of leaded petrol — a far cry from BSE, but in the news again owing to the fuel crisis that has swept across Europe over the past few months.

CELL ADHESION

How to lead a double life Just like W. E. Hill’s famous optical illusion (see picture), ephrins have mastered the art of being simultaneously attractive and repulsive. Three papers, in Nature, EMBO Journal and The Journal of Biological Chemistry provide clues as to how these signalling molecules lead their double life. Eph receptor tyrosine kinases and their membrane-tethered ligands, the ephrins, provide guidance cues for developing neurons and blood vessels. The classical view of ephrin signalling is that it mediates repulsion between ephrin-expressing and Ephexpressing cells, but this relationship doesn’t always hold true. Johan Holmberg and colleagues found such a case when they knocked out the gene for ephrin-A5 in mice: some of the mice lacked brains because the neural tube had failed to close at the cranial end — probably due to failure of an adhesive, rather than a repulsive, signal. In situ hybridization in wild-type mouse embryos revealed that ephrinA5 and three splice variants of its receptor, EphA7, are expressed at the edges of the cranial neural folds as the neural tube closes, and cells dissected from them adhered more tightly to EphA7-coated surfaces if they expressed ephrin-A5. But do all the EphA7 splice variants behave in the same way? In chemotactic assays,

cells expressing full-length EphA7 actually repelled ephrin-A5, but when expression of a truncated splice form of EphA7 (EphA7-T1) was switched on, the repulsive effect was blocked. But there’s more to this response than just inhibition of repulsion, because when cells expressing the two EphA7 variants were plated out on confluent layers of a mixture of ephrin-A5+ and ephrin-A5– cells, the cells expressing the EphA7-T1 grew preferentially on the ephrin-A5+ cells, whereas those expressing the full-length receptor preferred ephrinA5– cells. The most likely mechanism is that the truncated receptor acts in a dominant-negative manner, a model supported by the finding that EphA7-T1 expression reduces tyrosine phosphorylation of the full-length receptor.

R N A LO C A L I Z AT I O N

She’s got a ticket to ride Ash1p is a transcriptional repressor necessary for matingtype switching in budding yeast, and its messenger RNA is transported into the bud where the protein is ultimately needed. To get there, the mRNA rides into the bud along actin tracks, carried by the myosin V-type motor Myo4p. But how does the

RNA hang on to the motor? Ralf-Peter Jansen and colleagues now report in the EMBO Journal that they have found the missing link. Genetic screens had determined that She2p and She3p are involved in ASH1 mRNA localization — but what exactly do they do? Jansen and colleagues found,

Another peculiarity of ephrin signalling is that it’s bidirectional: engagement of Eph receptors transmits signals to the ligand-expressing cells, as well as the receptor-expressing cells. This is all the more intriguing for the A-type ephrins, which are tethered by a glycosylphosphatidylinositol anchor. How can this transmit signals to the cell’s interior? Davy and Robbins find that ephrinA5, activated by the extracellular

through a biochemical approach, that She2p is the long-sought RNA-binding protein, specific for ASH1 mRNA. Furthermore, they showed that She3p is an adaptor that binds She2p through its carboxyl terminus and Myo4p through its amino terminus. Whereas She3p is associated with this myosin motor constitutively, and can be transported into the bud even in the absence of ASH1 mRNA, She2p needs to bind ASH1 mRNA to interact efficiently with She3p.

Alison Mitchell

162

www.nature.com/reviews/molcellbio

| DECEMBER 2000 | VOLUME 1

HIGHLIGHTS domain of the EphA5 receptor, stimulates adhesion to fibronectin and laminin, with subsequent morphological changes. These effects are blocked by an antibody against β1integrin. Experiments with inhibitors implicate mitogen-activated protein kinases and Src-family protein kinases in this response. Huai and Drescher have used a similar system to fish for molecules downstream of ephrin-A activation, and find a mysterious 120 kDa protein that becomes phosphorylated on a tyrosine residue before integrin-mediated adhesion occurs. We’re beginning to paint — albeit in broad brushstrokes — a picture of how ephrins and Eph receptors communicate their mixed messages, but now we must focus in on the fine detail. Are the adhesive effects of truncated Eph receptors due to ‘reverse’ signalling to ephrin-A-expressing cells? What’s the order of events upon ephrin activation, and what’s the identity of the 120 kDa stranger lurking in the shadows? Cath Brooksbank References and links ORIGINAL RESEARCH PAPER Holmberg, J.,

Clarke, D. L. & Frisen, J. Regulation of repulsion versus adhesion by different splice forms of an Eph receptor. Nature 408, 203–206 (2000) | Davy, A. & Robbins, S. M. Ephrin-A5 modulates cell adhesion and morphology in an integrin-dependent manner. EMBO J. 19, 5396–5405 (2000) | Huai, J. & Drescher, U. An Ephrin A-dependent signalling pathway controls integrin function and is linked to the tyrosine phosphorylation of a 120 kDa protein. J. Biol. Chem. (2000) FURTHER READING Frisen, J., Holmberg, J. & Barbacid, M., Ephrins and their Eph receptors: multitalented directors of embryonic development. EMBO J. 18, 5159–5165 (1999)

So it seems that the transport of ASH1 mRNA is pretty much solved. But how many other mRNAs might be localized in yeast, and should we expect similar chains of linkers for each of them? Raluca Gagescu References and links ORIGINAL RESEARCH PAPER Böhl, F. et al. She2p, a novel RNA-binding protein tethers ASH1 mRNA to the Myo4p myosin motor via She3p. EMBO J. 19, 5514–5524 (2000) FURTHER READING Jansen, R. P. RNAcytoskeletal associations. FASEB J. 13, 455–466 (2000)

CIRCADIAN RHYTHMS

D E V E LO P M E N T

Working the night shift

Divide and rule

Circadian clocks are regulated with the precision of a Swiss watch. They continue to tick without external cues, but they can also be reset (or ‘entrained’) by environmental signals such as light–dark cycles. But how? In December’s Nature Neuroscience, Paolo SassoneCorsi, David Allis and colleagues link light entrainment to the dynamic remodelling of chromatin. In mammals, the nerve centre of the circadian clock is the hypothalamic suprachiasmatic nucleus (SCN). Animals kept in darkness and given a pulse of light during the ‘subjective night’ (the time of day corresponding to the dark period in a normal light–dark cycle) show a phase-shifting of normal rhythms, accompanied by the expression of various clock and immediate-early genes in the SCN. These changes in gene expression are thought to be responsible for the light entrainment. What controls the dynamic regulation of these genes at the chromosomal level? To investigate, Sassone-Corsi and colleagues studied the effects of a night-time light pulse on phosphorylation of histone H3 — a central event in the remodelling of chromatin. They observed such light-induced phosphorylation in the SCN of mice, but not in other structures tested. The authors next showed that a time course of H3 phosphorylation parallels the induction profile of an early-response gene — c-fos — in the same SCN neurons, indicating that the two events are linked. Moreover, when mice were given baclofen, a drug that inhibits light-induced phase-shifts during the subjective night, both phosphorylation of histone H3 and expression of c-fos were reduced. Again, the implication is that one pathway controls both events. These results may indicate, conclude the authors, that “dynamic chromatin remodelling in the SCN occurs in response to a physiological stimulus in vivo”. And the next step in unravelling this complex molecular clockwork will be to identify the light-induced kinase that is responsible.

Cell-cycle control in eggs and early embryos has its own set of rules. In Drosophila melanogaster, the first mitosis is delayed until after fertilization, after which S phase and M phase are alternated for thirteen simplified cell cycles. Maternal mutations in three genes — pan gu (png), plutonium (plu) and giant nuclei (gnu) — disrupt this control, yielding eggs that replicate their DNA before fertilization, and embryos with fewer, larger nuclei; but how? A paper in Development provides the first mechanistic insights into this conundrum.

“In the beginning, the heavens and earth were still one and all was chaos. The universe was like a big black egg, carrying Pan Gu inside itself. After 18 thousand years Pan Gu woke from a long sleep. He felt suffocated, so he took up a broadax and wielded it with all his might to crack open the egg.” Ancient Chinese legend Fenger and colleagues found that png encodes a serine/threonine protein kinase expressed in the early embryo. It binds PLU and the complex has kinase activity in vitro. How does this complex control the cell cycle? A clue came from examining png mutant embryos: these have lower levels of cyclins A and B, and lower CDC2 kinase activity, so they carry on synthesizing DNA but can’t undergo mitosis. png, therefore, is needed to limit S phase and promote mitosis, which can be achieved by maintaining mitotic cyclin levels. However, no direct interaction between PNG–PLU and cyclins or CDC2 was found. Instead, the authors speculate, PNG–PLU might stabilize cyclins by blocking their access to the protein degradation machinery. Many questions remain; for example, what are PNG’s substrates? And does GNU interact with PNG–PLU? But one mystery has been solved: in vertebrates, the dual-function kinase Mos (see the Timeline by Yoshio Masui on page 228) prevents division of unfertilized eggs. Invertebrates don’t have Mos, but we now know that PNG can carry out some of its functions in flies.

Alison Mitchell References and links

Cath Brooksbank References and links

ORIGINAL RESEARCH PAPER Crosio, C. et al. Light induces chromatin

ORIGINAL RESEARCH PAPER Fenger, D. D. et al. PAN GU: a

modification in cells of the mammalian circadian clock. Nature Neurosci. 3, 1241–1247 (2000)

protein kinase that inhibits S phase and promotes mitosis in early Drosophila development. Development 127, 4763–4774 (2000)

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 1 | DECEMBER 2000 | 1 6 3

HIGHLIGHTS

WEB WATCH Sweet talk Do you remember what the difference is between a proteoglycan and a glycoprotein? If you don’t, no need to panic — go to Glycoforum and you’ll find out. If you do, go to Glycoforum anyway and learn something else, as the site contains loads of information. The website originates in Japan, and it’s available in both Japanese and English. Although some of the links are of limited usefulness if you live outside Japan, the site itself is accessible to English speakers. Once your eyes become accustomed to the bright yellow of the homepage, you’ll find all the important links grouped in one place. In fact, it’s one of the few thoroughly ‘clickable’ pages on the site, most others being built like textbook chapters — with the net advantage that you can download the figures to make slides. Undoubtedly, the pièce de résistance is GlycoWord, offering a comprehensive overview of all glycosciencerelated topics. This section is the answer to your prayers if you are preparing a course on lectins or proteoglycans, and it even contains sections about pathology and technology. A more specialized complement to this section is Hyaluronan Today, which features up-to-date information on all aspects of hyaluronan biology, from its structure to its involvement in morphogenesis and tissue remodelling. There are 15 chapters in this section so far, and new chapters are added regularly. The site also provides useful information about meetings as well as a few longish interviews with leaders in the field. The last section, Topics, lacks a bit of focus, featuring for example meeting programmes right next to the product catalogue of Seikagaku Corporation — a company that has strong links with Glycoforum.

Raluca Gagescu

164

P 5 3 R E G U L AT I O N

Tipping the balance Such is the mix-and-match nature of cell biology that many different modes of regulation come up time and again. Take, for example, the addition or removal of an acetyl group, which is a well-known means of regulating histones and turning gene expression on or off. But acetylation can also be used to control the activity of p53, and in the 16 November issue of Nature Wei Gu and colleagues describe one

way in which this process might be regulated. Acetylation by p300 stimulates the ability of p53 to bind DNA in a sequence-specific manner. Because p53 acetylation is enhanced when cells are treated with deacetylase inhibitors, there is likely to be a balance between the two processes, with the deacetylases fine-tuning the levels of acetylated p53. If this is the case, reasoned Gu and colleagues, p53 would be expected to interact physically with a deacetylase complex. They used a glutathione S-transferase (GST)–p53 affinity column to test this hypothesis and, among the eluted proteins, they pulled down the human histone deacetylase HDAC1. The two proteins do not

interact directly, however, as purified recombinant HDAC1 failed to bind p53. So what mediates the interaction? Gu and co-workers purified complexes containing HDAC1, then ran these fractions down the GST–p53 column. Only one protein remained bound after dissociation of the HDAC1 complex, and the authors named this PID (for ‘p53 target protein in the deacetylase complexes’). They then showed, by immunoprecipitation, that the interaction between p53 and PID is specific and direct both in vitro and in vivo. Sequence analysis revealed that PID is homologous to human MTA1, a protein found in the socalled NuRD complexes involved in nucleosome remodelling and histone deacetylation. Confirmation that the p53/PID/HDAC1 complex, too, is a

D I F F E R E N T I AT I O N

Molecular alchemy For over 1,500 years, alchemists struggled to convert base metals into gold. But their search for the legendary ‘philosopher’s stone’ that could make such a transformation resulted in failure and the slide of this pseudoscience into disrepute. Not so in cell biology, however. In the December issue of Nature Cell Biology, David Tosh and colleagues describe an equally startling transition — the conversion of pancreatic cells into hepatocytes, with no intervening cell division. The interconversion of differentiated cells has been observed before. For example, hepatic foci appear in the rat pancreas after various experimental treatments, and Tosh and colleagues chose to study this process in vitro. They used a synthetic glucocorticoid called dexamethasone to convert a rat pancreatic cell line into hepatocyte foci. Using an assay based on green-fluorescent protein, the authors conclude that this conversion occurs

directly from an exocrine cell type to a hepatocyte — a process known as transdifferentiation. Although they cannot rule out the possibility that an intermediate stem cell is involved, this seems unlikely. Moreover, as the transition is not blocked by 5-bromo-2′deoxyuridine (a thymine analogue that can be incorporated into DNA), the authors believe that it need not involve cell division. What, then, is responsible for this transdifferentiation? To find out, Tosh and colleagues studied the expression of transcription factors associated with hepatic differentiation. One such factor, CCAAT-enhancer binding protein β (C/EBPβ), was not detectable in the pancreatic cell line but it could be induced after treatment of these cells with dexamethasone. And, as the figure shows, C/EBPβ could induce the pancreatic cells to transdifferentiate (C/EBPβ is labelled green, with the liver-cell marker glucose-6-phosphatase in red). Tosh and co-workers confirmed

many of their results using an in vitro system that more closely resembles the situation in vivo (pancreatic buds isolated from mouse embryos). Nonetheless, they point out that some findings need further clarification. For example, the authors do not claim that glucocorticoids are involved in the formation of hepatic foci in vivo; only that such treatment

www.nature.com/reviews/molcellbio

| DECEMBER 2000 | VOLUME 1

HIGHLIGHTS

NuRD complex came with the identification of other members, among their number a chromatin-remodelling protein. The final test was to see whether PID is involved in deacetylation in vivo. As expected, expression of PID inhibited p53 acetylation in an HDAC1-dependent manner. It also affected the cellular functions of p53; for example, treatment with PID blocked p53-mediated growth inhibition, expression of endogenous p21 and apoptosis. It seems, then, that Gu and colleagues have identified a protein that tips the balance towards the deacetylation — and, hence, inactivation — of p53, with obvious implications for its role in human cancer. Alison Mitchell References and links ORIGINAL RESEARCH PAPER Luo, J. et al.

Deacetylation of p53 modulates its effect on cell growth and apoptosis. Nature 408, 377–381 (2000)

probably initiates the same molecular pathway. But in starting to identify some of the molecular elements that may be involved in this pathway, the authors may well have struck gold. Alison Mitchell References and links ORIGINAL RESEARCH PAPER Shen, C.-N.,

Slack, J. M. W. & Tosh, D. Molecular basis of transdifferentiation of pancreas to liver. Nature Cell Biol. 2, 879–887 (2000)

IN BRIEF

Rac-GFP

T- C E L L S I G N A L L I N G

ERK5 is a novel type of mitogen-activated protein kinase containing a transcriptional activation domain. FRET

Kasler, H. G., Victoria, J., Duramad, O. & Winoto, A. Mol. Cell. Biol. 20, 8382–8389 (2000)

TECHNIQUE

A certain FLAIR Small GTPases of the Rho family regulate vital cellular processes by controlling actin polymerization. As is the case for other signalling molecules, the effects of these GTPases are very focused in space and time. Wouldn’t it be useful, then, to quantify their spatio–temporal dynamics of activation? Klaus Hahn and colleagues now report in Science how to do this. The authors expressed Rac tagged with green-fluorescent protein (GFP), and microinjected at the same time PDB — a domain of PAK1 that specifically binds to active Rac–GTP — labelled with the dye Alexa. The sites of attachment of the fluorophores were chosen so that, when Rac and PDB interacted, GFP and Alexa would be close enough for fluorescence resonance energy transfer (FRET) to occur. With this experimental set-up, the authors could quantify in real time, in the same moving cell, the changing localization of Rac (GFP–Rac) and of Rac activation (FRET), and found them remarkably uncorrelated. The method was called FLAIR, for ‘fluorescence activation indicator for Rho proteins’. And, as the name says, there is no reason why it should not work for other GTPases of the Rho family, if an appropriate binding partner is chosen instead of PDB. Knowing the when and where of actin polymerization should open many doors for the study of actin-dependent processes, such as cell motility, shape generation or phagocytosis. Raluca Gagescu References and links ORIGINAL RESEARCH PAPER Kraynov, V. S. et al. Localized Rac activation dynamics visualized in living cells. Science 290, 333–337 (2000)

Mitogen-activated protein kinases (MAPKs) can translocate to the nucleus and activate transcription, but their effects have been assumed to be indirect. Kasler and colleagues now characterize a new type of MAPK — ERK5 — that has a potent transcriptional activation domain at its carboxyl terminus, as well as another domain that binds the transcription factor MEF2D. In T cells, maximal activation of MEF2D in response to calcium requires both domains and is blocked by the corepressor Cabin 1. CELL DIVISION

Protein kinase C signalling mediates a program of cell cycle withdrawal in the intestinal epithelium. Frey, M. R. et al. J. Cell Biol. 151, 763–777 (2000)

Why do most cells stop dividing when they differentiate? In developing intestinal epithelial cells, protein kinase C is activated as the cells climb the intestinal crypts. Frey at al. now couple activation of protein kinase C to a coordinated programme of downstream events: downregulation of mitotic cyclins and altered expression and phosphorylation of retinoblastoma protein and its relatives sends these cells down the pathway to quiescence. C E L L U L A R M I C R O B I O LO G Y

A PEST-like sequence in listeriolysin O essential for Listeria monocytogenes pathogenicity. Decatur, A. L. & Portnoy, D. A. Science 290, 992–995 (2000)

After Listeria monocytogenes is engulfed by its host cell, it lyses its phagocytic vacuole with the help of a secreted pore-forming toxin called listeriolysin O. This allows the bacterium to escape into the cytosol, where it can rapidly multiply. But what stops the toxin from destroying its home by lysing the plasma membrane too? The answer is simple: listeriolysin O contains a PEST-like sequence that targets it for degradation by the proteasome as soon as it comes in contact with the cytosol. P R OT E I N M E T H Y L AT I O N

Carboxyl methylation of the phosphoprotein phosphatase 2A catalytic subunit promotes its functional association with regulatiory subunits in vivo. Wu, J. et al. EMBO J. 19, 5672–5681 (2000)

Several proteins can be modified by methylation of their carboxyl termini, but does this alter their function? Wu and colleagues show that it does — at least for protein phosphatase 2A (PP2A). They identify the yeast enzymes responsible for adding and removing the methyl group and, by manipulating these enzymes genetically, show that methylation alters the affinity of PP2A for different regulatory subunits.

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 1 | DECEMBER 2000 | 1 6 5

HIGHLIGHTS

IN BRIEF

D N A R E PA I R

A taxing question

MEIOSIS

Direct coupling between meiotic DNA replication and recombination initiation. Borde, V., Goldman, A. S. & Lichten, M. Science 290, 806–809 (2000)

During meiosis, information is swapped between parental chromosomes by homologous recombination. Initiation of this process requires a double-stranded DNA break (DSB) which, according to this paper, is introduced in a replication-dependent manner. The authors show that by delaying replication of a chromosomal segment, the formation of a DSB can be delayed in that segment. C H R O M O S O M E B I O LO G Y

Cleavage of cohesin by the CD clan protease separin triggers anaphase in yeast. Uhlmann, F. et al. Cell 103, 375–386 (2000)

Disjunction of homologous chromosomes in meiosis I depends on proteolytic cleavage of the meiotic cohesin Rec8 by separin. Buonomo, S. B. C. et al. Cell 103, 387–398 (2000)

Two distinct pathways remove mammalian cohesin from chromosome arms in prophase and from centromeres in anaphase. Waizenegger, I. C. et al. Cell 103, 399–410 (2000)

This trio of papers — two from Kim Nasmyth’s lab and one from Jan-Michael Peters and colleagues — tackle the question of how sister-chromatid cohesion is regulated. The first shows that separin, a conserved protein responsible for cleaving the Scc1 subunit of cohesin, is a cysteine protease related to caspases. Moreover, in vitro it alone is enough to cleave Scc1, an event that triggers sister chromatid separation. The second paper shows that cleavage of Rec8 — the meiotic equivalent of Scc1 — by separin at two different sites is necessary for the resolution of chiasmata during meiosis. Finally, Waizenegger et al. propose that, in vertebrates, cohesin is removed from chromosome arms by a different, cleavageindependent mechanism to the one that removes centromeric cohesin and involves cleavage of Scc1. C Y TO S K E L E TO N

Dynein, dynactin, and kinesin II’s interaction with microtubules is regulated during bidirectional organelle transport.

How does mutation of a protein involved in the response to DNA damage lead to defects in the nervous system? Reporting in Genes and Development, Peter McKinnon and co-workers describe a mechanism that might, they say, normally act during development to eliminate neural cells with genomic damage. Several human syndromes with defective responses to DNA damage also lead to neurological lesions, and the best studied of these is the rare disorder ataxia-telangiectasia. Mutation of the protein responsible, the ATM kinase (reviewed by Kastan and Lim on page 179 of this issue), results in progressive neurodegeneration. But how? To find out, McKinnon and colleagues drew on their knowledge of DNA ligase IV (Lig4) — a molecular glue that binds double-stranded DNA breaks (DSBs), particularly during the process of V(D)J recombination. Mice with no functional Lig4 show widespread apoptosis in the developing nervous system, as well as embryonic lethality and defects in V(D)J recombination and lymphocyte development. The lack of Lig4 probably allows DSBs to accumulate and, given that ATM acts as a sensor for DSBs, the authors wondered whether these lesions might activate ATM (designated Atm in the mouse). To answer this question they turned it on its head — if Atm is activated in response to a lack of Lig4, then might a lack of Atm rescue the Lig4-null phenotype? McKinnon and colleagues generated Atm−/−Lig4−/− double-knockout mice, and found that, in contrast to Lig4−/− single knockouts, these mice showed no apoptosis in the embryonic nervous system. Moreover, most of the processes required for correct neural development (as measured with markers for neuronal differentiation) were normal in the Atm−/−Lig4−/− mice. They were smaller than their wild-type littermates, however, and they also died roughly two days after birth. The authors then tested whether the observed defects in the immune systems of Lig4−/− mice were rescued in the double knockouts. And they weren’t — these animals showed T-cell defects and an almost complete lack of CD4+CD8+ thymocytes. McKinnon and co-workers believe that this result reflects the tissue-specific functionality of Atm, showing that the observed neuronal rescue is highly selective. Plenty of questions remain. One is how these findings tie in with other studies showing that a lack of p53 also rescues the embryonic lethality in Lig4−/− mice, albeit to a differing degree (for example, p53−/−/Lig4−/− mice survive until six weeks of age). Another is whether other types of DNA lesion can trigger apoptosis in developing neurons, a possibility that seems very likely. Finally, McKinnon and co-workers point out that it’s remarkable how development can proceed so completely in the Atm−/−Lig4−/− mice — after all, their neurons must contain many unrepaired DSBs. Perhaps, then, it’s no surprise that the mice die so soon after birth. And thinking about these results in the context of the human disease ataxia-telangiectasia, a lack of ATM probably allows cells with endogenously produced DSBs to become part of the nervous system. Subsequent malfunctioning owing to this genomic damage would then lead to the observed neurodegeneration.

Reese, E. L. & Haimo, L. T. J. Cell Biol. 151, 155–165 (2000)

Alison Mitchell

Dynein and kinesin motors transport organelles to opposite ends of microtubules, but it is a mystery what controls the net direction of the vesicles. This paper shows that dynein, dynactin and kinesin II are continuously associated with pigmented organelles in Xenopus melanophores, indicating that association of the motors with the organelle is not regulated. The direction is, in fact, determined by controlling the binding of motors to microtubules, and this probably occurs through phosphorylation.

166

References and links ORIGINAL RESEARCH PAPER Lee, Y.

et al. Defective neurogenesis resulting from DNA ligase IV deficiency requires Atm. Genes Dev. 14, 2576–2580 (2000) FURTHER READING Lieber, M. R. The biochemistry and biological significance of nonhomologous DNA end joining: an essential repair process in multicellular eukaryotes. Genes Cells 4, 77–85 (1999)

www.nature.com/reviews/molcellbio

| DECEMBER 2000 | VOLUME 1

α4 α3

α5

α2

α7

α1

Without PA26

taining an α3-subunit lacking its amino terminus. This mutant 20S particle chomps its way through peptides without assistance from 19S particles, and its crystal structure reveals how: unlike in the wild-type 20S particle, where the seven amino-terminal tails have a fixed structure, the chamber’s entrance is disordered in the mutant. The α3-tail therefore calls the others to order and, remarkably, it can do so even if it’s added as a separate peptide. One of the proteasome’s tasks is to process antigens for presentation to T cells, but the optimal size of processed antigen is 8–10 amino acids — larger than the typical proteasome product. Interferon-γ induces the production of subunits that make proteasomes better suited to antigenic peptide production. One of these is trypanosome PA26 (PA28 in humans), which forms a heptameric 11S particle that can substitute for the 19S particle (although, unlike the 19S particle, it doesn’t recognize the ubiquitin tags that target proteins for destruction). Frank Whitby and colleagues now show how the 11S particle opens the proteasome’s gate. By bind-

T R A N S L AT I O N

Unmasking the effects of CPEB The best way to ensure the safe delivery of a protein is to produce it as close as possible to where it’s needed. Reporting in Cell, Joel D. Richter and colleagues describe how this may be done for cyclin B1. By localizing the proteins responsible for translation of cyclin B1 messenger RNA to the mitotic spindle, production of this key player in cell division can be tightly controlled. And the results indicate that this regulation may be necessary for integrity of the mitotic apparatus. Translation of cyclin B1 mRNA is regulated by cytoplasmic polyadenylation. This process is critical for the activation of different maternally inherited mRNAs during early development in many animals. It has been extensively studied in Xenopus oocytes where, in response to progesterone stimulation, the poly(A) tails of certain mRNAs (encoding, among them, several cyclins) are elongated. A central player in polyadenylation is the cytoplasmic polyadenylation element-binding protein (CPEB), which recruits a factor that

α3

N α6

An ancient Greek legend tells that the gate to Hades was guarded by Cerberus, a many-headed dog. But what guards the proteasome — a cellular underworld where proteins are sent to their deaths? Two structural studies, published in Nature Structural Biology and Nature, characterize the many-tailed beast at the proteasome’s gate, and reveal the trick that the cell uses to tame it. The working proteasome consists of a 20S core particle with a 19S regulatory particle at one or both ends. In the 20S core, four rings of seven subunits form a hollow chamber, with proteolytic β-subunit rings in the centre and α-subunit rings sealing each end. The seal itself, comprising a meshwork of highly divergent α-subunit amino-terminal tails, is broken when 19S particles are added, but we can only guess at how. Michael Groll and colleagues suspected that the α3-subunit’s tail holds the key to the gate, because it is the only tail that contacts all the others. They confirmed their suspicions by solving the structure of a 20S proteasome con-

α5

Reverse turn

P R OT E I N M E TA B O L I S M

Gates of destruction

α4

α2

promotes the interaction between poly(A) polymerase and the end of the mRNA. Polyadenylation in turn triggers translation, and a key to this switch is maskin — a protein that was initially identified on the basis of its specific immunoprecipitation with CPEB. Given that some mRNAs are concentrated in certain regions of Xenopus oocytes, Richter and colleagues wondered whether CPEB and maskin might be involved in mRNA localization as well as translation. To test this, they immunostained Xenopus oocytes at various stages of development with antibodies against the two proteins and found that both were especially concentrated at the cortex of the animal pole during late stages of oocyte development. Surprisingly, though, in the early embryo CPEB and maskin seemed to localize to structures resembling the mitotic spindle. Closer analysis using antibodies against αtubulin, CPEB and maskin confirmed this suspicion. At metaphase, the authors observed a gradient of CPEB and maskin along the

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

α2

α6

α7

α1

Bound to PA26

ing a pocket in the interfaces between the αsubunits, the 11S particle puts pressure on a reverse turn at the end of each α-subunit’s amino-terminal tail. This trips a conformational switch, causing the tails to point towards the 11S particle instead of obscuring the gate (see picture). By grabbing the dog by its tails, then, the 11S subunit opens the gates of destruction. Both studies show that, irrespective of whether the α-subunits are in their open or closed conformations, the structures of the βsubunits remain unchanged, refuting the idea that gating might allosterically activate the the β-subunits. Does the 19S subunit use the same trick to open the gates? We’ll need a hero bearing crystals before that legend can be told. Cath Brooksbank References and links ORIGINAL RESEARCH PAPER Groll, M. et al. A gated

channel into the proteasome core particle. Nature Struct. Biol. 7, 1062–1067 (2000) | Whitby, F. G. et al. Structural basis for the activation of 20S proteasomes by 11S regulators. Nature 408, 115–120 (2000) FURTHER READING Pickart, C. M. & VanDemark, A. P. Opening doors into the proteasome. Nature Struct. Biol. 7, 999–1001 (2000)

length of the spindles, peaking at the area around the centrosomes. So CPEB and maskin localize to spindles and centromeres, but what about the mRNAs that they regulate? Using in situ hybridization, the authors next showed that cyclin B1 mRNA is also localized to the animal pole of Xenopus oocytes and, more specifically, to spindles. This localization depends on CPEB, as cyclin synthesis was blocked in one-cell embryos injected with an antibody against CPEB. This treatment also caused embryos to divide three to five times more slowly than controls, and many of them showed spindle defects. The conclusion, then, is that CPEB and maskin regulate the translation of cyclin B1, a process that is important not only for integrity of the mitotic apparatus, but for cell division as a whole. Local delivery, it seems, really is the safest option. Alison Mitchell References and links ORIGINAL RESEARCH PAPER Groisman, I. et al. CPEB,

maskin, and cyclin B1 mRNA at the mitotic apparatus: implications for local translational control of cell division. Cell 103, 435–447 (2000) FURTHER READING Richter, J. D. Cytoplasmic polyadenylation in development and beyond. Microbiol. Mol. Biol. Rev. 63, 446–456 (1999)

VOLUME 1 | DECEMBER 2000 | 1 6 7

α2

REVIEWS HOW CELLS READ TGF-β SIGNALS Joan Massagué Cell proliferation, differentiation and death are controlled by a multitude of cell–cell signals, and loss of this control has devastating consequences. Prominent among these regulatory signals is the transforming growth factor-β (TGF-β) family of cytokines, which can trigger a bewildering diversity of responses, depending on the genetic makeup and environment of the target cell. What are the networks of cell-specific molecules that mould the TGF-β response to each cell’s needs? SMADS

A family of transcription factors that mediate TGF-β signals. The term SMAD is derived from the founding members of this family, the Drosophila protein MAD (Mothers Against Decapentaplegic) and the Caenorhabditis elegans protein SMA (Small body size). GS REGION

Regulatory region in TGF-β receptors. UBIQUITYLATION

The attachment of the protein ubiquitin to lysine residues of other molecules, often as a tag for their rapid cellular degradation. PROTEASOME

Protein complex responsible for degrading intracellular proteins that have been tagged for destruction by the addition of ubiquitin.

Cell Biology Program and Howard Hughes Medical Institute, Memorial SloanKettering Cancer Center, Box 116, 1,275 York Avenue, New York, New York 10021, USA. e-mail: j-massague@ski.mskcc.org

Two emerging concepts are beginning to shed light on how cells read cytokine signals. First, it is increasingly clear that cytokine signalling pathways are not insulated devices but segments in a dense network. Through several crosstalk and feedback links, the network monitors and adjusts the activity of each constituent pathway and, in so doing, determines the nature and timing of the signals conveyed. Second, the pathway provides the cell with information about the arrival of a certain cue, but does not provide precise instructions. The cell, more than the pathway, determines the outcome of the signal. In the case of TGF-β, the task of the pathway is to place an activated SMAD transcription factor — a signal mediator — in the nucleus. The cell then decides which of many potential target genes will respond to the SMAD through the action of components that take SMADs to specific target genes. Sensing and propagating TGF-β signals

The TGF-β family includes many related factors that have diverse functions during embryonic development and adult tissue homeostasis [online resource] (reviewed in REFS 1–4). TGF-β and related factors use a simple mechanism to signal to the nucleus (FIG. 1). They bind to membrane receptors that have a cytoplasmic serine/threonine kinase domain. Binding of the ligand causes the assembly of a receptor complex that phosphorylates proteins of the SMAD family — closely related proteins that bind DNA and recruit transcriptional co-activators or corepressors [online resource] (REF. 1 and references therein). Phosphorylation causes SMADs to move into the nucleus, where they assemble complexes that directly control gene expression. So, at its simplest, the basic TGF-β signalling engine consists

of a receptor complex that activates SMADs and a SMAD complex that controls transcription. Key aspects of this signal-transduction process have been worked out (FIG. 1). Each ligand of the TGF-β family binds to specific pairs of receptor serine/threonine kinases, belonging to groups known as the type I and type II receptors, respectively [online resource] (reviewed in REF. 1). Most mammalian cells express different members of this receptor family, some of which may be shared by different TGF-β ligands. When the ligand binds, it acts as a dimeric assembly factor, bringing together two type I and two type II receptors. In this complex, receptor II exerts its only known function, which is to activate receptor I by phosphorylation of the GS REGION5,6 (BOX 1). The type I receptor then phosphorylates SMAD proteins that propagate the signal (FIG. 1). One mechanism for switching off the TGF-β signal involves SMAD UBIQUITYLATION in the nucleus, followed by PROTEASOME-mediated degradation of the SMAD protein7 (FIG. 1). A separate ubiquitylation mechanism controls the basal level of SMAD through the UBIQUITIN LIGASE SMAD ubiquitylation regulatory factor (SMURF1)8. A more common mechanism for returning a phosphorylated mediator to its basal state, protein dephosphorylation, has not been described for the deactivation of SMADs. SMADs function as signal transducers of TGF-β family members in organisms ranging from worms to humans [online resource]1,3,9. In vertebrates, the type I receptors for bone morphogenetic proteins (BMPs), growth and differentiation factors (GDFs) and antiMüllerian hormone/Müllerian inhibiting substance (AMH/MIS) signal through SMAD1 or the closely related SMAD5 and SMAD8, whereas the receptors for

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 1 | DECEMBER 2000 | 1 6 9

REVIEWS

TGF-β TGF-β receptor II

Plasma membrane

ATP

ADP

SARA R-SMAD

R-SMAD Co-SMAD

SMAD

Nuclear pore complex Nuclear membrane

+ Cofactor

Ubiquitination and degradation

Co-activator/ repressor

Co-SMAD R-SMAD DNA

Figure 1 | The basic SMAD pathway. Receptor-regulated SMAD transcription factors (RSMADs) require transforming growth factor-β (TGF-β)-induced phosphorylation to assemble transcription regulatory complexes with partner SMADs (co-SMADs). R-SMADs can move into the nucleus on their own but, to be accessible to membrane receptors, R-SMADs are tethered in the cytoplasm by proteins such as SARA (SMAD anchor for receptor activation). Receptor activation occurs when TGF-β induces the association of two type I and two type II receptors. Both receptor components have a serine/threonine protein kinase domain in the cytoplasmic region. In the basal state, the type I receptor is kept inactive by a wedge-shaped GS region, which presses against the kinase domain, dislocating its catalytic centre103. In the ligandinduced complex, the type II receptor phosphorylates the GS domain and this activates the type I receptor, which catalyses R-SMAD phosphorylation. Phosphorylation decreases the affinity of R-SMADs for SARA and increases their affinity for co-SMADs. The resulting SMAD complex is free to move into the nucleus and competent to associate with transcriptional coactivators or corepressors. SMADs can contact DNA, but effective binding to particular gene regulatory sites is enabled by specific DNA-binding cofactors. R-SMADs that move into the nucleus may return to the cytoplasm, but their ubiquitylation- and proteasome-dependent degradation in the nucleus provide a way to terminate TGF-β responses7.

UBIQUITIN LIGASE

An enzyme that couples the small protein ubiquitin to lysine residues on a target protein, marking that protein for destruction by the proteasome. NUCLEAR LOCALIZATION SIGNAL

A 7–9 residue sequence within a protein, rich in basic residues, which acts as a signal for localization of the protein within the nucleus.

170

TGF-β, Activin and Nodal do so through SMAD2 and SMAD3. These are collectively referred to as receptorphosphorylated SMADs (R-SMADs). Receptor-mediated phosphorylation at carboxy-terminal serine residues increases the affinity of R-SMADs for a particular member of the family, SMAD4 (REF. 10). The SMAD4 protein (also known as DPC4, deleted in pancreatic carcinoma locus 4)11,12 (FIG. 1) functions as a shared partner (or coSMAD) of R-SMADs and is required for active transcriptional complexes to assemble12–15. R-SMADs bind the transcriptional co-activators p300 and CBP (CREBbinding protein; where CREB stands for cyclic AMPregulated enhancer-binding protein)16 (reviewed in REF. 17). One function of SMAD4 may be to allow this recruitment through an interaction with the protein MSG-1 (which stands for melanocyte-specific gene, although it is also expressed in other cell types), which may function as a p300 adaptor18.

In the basal state, SMADs are retained in the cytoplasm. In the case of SMAD2, this retention is mediated by interactions with the SMAD anchor for receptor activation, SARA19 (FIG. 1). In addition to limiting SMAD movements, the contact with SARA occludes a region of SMAD2 that mediates nuclear import10. Receptor-mediated phosphorylation not only increases the affinity of SMAD2 for SMAD4 but also decreases its affinity for SARA, releasing SMAD2 and unmasking its nuclear import function, which leads to rapid accumulation of the activated SMAD2 in the nucleus10 (FIG. 1). Adjacent to the SMAD-binding domain, SARA contains a FYVE domain, a type of protein fold that, in other proteins, contributes to their anchoring to endosome membranes19. This raises the possibility that the TGF-β receptor may have to undergo internalization to reach the SARA-bound SMAD substrate. Proteins other than SARA may also control SMAD access to receptors and movement to the nucleus. The specific contribution of SARA to these processes needs to be assessed through genetic ablation of its function. The SMAD nuclear import process that SARA inhibits is independent of the NUCLEAR LOCALIZATION SIGNAL (NLS) import pathway10, which imports proteins that contain a lysine- and arginine-rich sequence known as the NLS . The NLS is recognized as an extended peptide loop by the nuclear import factor importin-α, as defined by crystallography and a nuclear import assay system20. In this assay system, SMAD import is independent of importin-α10. Whether SMADs also use the classical NLS pathway is still uncertain. This possibility is raised by the presence of a conserved lysine-rich sequence in SMADs that, when mutated in SMAD2, prevents its nuclear import21. However, mutations at, or near, this sequence in other SMADs can extensively disrupt both the conformation of the protein and several of its functions22,23. Crystallographic studies have shown that this lysine-rich sequence forms an α-helix24, which is structurally incompatible with recognition by importin-α20. Indeed, SMAD2 does not bind to importin-α21. SMAD4 also has an intrinsic, agonist-independent nuclear import function, but is kept out of the nucleus by a nuclear export signal25. The related factor SMAD4β lacks this signal and is constitutively nuclear26. SMADs can therefore be regarded as nuclear proteins that are kept in the cytoplasm so that they are accessible to receptors that are activated at the plasma membrane. Specificity in signal transduction

The cellular context is central in determining which genes will respond to an activated SMAD complex as it arrives in the nucleus. However, the choice of target genes is circumscribed by the competence of each RSMAD protein. Both SMAD1 and SMAD2 (and the other members of either subgroup) are competent to access separate sets of target genes27. By specifically engaging SMADs from one subgroup or the other, the TGF-β branch (the TGF-βs, Activins and Nodals) and the BMP branch (BMPs, GDFs and MIS) of the TGF-β family direct the cell to choose from different sets of potential target genes (FIG. 2). www.nature.com/reviews/molcellbio

| DECEMBER 2000 | VOLUME 1

REVIEWS An example of the dichotomy between the SMAD1dependent and SMAD2-dependent pathways is provided by their opposing, albeit complementary, functions during Xenopus laevis embryogenesis28,29. In response to Nodal-related factors (Xnr-1, Xnr-2 and Vg1), SMAD2

Box 1 | A structural view of the SMAD signalling pathway MH1 domain

Linker

MH2 domain

SSXS

SMAD2 MH2 domain SSXS

TGF-β type I receptor cytoplasmic domain αH-1

L45 loop GS region

αH-2 L3 loop

DNA-binding cofactor (FAST, Mixer) SMAD3 MH1–DNA complex

The crystal structure of the cytoplasmic region of the TGF-β type I receptor103 reveals that, in the basal state, this receptor is maintained inactive by a wedge-shaped GS region (green) that presses against the kinase domain (blue), dislocating its catalytic centre. Phosphorylation of the GS region by the TGF-β type II receptor activates the type I receptor kinase, allowing it to phosphorylate SMAD proteins at the carboxyterminal sequence SSXS. The SMADs contain conserved amino-terminal and carboxy-terminal regions known as the MH1 domain and MH2 domain, respectively (for ‘Mad homology’ in reference to the first identified family member, the Drosophila Mad gene product). The MH1 and MH2 domains form globular structures24,104 and are linked by a region that is less well conserved among SMADs. This linker harbours mitogen-activated protein kinase phosphorylation sites73,74 and sites for recognition by the ubiquitin ligase SMURF1 (REF. 8). In the basal state, SMADs stay in the cytoplasm. The SMAD2 protein is retained in the cytoplasm by an interaction with the protein SARA (SMAD anchor for receptor activation)19. The crystal structure of the SMAD2 MH2 domain bound to SARA shows that this contact is mediated by several contiguous hydrophobic interactions along the back of the MH2 domain (not shown)105. The crystal structure of the SMAD4 MH2 domain shows that it forms a homotrimer in the basal state (not shown), and several tumour-derived mutations affect critical residues in the monomer–monomer interfaces of this trimer104. When the activated TGF-β receptor recognizes R-SMADs, the specificity of this recognition is determined by the sequence of the L45 loop on the receptor kinase domain (in red circle) and the sequence of the L3 loop (purple) in the SMAD MH2 domain34. The L3 loop is a short, conserved sequence that differs in only two amino acids between the SMAD1,5,8 subgroup and the SMAD2,3 subgroup. The differences in surface structures between these two versions of the L3 loop are sufficient for SMAD discrimination by the receptor27. SMAD1 recognition by receptors of the ALK1 subgroup also requires the α-helix 1. Receptor-mediated phosphorylation releases SMADs from cytoplasmic anchors10,19. It also exposes a SMAD nuclear import signal and augments the affinity of R-SMADs for SMAD4. Once in the nucleus, the activated SMADs contact DNA through the MH1 domain106 and activate transcription through the MH2 domain107. The crystal structure of the SMAD3 MH1 domain bound to cognate oligonucleotides has shown that this contact occurs through a β-hairpin structure (red)24. The MH2 domain is responsible for interaction with partner SMADs12, various DNA-binding cofactors13,38,39, and transcriptional co-activators15,16 and corepressors49. Other DNAbinding cofactors may interact with the MH1 domain41,43. The sequence of α-helix 2 (αH-2, blue) specifies the interaction with various DNA-binding cofactors27,39.

activates expression of dorsal MESODERM marker genes and gives rise to dorsal mesoderm, whereas SMAD1, in response to BMP4, induces a different set of genes that specify the formation of ventral mesoderm and suppress neural fates. Ectopic expression of SMAD1 or SMAD2 mimics the effect of these two groups of agonists, giving rise to ectopic ventral or dorsal structures, respectively. Another example is provided by the formation of left–right asymmetry in vertebrate embryos by an interplay between BMP and Nodal signals30–32. How does the system ensure specificity and avoid crossreactivity in the structurally related receptor–SMAD interactions of these two functionally different pathways? The answer lies in interacting protein modules — a common solution to signal-transduction problems. Discrete structural elements on the kinase domain of the receptor and the MAD homology (MH2) domain of the SMAD dictate the specificity of their interactions, and therefore the choice of SMADs by the receptors (BOX 1). The structures in question are the L45 loop on the receptor kinase and the L3 loop on the SMAD MH2 domain27,33,34. The L45 loop is of identical sequence within the TGF-β and BMP subgroups, but differs by three residues between these two subgroups. Likewise, the L3 loop differs by only two amino acids between the SMAD1 and SMAD2 subgroups. These differences seem sufficient to enforce discrimination in receptor–SMAD interactions. It would be an oversimplification to consider the receptors (or SMADs) within a subgroup as functionally equivalent molecules. Important differences exist, an example of which is provided by SMAD2 and SMAD3. These two proteins have nearly identical primary structure but SMAD3 lacks an insert that is present in the most common splice form of SMAD2 (REF. 35). This insert, adjacent to the DNA-binding element of the MH1 domain, prevents the direct contact of SMAD2 with DNA24. This singularity is likely to underlie some of the functional differences between SMAD2 and SMAD3 (REF. 36). Type I receptors within the same subgroup may also have functional differences in ligand-binding specificity, SMAD-phosphorylation kinetics, interaction with regulators or activation of alternative pathways. Establishing a cellular context

An activated R-SMAD–SMAD4 complex arriving in the nucleus is faced with a choice of many potential target genes and an inability to bind to most of them on its own. SMADs can recognize the DNA sequence CAGAC (and certain G+C-rich sequences), but their affinity seems too low to support unassisted binding to DNA under physiological conditions24. If their affinity for the simple CAGAC sequence were higher, SMADs would decorate the entire chromosome. Other factors must therefore be required for selective binding of an activated SMAD complex to target ENHANCER ELEMENTS. Hypothetically, this requirement could be fulfilled by the cooperative binding of a SMAD oligomer to contiguous CAGAC sequences located at the right distance from each other. A gene regulated in this way could be activated by SMADs regardless of the cell type. The SMAD7 gene might be an example of this37. Most

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 1 | DECEMBER 2000 | 1 7 1

REVIEWS

MESODERM

The middle of the three embryonic germ layers, and the source of structures including bone, muscle, connective tissue and dermis. MH1 AND MH2 DOMAIN

Conserved amino-terminal and carboxy-terminal globular domains, respectively, of SMAD proteins. ENHANCER ELEMENT

Sequence in the regulatory region of a gene, recognized by factors that enhance the activity of the transcriptional promoter.

SMAD-mediated gene responses, however, are cell-type dependent. Growing evidence indicates that activated SMADs achieve high affinity in their interactions with DNA by associating with partner DNA-binding cofactors (FIG. 1) — structurally diverse proteins that have in common the ability to simultaneously contact an R-SMAD and a specific DNA sequence (FIG. 3) (reviewed in REF. 17). The fact that these proteins are functionally expressed in some cell types but not in others provides a basis for the cell-type specificity of TGF-β family gene responses13,38. Furthermore, SMAD DNA-binding cofactors contribute to the segregation of the SMAD1 and SMAD2 pathways by specifically interacting with SMADs from only one group27,38,39. And, by requiring the presence of a specific DNA sequence at an exact

distance from a CAGAC site, SMAD DNA-binding cofactors provide target-gene selectivity38,40,41. Adaptors and partners

One group of SMAD cofactors seem to function purely as DNA-binding adaptors. Examples of this group include the proteins OAZ (Olf-associated zinc finger) for SMAD1 (REF. 38), and FAST (forkhead activin signal transducer)42 and Mixer39 for SMAD2 (FIG. 3). These factors lack intrinsic transcriptional activity, and their role is to buttress the DNA contact of an activated SMAD complex. The selective recognition between these cofactors and the SMADs involves a SMAD-interaction domain (conserved in FAST and Mixer but different in OAZ) and a protruding element (α-helix 2) on the MH2 domain of SMADs27,38,39 (BOX 1). As with the recep-

Ligands TGF-β Activin BMP BMP BMP

Accessory receptors Betaglycan Endoglin Crypto

LAP Follistatin Caronte Gremlin Noggin

Thrombospondin Activin SHH SHH BMP

DNA-binding cofactors II Signalling receptors

IIII

Co-SMAD

R-SMAD

β-catenin SMAD1 SMAD2 SMAD2 JAK1/2 JNK

LEF1/TCF CBFA1 CBFA3 Mixer STAT3 JUN

FKBP12 FKBP12.6

Wnt BMP TGF-β Xnr LIF TNF-α

UBCH5 SNIP1

BAMBI Cofactor

SMURF1 SMAD7 SMAD6 ERK BMP TGF-β TNF-α IFN-γ EGF

SMAD1 SMAD2 NFκB STAT1

Co-SMAD R-SMAD

Co-activators p300, CBP

BF1 Ras Nucleus

TGIF Ski/SnoN Corepressors

TGF-β

Figure 2 | A network controlling a pathway. The transforming growth factor-β (TGF-β) signalling pathway receives regulatory inputs in both the pre-receptor phase and the post-receptor phase. Signalling pathways activated by diverse agonists (blue arrows) augment (green arrows) or inhibit (red T-shapes) the TGF-β system at the level of the ligands, the receptors, the SMADs, the DNA-binding cofactors, or the transcriptional co-activators or corepressors (REFS. 66,67 and references therein). LAP sequesters TGF-β, Follistatin sequesters Activin, and the structurally diverse proteins Caronte, Gremlin and Noggin sequester BMPs. The small proteins FKBP12 or FKBP12.6 bind to the GS region of type I receptors in the basal state, blocking access to activators. BAMBI is a pseudoreceptor that interferes with receptor complex formation. SMAD6 and SMAD7 are inhibitory SMADlike decoys. TGIF, SKI and SnoN are SMAD corepressors. Various signals control the activity of these regulators, providing feedback control (note regulation by BMP, TGF-β and Activin signals) and crosstalk control (note regulation by SHH, TNF-α, IFN-γ and EGF). Various signals positively control the activity of DNA-binding cofactors (LEF1/TCF column) that cooperate with activated SMADs to assemble gene-specific transcriptional complexes. Other regulators may provide other routes for feedback and crosstalk, but the signals controlling this activity remain unknown. These regulators include the accessory receptor Betaglycan, the putative accessory receptors Endoglin and Crypto, the cytoplasmic ubiquitin ligase SMURF1, UBCH5-coupled mediators of SMAD ubiquitylation in the nucleus, the forkhead inhibitor of FAST-SMAD interaction BF1 (REF. 108), the inhibitor of SMAD–p300 interaction SNIP1 (REF. 109), and others66,67. (BAMBI, BMP and Activin membrane-bound inhibitor; BF1, brain factor 1 ;BMP, bone morphogenetic protein; CBFA, core-binding factor A; CBP, CREB-binding protein; CREB, cyclic AMP-regulated enhancer binding protein; EGF, epidermal growth factor; FKBP12, FK506-binding protein of 12 kDa; FKBP12.6, FK506-binding protein of 12.6 kDa; IFN-γ, interferon-γ; JAK, janus kinase; LAP, latency-associated protein; LIF, leukaemia inhibitory factor; NFκB, nuclear factor of κlight polypeptide gene enhancer in B cells; R-SMAD, Receptor-regulated SMAD transcription factor; SHH, sonic hedgehog; STAT, signal transducer and activator of transcription; UBCH5, ubiquitylation cofactor human 5 ; SMURF, SMAD ubiquitylation regulatory factor;SNIP1, SMAD nuclear-interacting protein 1; SKI, Sloan-Kettering Institute proto-oncogene; SnoN, Ski-related novel gene N; TGIF, TG3-interacting factor; TNF-α, tumour necrosis factor-α; Xnr, Xenopus nodal-related)

172

www.nature.com/reviews/molcellbio

| DECEMBER 2000 | VOLUME 1

REVIEWS

Cofactor Target

Function Ventral mesoderm specification by BMP in Xenopus Osteoblast differentiation by BMP in human, mouse Visceral mesoderm formation by Dpp in Drosophila Endoderm formation by Dpp in Drosophila

OAZ BMP, SMAD1, CBFA1? DPP Tinman inputs MAD CREB

Vent.2 Osteocalcin? Tinman Ubx

FAST TGF-β, Nodal, Mixer Activin SMAD2,3 TFE3 inputs CBFA3 Jun Lef1/TCF

Mix.2 Mesoderm specification by Nodal in Xenopus Nodal, Lefty2 Left plate mesoderm formation by Nodal in mouse Goosecoid Anterior mesoderm induction by Nodal in mouse PAI-1 Plasminogen system control by TGF-β in human, mouse IgA Immunoglobulin A class switching by TGF-β in human c-Fos Diverse TGF-β responses Xtwn Mesoendoderm differentiation by Nodal in Xenopus

Figure 3 | SMADs interact with DNA-binding cofactors to specify TGF-β gene responses. R-SMADs depend on a structurally diverse group of DNA-binding cofactors for their interaction with particular target genes (REF. 17 and references therein). Distinct groups of R-SMADs use different subsets of DNA-binding cofactors, providing pathway specificity to the gene response. In vertebrates, OAZ and CBFA1 interact with SMAD1 in response to BMP signals, and the indicated factors interact with SMADs 2 and 3 in response to TGF-β, Activin or Nodal signals. In Drosophila the SMAD1 orthologue MAD is thought to interact with Tinman and CREB in response to the BMP-like factor DPP. Many of these cofactors are expressed only in certain cell types, conferring cell-type specificity to a SMAD-mediated response. (BMP, bone morphogenetic protein; CBFA, core-binding factor A; CREB, cyclic AMP-regulated enhancerbinding protein; Dpp, decapentaplegic; FAST, forkhead activin signal transducer; IgA, Immunoglobulin type A; LEF1, lymphoid enhancer-binding factor 1; MAD, Mothers against decapentaplegic; OAZ, Olf-associated zinc finger; PAI-1, plasminogen activator inhibitor-1; TFE3, transcription factor binding to immunoglobulin heavy constant mu enhancer 3; Ubx, ultrabithorax; Xtwn, Xenopus Twin.)

tor–SMAD interaction, the selectivity of the SMAD–cofactor interaction is dictated by a handful of subtype-specific amino-acid residues at the contact site. Another group of proteins facilitating SMAD binding to target promoters are transcription factors that have their own ability to recruit co-activators and function independently of SMADs in other contexts. Examples include JUNB43, TFE3 (transcription factor binding to immunoglobulin heavy constant mu enhancer 3)41, core-binding factor A/acute myeloneous leukaemia (CBFA/AML) proteins44–46 and lymphoid enhancer-binding factor 1/T-cell-specific factor (LEF1/TCF)47,48 (FIG. 3). Some of these factors are themselves regulated by extracellular signals, providing a basis for integration of different inputs at the transcriptional level (FIG. 2). For example, the mediator of Wnt/β-catenin signalling, LEF1/TCF, cooperates with SMADs in the activation of Xtwn (Xenopus Twin) in Xenopus in response to Nodal-related signals47. When SMADs bind to DNA with the help of a DNAbinding adaptor, they may be solely responsible for recruiting co-activators. When they bind to DNA in collaboration with other transcription factors that have their own co-activator-recruiting ability, SMADs perhaps return the favour by modifying the composition of coactivator complexes or the kinetics of their assembly. The individual contributions of the SMADs and their partner transcription factors in such complexes, and the value added by their collaboration, are important questions. Choosing co-activators and corepressors HYPOMORPHIC ALLELE

A mutant gene having a similar but weaker function than the wild-type gene.

A second important choice that must be made in the nucleus is whether an activated SMAD complex will activate or repress gene expression. It turns out that SMADs

can recruit not only transcriptional co-activators but also corepressors. SMADs bind the transcriptional corepressors TG3-interacting factor (TGIF)49, Sloan-Kettering Institute proto-oncogene (SKI) and its relative Ski-related novel gene N (SnoN)50,51. These proteins bind histone deacetylases (HDACs), whose effects generally lead to chromatin condensation at target-gene promoters, thereby opposing the action of the histone acetyltransferase (HAT) activities associated with the co-activators p300 and CBP17. TGIF binds to SMAD2 and SMAD3 in competititon with p300, so the relative levels of p300 and TGIF in a cell influence whether a SMAD complex will bring one or the other to a target gene49, and the same may be true of SKI and SnoN. TGF-β inhibits the expression of many genes52, and these responses might be mediated by SMADs recruiting corepressors to the regulatory regions of some of these genes. The TGIF, SKI and SnoN proteins may have different roles as SMAD corepressors. Both SKI and SnoN are degraded soon after cell stimulation with TGF-β, but accumulate again later51,53. So SKI and SnoN may keep SMADs from activating transcription in the basal state, and may help terminate TGF-β action later (FIG. 2). TGIF, on the other hand, seems to set the maximal level to which the TGF-β signal can activate transcription49. TGIF levels may be increased by signalling through extracellular-signal-regulated kinase (ERK), a member of the mitogen-activated protein kinase (MAPK) pathway54. The effect of TGIF mutations in humans illustrates the importance of adjusting the levels of this corepressor. TGIF mutations cause holoprosencephaly (HPE), a genetic disorder with defective bifurcation of anterior neural structures, leading to cyclopia (single eye) and a single cerebral ventricle55. Mutations in TGIF that are associated with HPE generally involve deletion of one copy of the gene, or a heterozygous mutation giving rise to a HYPOMORPHIC ALLELE. So even a slight reduction in the activity of TGIF can have devastating developmental consequences. Furthermore, TGIF may mediate gene repression through the Nodal/SMAD2 pathway, as mutations in zebrafish cyclops (a nodal homologue)56 or combined heterozygous null mutations in Nodal and SMAD2 in the mouse57 also give rise to cyclopia. Enlisting MAPK pathways

SMADs are the only TGF-β receptor substrates and signal transducers known so far, and they are central in most actions of the TGF-β family. However, persistent reports indicate that TGF-β and BMP may also send signals through MAPK pathways. The effects of TGF-β and BMP on the Jun N-terminal kinase (JNK), p38 and ERK MAPKs vary extensively in kinetics, magnitude and kinase subtype, depending on the conditions, and are present in only some of many cell lines surveyed. In some cases, these kinases are activated by TGF-β with slow kinetics, indicating that these might be delayed, indirect effects. In other cases, however, the activation is rapid, indicating that it may not be secondary to transcriptional events. TGF-β has been reported to activate JNK rapidly through MKK4 (MAP kinase kinase 4) in a

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 1 | DECEMBER 2000 | 1 7 3

REVIEWS fibrosarcoma cell line58 or p38 through MKK3 in lung and kidney epithelial cell lines59,60 (FIG. 4). The physiological relevance of these effects is uncertain, partly because there is insufficient genetic evidence to establish their roles in vivo61. We also do not know the nature of the biochemical link between TGF-β receptors and the MKKs, and why this link is present in some conditions but not in many others. One contender for this link is TAK1 (TGF-β activated kinase), a MAPK kinase kinase (MAPKKK) family member62,63 that is also implicated in the unrelated interleukin-1 and Wnt pathways. Members of the Rho family of small GTPases have also been proposed in the coupling of TGF-β receptors to JNK activation64. The ability of TGF-β to increase the activity of AP-1

IMAGINAL DISC

Single-cell layer epithelial structures of the Drosophila larva that give rise to wings, legs and other appendages.

Growth factor

Cytokine

Growth factor receptor

TGF-β TGF-β receptor II

Cytokine receptor

Plasma membrane

GRB2

XIAP HPK1

SOS

RAS +

TAK1

R-SMAD

A signalling network at play

– MEK

MKK4

MKK3

JNK

p38

– Ubiquitination and degradation

ERK

Nuclear pore complex

Nuclear membrane TGIF Cofactor

–

+ P

Co-SMAD R-SMAD

Fos

Jun

ATF2 DNA

Figure 4 | Crosstalk between the SMAD and mitogen-activated protein kinase pathways. The three principal MAPK pathways in mammalian cells may affect the SMAD pathway through various mechanisms. The Ras–MEK–ERK pathway can decrease TGF-β receptor levels by controlling expression, attenuate SMAD accumulation in the nucleus by phosphorylating SMADs in the linker region and increase the level of the SMAD corepressor TGIF by stabilizing this protein. The MKK4/JNK and MKK3/p38 pathways, which can be activated by various cytokines, enhance the activity of Jun and ATF2 transcription factors that may cooperate with SMADs through direct physical contacts. In certain cell types and conditions, the MKK4/JNK and MKK3/p38 pathways are reportedly activated by TGF-β itself, and the proteins XIAP, HPK1 and TAK1 might be involved in this link. The direct nature and physiological relevance of these interactions remain to be established. (ATF2, activating transcription factor 2; ERK, extracellular-signal-regulated kinase; GRB2, growth factor receptor-binding proteins 2; JNK, Jun amino-terminal kinase; XIAP, Xenopus inhibitor of apoptosis; HPK1, haematopoietic progenitor kinase 1; TAK1, TGF-β-activated kinase; MAPK, mitogen-activated protein kinase; MKK, MAPK kinase; R-SMAD, Receptorregulated SMAD transcription factors; sos, son of sevenless; TGF-β, transforming growth factor-β.)

174

(Jun–Fos) complexes through phosphorylation of c-Jun by JNK, or the activity of CREB complexes through phosphorylation of activating-transcription factor 2 (ATF2) by p38 could, in principle, result in activation of AP-1 or CREB-target genes. Indeed, induction of the fibronectin gene by TGF-β might be mediated partly by JNK and partly by SMADs58. Interestingly, several reports have indicated43,59,60,65 that TGF-β-activated MAPKs may converge on SMADs, alongside the direct effect of TGF-β receptors on these proteins. In vitro at least, activated SMADs can associate with Jun or ATF2 complexes. Furthermore, SMADs have been reported to be phosphorylated by JNK, although the phosphorylation sites remain to be identified64. If the MAPK pathways can modify the activity of SMAD complexes, it follows that classical activators of these pathways, such as cytokines (for example, tumour necrosis factor-α (TNF-α)) and cellular stress (for example, ionizing radiation or hyperosmolarity), could also modulate the activity of SMAD transcriptional complexes. And reciprocally, TGF-β-induced SMAD interactions with AP-1 and CREB could modify the cellular response to other cytokines and stress signals. There is a clear need to delineate the extent and relevance of these interactions under physiological conditions.

As a segment of the signalling networks in the cell, the SMAD pathway provides information that is read as a function of other inputs supplied by the rest of this network. Some inputs act directly on the TGF-β receptor and SMAD signal-transduction process, rapidly controlling the activity of its various components (FIG. 2). This process is generically known as ‘signalling crosstalk’ and can lead to the rapid attenuation or even cancellation of SMAD signalling activity66,67. Interestingly, few examples have emerged so far of crosstalk increasing the biochemical activity of the TGF-β receptors or the SMADs. It could be that once the TGF-β receptors are activated, most of the necessary adjustments to the signal-transduction process are for the purpose of bridling the powerful activity of the SMAD pathway. Various stimuli are reported to control the expression of TGF-β family receptors. Signalling by the growth factor hedgehog (HH) represses expression of the Decapentaplegic (DPP) receptor thickveins in a central region of the Drosophila melanogaster wing IMAGINAL DISC, thereby shaping the contours of the DPP signalling field68. An example of a molecule that directly interferes with the function of the TGF-β-receptor is SMAD7 (FIG. 2). The SMAD7 protein is a divergent member of the SMAD family that can bind TGF-β and BMP receptors and interfere with phosphorylation of R-SMADs69,70. In some cell types, SMAD7 expression is increased in response to the proinflammatory cytokines interferon-γ (IFN-γ) acting through the janus kinase (JAK) /signal transducer and activator of transcription (STAT) signalling pathway71 and TNF-α acting through the NF-κB (nuclear factor of κ-light polypeptide gene enhancer in B cells) pathway72 (FIG. 2). The resulting surge in SMAD7 levels interferes with TGF-β receptor phosphorylation www.nature.com/reviews/molcellbio

| DECEMBER 2000 | VOLUME 1

REVIEWS

CHONDROCYTE

Differentiated cell of cartilage tissue. OSTEOBLAST

A mesenchymal cell with capacity to differentiate into bone tissue. ECTODERM

The outer of the three embryonic germ layers, which gives rise to epidermis and neural tissue.

of SMAD3 and inhibits SMAD signalling. These effects of IFN-γ and TNF-α fall within the framework of a general antagonism between these cytokines and TGF-β in the regulation of immune-cell functions1. The activity of the TGF-β and BMP pathways is also regulated by crosstalk at the level of the activated SMADs (FIG. 2). Activation of the Ras–MEK–ERK pathway by growth factor receptor signals, or by oncogenic mutation of ras, leads to ERK-mediated phosphorylation of SMAD1, SMAD2 and SMAD3 (REFS 73,74). ERKs phosphorylate SMADs at specific sites in the region linking the MH1 and MH2 domains. In response to epidermal growth factor (EGF), these phosphorylations attenuate the accumulation of activated SMADs in the nucleus at low levels of TGF-β stimulation74. At higher levels of TGF-β stimulation, ERK-phosphorylated SMADs do accumulate abundantly in the nucleus, but several responses to SMADs remain altered, indicating that ERK activation has other effects on SMAD signalling. Crosstalk between the SMAD and MAPK pathways can work in the other direction too. For example, during the specification of cardiac and muscle cell fates in Drosophila, DPP-activated MAD (mothers against decaplentaplegic, the SMAD1 orthologue), in cooperation with wingless-activated TCF, and two other transcription factors (Twist and Tinman), acts on an enhancer region of the muscle identity gene evenskipped (eve)75. This enables eve to respond to Pointed, a transcription factor that is activated by the Drosophila EGF receptor through the Ras/MAPK pathway. In other words, SMAD and wingless signals together allow a gene to respond to a Ras/MAPK signal. Feedback

Another important source of regulation of the intracellular TGF-β pathway is by negative feedback, and this occurs at several levels (FIG. 2). At the level of the receptors, BMP exerts negative-feedback control through the protein BAMBI (BMP and Activin membrane-bound inhibitor)76 — a truncated, kinase-deficient type I receptor that interacts with BMP receptors and interferes with their activation. During Xenopus embryogenesis, Bambi expression overlaps BMP4 expression and requires BMP signalling. Bambi is therefore considered to be a negative-feedback regulator of the BMP pathway. Other forms of negative feedback are known (FIG. 4). Some are mediated by antagonistic SMADs. In addition to IFN-γ and TNF-α, inducers of SMAD7 expression include TGF-β and BMP themselves70, and this mechanism is conserved in Drosophila77. Another antagonistic SMAD, SMAD6, specifically binds to activated SMAD1 (REF. 78) and acts as a selective inhibitor of BMP signalling in vivo78–80. SMAD6 expression is elevated by BMP in vivo79, and SMAD6-defective mice develop bony tissue within the aorta walls81. Both these observations indicate that SMAD6 functions in BMP negative feedback. At higher levels of overexpression, SMAD6 can also bind to and block diverse TGF-β family receptors82, but whether this occurs at physiological levels of SMAD6 remains unknown. The increased expression of SKI and SnoN at late stages of TGF-β stimulation

(described above) might also limit the duration of the TGF-β response51,53. Extracellular control

What controls the extracellular phase of TGF-β pathways? Membrane receptors represent not the start but the midpoint in the cytokine pathway. The function of the cytokine — and so the opportunities for its regulation — start with its production and secretion. Numerous proteins such as latency-associated protein (LAP), Follistatin, Noggin, Chordin and the related factors Caronte, Cerberus and Gremlin have been identified that bind TGF-β family factors and prevent their contacts with signalling receptors (reviewed in REF. 66). Several membrane-anchored factors, including Betaglycan, Endoglin and Crypto seem to facilitate ligand binding to TGF-β family receptors1,2. Some of these have a dual role. For example, Betaglycan facilitates TGF-β binding to its signalling receptors83, but also facilitates the binding of Inhibin — an Activin antagonist — to Activin receptors, decreasing cell stimulation by Activin84. Although cells read TGF-β family signals through membrane receptors and intracellular networks, the ability of a cell to exert control over the input may extend all the way to the pre-receptor phase of the cytokine pathway. Indeed, several of the proteins that control the activation of TGF-β family members or their access to membrane receptors may function as mediators of feedback or crosstalk control (FIG. 2). Examples of regulation at this level are provided by the BMP-sequestering factor Caronte, whose expression on the left lateral plate mesoderm is induced by SHH signalling in the chick embryo30,31. The BMP inhibitor Gremlin is involved in a similar regulatory loop in response to SHH signals during the outgrowth and patterning of the vertebrate limb85. The expression of another BMP inhibitor, Noggin, is increased by BMP in 86 CHONDROCYTE and OSTEOBLAST cultures , and that of the Activin antagonist Follistatin is increased by Activin in anterior pituitary cell cultures87. A complex web of interacting proteins are implicated in the activation of latent TGF-β (reviewed in REFS 1,66), and there is compelling genetic evidence that at least one of these, Thrombospondin, has such a function in vivo88. Quantitative signals, qualitative outputs

Inputs that control the level of a SMAD signal can have both quantitative and qualitative effects on the response. As well as adjusting the amplitude or duration of a SMAD response, some inputs can activate different sets of genes at different signal thresholds. This is because certain genes are activated only by levels of signal that are intermediate. A classical example of this phenomenon is provided by the different cell fates that are induced by different concentrations of Activin added to Xenopus ECTODERM explants89. The induction of different cell fates is a function of the level of Activin/Nodal receptor occupancy by ligand90 and can be partly recapitulated by injection of increasing amounts of SMAD2 transcript28. The mechanistic basis for the differential regulation of target genes by different levels of SMAD signal remains unknown.

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 1 | DECEMBER 2000 | 1 7 5

REVIEWS

Epithelial cell

B cell

Osteoblast

Xenopus mesoendoderm

TGF-β

BMP

Xnr

SMAD3

Smad1

Smad2

β-catenin

Jun

Cbfa3

Cbfa1

Mixer

Lef1/Tcf

AP1

CBFA3

CBFA1

Mixer

Wnt

Xtwn Xtwn

PAI-1 other genes

Osteogenic genes

IgCα

goosecoid

Figure 5 | Self-modifying signals. Several cases illustrate the idea that SMAD signalling at one stage can determine the response to SMAD signalling at another. In epithelial cells and other cell types, transforming growth factor-β (TGF-β)/SMAD signalling induces the expression of c-jun and JUNB, which encode components of AP-1 transcriptional complexes that may cooperate with SMADs in the activation of pai-1 (plasminogen activator inhibitor-1) and other regulators of the extracellular matrix. In B cells, TGF-β /SMAD signalling activates expression of CBFA3 (core-binding factor A ) which, in turn, may cooperate with SMAD3 in the activation of the immunoglobulin-α constant region promoter, prompting immunoglobulin A (IgA) class switching. In osteoblasts, BMP/SMAD signals induce expression of CBFA1, a mediator of osteoblast differentiation in cooperation with SMAD1. In the development of dorsal mesoderm in Xenopus, Nodal-related (Xnr) signals acting through SMAD2 augment the expression of Mixer, whose product associates with SMAD2 to mediate goosecoid activation. This occurs in cooperation with Xtwn (Xenopus Twin) whose own expression is induced by SMAD2–SMAD4 in cooperation with Wnt-activated lymphoid enhancer-binding factor 1/T cell-specific factor (LEF1/TCF).

ASTROCYTES

Star-shaped cells that support the tissue of the central nervous system. IMMUNOGLOBULIN-α CONSTANT REGION

Region of an antibody molecule that is constant within — and defines — each of the basic classes of immunoglobulin. ANTIBODY CLASS SWITCHING

Process by which the region of an immunoglobulin heavy-chain gene that encodes the antigen recognizing (variable) portion is recombined with the constant region of a different immunoglobulin class. MESENCHYME

Loosely organized, undifferentiated mesodermal cells. MESOENDODERM

Gives rise to both the mesoderm and the endodermal tissue of the embryo. DORSAL MARGINAL ZONE

Region of the Xenopus embryo that gives rise to the dorsal mesoderm. GASTRULA

Multilayered embryo with an outer cell layer (ectoderm), an inner cell layer (endoderm), and an intermediate cell layer (mesoderm).

176

The ERK-activated Ras pathway deserves special mention here because it can modify TGF-β signalling pathways at different levels (FIG. 4). In the normal control of cell proliferation, the antimitogenic effect of TGF-β is dominant over the effect of mitogenic factors that signal through Ras. However, a hyperactive Ras pathway (for example, in cells expressing a ras oncogene) can effectively counteract the antiproliferative activity of TGF-β by a combination of mechanisms that include downregulation of TGF-β receptors91, attenuation of SMAD accumulation in the nucleus73,74, and perhaps other mechanisms (FIG. 4). In contrast, Ras signals strongly cooperate with SMADs during mesoderm induction in Xenopus92,93 and invasion by mouse and human carcinoma cells94. In other words, Ras can determine the overall response of the cell to TGF-β signals. Inputs that meet SMADs downstream of the signal-transduction process, at the level of target gene promoters, function as qualitative modifiers of SMAD responses. Many developmental processes are under the joint control of the SMAD pathway with the Wnt, hedgehog or Ras pathways3. Concrete examples are provided by inputs that activate transcription factors that collaborate with SMADs to activate certain target genes (FIG. 2). This is the case with LEF1/TCF, which is activated by the Wnt/β-catenin pathway and can physically interact with SMAD to activate certain promoters47,48. In a related example, STAT3, activated by the cytokine leukaemia inhibitory factor (LIF), and SMAD1, activated by BMP, can be bridged by the coactivator p300, and this may underlie the synergy of the cytokines during the differentiation of mouse neuroepithelial cell cultures into ASTROCYTES95.

Self-modifying signals

The response to a SMAD-activating signal at one stage can determine the response to the same signal at another (FIG. 5). For example, if TGF-β can induce the expression of Jun family members65 and increase the activity of JNKs, and if Jun-containing AP-1 complexes can modify the activity of SMAD complexes as mentioned above, in principle at least, TGF-β could change the outcome of its own SMAD pathway by affecting the level of AP-1 in the cell. These considerations also apply to CBFA/AML transcription factors (CBFA1, CBFA2 and CBFA3; also known as AML3, AML1, and AML2, respectively) as both targets and partners of activated SMADs. CBFA3 is implicated in the activation of the germline IMMUNOGLOBULIN-α CONSTANT REGION (IgCα) gene by TGF-β, which stimulates ANTIBODY CLASS SWITCHING to IgA immunoglobulin in B cells96. The CBFA3 protein associates with SMAD3–SMAD4, and the resulting complex recognizes and stimulates the IgCα promoter45,46. But it turns out that TGF-β increases the expression of CBFA3 in spleen B cells46, thereby setting the stage for a subsequent collaboration of CBFA3 with SMADs (FIG. 5). Furthermore, CBFA1 is critical for osteoblast differentiation and bone formation, as shown by the absence of bone in Cbfa1 null mice97,98 and the heritable disorder cleidocranial dysplasia caused by CBFA1 mutation in humans99,100. CBFA1 expression is increased by BMP7 in MESENCHYMAL cells44, which may, in turn, augment the ability of BMP-activated SMAD1 to cooperate with CBFA1 in transcriptional activation of candidate osteoblast differentiation genes46. An elaborate example of a self-modifying signalling process is provided by the regulation of goosecoid in Xenopus embryos (FIG. 5). goosecoid is a MESOENDODERMAL gene whose expression is fairly restricted to the DORSAL MARGINAL ZONE of the early GASTRULA (REF. 39 and references therein). Activation of goosecoid in vivo largely depends on the synergistic interaction between a proximal enhancer element that responds to the transcription factor Xtwn101 and a distal enhancer element that responds to SMAD2 together with Mixer39. Both Mixer and Xtwn are put in place partly by SMAD2 signalling. Mixer is induced by an Activin-like factor acting through SMAD2, possibly with help from FAST102. Xtwn expression, on the other hand, is activated by Wnt signalling through β-catenin and LEF1/TCF in cooperation with the SMADs47,48 (FIG. 5). So, signalling through SMAD2 seems to have the capacity to do both — create a context for accurate goosecoid activation and then execute this activation. A satisfying challenge

The elucidation of the TGF-β/SMAD pathway — from cytokine to target gene — and the simultaneous identification of several regulators and signals controlling this pathway, is both satisfying and challenging. It is satisfying because it provides, in concrete molecular terms, a framework for understanding how the cellular context determines a cell’s response to TGF-β family signals. The response depends as much on the activity of partners and modulators that exert control and determine targets www.nature.com/reviews/molcellbio

| DECEMBER 2000 | VOLUME 1

REVIEWS as it does on the basic TGF-β signal itself. The network of partners, modulators and regulatory inputs is obviously complex. In principle, this complexity could be managed by assessing the levels and activity of the entire set of proteins impinging on the signal–response transaction. Profiling the entire set of proteins that are relevant to this process for a given cell type and set of conditions is challenging, but possible with current technology. Profiling the post-translational modifications and other variables affecting the activity of these proteins is more difficult, but equally relevant to those who wish to gain a holistic view of a signalling network. More challenging still is the ease with which the activity of these components, and so of the whole signalling network, varies depending on the environment of the cell. To better understand TGF-β action and how it is affected in disorders and in response to bioactive agents, it will be necessary to achieve a progressively more integrated view of this signalling system. And given the effort that this takes, it will also be necessary to focus on physiological conditions that matter.

1. 2. 3. 4.

10.

11.

12.

13.

14.

15.

16.

Massagué, J. TGFβ signal transduction. Annu. Rev. Biochem. 67, 753–791 (1998). Schier, A. F. & Shen, M. M. Nodal signalling in vertebrate development. Nature 403, 385–389 (2000). Whitman, M. SMADs and early developmental signaling by the TGFβ superfamily. Genes Dev. 12, 2445–2462 (1998). Massagué, J., Blain, S. W. & Lo, R. S. TGF-β signaling in growth control, cancer and heritable disorders. Cell 103, 295–309 (2000). Wrana, J. L., Attisano, L., Wieser, R., Ventura, F. & Massagué, J. Mechanism of activation of the TGF-β receptor. Nature 370, 341–347 (1994). Elucidation of the mechanisms of receptor activation based on a combined biochemical and genetic approach. Wieser, R., Wrana, J. L. & Massagué, J. GS domain mutations that constitutively activate TβR-I, the downstream signaling component in the TGF-β receptor complex. EMBO J. 14, 2199–2208 (1995). Lo, R. S. & Massagué, J. Ubiquitin-dependent degradation of TGF-β-activated SMAD2. Nature Cell Biol. 1, 472–478 (1999). Zhu, H., Kavsak, P., Abdollah, S., Wrana, J. L. & Thomsen, G. H. A SMAD ubiquitin ligase targets the BMP pathway and affects embryonic pattern formation. Nature 400, 687–693 (1999). Raftery, L. A. & Sutherland, D. J. TGF-β family signal transduction in Drosophila development: from Mad to SMADs. Dev. Biol. 210, 251–268 (1999). A chronicle of the discovery of MAD, the founding member of the SMAD family. Xu, L., Chen, Y. G. & Massagué, J. The nuclear import function of SMAD2 is masked by SARA and unmasked by TGFβ-dependent phosphorylation. Nature Cell Biol. 2, 559–562 (2000). Hahn, S. A. et al. DPC4, a candidate tumor suppressor gene at human chromosome 18q21.1. Science 271, 350–353 (1996). Lagna, G., Hata, A., Hemmati-Brivanlou, A. & Massagué, J. Partnership between DPC4 and SMAD proteins in TGFβ signalling pathways. Nature 383, 832–836 (1996). Reveals the role of SMAD4 as a shared partner of receptor-activated SMADs. Chen, X. et al. SMAD4 and FAST-1 in the assembly of activin-responsive factor. Nature 389, 85–89 (1997). Delineates the role of FAST as a SMAD DNA-binding cofactor Liu, F., Pouponnot, C. & Massagué, J. Dual role of the SMAD4/DPC4 tumor suppressor in TGFβ-inducible transcriptional responses. Genes Dev. 11, 3157–3167 (1997). de Caestecker, M. P. et al. The SMAD4 activation domain (SAD) is a proline-rich, p300-dependent transcriptional activation domain. J. Biol. Chem. 275, 2115–2122 (2000). Janknecht, R., Wells, N. J. & Hunter, T. TGF-β-stimulated

cooperation of SMAD proteins with the coactivators CBP/p300. Genes Dev. 12, 2114–2119 (1998). 17. Massagué, J. & Wotton, D. Transcriptional control by the TGF–β/SMAD signaling system. EMBO J. 19, 1745–1754 (2000). 18. Yahata, T. et al. The MSG1 non-DNA-binding transactivator binds to the p300/CBP coactivators, enhancing their functional link to the SMAD transcription factors. J. Biol. Chem. 275, 8825–8834 (2000). 19. Tsukazaki, T., Chiang, T. A., Davison, A. F., Attisano, L. & Wrana, J. L. SARA, a FYVE domain protein that recruits SMAD2 to the TGFβ receptor. Cell 95, 779–791 (1998). Identification of a cytoplasmic regulator of SMAD movement to TGF-β receptors. 20. Conti, E., Uy, M., Leighton, L., Blobel, G. & Kuriyan, J. Crystallographic analysis of the recognition of a nuclear localization signal by the nuclear import factor karyopherinα. Cell 94, 193–204 (1998). 21. Xiao, Z., Liu, X. & Lodish, H. F. Importin-β mediates nuclear translocation of SMAD 3. J. Biol. Chem. 275, 23425–23428 (2000). 22. Morén, A., Itoh, S., Moustakas, A., Dijke, P. & Heldin, C. H. Functional consequences of tumorigenic missense mutations in the amino-terminal domain of SMAD4. Oncogene 19, 4396–4404 (2000). 23. Jones, J. B. & Kern, S. E. Functional mapping of the MH1 DNA-binding domain of DPC4/SMAD4. Nucleic Acids Res. 28, 2363–2368 (2000). 24. Shi, Y. et al. Crystal structure of a SMAD MH1 domain bound to DNA: Insights on DNA-binding in TGF-β signaling. Cell 94, 585–594 (1998). 25. Watanabe, M., Masuyama, N., Fukuda, M. & Nishida, E. Regulation of intracellular dynamics of SMAD4 by its leucine-rich nuclear export signal. EMBO Rep. 1, 176–182 (2000). 26. Masuyama, N., Hanafusa, H., Kusakabe, M., Shibuya, H. & Nishida, E. Identification of two SMAD4 proteins in Xenopus. Their common and distinct properties. J. Biol. Chem. 274, 12163–12170 (1999). 27. Chen, Y. G. et al. Determinants of specificity in TGF-β signal transduction. Genes Dev. 12, 2144–2152 (1998). 28. Graff, J. M., Bansal, A. & Melton, D. A. Xenopus Mad proteins transduce distinct subsets of signals for the TGF-β superfamily. Cell 85, 479–487 (1996). The opposing but complementary roles of SMAD1 and SMAD2 are revealed. 29. Baker, J. & Harland, R. M. A novel mesoderm inducer, mMadr-2, functions in the activin signal transduction pathway. Genes Dev. 10, 1880–1889 (1996). 30. Rodriguez Esteban, C. et al. The novel Cer-like protein Caronte mediates the establishment of embryonic left–right asymmetry. Nature 401, 243–251 (1999). 31. Yokouchi, Y., Vogan, K. J., Pearse, R. V. & Tabin, C. J. Antagonistic signaling by Caronte, a novel Cerberusrelated gene, establishes left–right asymmetric gene

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

expression. Cell 98, 573–583 (1999). 32. Saijoh, Y. et al. Left–right assymetric expression of lefty2 and nodal is induced by a signaling pathway that includes the transcription factor FAST2. Mol. Cell 5, 35–47 (2000). References 31 and 32 report on regulation of BMP activity by Hedgehog in the extracellular space. 33. Feng, X. H. & Derynck, R. A kinase subdomain of transforming growth factor-β (TGF-β) type I receptor determines the TGF-β intracellular signaling activity. EMBO J. 16, 3912–3922 (1997). 34. Lo, R. S., Chen, Y. G., Shi, Y. G., Pavletich, N. & Massagué, J. The L3 loop: a structural motif determining specific interactions between SMAD proteins and TGF-β receptors. EMBO J. 17, 996–1005 (1998). 35. Yagi, K. et al. Alternatively spliced variant of SMAD2 lacking exon 3. Comparison with wild-type SMAD2 and SMAD3. J. Biol.Chem. 274, 703–709 (1999). 36. Ashcroft, G. S. et al. Mice lacking SMAD3 show accelerated wound healing and an impaired local inflammatory response. Nature Cell Biol. 1, 260–266 (1999). 37. Denissova, N. G., Pouponnot, C., Long, J., He, D. & Liu, F. Transforming growth factor β-inducible independent binding of SMAD to the SMAD7 promoter. Proc. Natl Acad. Sci. USA 97, 6397–6402 (2000). 38. Hata, A. et al. OAZ uses distinct DNA- and protein-binding zinc fingers in separate BMP–SMAD and Olf signaling pathways. Cell 100, 229–240 (2000). 39. Germain, S., Howell, M., Esslemont, G. M. & Hill, C. S. Homeodomain and winged-helix transcription factors recruit activated SMADs to distinct promoter elements via a common SMAD interaction motif. Genes Dev. 14, 435–451 (2000). 40. Yeo, C. Y., Chen, X. & Whitman, M. The role of FAST-1 and SMADs in transcriptional regulation by activin during early Xenopus embryogenesis. J. Biol. Chem. 274, 26584–26590 (1999). 41. Hua, X., Miller, Z. A., Wu, G., Shi, Y. & Lodish, H. F. Specificity in transforming growth factor β-induced transcription of the plasminogen activator inhibitor-1 gene: interactions of promoter DNA, transcription factor muE3, and SMAD proteins. Proc. Natl Acad. Sci. USA 96, 13130–13135 (1999). 42. Chen, X., Rubock, M. J. & Whitman, M. A transcriptional partner of MAD proteins in TGF-β signalling. Nature 383, 691–696 (1996). 43. Zhang, Y., Feng, X. H. & Derynck, R. SMAD3 and SMAD4 cooperate with c-Jun/c-Fos to mediate TGF-β-induced transcription. Nature 394, 909–913 (1998). 44. Tsuji, K., Ito, Y. & Noda, M. Expression of the PEBP2αA/AML3/CBFA1 gene is regulated by BMP4/7 heterodimer and its overexpression suppresses type I collagen and osteocalcin gene expression in osteoblastic and nonosteoblastic mesenchymal cells. Bone 22, 87–92 (1998).

VOLUME 1 | DECEMBER 2000 | 1 7 7

REVIEWS

45. Pardali, E. et al. SMAD and AML proteins synergistically confer transforming growth factor β1 responsiveness to human germ-line IgA genes. J. Biol. Chem. 275, 3552–3560 (2000). 46. Hanai, J. et al. Interaction and functional cooperation of PEBP2/CBF with SMADs. Synergistic induction of the immunoglobulin germline Cα promoter. J. Biol. Chem. 274, 31577–31582 (1999). 47. Nishita, M. et al. Interaction between Wnt and TGF-β signalling pathways during formation of Spemann’s organizer. Nature 403, 781–785 (2000). Evidence for direct cooperation between mediators of the TGF-β and Wnt pathways. 48. Labbe, E., Letamendia, A. & Attisano, L. Association of SMADs with lymphoid enhancer binding factor 1/T cellspecific factor mediates cooperative signaling by the transforming growth factor-β and wnt pathways. Proc. Natl Acad. Sci. USA 97, 8358–8363 (2000). 49. Wotton, D., Lo, R. S., Lee, S. & Massagué, J. A SMAD transcriptional corepressor. Cell 97, 29–39 (1999). The first SMAD transcriptional corepressor and its competition with co-activators. 50. Luo, K. et al. The ski oncoprotein interacts with the SMAD proteins to repress TGF-β signaling. Genes Dev. 13, 2196–2206 (1999). 51. Sun, Y., Liu, X., Ng-Eaton, E., Lodish, H. F. & Weinberg, R. A. SnoN and Ski protooncoproteins are rapidly degraded in response to transforming growth factor-β signaling. Proc. Natl Acad. Sci. USA 96, 12442–12447 (1999). References 50 and 51 report on the proto-oncogene SKI as a corepressor of SMADs in the basal state. 52. Perou, C. M. et al. Distinctive gene expression patterns in human mammary epithelial cells and breast cancers. Proc. Natl Acad. Sci. USA 96, 9212–9217 (1999). 53. Stroschein, S. L., Wang, W., Zhou, S., Zhou, Q. & Luo, K. Negative feedback regulation of TGF-b signaling by the SnoN oncoprotein. Science 286, 771–774 (1999). 54. Lo, R. S., Wotton, D. & Massagué, J. EGF signaling via Ras stabilizes the SMAD transcriptional corepressor TGIF. EMBO J. (in the press). 55. Gripp, K. W. et al. Mutations in TGIF cause holoprosencephaly and link nodal signalling to human neural axis determination. Nature Genet. 25, 205–208 (2000). 56. Sampath, K. et al. Induction of the zebrafish ventral brain and floorplate requires cyclops/nodal signalling. Nature 395, 185–189 (1998). 57. Nomura, M. & Li, E. SMAD2 role in mesoderm formation, left–right patterning and craniofacial development. Nature 393, 786–790 (1998). 58. Hocevar, B. A., Brown, T. L. & Howe, P. H. TGF-β induces fibronectin synthesis through a c-Jun N-terminal kinasedependent, SMAD4-independent pathway. EMBO J. 18, 1345–1356 (1999). An example of rapid activation of JNK by TGF-β and its effects on gene expression. 59. Hanafusa, H. et al. Involvement of the p38 mitogenactivated protein kinase pathway in transforming growth factor-β-induced gene expression. J. Biol. Chem. 274, 27161–27167 (1999). 60. Sano, Y. et al. ATF-2 is a common nuclear target of SMAD and TAK1 pathways in transforming growth factor-β signaling. J. Biol. Chem. 274, 8949–8957 (1999). 61. Takatsu, Y. et al. TAK1 participates in c-Jun N-terminal kinase signaling during Drosophila development. Mol. Cell. Biol. 20, 3015–3026 (2000). 62. Shibuya, H. et al. TAB1: an activator of the TAK1 MAPKKK in TGF-β signal transduction. Science 272, 1179–1182 (1996). 63. Yamaguchi, K. et al. XIAP, a cellular member of the inhibitor of apoptosis protein family, links the receptors to TAB1–TAK1 in the BMP signaling pathway. EMBO J. 18, 179–187 (1999). 64. Engel, M. E., McDonnell, M. A., Law, B. K. & Moses, H. L. Interdependent SMAD and JNK signaling in transforming growth factor-β-mediated transcription. J. Biol. Chem. 274, 37413–37420 (1999). 65. Wong, C. et al. SMAD3–SMAD4 and AP-1 complexes

178

66. 67.

68.

69.

70.

71.

72.

73.

74.

75.

76. 77.

78.

79.

80.

81.

82. 83.

84.

85.

86.

87.

synergize in transcriptional activation of the c-Jun promoter by transforming growth factor-β. Mol. Cell. Biol. 19, 1821–1830 (1999). Massagué, J. & Chen, Y. G. Controlling TGF-β signaling. Genes Dev. 14, 627–644 (2000). ten Dijke, P., Miyazono, K. & Heldin, C. H. Signaling inputs converge on nuclear effectors in TGF-β signaling. Trends Biochem. Sci. 25, 64–70 (2000). Tanimoto, H., Itoh, S., ten Dijke, P. & Tabata, T. Hedgehog creates a gradient of DPP activity in Drosophila wing imaginal discs. Mol. Cell 5, 59–71 (2000). Crosstalk between the Hedgehog and the BMP system at the level of Decapentaplegic receptors. Hayashi, H. et al. The MAD-related protein SMAD7 associates with the TGFβ receptor and functions as an antagonist of TGFβ signaling. Cell 89, 1165–1173 (1997). Nakao, A. et al. Identification of SMAD7, a TGFβ-inducible antagonist of TGF-β signaling. Nature 389, 631–635 (1997). First evidence of antagonistic SMADs relaying feedback signals. Ulloa, L., Doody, J. & Massagué, J. Inhibition of transforming growth factor-β/SMAD signalling by the interferon-γ/STAT pathway. Nature 397, 710–713 (1999). Negative regulation of the SMAD pathway by the STAT pathway. Bitzer, M. et al. A mechanism of suppression of TGFβ/SMAD signaling by NFκB/RelA. Genes Dev. 14, 187–197 (2000). Kretzschmar, M., Doody, J. & Massagué, J. Opposing BMP and EGF signalling pathway converge on the TGFβ family mediator SMAD1. Nature 389, 618–622 (1997). Negative effects of Ras/ERK pathway on SMAD function. Kretzschmar, M., Doody, J., Timokhina, I. & Massagué, J. A mechanism of repression of TGF-β/SMAD signaling by oncogenic ras. Genes Dev. 13, 804–816 (1999). Halfon, M. S. et al. Ras pathway specificity is determined by the integration of multiple signal-activated and tissuerestricted transcription factors. Cell 103, 63–74 (2000). Onichtchouk, D. et al. Silencing of TGF-β signalling by the pseudoreceptor BAMBI. Nature 401, 480–485 (1999). Tsuneizumi, K. et al. Daughters against dpp modulates dpp organizing activity in Drosophila wing development. Nature 389, 627–631 (1997). Hata, A., Lagna, G., Massagué, J. & Hemmati-Brivanlou, A. SMAD6 inhibits BMP/SMAD1 signaling by specifically competing with the SMAD4 tumor suppressor. Genes Dev. 12, 186–197 (1998). Yamada, M., Szendro, P. I., Prokscha, A., Schwartz, R. J. & Eichele, G. Evidence for a role of SMAD6 in chick cardiac development. Dev. Biol. 215, 48–61 (1999). Zhao, J., Shi, W., Chen, H. & Warburton, D. SMAD7 and SMAD6 differentially modulate transforming growth factorβ-induced inhibition of embryonic lung morphogenesis. J. Biol. Chem. 275, 23992–23997 (2000). Galvin, K. M. et al. A role for SMAD6 in development and homeostasis of the cardiovascular system. Nature Genet. 24, 171–174 (2000). Imamura, T. et al. SMAD6 inhibits signalling by the TGFβ superfamily. Nature 389, 622–626 (1997). López-Casillas, F., Wrana, J. L. & Massagué, J. Betaglycan presents ligand to the TGF-β signaling receptor. Cell 73, 1435–1444 (1993). Lewis, K. A. et al. Betaglycan binds inhibin and can mediate functional antagonism of activin signalling. Nature 404, 411–414 (2000). The surprising versatility of a TGF-β accessory receptor is revealed. Zúñiga, A., Haramis, A. P., McMahon, A. P. & Zeller, R. Signal relay by BMP antagonism controls the SHH/FGF4 feedback loop in vertebrate limb buds. Nature 401, 598–602 (1999). Gazzerro, E., Gangji, V. & Canalis, E. Bone morphogenetic proteins induce the expression of noggin, which limits their activity in cultured rat osteoblasts. J. Clin. Invest. 102, 2106–2114 (1998). Bilezikjian, L. M., Corrigan, A. Z., Blount, A. L. & Vale, W. W. Pituitary follistatin and inhibin subunit messenger ribonucleic

88. 89.

90.

91.

92.

93.

94.

95.

96.

97.

98.

99.

100.

101.

102.

103.

104.

105.

106.

107.

108.

109.

acid levels are differentially regulated by local and hormonal factors. Endocrinology 137, 4277–4284 (1996). Crawford, S. E. et al. Thrombospondin-1 is a major activator of TGF-β1 in vivo. Cell 93, 1159–1170 (1998). Green, J. B. A., New, H. V. & Smith, J. C. Responses of embryonic Xenopus cells to activin and FGF are separated by multiple dose thresholds and correspond to distinct axes of the mesoderm. Cell 71, 731–739 (1992). Dyson, S. & Gurdon, J. B. The interpretation of position in a morphogen gradient as revealed by occupancy of activin receptors. Cell 93, 557–568 (1998). A saga showing the morphogenetic potential of activin signalling during Xenopus embryogenesis. Zhao, J. & Buick, R. N. Regulation of transforming growth factor-β receptors in H-ras oncogene-transformed rat intestinal epithelial cells. Cancer Res. 55, 6181–6188 (1995). Cornell, R. A. & Kimelman, D. Activin-mediated mesoderm induction requires FGF. Development 120, 453–462 (1994). LaBonne, C. & Whitman, M. Mesoderm induction by activin requires FGF-mediated intracellular signals. Development 120, 463–472 (1994). Oft, M. et al. TGF-β1 and Ha-Ras collaborate in modulating the phenotypic plasticity and invasiveness of epithelial tumor cells. Genes Dev. 10, 2462–2477 (1996). TGF-β is shown to instigate tumorigenic behaviour in transformed mammary cells. Nakashima, K. et al. Synergistic signaling in fetal brain by STAT3–SMAD1 complex bridged by p300. Science 284, 479–482 (1999). Shi, M. J. & Stavnezer, J. CBF α3 (AML2) is induced by TGF-β1 to bind and activate the mouse germline Igα promoter. J. Immunol. 161, 6751–6760 (1998). Otto, F. et al. Cbfa1, a candidate gene for cleidocranial dysplasia syndrome, is essential for osteoblast differentiation and bone development. Cell 89, 765–771 (1997). Komori, T. et al. Targeted disruption of Cbfa1 results in a complete lack of bone formation owing to maturational arrest of osteoblasts. Cell 89, 755–764 (1997). Lee, B. et al. Missense mutations abolishing DNA binding of the osteoblast-specific transcription factor OSF2/CBFA1 in cleidocranial dysplasia. Nature Genet. 16, 307–310 (1997). Mundlos, S. et al. Mutations involving the transcription factor CBFA1 cause cleidocranial dysplasia. Cell 89, 773–779 (1997). Laurent, M. N., Blitz, I. L., Hashimoto, C., Rothbacher, U. & Cho, K. W. The Xenopus homeobox gene twin mediates Wnt induction of goosecoid in establishment of Spemann’s organizer. Development 124, 4905–4916 (1997). Henry, G. L. & Melton, D. A. Mixer, a homeobox gene required for endoderm development. Science 281, 91–96 (1998). Huse, M., Chen, Y. G., Massagué, J. & Kuriyan, J. Crystal structure of the cytoplasmic domain of the type I TGF-β receptor in complex with FKBP12. Cell 96, 425–436 (1999). Shi, Y., Hata, A., Lo, R. S., Massagué, J. & Pavletich, N. P. A structural basis for mutational inactivation of the tumour suppressor SMAD4. Nature 388, 87–93 (1997). Wu, G. et al. Structural basis of SMAD2 recognition by the SMAD Anchor for Receptor Activation. Science 287, 92–97 (2000). Kim, J., Johnson, K., Chen, H. J., Carroll, S. & Laughon, A. Drosophila Mad binds to DNA and directly mediates activation of vestigial by Decapentaplegic. Nature 388, 304–308 (1997). Liu, F. et al. A human Mad protein acting as a BMPregulated transcriptional activator. Nature 381, 620–623 (1996). Dou, C. et al. BF-1 interferes with transforming growth factor-β signaling by associating with SMAD partners. Mol. Cell. Biol. 20, 6201–6211 (2000). Kim, R. H. et al. A novel SMAD nuclear interacting protein, SNIP1, suppresses p300-dependent TGF-β signal transduction. Genes Dev. 14, 1605–1616 (2000).

www.nature.com/reviews/molcellbio

| DECEMBER 2000 | VOLUME 1

REVIEWS

THE MANY SUBSTRATES AND FUNCTIONS OF ATM Michael B. Kastan and Dae-sik Lim As its name suggests, the ATM — ‘ataxia-telangiectasia, mutated’ — gene is responsible for the rare disorder ataxia-telangiectasia. Patients show various abnormalities, mainly in their responses to DNA damage, but also in other cellular processes. Although it is hard to understand how a single gene product is involved in so many physiological processes, a clear picture is starting to emerge. Ataxia-telangiectasia is a rare, autosomal-recessive inherited disorder with a complex clinical phenotype (BOX 1). The gene responsible, ATM, was cloned in 1995 (REF. 1) (BOX 2), and found to encode a 370-kDa protein kinase. ATM belongs to a family of proteins, conserved from yeast to humans2, that regulate cell-cycle checkpoints and are involved in DNA repair and recombination. This fits with the clinical phenotype of ataxiatelangiectasia — cells from patients show abnormal responses to ionizing radiation, checkpoint alterations in the cell cycle, and increased chromosomal breakage and telomere end fusions. However, their responses to ultraviolet irradiation and base-damaging agents are relatively normal, indicating that ATM acts specifically in the cellular responses to ionizing radiation and DNA double-stranded breaks (DSBs), rather than having a more general function in DNA repair. But various other abnormalities in patients have raised the possibility that ATM may also be involved in other cellular processes3. So far, studies on ATM function indicate that the complex spectrum of abnormalities in cells and patients lacking ATM reflects the relatively large number of substrates of ATM in various pathways. Department of Hematology–Oncology, Saint Jude Children’s Research Hospital, D1034, 332 North Lauderdale Street, Memphis, Tennessee 38105, USA. Correspondence to M.B.K. e-mail: michael.kastan@stjude.org

The ATM family

The ATM protein has a carboxy-terminal sequence with significant homology to the catalytic domain of phosphatidylinositol-3-OH kinase (PI(3)K). ATM-like proteins from several organisms2 have a similar size and structure (TABLE 1), and they all contain PI(3)Klike domains. So we might expect them to be lipid kinases. But some of them — such as the DNA-depen-

dent protein kinase (DNA-PK) and mTOR (target of rapamycin) — have been shown to be serine/threonine protein kinases4,5. What, then, are the downstream targets of ATM? In vitro, ATM can phosphorylate the tumour supressor protein p53 at serine 15. Moreover, activity of the enzyme increases after cells are exposed to ionizing radiation or radio-mimetic drugs6,7, and can be abolished by mutation at two conserved amino acids. In other words, ATM has bona fide intrinsic protein kinase activity. So far, however, it has not been reported to have measurable lipid kinase activity. The biochemical properties of ATM have been compared in vitro with those of two related mammalian proteins — ATR (ataxia-telangiectasia and Rad3-related kinase) and DNA-PK — and there are some differences8. For example, whereas the activities of ATM and ATR depend on Mn2+, DNA-PK activity is unaffected by changes in this divalent cation6,8,9. Many members of the ATM family seem to require Mn2+ for their protein-kinase activity — indeed, the p110 catalytic subunit of PI(3)K also has a Mn2+-dependent serine/threonine kinase activity in addition to its lipidkinase activity10. However, all of these cofactor requirements are based on in vitro assays, and this may not be the case in vivo. Another biochemical difference between these three similar enzymes is their dependence on DNA ends and the Ku proteins for optimal activity. The Ku proteins (KU70 and KU80) bind specifically to double-stranded DNA ends and enhance the activity of DNA-PK (probably by promoting its association with the DNA ends)4.

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 1 | DECEMBER 2000 | 1 7 9

REVIEWS

Box 1 | Clinical phenotypes of ataxia-telangiectasia Clinical phenotype • Cerebellar ataxia: progressive neuronal degeneration • Immunological dysfunction: low IgA and IgE; normal V(D)J recombination • Cancer predisposition: lymphoma and leukaemia; chromosomal translocation • Hypogonadism • Sensitivity to ionizing radiation • Premature ageing • Increased alpha-fetoprotein • Small stature • Insulin resistance Cellular phenotype • DNA damage and repair: chromosomal instability; cell-cycle checkpoint defects in G1, S (radio-resistant DNA synthesis) and G2/M; sensitivity to ionizing radiation, increased chromosomal breakages, telomere end-to-end fusions; DNA repair; subtle, but normal kinetics • Other abnormalities: cytoskeletal defects, abnormalities in the plasma membrane, an increased number of lysosomes; high levels of trophic factors needed for growth; defects in intracellular signalling

REPLICATION PROTEIN A

A single-stranded DNAbinding factor that is essential for DNA repair, recombination and replication. V(D)J RECOMBINATION

A specialized form of recombination that assembles the genes that encode lymphocyte antigen receptors from variable (V), diversity (D) and joining (J) gene segments. DNA double-strand breaks are introduced between the V, D and J segments and DNA repair proteins then join the segments together.

This makes sense, as DNA-PK is involved in repairing DSBs. Because ATM is also activated by agents that cause DSBs, and because it seems to be a critical signalling molecule in response to such breaks, it would be no surprise to discover that its activity is also affected by DNA ends or the Ku proteins. But neither immunoprecipitated nor purified ATM seems to be directly activated by DNA ends6,8,9. However, purified ATM bound to DNA ends has been observed by atomic-force microscopy11. Moreover, ATM-dependent phosphorylation of REPLICATION PROTEIN A (RPA) can be stimulated by exogenous DNA ends9,12. Perhaps, by binding to DNA, the RPA undergoes a conformational change that makes it a better substrate for ATM. Given that the purification method of ATM, the source or amount of substrate and cofactors, and the in vitro assay system could all influence the apparent effects of DNA ends on the activity of ATM, the picture is still not clear. Furthermore, it is not clear whether ATR is activated by DNA ends. As well as the distinct biochemical properties of ATM, ATR and DNA-PK, mice lacking these three enzymes show phenotypic differences. For example,

Table 1 | ATM-related proteins Organism

Protein

Functions

Schizosaccharomyces pombe67 Saccharomyces cerevisiae Saccharomyces cerevisiae102

Rad3p Mec1p Tel1p

DNA-damage response pathways, transcriptional regulation, cell-cycle arrest, DNA repair, maintenance of telomeres

Drosophila melanogaster103

MEI-41

Irradiation-induced delay of entry into mitosis

Vertebrate4

DNA-PK

Ser/Thr kinase involved in DNA repair and V(D)J recombination

Vertebrate5

mTOR/FRAP Ser/Thr kinase. Regulates protein synthesis in response to growth factors or hormones

101,104,105

Vertebrate

180

ATR

DNA-damage response pathways

mice with mutations in the gene encoding the DNA-PK catalytic subunit (DNA-PKCS) have severe immunodeficiencies (owing to a defect in V(D)J RECOMBINATION), but a low frequency of lymphoid malignancies4. In contrast, both mice and people lacking ATM have a relatively mild immunodeficiency, but many lymphoid malignancies13,14. Deletion of ATR, by contrast, results in an embryonic-lethal phenotype, indicating that ATR is an essential gene15. Other differences among this gene family are apparent in their roles in the cell-cycle-checkpoint response to ionizing radiation. Checkpoints are surveillance systems that maintain genomic integrity after damage of DNA, or other cellular macromolecules, by agents such as ionizing radiation or ultraviolet light. These checkpoints either result in programmed cell death (apoptosis) to eliminate the damaged cell, or they stall the cell cycle to prevent replication of potentially damaged DNA. The cell-cycle-checkpoint responses to ionizing radiation are relatively normal in cells lacking DNA-PKCS16–18, but they are markedly abnormal in ATM-deficient cells19–22. The biochemical and phenotypic differences between these three related enzymes indicate that they probably have distinct cellular substrates or respond to different signals (FIG. 1). In terms of substrates, a general consensus motif has been defined for ATM8,23 — ATM phosphorylates a serine or threonine residue only if it is followed by a glutamine (the ‘SQ/TQ’ motif). Positively charged amino acids near to the target serine/threonine generally seem to diminish phosphorylation, whereas hydrophobic or negatively charged amino acids enhance it8. And although ATM, ATR and DNA-PK have similar target sequence preferences, even crude in vitro kinase assays can detect differences in their substrate specificities8. The next level in understanding how ATM functions is to show these in vitro substrate specificities in vivo (TABLE 2). Cell-cycle checkpoints

Much of the research into ATM has focused on its effects on cell-cycle checkpoints because, as noted above, these are disrupted in cells that lack ATM. The cell cycle can potentially arrest at several stages, but it usually stalls before S phase or during S phase (known as the G1 and S-phase checkpoints, respectively) to prevent the cell from progressing through DNA replication, or before mitosis (the G2 checkpoint) to prevent aberrant segregation of damaged chromosomes. Failure of these mechanisms would be expected to result in genomic instability and cancer predisposition. Cells from people with ataxia-telangiectasia have specific defects in response to DNA damage induced by ionizing radiation24. In response to ionizing radiation, these ATM-defective cells show abnormal cell-cycle arrests in G1, S phase and G2 (REFS 19–22). By characterizing protein targets of the ATM kinase, we have gained insight into the molecular mechanisms that underlie the cell-cycle checkpoints induced by ionizing radiation. The G1 checkpoint. Arrest of the cell cycle at the G1 checkpoint is caused by DNA damage, which induces increased cellular levels of p53 (REF. 25). The signal for www.nature.com/reviews/molcellbio

| DECEMBER 2000 | VOLUME 1

REVIEWS

Table 2 | Substrates and sites for ATM in vivo Protein

Sequence

Amino acid

Cell function

p53 (REFS 6,7)

PPLSQETFS

Ser 15

G1 checkpoint

p95/NBS1 (REFS 52–55)

ITNSQTLIP PSLSQGVSV

Ser 278 Ser 343

Radiosensitivity, foci S checkpoint

MDM2 (REF. 37)

EDYSQPSTS

Ser 395

G1 checkpoint

CHK2 (REFS 43,106,107)

TVSTQELYS

Thr 68

G1 checkpoint, others?

BRCA1 (REFS 62,108)

GLSSQSDIL QHGSQPSNS VLTSQKSSE NYPSQEELI

Ser 1387 Ser 1423 Ser 1457 Ser 1524

TBD* TBD TBD TBD

CtIP88

ADLSQYKMD DSFSQAADE

Ser 664 Ser 745

TBD (through BRCA1) TBD (through BRCA1)

4E-BP1 (REF. 98)

GEESQFEMD

Ser 112

Insulin response

*TBD, to be determined.

CHK2

A serine/threonine protein kinase that has an important function in cell-cycle regulation in response to DNA damage.

p53 induction after ionizing radiation is broken DNA, although other cellular stresses — such as oxygen starvation and depletion of ribonucleotide triphosphates — can also increase its levels (REF. 26). Through its transcription-factor activity, p53 activates production of p21 (also known as WAF1 or CIP1) (FIG. 2), which, in turn, inhibits cyclin E and its partner CDK2 — a complex required for progression of the cell cycle from G1 to S phase. This is the only DNA damage checkpoint that requires p53 for its initiation. How is this pathway linked to ATM? The first clue was the observation that cells from patients with ataxiatelangiectasia show poor induction of p53 after ionizing radiation19. However, mechanistic insights into how ATM signals to p53 initially came from studies of p53, not ATM. Ionizing radiation was found to induce phosphorylation of p53 on serine 15 (REFS 28,29) and this phosphorylation was diminished in ATM-deficient cells29, indicating that ATM is needed. Subsequently, in vitro ATM kinase-assays provided evidence that ATM directly phosphorylates p53 on this same site, Ser 15 (REFS 6,7) (but for a twist in this tale, see BOX 3). It was not clear, however, whether this phosphorylation is involved in the induction of p53. In fact, mutation of Ser 15 to alanine — effectively preventing the phosphorylation of p53 — has little effect on the halflife of p53 or on its ability to cause cell-cycle arrest or apoptosis30–32. It has been suggested that phosphorylation of Ser 15 might facilitate other post-translational modifications of p53, and in vitro data support this

Box 2 | The search for ATM Many laboratories tried to clone the gene that is defective in ataxia-telangiectasia by functional complementation of radiosensitivity in cells from patients, but all of these attempts failed. Linkage analysis in family studies led to the locus of the mutated gene being mapped to chromosome 11q22–23, and subsequent finer localization to an 850-kb region on chromosome 11q23.1 (REF. 100). The eventual identification of ATM relied on a positional-cloning technique and sequencing of expressed genes in this mapped region, and was a breakthrough in our ability to investigate the biology of ataxia-telangiectasia1. The failure of the original complementation cloning approach was explained by the large size of the ATM messenger RNA — over 13 kb in length. This precluded its presence as a full-length complementary DNA in expression libraries used at the time.

concept with respect to acetylation of lysine 382 (REF. 33). Indeed, phosphorylation of Ser 15 seems to affect the ability of p53 to bind to another transcription factor, p300, and it also modulates the transactivation of target genes by p53 (REF. 32). So how does ATM induce p53 after ionizing radiation? If phosphorylation of Ser 15 does not affect the levels of p53, some other target of the ATM kinase must be responsible. There do not seem to be other target sites on p53 for phosphorylation by ATM, so the theory has been that another protein is affected by ATM, and that this protein modulates the levels of p53. The best candidate was a protein called MDM2, an important regulator of p53 degradation34,35, and last year Khosravi et al.36 showed an ATM-dependent phosphorylation of MDM2 after ionizing radiation (FIG. 2). So the current hypothesis is that, after ionizing radiation, phosphorylation of Ser 395 in MDM2 by ATM decreases the ability of MDM2 to shuttle p53 from the nucleus to the cytoplasm, thereby allowing the p53 protein to accumulate37. Ionizing radiation Other types of DNA damage

ATR complex

DSBs KU70 KU80 DN A -P

? ? ?

ATM complex

M G2

? ?

or NH HEJ EJ? ?

or N

Cell-cycle checkpoints

DNA repair (HR/NHEJ)

Cell-cycle arrest or apoptosis

Radiation resistance

G1 R

Figure 1 | Function of ATM in cell-cycle checkpoints and DNA repair pathways. Exposure of cells to ionizing radiation results in arrest of cell-cycle progression at several points. In the G1 phase of the cycle, the arrest occurs before or at the ‘restriction point’ (R). Cells irradiated in S phase arrest immediately at any point throughout S phase (arrowheads), apparently as a function of inhibition of replicon initiation. Irradiated cells arrest in G2, but because irradiated cells continue to enter mitosis after irradiation for up to 30 minutes, this arrest (grey) must occur in G2 before the G2/M border. Cells lacking ATM fail to show all of these arrests after ionizing radiation. DNA breaks can be repaired by at least two pathways — homologous recombination (HR) and non-homologous end-joining (NHEJ). ATM functions in cell-cycle checkpoints after ionizing radiation and has an undefined role in DNA repair, whereas ATR is involved in cellcycle checkpoints in response to various types of DNA damage. The catalytic subunit of DNA-dependent protein kinase (DNA-PKCS), aided by KU70 and KU80, functions in NHEJ. Associated cofactors of ATM and ATR, which may be required for their interaction with DNA ends, have not yet been found.

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 1 | DECEMBER 2000 | 1 8 1

REVIEWS

Ionizing radiation DSBs ATR

ATM* ?

T68

CHK2 S15 S20 S395

p53

T68

CHK2

MDM2

CHK1

BRCA1 MRE11 RAD50

S216

S343 S395

S15 S20

p53

MDM2

14-3-3

NBS1

? Replication machinery

p21

CDC25C

14-3-3 CDC25C Sequestered in cytoplasm

Tyr-p CDC2

CDC2

CDK2–cyclin E G1

Figure 2 | ATM and other molecular signals controlling cell-cycle checkpoints after ionizing radiation. The specific activity of ATM is increased after introduction of DNA doublestrand breaks (DSBs) in DNA through ionizing radiation or other means. G1: Activated ATM (ATM*) directly phosphorylates three proteins involved in controlling p53 functions or levels — p53 (serine 15), CHK2 (threonine 68) and MDM2 (serine 395). The CHK2 kinase is thought to be activated by ATM and it, in turn, phosphorylates p53 on serine 20. This phosphorylation event and the phosphorylation of MDM2 both seem to inhibit binding of MDM2 to p53 and should result in an increase in the level of p53 protein. The increased level of p53 protein transcriptionally induces p21, which inhibits CDK2–cyclin E, and causes arrest in the G1 phase of the cycle. S: Activated ATM also phosphorylates NBS1 (serine 343) and this phosphorylation event is required for the ionizing radiation-induced S-phase arrest. NBS1 exists in a complex with MRE11, RAD50 and BRCA1. The potential role of these proteins in the S-phase arrest remains to be clarified. In addition, CHK2 may also be involved in this pathway after activation by ATM through phosphorylation of BRCA1 or NBS1. G2: Details have not been worked out clarifying the downstream targets of ATM at the G2 checkpoint. CHK2 and CHK1 may possibly be targets for ATM and ATR in the G2–M checkpoint pathway, respectively. CDC25C and 14-3-3 have been implicated in regulation of CDC2 kinase and progression through G2. The dashed arrows and question marks represent possible signalling steps and the solid arrows represent reported phosphorylation events. Animated online

RADIORESISTANT DNA SYNTHESIS

A failure of the rapid decrease in DNA synthesis in ataxiatelangiectasia cells that occurs in normal cells after ionizing radiation. RESTRICTION POINT

A point late in the G1 stage of the cell cycle at which mammalian cells become committed to entry into S phase, even without other growth factors. REPLICON

A structural complex at which replication of DNA occurs. HYDROXYUREA

A inhibitor of ribonucleotide reductase that blocks replication during S phase by preventing nucleotide synthesis.

182

There is a final twist in the ATM signalling pathway at the G1 checkpoint31,38. Ionizing radiation induces phosphorylation of another serine residue in p53 — Ser 20 — in an ATM-dependent manner. Interestingly, Ser 20 is in the middle of the domain of p53 that binds to MDM2, and mutation of this residue abolishes the p53 stabilization in response to DNA damage31. This has led to the suggestion that phosphorylation of p53 on Ser 20 after ionizing radiation could result in stabilization of p53 by disrupting its interaction with MDM2. Despite this dependence, neither ATM nor ATR can phosphorylate Ser 20 in vitro6. So Ser 20 is thought to be phosphorylated after ionizing radiation by some other kinase which, in turn, is directly or indirectly activated by ATM. The CHK2 kinase has now been implicated in this step39–41. In vitro, CHK2 can phosphorylate p53 on Ser 20 (REFS 39–41); CHK2 is phosphorylated in an ATMdependent manner after ionizing radiation42; and ATM can directly phosphorylate CHK2 in vitro43. The S-phase checkpoint. One of the first abnormalities to be characterized in ATM-deficient cells was a failure to

arrest DNA synthesis after ionizing radiation20,21. This phenomenon was called RADIORESISTANT DNA SYNTHESIS (RDS). Although the G1 checkpoint also results in a decrease in DNA replication, the S-phase checkpoint is a distinct process (FIG. 2). For example, whereas p53-dependent G1 arrest occurs within G1, before the RESTRICTION 44,45 POINT , RDS is a measure of DNA synthesis soon after irradiation. So it reflects cells that are already in S phase, rather than cells entering S phase from G1, and it does not depend on p53 (REFS 46,47). It is thought to reflect the inhibition of new REPLICON initiation21, and it is not synonymous with the S-phase arrest pathway induced by treatment with HYDROXYUREA. How is the ATM kinase involved in S-phase arrest? A potential substrate, identified in an in vitro screen for the ATM consensus target sequence, was p95/NBS1 (REF. 8). The gene that encodes this protein is mutated in a disorder that is similar to ataxia-telangiectasia — Nijmegen breakage syndrome (NBS). In both cases, patients show radiation sensitivity, RDS, a predisposition to cancer and chromosomal instability48. However, the neurological abnormality in NBS patients is a stable MICROCEPHALY, rather than progressive ATAXIA. Cloning of the p95/NBS1 gene49,50 revealed the protein to be part of a complex that also contains RAD50 and MRE11. This complex is known to bind to, and potentially help repair, DSBs49,51. Given that NBS1 is part of a complex that localizes to DSBs, and that ATM is activated by such strand breaks in cells, could the NBS1 complex somehow signal to ATM? No, because ATM is still activated and the p53 signalling pathway is still induced in cells lacking NBS52. However, ATM and NBS1 have been linked in a common signalling pathway. It turns out that the in vitro target site on NBS1 that is phosphorylated by the ATM kinase — Ser 343 — is also a target site in vivo 52–55. This phosphorylation event has now been shown to be required for the ionizing radiation-induced S-phase checkpoint52,54 (FIG. 2), although we do not yet know how it contributes to a transient inhibition of replicon initiation. The S-phase checkpoint may also require other factors. For example, germline mutation of MRE11 causes another syndrome, ataxia-telangiectasia-like disorder, the symptoms of which include radiosensitivity, RDS and neurological degeneration56. So normal MRE11 function may be required for this checkpoint. In addition, BRCA1 can be isolated in a complex with NBS1/MRE11/RAD50, and could also be part of this signalling pathway57,58. Finally, based on extrapolation from yeast signalling pathways, it is reasonable to predict that CHK2 may also participate in this checkpoint pathway. The two yeast ATM homologues (Mec1p and Rad3p; TABLE 1) and the CHK2 homologues (Rad53p and Cds1p) are all required for the DNA damage-induced S-phase checkpoint in yeast59,60. In fact, the mammalian signalling pathways are turning out to be remarkably similar to those found in yeast61. Because BRCA1 and CHK2 are also substrates of ATM43,62 (FIG. 2), and MRE11 contains a reasonable ATM target site8, ATM may control the Sphase checkpoint by phosphorylating several proteins in www.nature.com/reviews/molcellbio

| DECEMBER 2000 | VOLUME 1

REVIEWS a complex — similar to the situation for ATM-mediated control of the G1 checkpoint (FIG. 2).

MICROCEPHALY

An abnormally small head. ATAXIA

Dyscoordination of gait and other movements controlled by cerebellum. BRCA1

A tumour-suppressor gene that is linked to hereditary early onset of breast and ovarian cancer. PROPIDIUM IODIDE ASSAY

A fluorescent DNAintercalating dye used to measure DNA content in flow cytometry assays. 14-3-3 PROTEIN

A regulatory protein that binds to phosphorylated forms of various proteins that are involved in signal transduction and cell-cycle control. BLASTOCYST

An early stage of embryonic development at which cells begin to commit to certain developmental lineages. HISTONE DEACETYLASE

An enzyme that removes the acetyl groups of core histones; its activity has an important function in transcriptional regulation and cell-cycle progression through alterations in chromatin structure.

The G2 checkpoint. As yet, none of the proteins that regulate progression from G2 into M in mammalian cells has been identified as an ATM target. This lack of progress may be due, in part, to the unusual nature of the G2 checkpoint defect in ATM-deficient cells, which seems to vary depending on the assay used. The progression of cells from G2 into M can be followed at early time points after ionizing radiation (roughly 30 to 90 minutes) by directly examining the number of mitotic cell. Such studies show that ataxia-telangiectasia cells do not have the normal cell-cycle arrest and that they continue to enter mitosis. (This rapid arrest in G2 is independent of p53.) However, if the cell-cycle distribution is examined, for example by PROPIDIUM IODIDE ASSAY at later times after ionizing radiation, ataxia-telangiectasia cells seem to have a prolonged G2 arrest22,63. This apparent discrepancy results from the fact that ataxia-telangiectasia cells irradiated in the G2 phase of the cell cycle fail to arrest in G2 (REFS 22,47,63). However, ataxia-telangiectasia cells irradiated in G1 or S phase not only arrest when they get to G2, but they arrest for a prolonged period of time. One possible explanation is that there are different signals for G2 arrest, depending on which phase of the cycle a cell is in when it is irradiated. ATM is thought to be critical for signalling to the arrest machinery only if cells are irradiated while they are in G2. For cells that enter G2 from G1 and S phase, a lack of ATM and subsequent lack of the S-phase checkpoint may result in a different type of DNA lesion that does not require ATM for signalling to the G2-arrest machinery. This parallel but independent signalling pathway may operate through the ATM-like protein ATR, which seems to be more important for responses to ultraviolet light and hydroxyurea than to ionizing radiation (BOX 3). The molecular mechanisms of G2–M checkpoint control are relatively well understood in yeast. They are almost certainly evolutionarily conserved in mammals, although definitive mechanistic controls are not clear. An intriguing model for G2 arrest in humans is shown in FIG. 2 (REF. 64). It begins with phosphorylation of the phosphatase CDC25, on Ser 216, by the CHK1 kinase. Phosphorylated CDC25 then binds to 14-3-3 PROTEIN and is sequestered in the cytoplasm. Because it is trapped in the cytoplasm, CDC25 cannot do its usual jobs in the nucleus — one of which is to dephosphorylate CDC2. Dephosphorylation of CDC2 usually allows progression from G2 into M phase, so this process is effectively blocked64. ATM might have a similar effect by activating the CHK2 kinase, which, like CHK1, phosphorylates CDC25. Genetic studies support the idea of functional roles for CHK1 or CHK2 in controlling the G2–M checkpoint. Embryonic stem cells that lack CHK2 cannot maintain an ionizing radiation-induced arrest in G2 (REF. 39). Moreover, CHK1-deficient embryonic stem cells and BLASTOCYSTS show a defective G2–M checkpoint in response to ionizing radiation, ultraviolet

light and hydroxyurea65,66. But how do ATM and its relative ATR fit into these signalling pathways? On the basis of the functional roles of Chk1p and Cds1p/Rad53p in yeast67, their human homologues (CHK1 and CHK2) are attractive targets for ATM in the G2–M checkpoint pathway. The yeast ATM-family members Mec1p and Rad3p are required for phosphorylation of Chk1p in response to DNA damage68,69, and activation of Rad53p/Cds1p by DNA damage or replication blocks depends on members of the PI(3)K family (Mec1p, Rad3p)60,70. However, although ATMdependent phosphorylation of CHK2 seems to function in the G1 checkpoint, it is not known whether this, or indeed ATR-dependent phosphorylation of CHK1, participates in the G2 arrest. BRCA1 — a target of both ATM and CHK2 — may also be involved in G2 checkpoint control71. In addition, as may occur at the G1 checkpoint, the ATM–CHK2 and ATR–CHK1 pathways may cooperate to control the G2–M checkpoint. It may turn out that a recurring theme in ATM regulation of ionizing radiationinduced checkpoints is that ATM phosphorylates several proteins in a complex and that it acts together with other kinases in the complex, such as CHK2. ATM targets in DNA repair

The hypersensitivity to ionizing radiation seen in ATMdeficient cells is not due simply to the abnormal cellcycle checkpoints47,72. Rather, it could result from abnormal repair of the DNA lesions introduced by ionizing radiation. Although defects in DSB repair have not been found73, ATM-deficient cells do show an increased frequency of chromosomal breaks after exposure to ionizing radiation47,74. So there seems to be some sort of chromosomal-break repair abnormality despite the normal kinetics of DSB repair75. This deficiency in DNA repair may be due to abnormal chromatin organization — the DNA needs to be remodelled to be repaired. Consistent with this idea, ATM is associated with chromatin and the nuclear matrix12; the rapid dephosphorylation of histone H1 in response to ionizing radiation depends on ATM76; and ATM binds to HISTONE DEACETY77 LASE . However, definitive ATM targets affecting the structure of chromatin have not been identified.

Box 3 | The ATR kinase Suspicions of another, previously unknown, kinase that could phosphorylate p53 on serine 15 arose because, in ATM-deficient cells, this residue could be phosphorylated some time after cells had been treated with ionizing radiation. Moreover, ultraviolet irradiation induced this same phosphorylation event, regardless of whether ATM was functional. The ATM-related protein, ATR, has since been shown to phosphorylate serine 15 in vitro6, and overexpression of a dominant-negative ATR can prevent both ultraviolet-induced and the late ionizing radiation-induced phosphorylation of this site101. However, it is not yet clear whether ATR is activated by ultraviolet light, analogously to the activation of ATM by ionizing radiation.

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 1 | DECEMBER 2000 | 1 8 3

REVIEWS

DSBs

ATM*

c-Abl CHK2

NBS1 RAD51

MRE11

RAD52 BRCA1 Others

RAD50 DNA-PKcs

HR repair pathway

NHEJ repair pathway RAD52 Ku end binding XRCC4 Ligase IV

RAD51

Repair/radiation sensitivity

Figure 3 | A role of ATM in DNA double-stranded break repair pathways. RAD51/RAD52 and BRCA1 are required for the homologous-recombinational (HR) repair pathway. The non-homologous end-joining (NHEJ) repair pathway depends on DNA-PKcs, the Ku proteins, XRCC4 and ligase IV. The NBS1/MRE11/RAD50 complex is thought be involved in both DNA double-stranded break (DSB) repair pathways. Although the mechanisms responsible for radiosensitivity in ataxia-telangiectasia cells have not been clarified, ATM seems to regulate several candidates involved in DSB repair (BRCA1, NBS1, MRE11 and so on). Dysfunction of these candidates in DSB repair pathway in ATM-deficient cells may explain radiation sensitivity and chromosomal abnormalities. Dashed arrows represent possible signalling steps between ATM and c-Abl and CHK2, and c-Abl and CHK2 with RAD51 and BRCA1, respectively.

DNA double-stranded breaks can be repaired in mammalian cells by at least two pathways (FIG. 3) — homologous recombination or non-homologous endjoining (NHEJ; also known as illegitimate recombinational repair)78. On the assumption that a subtle repair defect could result from an aberrant recombinational pathway, abnormalities in these recombinational pathways have been sought in ATM-deficient cells. High rates of intrachromosomal recombination have been found in ATM-deficient cells79,80, as well as error-prone recombination and increased levels of extrachromosomal recombination81. It is still not known, however, which repair pathway is abnormal in these cells or how ATM would affect such a pathway. Although NHEJ was thought to be predominant in mammalian cells,ATM has recently been linked82,83 to homologous recombination. The ATM kinase could also modulate radiosensitivity through its interactions with BRCA1, NBS1 or CHK2 (FIG. 3). BRCA1 forms a complex with RAD51 (REF. 84), an essential protein in homologous-recombinational repair after DNA damage. Moreover, cells lacking BRCA1 are hypersensitive to ionizing radiation and defective in transcription-coupled repair and homologous-recombinational repair85,86. As observed with the checkpoint controls, the effects of ATM on BRCA1 could be both direct and indirect, because CHK2 also

184

phosphorylates BRCA1. This phosphorylation is required for the release of BRCA1 from CHK2, and is important for cellular viability after ionizing radiation87. Furthermore, ATM phosphorylates CtIP (CtBP-interacting protein), a protein that binds to and inhibits BRCA1 function, resulting in the release and apparent activation of BRCA1 (REF. 88). Human cells mutated for NBS1 or MRE11 (REFS 48,89), and yeast mutated for any one of the Mre11p, Rad50p and Xrs2p (an NBS1 analogue) complex, are radiosensitive90,91. So ATM could also affect repair and radiosensitivity through this complex. This potential mechanism is particularly attractive owing to the linkage of MRE11 and NBS1 to both NHEJ and recombinational repair. Finally, because ATM is also found in a large protein complex with BRCA1 and numerous DNA repair-associated proteins (p95/MRE11/RAD50, MSH2, MSH6, MLH and BLM)58, there are many other potential mechanisms by which ATM could affect DNA repair and radiosensitivity. Other targets for ATM

Patients with ataxia-telangiectasia show a progressive neurodegeneration that could, in theory, result from deficiencies in dealing with DSBs in neuronal cells. If this turns out to be the case, then any of the substrates discussed above in the DNA damage signalling pathways could also be involved in the neurodegeneration phenotype. The recent observation89 that patients with mutations in MRE11 have an ataxia-telangiectasia-like phenotype that includes some ataxic symptoms would be consistent with this possibility. In addition, the observation that mice deficient in the repair enzyme ligase IV have a defect in neurological development that is ATM dependent92 indicates both that DNA damage normally occurs during development of the nervous system and that ATM signalling is an important determinant of neuronal cell survival after this damage. However, the involvement of ATM in other physiological pathways could also account for the neurological symptoms of ataxia-telangiectasia. Unfortunately, it has been difficult to study the role of ATM in the nervous system experimentally. For example, ATM-deficient mice have, at worst, a mild neurological abnormality14, and the progressive neuronal death in humans seems to be most evident in the sparse population of cerebellar Purkinje cells. Could ATM have extra functions that are restricted to certain cell types, such as neurons? The suspicion that there are other functions and targets for ATM comes from data linking it to c-Abl93 and the NFκB signalling pathway94; the observation that it can bind to a vesicular protein, β-adaptin95; and the fact that it is found mainly in the cytoplasm of neurons96,97. Interestingly, one cytoplasmic function and target for ATM has been reported — in some cell types, insulin treatment activates the ATM kinase and stimulates phosphorylation of the translational regulatory protein, 4E-BP1 (REF. 98). This observation provides a potential link to the unusual insulin-resistance phenotype previously reported in ataxia-telangiectasia patients. It remains to be seen whether ATM participates in other www.nature.com/reviews/molcellbio

| DECEMBER 2000 | VOLUME 1

REVIEWS cytokine signalling pathways and whether this contributes to the neurological (or other) abnormalities (BOX 1). However, it is difficult to explain the cell-membrane alterations, insulin signalling abnormalities and cytoskeletal defects in ATM-deficient cells3 if the problem is just with DNA-damage responses. So, although alterations in DNA-damage responses seem to be the most straightforward explanation for the neurodegeneration observed in ataxia-telangiectasia patients, it is also conceivable that the neuronal dysfunction results instead from alterations in cell membranes or cell signalling that depend on ATM. As with the approaches used to clarify the role of ATM in DNA-damage responses, insights along these lines will probably develop from identification of the ATM substrates for these functions. Many substrates, many functions

ATM is a protein kinase that seems to have many substrates. Because numerous cellular processes are affected by ATM, when this protein is missing the result is a varied disease phenotype in mice and humans. Several targets participate in DNA-damageresponse pathways and account for some of the checkpoint abnormalities seen in ATM-deficient cells. Other substrates will probably be identified that further clarify these checkpoint pathways and explain the effects of ATM on DNA repair or recombination. One recurring theme in these targets may be the fact that ATM tends to phosphorylate several proteins in a complex, perhaps to achieve functional endpoints from several overlapping mechanisms. The story may get even more complex if ATM has bona fide targets that are not involved in DNA-damage-response path-

Savitsky, K. et al. A single ataxia telangiectasia gene with a product similar to PI-3 kinase. Science 268, 1749–1753 (1995). The report of the cloning of the ATM gene after a twodecade search. 2. Hunter, T. When is a lipid kinase not a lipid kinase? When it is a protein kinase. Cell 83, 1–4 (1995). 3. Lavin, M. F. & Shiloh, Y. The genetic defect in ataxiatelangiectasia. Annu. Rev. Immunol. 15, 177–202 (1997). 4. Smith, G. C. & Jackson, S. P. The DNA-dependent protein kinase. Genes Dev. 13, 916–934 (1999). 5. Abraham, R. T. Mammalian target of rapamycin: immunosuppressive drugs uncover a novel pathway of cytokine receptor signaling. Curr. Opin. Immunol. 10, 330–336 (1998). 6. Canman, C. E. et al. Activation of the ATM kinase by ionizing radiation and phosphorylation of p53. Science 281, 1677–1679 (1998). 7. Banin, S. et al. Enhanced phosphorylation of p53 by ATM in response to DNA damage. Science 281, 1674–1677 (1998). References 6 and 7 reported that ATM directly phosphorylates p53 and is activated by DNA damage. 8. Kim, S. T., Lim, D. S., Canman, C. E. & Kastan, M. B. Substrate specificities and identification of putative substrates of ATM kinase family members. J. Biol. Chem. 274, 37538–37543 (1999). 9. Chan, D. W. et al. Purification and characterization of ATM from human placenta. A manganese-dependent, wortmannin-sensitive serine/threonine protein kinase. J. Biol. Chem. 275, 7803–7810 (2000). 10. Dhand, R. et al. PI 3-kinase is a dual specificity enzyme: autoregulation by an intrinsic protein-serine kinase activity. EMBO J. 13, 522–533 (1994). 11. Smith, G. C. et al. Purification and DNA binding properties of the ataxia-telangiectasia gene product ATM. Proc. Natl Acad. Sci. USA 96, 11134–11139 (1999).

ways. Although ataxia-telangiectasia is rare, the gene product mutated in this disease seems to mediate many critical cellular processes and will provide mechanistic insights into these general processes. For example, because ATM must somehow signal to the DNA replication machinery to arrest S-phase progression rapidly after irradiation, studies of ATM may help elucidate the replication machinery in a manner analogous to that seen with DNA repair proteins and the transcriptional machinery99. In addition to mechanistic insights, there should be clinical benefits. Therapeutic interventions based on these insights — which seek to reduce some of the clinical abnormalities in the disease — would be more than welcome. However, the clinical benefits are likely to extend to a much larger group, including cancer patients in general. For example, inhibition of ATM or other proteins in its pathways are potential targets for increasing the radiosensitivity of tumour cells. In the end, characterization of the mechanism underlying action of this large gene product that is mutated in a rare human disease will probably add to our understanding of several basic cellular processes and has the potential to result in new clinical approaches. Links DATABASE LINKS Ataxia-telangiectasia | ATM | PI(3)K |

MTOR | DNA-PK | p53 | KU70 | KU80 | RPA | ATR | p21 | CDK2 | p300 | MDM2 | CHK2 | NBS1 | Nijmegen breakage syndrome | RAD50 | MRE11 | ataxiatelangiectasia-like disorder | BRCA1 | CDC25 | CHK1 | MSH2 | MSH6 | MLH | BLM | ligase IV | c-Abl | NF-κB | β-adaptin | 4E-BP1

12. Gately, D. P., Hittle, J. C., Chan, G. K. & Yen, T. J. Characterization of ATM expression, localization, and associated DNA-dependent protein kinase activity. Mol. Biol. Cell 9, 2361–2374 (1998). 13. Taylor, A. M., Metcalfe, J. A., Thick, J. & Mak, Y. F. Leukemia and lymphoma in ataxia telangiectasia. Blood 87, 423–438 (1996). 14. Barlow, C. et al. Atm-deficient mice: a paradigm of ataxia telangiectasia. Cell 86, 15–71 (1996). The generation of the first ataxia-telangiectasiaknockout mouse with many, but not all, of the features of the human disease. 15. Brown, E. J. & Baltimore, D. ATR disruption leads to chromosomal fragmentation and early embryonic lethality. Genes Dev. 14, 397–402 (2000). 16. Komatsu, K., Yoshida, M. & Okumura, Y. Murine scid cells complement ataxia-telangiectasia cells and show a normal post-irradiation response of DNA synthesis. Int. J. Radiat. Biol. 63, 725–730 (1993). 17. Huang, L. C., Clarkin, K. C. & Wahl, G. M. p53-dependent cell cycle arrests are preserved in DNA-activated protein kinase-deficient mouse fibroblasts. Cancer Res. 56, 2940–2944 (1996). 18. Allalunis-Turner, J., Barron, G. M. & Day, R. S. Intact G2phase checkpoint in cells of a human cell line lacking DNAdependent protein kinase activity. Radiat. Res. 147, 284–287 (1997). 19. Kastan, M. B. et al. A mammalian cell cycle checkpoint pathway utilizing p53 and GADD45 is defective in ataxiatelangiectasia. Cell 71, 587–597 (1992). The first linkage of the ataxia-telangiectasia gene product with the p53 pathway. 20. Houldsworth, J. & Lavin, M. F. Effect of ionizing radiation on DNA synthesis in ataxia telangiectasia cells. Nucleic Acids Res. 8, 3709–3720 (1980). 21. Painter, R. B. & Young, B. R. Radiosensitivity in ataxiatelangiectasia: a new explanation. Proc. Natl Acad. Sci.

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

USA 77, 7315–7317 (1980). 22. Beamish, H. & Lavin, M. F. Radiosensitivity in ataxiatelangiectasia: anomalies in radiation-induced cell cycle delay. Int. J. Radiat. Biol. 65, 175–184 (1994). 23. O’Neill, T. et al. Utilization of oriented peptide libraries to identify substrate motifs selected by ATM. J. Biol. Chem. 275, 22719–22727 (2000). 24. Morgan, S. E. & Kastan, M. B. p53 and ATM: cell cycle, cell death, and cancer. Adv. Cancer Res. 71, 1–25 (1997). 25. Kastan, M. B., Onyekwere, O., Sidransky, D., Vogelstein, B. & Craig, R. W. Participation of p53 protein in the cellular response to DNA damage. Cancer Res. 51, 6304–6311 (1991). 26. Giaccia, A. J. & Kastan, M. B. The complexity of p53 modulation: emerging patterns from divergent signals. Genes Dev. 12, 2973–2983 (1998). 27. el-Deiry, W. S. et al. WAF1, a potential mediator of p53 tumor suppression. Cell 75, 817–825 (1993). 28. Shieh, S. Y., Ikeda, M., Taya, Y. & Prives, C. DNA damageinduced phosphorylation of p53 alleviates inhibition by MDM2. Cell 91, 325–334 (1997). 29. Siliciano, J. D. et al. DNA damage induces phosphorylation of the amino terminus of p53. Genes Dev. 11, 3471–3481 (1997). 30. Ashcroft, M., Kubbutat, M. H. & Vousden, K. H. Regulation of p53 function and stability by phosphorylation. Mol. Cell. Biol. 19, 1751–1758 (1999). 31. Chehab, N. H., Malikzay, A., Stavridi, E. S. & Halazonetis, T. D. Phosphorylation of Ser-20 mediates stabilization of human p53 in response to DNA damage. Proc. Natl Acad. Sci. USA 96, 13777–13782 (1999). 32. Dumaz, N. & Meek, D. W. Serine 15 phosphorylation stimulates p53 transactivation but does not directly influence interaction with HDM2. EMBO J. 18, 7002–7010 (1999). 33. Sakaguchi, K. et al. DNA damage activates p53 through a phosphorylation-acetylation cascade. Genes Dev. 12,

VOLUME 1 | DECEMBER 2000 | 1 8 5

REVIEWS

2831–2841 (1998). 34. Haupt, Y., Maya, R., Kazaz, A. & Oren, M. Mdm2 promotes the rapid degradation of p53. Nature 387, 296–299 (1997). 35. Kubbutat, M. H., Jones, S. N. & Vousden, K. H. Regulation of p53 stability by Mdm2. Nature 387, 299–303 (1997). 36. Khosravi, R. et al. Rapid ATM-dependent phosphorylation of MDM2 precedes p53 accumulation in response to DNA damage. Proc. Natl Acad. Sci. USA 96, 14973–14977 (1999). 37. Maya, R. et al. ATM-dependent phosphorylation of Mdm2 on serine 394: role in p53 activation by DNA damage. Genes Dev. (submitted). 38. Shieh, S. Y., Taya, Y. & Prives, C. DNA damage-inducible phosphorylation of p53 at N-terminal sites including a novel site, Ser20, requires tetramerization. EMBO J. 18, 1815–1823 (1999). 39. Hirao, A. et al. DNA damage-induced activation of p53 by the checkpoint kinase Chk2. Science 287, 1824–1827 (2000). Development of Chk2-knockout mouse cells and linkage of Chk2 to p53 and the G1-checkpoint pathway. 40. Chehab, N. H., Malikzay, A., Appel, M. & Halazonetis, T. D. Chk2/hCds1 functions as a DNA damage checkpoint in G(1) by stabilizing p53. Genes Dev. 14, 278–288 (2000). 41. Shieh, S. Y., Ahn, J., Tamai, K., Taya, Y. & Prives, C. The human homologs of checkpoint kinases Chk1 and Cds1 (Chk2) phosphorylate p53 at multiple DNA damageinducible sites. Genes Dev. 14, 289–300 (2000). References 40 and 41 show that CHK2 phosphorylates p53 on serine 20 and contributes to its stabilization in an ATM-dependent manner. 42. Matsuoka, S., Huang, M. & Elledge, S. J. Linkage of ATM to cell cycle regulation by the Chk2 protein kinase. Science 282, 1893–1897 (1998). Demonstration that CHK2 is phosphorylated and regulated by ATM. 43. Zhou, B. B. et al. Caffeine abolishes the mammalian G(2)/M DNA damage checkpoint by inhibiting ataxiatelangiectasia-mutated kinase activity. J. Biol. Chem. 275, 10342–10348 (2000). 44. Slebos, R. J. et al. p53-dependent G1 arrest involves pRBrelated proteins and is disrupted by the human papillomavirus 16 E7 oncoprotein. Proc. Natl Acad. Sci. USA 91, 5320–5324 (1994). 45. Di Leonardo, A., Linke, S. P., Clarkin, K. & Wahl, G. M. DNA damage triggers a prolonged p53-dependent G1 arrest and long-term induction of Cip1 in normal human fibroblasts. Genes Dev. 8, 2540–2551 (1994). 46. Larner, J. M., Lee, H. & Hamlin, J. L. Radiation effects on DNA synthesis in a defined chromosomal replicon. Mol. Cell. Biol. 14, 1901–1908 (1994). 47. Morgan, S. E., Lovly, C., Pandita, T. K., Shiloh, Y. & Kastan, M. B. Fragments of ATM which have dominant-negative or complementing activity. Mol. Cell. Biol. 17, 2020–2029 (1997). 48. Shiloh, Y. Ataxia-telangiectasia and the Nijmegen breakage syndrome: related disorders but genes apart. Annu. Rev. Genet. 31, 635–662 (1997). 49. Carney, J. P. et al. The hMre11/hRad50 protein complex and Nijmegen breakage syndrome: linkage of doublestrand break repair to the cellular DNA damage response. Cell 93, 477–486 (1998). 50. Varon, R. et al. Nibrin, a novel DNA double-strand break repair protein, is mutated in Nijmegen breakage syndrome. Cell 93, 467–476 (1998). References 49 and 50 reported the cloning of the NBS1 gene and demonstrated its interactions with MRE11 and RAD50. 51. Paull, T. T. & Gellert, M. Nbs1 potentiates ATP-driven DNA unwinding and endonuclease cleavage by the Mre11/Rad50 complex. Genes Dev. 13, 1276–1288 (1999). 52. Lim, D. S. et al. ATM phosphorylates p95/nbs1 in an Sphase checkpoint pathway. Nature 404, 613–617 (2000). Demonstration that ATM activation is not dependent on NBS1, but rather that ATM phosphorylates NBS1 and that this phosphorylation is important for the ionizing-radiation-induced S-phase checkpoint. 53. Gatei, M. et al. ATM-dependent phosphorylation of nibrin in response to radiation exposure. Nature Genet. 25, 115–119 (2000). 54. Zhao, S. et al. Functional link between ataxia-telangiectasia and Nijmegen breakage syndrome gene products. Nature 405, 473–477 (2000). 55. Wu, X. et al. ATM phosphorylation of Nijmegen breakage syndrome protein is required in a DNA damage response. Nature 405, 477–482 (2000). 56. Stewart, G. S. et al. The DNA double-strand break repair

186

gene hMRE11 is mutated in individuals with an ataxiatelangiectasia-like disorder. Cell 99, 577–587 (1999). 57. Zhong, Q. et al. Association of BRCA1 with the hRad50–hMre11–p95 complex and the DNA damage response. Science 285, 747–750 (1999). 58. Wang, Y. et al. BASC, a super complex of BRCA1associated proteins involved in the recognition and repair of aberrant DNA structures. Genes Dev. 14, 927–939 (2000). 59. Paulovich, A. G. & Hartwell, L. H. A checkpoint regulates the rate of progression through S phase in S. cerevisiae in response to DNA damage. Cell 82, 841–847 (1995). 60. Lindsay, H. D. et al. S-phase-specific activation of Cds1 kinase defines a subpathway of the checkpoint response in Schizosaccharomyces pombe. Genes Dev. 12, 382–395 (1998). 61. Elledge, S. J. Cell cycle checkpoints: preventing an identity crisis. Science 274, 1664–1672 (1996). 62. Cortez, D., Wang, Y., Qin, J. & Elledge, S. J. Requirement of ATM-dependent phosphorylation of brca1 in the DNA damage response to double-strand breaks. Science 286, 1162–1166 (1999). 63. Scott, D., Spreadborough, A. R. & Roberts, S. A. Radiation-induced G2 delay and spontaneous chromosome aberrations in ataxia-telangiectasia homozygotes and heterozygotes. Int. J. Radiat. Biol. 66, S157–S163 (1994). 64. Peng, C. Y. et al. Mitotic and G2 checkpoint control: regulation of 14-3-3 protein binding by phosphorylation of Cdc25C on serine-216. Science 277, 1501–1505 (1997). Demonstration of a G2-checkpoint control mechanism involving 14-3-3 protein binding regulated by phosphorylation of CDC25C. 65. Liu, Q. et al. Chk1 is an essential kinase that is regulated by Atr and required for the G(2)/M DNA damage checkpoint. Genes Dev. 14, 1448–1459 (2000). 66. Takai, H. et al. Aberrant cell cycle checkpoint function and early embryonic death in Chk1(–/–) mice. Genes Dev. 14, 1439–1447 (2000). 67. Rhind, N. & Russell, P. Mitotic DNA damage and replication checkpoints in yeast. Curr. Opin. Cell Biol. 10, 749–758 (1998). 68. Walworth, N. C. & Bernards, R. Rad-dependent response of the chk1-encoded protein kinase at the DNA damage checkpoint. Science 271, 353–356 (1996). 69. Sanchez, Y. et al. Control of the DNA damage checkpoint by chk1 and rad53 protein kinases through distinct mechanisms. Science 286, 1166–1171 (1999). 70. Sanchez, Y. et al. Regulation of RAD53 by the ATM-like kinases MEC1 and TEL1 in yeast cell cycle checkpoint pathways. Science 271, 357–360 (1996). 71. Xu, X. et al. Centrosome amplification and a defective G2–M cell cycle checkpoint induce genetic instability in BRCA1 exon 11 isoform-deficient cells. Mol. Cell 3, 389–395 (1999). 72. Verhaegh, G. W. et al. A gene that regulates DNA replication in response to DNA damage is located on human chromosome 4q. Am. J. Hum. Genet. 57, 1095–1103 (1995). 73. McKinnon, P. J. Ataxia-telangiectasia: an inherited disorder of ionizing-radiation sensitivity in man. Progress in the elucidation of the underlying biochemical defect. Hum. Genet. 75, 197–208 (1987). 74. Cornforth, M. N. & Bedford, J. S. On the nature of a defect in cells from individuals with ataxia-telangiectasia. Science 227, 1589–1591 (1985). 75. Pandita, T. K. & Hittelman, W. N. Initial chromosome damage but not DNA damage is greater in ataxia telangiectasia cells. Radiat. Res. 130, 94–103 (1992). 76. Guo, C. Y., Wang, Y., Brautigan, D. L. & Larner, J. M. Histone H1 dephosphorylation is mediated through a radiation-induced signal transduction pathway dependent on ATM. J. Biol. Chem. 274, 18715–18720 (1999). 77. Kim, G. D. et al. Sensing of ionizing radiation-induced DNA damage by ATM through interaction with histone deacetylase. J. Biol. Chem. 274, 31127–31130 (1999). 78. Karran, P. DNA double strand break repair in mammalian cells. Curr. Opin. Genet. Dev. 10, 144–150 (2000). 79. Meyn, M. S. High spontaneous intrachromosomal recombination rates in ataxia- telangiectasia. Science 260, 1327–1330 (1993). 80. Bishop, A. J., Barlow, C., Wynshaw-Boris, A. J. & Schiestl, R. H. Atm deficiency causes an increased frequency of intrachromosomal homologous recombination in mice. Cancer Res. 60, 395–399 (2000). 81. Luo, C. M. et al. High frequency and error-prone DNA recombination in ataxia telangiectasia cell lines. J. Biol. Chem. 271, 4497–4503 (1996). 82. Chen, G. et al. Radiation-induced assembly of Rad51 and

Rad52 recombination complex requires ATM and c-Abl. J. Biol. Chem. 274, 12748–12752 (1999). 83. Morrison, C. et al. The controlling role of ATM in homologous recombinational repair of DNA damage. EMBO J. 19, 463–471 (2000). 84. Scully, R. et al. Association of BRCA1 with Rad51 in mitotic and meiotic cells. Cell 88, 265–275 (1997). 85. Gowen, L. C., Avrutskaya, A. V., Latour, A. M., Koller, B. H. & Leadon, S. A. BRCA1 required for transcription-coupled repair of oxidative DNA damage. Science 281, 1009–1012 (1998). 86. Moynahan, M. E., Chiu, J. W., Koller, B. H. & Jasin, M. Brca1 controls homology-directed DNA repair. Mol. Cell 4, 511–518 (1999). 87. Lee, J. S., Collins, K. M., Brown, A. L., Lee, C. H. & Chung, J. H. hCds1-mediated phosphorylation of BRCA1 regulates the DNA damage response. Nature 404, 201–204 (2000). 88. Li, S. et al. Functional link of BRCA1 and ataxia telangiectasia gene product in DNA damage responses. Nature 406, 210–215 (2000). 89. Stewart, G. S. et al. The DNA double-strand break repair gene hMRE11 is mutated in individuals with an ataxiatelangiectasia-like disorder. Cell 99, 577–587 (1999). 90. Ivanov, E. L., Korolev, V. G. & Fabre, F. XRS2, a DNA repair gene of Saccharomyces cerevisiae, is needed for meiotic recombination. Genetics 132, 651–664 (1992). 91. Ajimura, M., Leem, S. H. & Ogawa, H. Identification of new genes required for meiotic recombination in Saccharomyces cerevisiae. Genetics 133, 51–66 (1993). 92. Lee, Y., Barnes, D. E., Lindahl, T. & McKinnon, P. J. Defective neurogenesis resulting from DNA ligase IV deficiency requires ATM. Genes Dev. 14, 2576–2550 (2000). 93. Baskaran, R. et al. Ataxia telangiectasia mutant protein activates c-Abl tyrosine kinase in response to ionizing radiation. Nature 387, 516–519 (1997). 94. Lee, S. J., Dimtchev, A., Lavin, M. F., Dritschilo, A. & Jung, M. A novel ionizing radiation-induced signaling pathway that activates the transcription factor NF-κB. Oncogene 17, 1821–1826 (1998). 95. Lim, D. S. et al. ATM binds to β-adaptin in cytoplasmic vesicles. Proc. Natl Acad. Sci. USA 95, 10146–10151 (1998). 96. Oka, A. & Takashima, S. Expression of the ataxiatelangiectasia gene (ATM) product in human cerebellar neurons during development. Neurosci. Lett. 252, 195–198 (1998). 97. Barlow, C. et al. ATM is a cytoplasmic protein in mouse brain required to prevent lysosomal accumulation. Proc. Natl Acad. Sci. USA 97, 871–876 (2000). 98. Yang, D. & Kastan, M. B. Participation of ATM in insulin signalling through phosphorylation of eIF-4E binding protein 1 (4E–BP1). Nature Cell Biol. 2, 893–898 (2000). 99. Meijer, M. & Smerdon, M. J. Accessing DNA damage in chromatin: insights from transcription. Bioessays 21, 596–603 (1999). 100. Gatti, R. A. et al. Genetic haplotyping of ataxiatelangiectasia families localizes the major gene to an approximately 850 kb region on chromosome 11q23. 1. Int. J. Radiat. Biol. 66, S57–S62 (1994). 101. Tibbetts, R. S. et al. A role for ATR in the DNA damageinduced phosphorylation of p53. Genes Dev. 13, 152–157 (1999). 102. Morrow, D. M., Tagle, D. A., Shiloh, Y., Collins, F. S. & Hieter, P. TEL1, an S. cerevisiae homolog of the human gene mutated in ataxia telangiectasia, is functionally related to the yeast checkpoint gene MEC1. Cell 82, 831–840 (1995). 103. Hari, K. L. et al. The mei-41 gene of D. melanogaster is a structural and functional homolog of the human ataxia telangiectasia gene. Cell 82, 815–821 (1995). 104. Cimprich, K. A., Shin, T. B., Keith, C. T. & Schreiber, S. L. cDNA cloning and gene mapping of a candidate human cell cycle checkpoint protein. Proc. Natl Acad. Sci. USA 93, 2850–2855 (1996). 105. Cliby, W. A. et al. Overexpression of a kinase-inactive ATR protein causes sensitivity to DNA-damaging agents and defects in cell cycle checkpoints. EMBO J. 17, 159–169 (1998). 106. Matsuoka, S. et al. Ataxia telangiectasia-mutated phosphorylates Chk2 in vivo and in vitro. Proc. Natl Acad. Sci. USA 97, 10389–10394 (2000). 107. Ahn, J.-Y., Schwarz, J. K., Piwnica–Worms, H. & Canman, C. E. Threonine 68 phosphorylation by ATM is required for efficient activation of Chk2 in response to ionizing radiation. Cancer Res. (in the press). 108. Gatei, M. et al. Role for ATM in DNA damage-induced phosphorylation of BRCA1. Cancer Res. 15, 3299–3304 (2000).

www.nature.com/reviews/molcellbio

| DECEMBER 2000 | VOLUME 1

REVIEWS

THREE WAYS TO MAKE A VESICLE Tomas Kirchhausen Cargo molecules have to be included in carrier vesicles of different forms and sizes to be transported between organelles. During this process, a limited set of proteins, including the coat proteins COPI, COPII and clathrin, carries out a programmed set of sequential interactions that lead to the budding of vesicles. A general model to explain the formation of coated vesicles is starting to emerge but the picture is more complex than we had imagined.

ENDOCYTIC PATHWAY

Macromolecules are taken up by invagination of the plasma membrane. They first arrive in early endosomes, then late endosomes, and finally lysosomes, where they are degraded by hydrolases. SECRETORY PATHWAY

Secretory or membrane proteins are inserted into the endoplasmic reticulum. They are then transported through the Golgi to the trans-Golgi network, where they are sorted to their final destination. FIBROBLAST

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

Actin-dependent process by which cells engulf large volumes of fluids.

Harvard Medical School, 200 Longwood Avenue, Boston, Massachusetts 02115, USA. e-mail: kirchhausen@crystal. harvard.edu

Transport of proteins and lipids along the ENDOCYTIC or SECRETORY PATHWAYS is a hallmark of eukaryotic cells. The membrane fluxes along these pathways are very large and rapid. A FIBROBLAST kept in resting conditions in a tissue culture plate internalizes an amount of membrane equivalent to the whole surface area of the cell in one hour. Inside the cell, it often takes only seconds for a carrier vesicle to move from the donor membrane to the acceptor organelle. But intracellular traffic is not only rapid, it is also very selective. Only a subset of the proteins and lipids in the donor membrane are allowed into the transport vesicle, effectively preventing the homogenization of membrane components and permitting membranous organelles to maintain distinct identities throughout the life of the cell. Considerable progress has been made towards understanding the molecular basis of membrane traffic1–8. A number of traffic pathways have been defined, their major protein components identified and the structures of several of the key components determined at atomic or molecular resolution. The best studied traffic pathways are those that use carrier vesicles that are clearly identifiable by their coats, made of the coatomer COPI, of COPII, or of clathrin and its partners (FIG. 1). During the formation of a vesicle, a limited set of coat proteins (TABLE 1) carries out a programmed set of sequential interactions that lead to budding from the parent membrane, uncoating, fusion with a target membrane and recycling of the coat components. There are clear similarities and differences between the ways that COPI, COPII and clathrin coats handle these steps (TABLE 2). The first group of reactions, leading to the specific recruitment of coat components to the corresponding donor membrane, form the initiation step. This step is

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

energy dependent and includes sorting of cargo to the forming coat. Coat propagation, the second step in the process, couples further addition of coat components and additional recruitment of cargo with invagination of the underlying membrane. When formation of the coat ends, the vesicle buds by scission of the neck connecting the deeply invaginated membrane to the donor surface. This is a relatively simple step for COPI- and COPIIcoated vesicles but involves a significant amount of regulation for clathrin-coated vesicles. Finally, during uncoating, the coat components are released so that membrane fusion can occur between the naked vesicle and the target organelle. This article discusses the emerging molecular rationale for coated-vesicle assembly for, in order of increasing complexity, COPII, COPI and clathrin. Pathways of membrane traffic

COPI and COPII vesicles traffic between the endoplasmic reticulum (ER) and the Golgi complex — COPI primarily from the Golgi to the ER and between Golgi cisternae, and COPII from the ER to the Golgi (FIG. 1). The clathrin pathway has two major routes, from the plasma membrane to the early endosome and from the Golgi to the endosome. Other structures have been observed in the cell that do not have COP or clathrin coats. Internalization from the plasma membrane can also occur via MACROPINOCYTOSIS, PHAGOCYTOSIS and probably through CAVEOLAE9. In the secretory pathway, poorly understood tube-like structures connect the Golgi with the plasma membrane and the ER10–13. In the endocytic pathway, tubules emerge from early endosomes and participate in recycling to the plasma membrane14. These carrier structures have been hard to study, partly because they tend to be heterogeneous. In some cases,

VOLUME 1 | DECEMBER 2000 | 1 8 7

REVIEWS molecular detail. Nevertheless, there is good agreement in the field that the major components have been identified and there is a good working model for the sequence in which they act (FIG. 2).

Clathrin coats Recycling endosome

Early endosome

Late endosome

Clathrin coats

Trans-Golgi

COPII coats

Cis-Golgi

COPI coats Lysosome

Endoplasmic reticulum (ER)

Figure 1 | The major membrane traffic pathways that use carrier vesicles coated with COPI, COPII and clathrin in eukaryotic cells. In the biosynthetic pathway, newly synthesized molecules are transported from the endoplasmic reticulum to the Golgi and from one cisterna of the Golgi to the next until they reach the trans-Golgi network. There, sorting occurs, directing traffic to the plasma membrane or to endosomes. In the endocytic pathway, macromolecules are internalized at the plasma membrane and forwarded to early endosomes, from where they are either recycled to the plasma membrane through recycling endosomes or forwarded towards degradation in late endosomes and lysosomes.

they might be used for large-scale movement of selectively captured membrane components12,15. The most important distinction between coated vesicles and other forms of membrane carriers is the presence of an identifiable protein coat, which assembles at a particular region of the membrane, locally deforming it. Although the events required are similar — tubes, macropinosomes and phagosomes also need to form, pinch off, reach their target and fuse with it just like coated vesicles — it is impossible to transfer our understanding of vesicle traffic to these other mechanisms. COPII

PHAGOCYTOSIS

Actin-dependent process, by which cells engulf external particulate material by extension and fusion of pseudopods around each particle. CAVEOLA

Flask-shaped invagination at the plasma membrane, possibly involved in the uptake of extracellular materials.

188

COPII components, and COPII-coated vesicles, were originally discovered in the yeast Saccharomyces cerevisiae using genetic approaches coupled with a cell-free assay that measured the transfer of a marker protein from the ER to the Golgi3,8. This pathway has a mammalian counterpart and most of the COPII components have been identified (TABLE 1). Newly synthesized proteins destined for secretion are sorted into COPII-coated vesicles at specialized regions of the ER, which in mammalian cells do not contain membrane-bound ribosomes. Biochemical analysis has led to a relatively detailed model for the mechanism of COPII vesicle formation. But as there is no structural information on any of the coat components, it is hard to claim that we understand the mechanism of even this relatively simple pathway in

| DECEMBER 2000 | VOLUME 1

Biochemical and cell biological observations. Although several as-yet-undefined membrane components are presumably necessary for efficient operation of the COPII pathway, vesicular transport can be minimally reconstituted using three cytosolic components containing a total of five proteins: the Sec23p–Sec24p complex, the Sec13p–Sec31p complex and the small GTPase Sar1p (REF. 16). These proteins support a cargo-carrying budding reaction from isolated ER membranes. Deletion of the genes for most of these components is lethal for yeast, highlighting the strict dependence on this pathway for ER-to-Golgi traffic. The events driven by these five cytosolic proteins, together with the membrane components required for targeting and fusion, are varied and complex (TABLE 2). Cargo capture, deformation of the budding membrane, scission to detach the forming vesicle from the donor membrane and coat release before fusion of the vesicle with the Golgi network all occur in this minimal system. The GTP-binding protein Sar1p is particularly important for the budding reaction because its activation initiates coat formation; it also recruits part of the ‘label’ needed for correct vesicle targeting and fusion. The GDP-bound form of Sar1p is normally cytosolic and is recruited to the ER membrane upon interaction with Sec12p, an ER-bound membrane protein that serves as the guanine exchange factor (GEF) for Sar1p (REF. 17). Sar1p–GTP then facilitates the association of the Sec23p–Sec24p complex with cargo proteins. The Sec23p–Sec24p complex is probably the component responsible for cargo recognition18,19 but the sorting signals recognized by the complex remain to be identified. Members of the p24 family of transmembrane proteins bind to Sec23p through a cytosolic diphenylalanine motif. As these proteins are required for efficient ER-to-Golgi traffic of some cargo proteins20, it is thought that they might serve as cargo adaptors21. In addition to recruiting the Sec23p–Sec24p complex, the GTP-bound form of Sar1p activates Sec23p to bind SNARE proteins involved in the specificity of targeting and in the fusion reaction of vesicles with acceptor membranes18. ER membranes with Sec23p–Sec24p and Sar1p can then recruit Sec13p–Sec31p (REF. 16). The complex is likely to act as a scaffold, very much like clathrin, to drive membrane deformation and to complete vesicle budding. Completing the cycle, Sec23p acts as a GTPase-activating protein (GAP) for Sar1p. It is thought that, after GTP hydrolysis, Sar1p–GDP is released, leading to uncoating before fusion of the vesicle to the target membrane and formation of a new coated vesicle. GTP hydrolysis by Sar1p is thus a timer, triggering uncoating at a suitable interval after coat formation. Studies on live cells. The model describing the formation of COPII vesicles brings together a number of bio-

www.nature.com/reviews/molcellbio

REVIEWS

Table I | Coat proteins and their main binding partners Protein

Mammalian

Yeast

Salient features

Sar1p

hSar1p

Sar1p

Small GTPase; Ras family

Sec13p–Sec31p

hSec13p hSec31p

Sec13p Sec31p

WD 40 repeats (β-propeller) WD 40 repeats (β-propeller)

Sec23p–Sec24p

hSec23p hSec24p ? ?

Sec23p Sec24p/Iss1p/Lst1p Sec12p Sec16p

Sequence homology with Sec24p; GAP for Sar1p Sequence homology with Sec23p GEF for Sar1p Membrane protein; forms a ternary complex with Sec23p–Sec24p

ARF1

yARF1/2/3

Small GTPase; Ras family

Coatomer

α-COP β-COP

Ret1p Sec26p

β′-COP γ-COP δ-COP ε-COP ζ-COP

Sec27p Sec21p Ret2p Sec28p Ret3p

WD 40 repeats (β-propeller) Binds ARF1 weak sequence identity to β-AP WD 40 repeats (β-propeller) Binds members of p24 family Weak sequence identity to µ-AP

ARFGAP

ARFGAPs

Glo3p

GAP for ARF

ARFGEF

ARFGEFs

Gea1p/Gea2p

GEF for ARF

Chc1p

LCa/b

Clc1p

Subunits polymerize into a triskelion; Atomic structures for several fragments are known: α-zigzags and β-propeller Unknown function

AP-1

γ β1 µ1 σ1

Apl4p Apl2p Apm1p Aps1p

γ-AP, β-AP and α-AP sequences are related

AP 2

αa/c β2 µ2

Apl3p Apl1p Apm4p

σ2

Aps2p

Atomic structure of C-terminal α-ear is known Atomic structure of C-terminal β-ear is known Atomic structure of part of µ2 interacting with YppØ sorting motifs is known σ2 sequence weakly related to N-terminal portion of µ’s

β-arrestin1/2

None

COPII

COPI

Weak sequence identity to σ-AP

Clathrin Clathrin

Adaptors

β-arrestin

σ1 sequence weakly related to N-terminal portion of µ’s

Atomic structure not solved but probably similar to known α-arrestin structure

Partners Amphiphysin

Amphiphysin

Rvs161p/167p

Binds clathrin, AP-2, dynamin

AP180

yAP180

Binds clathrin; regulates size of neuronal vesicles

ARF1

Arf1p/2p

Helps recruit AP-1

Auxilin

Auxilin1/2

Aux1p

Contains J domain and is cofactor for Hsc70 uncoating ATPase; binds clathrin

Dynamin

Large GTPase; fission of necks in deeply invaginated clathrin pits

Endophilin

Fission; membrane deformation by changes in lipid composition at neck of deeply invaginated clathrin pit

Epsin

Ent1p/2p

Eps15

Eps15/15R

Pan1p

Binds AP-2; located at rim of clathrin coated pits; mainly excluded from clathrin coated vesicles

GGA

Gga1p/Gga2p

Binds γ-synergin; traffic regulation from Golgi to lysosome/vacuole

Intersectin

Intersectin1/2

Synaptojanin

Sjl1p/3p

Phosphoinositide 5′-phosphatase; role in coat release

Synaptotagmin

Synaptotagmin None

Ca2+ sensor; binds AP-2

Syndapin I

None

Binds N-WASP; link to actin network/signalling cascade ?

Uncoating ATPase

Hsc70

Ssa1p/2p

Dissociation of clathrin from coats

γ-synergin

Binds AP-1; unknown function

(AP, adaptor protein; ARF, ADP-ribosylation factor; ARFGAP, ADP-ribosylation factor GTPase activating protein; ARFGEF, ADP-ribosylation factor guanine exchange factor; COP, coatomer protein; Eps15, EGF receptor pathway substrate clone 15; GAP, GTPase activating protein; GEF, guanine exchange factor; N-WASP, neuronal Wiscott–Aldrich syndrome protein.)

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 1 | DECEMBER 2000 | 1 8 9

REVIEWS

GREEN-FLUORESCENT PROTEIN

Autofluorescent protein originally identified in the jellyfish Aequorea victoria.

chemical and genetic observations. The first efforts to visualize the budding of COPII vesicles in live cells yielded surprising results. The stably expressed human homologue of Sec13p and the transiently expressed human homologue of Sec24p, both tagged with the GREEN-FLUORESCENT PROTEIN (GFP), were detected by timelapse fluorescence microscopy22,23. They were found in

stable and nearly immobile bright spots associated with ER sites devoid of ribosomes, presumably where COPII vesicles form and bud. It is difficult to reconcile the stability of these spots with a model in which the recruitment of cytosolic Sec13p to the ER membrane is concurrent with formation and budding of COPII-coated vesicles. It is possible

Table 2 | Steps in coat formation COPII

COPI

Clathrin

1. GTPase activation and membrane binding

Sar1p–GDP is recruited to ER by transmembrane GEF (Sec12p) and converted to Sar1p–GTP

ARF1– GDP is recruited to Golgi by soluble GEFs (Gea1p and Gea2p), probably bound to Golgi membranes, and is converted to ARF1–GTP

TGN ARF1–GTP recruited to trans-Golgi membrane; mechanism is unknown Plasma membrane ?

2. Cargo and v-SNARE recruitment

a. Sar1p–GTP recruits Sec23p–Sec24p complex b. Sec23p–Sec24p binds to members of the p24 protein family of possible cargo receptors, and together with Sar1p binds v-SNAREs (Bet1p, Bos1p)

a. ARFGAP binds to ARF1–GTP and to the transmembrane KDEL-receptor b. ARF–GTP together with cargo proteins containing C-terminal KKXX (or KXKXX) motifs recruit cytosolic COPI coatomer

TGN a. ARF1–GTP recruits AP-1 adaptor to trans-Golgi membrane b. AP-1 binds to membrane receptors containing YppØ and LL motifs Plasma membrane a. ATP, GTP and phosphoinositides required to recruit AP-2 to plasma and/ or endosomal membranes; mechanism is unknown b. AP-2 binds to membrane protein synaptotagmin c. AP-2 binds to membrane receptors containing YppØ and LL motifs

3. Start of coat assembly

Cytosolic Sec13p–Sec31p complex binds to prebound Sec23p–Sec24p

I. Initiation

Cytosolic clathrin recruited to pre-bound AP-1 or AP-2 adaptors

II. Propagation 1. Loss of GTPase

Sar1p–GTP hydrolysis increased15–30-fold by Sec23p; Sar1p–GDP released and used in further cycles

ARF–GTP hydrolysis increased 1,000-fold by ARFGAP and COPI; ARF–GTP released and used in further cycles Hydrolysis rate depends on sequence of C-terminal motif in cargo

2. Further cargo recruitment and coat assembly

Growth of coats by sequential incorporation of other coat elements Membrane-bound cargo proteins diffuse laterally and are captured by the forming coats

Not known

Plasma membrane a. Association of AP-2 with clathrin or with phosphoinositides increases the affinity of AP-2 for YppØ motifs b. β-arrestin recruits seven-transmembrane G-coupled receptors to clathrin coats c. Growing edge of clathrin lattice contains AP-2 bound to Eps15

3. Membrane deformation Continuous process that is coupled to the growth of the coat III. Vesicle budding No other proteins are required in vitro; energy for membrane scission provided by coat polymerization

No other proteins are required in vitro; energy for membrane scission provided by coat polymerization

a. Amphiphysin binds to clathrin and AP-2 and acts as a dynamin receptor b. Dynamin–GDP is recruited to the neck of the budding vesicle and polymerizes into a dynamin ring; dynamin–GEF is unknown c. Endophilin is recruited to the ring. Membrane deformation and fission is facilitated by the coupling of the acyl transferase activity of endophilin and the neck constriction imparted by the dynamin ring d. Dynamin (in the rings) acts as its own dynamin– GAP; GTP hydrolysis releases dynamin for another cycle

Spontaneous ? (following GTP hydrolysis of Sar1p)

Spontaneous ? (following GTP hydrolysis of ARF1 activated by ARFGAP at time of cargo recognition or by second ARFGAP located close to the target membrane)

Plasma membrane a. Hsc70–ATP and auxilin bind to clathrin coats and drive disassembly, presumably by a clockwise twist imparted on clathrin triskelions (regulation exists but the mechanism is unknown) b. A fraction of AP-2 can be dissociated from membranes by Hsc70–ATP and an unknown cytosolic factor c. Phosphorylation of AP-2 (β-subunit) prevents its association with clathrin d. Synaptojanin required for efficient release of coats

IV. Uncoating

(AP, adaptor protein; ARF1, ADP-ribosylation factor 1; ARFGAP, ADP-ribosylation factor GTPase activating protein; Eps, EGF receptor pathway substrate; GAP, GTPase activating protein; GEF, guanine exchange factor.)

190

| DECEMBER 2000 | VOLUME 1

www.nature.com/reviews/molcellbio

REVIEWS that Sec13p and other COPII components accumulate in specific regions of the ER membrane, which can then act as reservoirs for the subsequent assembly of a COPII coat with the curvature needed for membrane fission and vesicle budding. The relatively low expression level of the described hSec13p–GFP would have made the detection of single COPII vesicles (as opposed to extended reservoirs) difficult. The development of algorithms and detectors that can distinguish weak signals from background noise and the generation of stronger chromophores would much improve our ability to analyse the dynamics of vesicular traffic. COPI

COPI-coated vesicles occur in various intracellular contexts1 and this versatility of function seems to correlate with a greater biochemical complexity compared with COPII vesicles (TABLE 1). Whereas COPII-dependent traffic is unidirectional (from ER membranes to the Golgi), the direction of COPI traffic is still a matter of controversy. COPI-coated vesicles seem to function primarily in Activation

Cargo capture

retrograde transport from the ER–Golgi intermediate compartment to the ER but they are also important in forward transport within the cisternae of the Golgi. The COPI coatomer is a complex of seven proteins (α, β, β′, γ, δ, ε and ζ). COPI-coated vesicles efficiently capture proteins carrying in their cytoplasmic carboxyterminal domain sorting signals of the form KKXX (the dilysine motif) or KXKXX (X is any amino acid). The KDEL receptor, a multiple-spanning membrane protein that binds and retrieves lumenal proteins containing the KDEL carboxy-terminal sequence, is also transported along this pathway. The γ subunit seems to be the component responsible for cargo recognition because it recognizes the KKXX and KXKXX motifs, but it is not known whether it also recognizes the KDEL receptor24. Members of the p24 protein family also interact with COPI coatomers in addition to COPII and might facilitate the recruitment of COPI coatomers to Golgi membranes25. The initial event in the COPI pathway that leads to recruitment of the coat requires the association of the Coat assembly D Pi

T GAP

D T

GTP GDP

Scission

Uncoating D

Sec12p

p24

GAP

Cargo

SNARE

Sec31p

Sec24p

Sec13p

Sec23p

D Sar1p–GDP

Sar1p–GTP

Figure 2 | The key steps in the formation of COPII-coated vesicles. Coat assembly is activated by the recruitment of Sar1p–GTP to the membrane. This allows the binding of the Sec23p–Sec24p complex and the recruitment of cargo. The Sec13p–Sec31p complex binds next, leading to membrane deformation. When the coat is complete, the vesicle buds. The GTPase activity of Sar1p is enhanced by Sec23p, which acts as a timer, leading to inactivation of Sar1p and uncoating. (GAP, GTPase activating protein.)

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 1 | DECEMBER 2000 | 1 9 1

REVIEWS

Activation

Cargo capture D

GTP GDP

αβ'ε

αβ'ε β

αβ'ε

T ?

Coat assembly

Scission Pi

GAP

αβ'ε γ δ β T

ζ Cargo

αβ'ε

γ ζ p24

α β 'ε γ δ β T

αβ'ε

Cargo receptor Soluble cargo

ARF1–GDP

γ ζ

α β 'ε

ARF1–GTP

αβ'ε

ARF1GAP

ARF1GEF

Uncoating D

GAP δ αβ 'ε β T γ

'ε αβ β T

'ε γ αβ TT β

SNARE

αβ′ε β

COPI coatomer

Figure 3 | The key steps in the formation of COPI-coated vesicles. Coat assembly is activated by the recruitment of ARF1–GTP to the membrane. This allows the binding of the COPI coatomer and the recruitment of cargo. GTP hydrolysis is slow when ARF1 is bound to its preferred cargo, allowing kinetic regulation of coat recruitment. Membrane deformation occurs at the same time as coat recruitment. When the coat is complete, the vesicle buds. The GTPase activity of ARF1 is enhanced by ARF1GAP, which acts as a timer, leading to inactivation of ARF1 and uncoating. (ARF1, ADP-ribosylation factor 1; ARF1GAP, ADP-ribosylation factor 1 GTPase activating protein; ARF1GEF, ADP-ribosylation factor 1 guanine exchange factor.)

MYRISTOYLATION

Covalent attachment of a hydrophobic myristoyl group to the amino-terminal glycine residue of a nascent polypeptide.

192

GTPase ARF1 (ADP-ribosylation factor 1) in its active form to the membrane (FIG. 3, TABLE 2). The ARF protein family has many members and targeting of ARF1 to the correct membrane involves specific association with its appropriate GEF. Several GEFs for ARF1 have been identified, one of which (known as ARF1GEF, ARNO3 or GRP1) seems to be specifically associated with the COPI pathway26. In contrast to the COPII-associated Sar1p, ARF1 is MYRISTOYLATED to allow its membrane association. In the GTP-bound state, the myristoyl group is exposed and ARF1 becomes membrane bound. When the GTP is hydrolysed, the protein undergoes a conformational change, developing a myristoylbinding pocket that covers the tail, solubilizing the protein. It has been suggested that the hydrolysis of GTP and release of ARF1 from the membrane act as a timer, triggering (as with Sar1p and COPII vesicles) the release of the other coat components and preparing the vesicle for fusion with its target membrane. The rate at which ARF1 hydrolyses GTP depends on its association with ARFGAP and the COPI complex27, both of which are required for full GTPase activation. An element of further regulation has recently been uncovered in this step. A synthetic peptide containing the carboxy-terminal FFXXRRXX sorting signal of the p24 protein hp24a, which binds to COPI, markedly reduces the ability of COPI to stimulate ARFGAP28.

| DECEMBER 2000 | VOLUME 1

However, other peptides that contain the FFXXKKXX sequence and bind to the same site in COPI do not affect the stimulation of GTP hydrolysis of ARF by COPI (FIG. 3). These data indicate that the inhibitory peptides may represent preferred cargo. When these peptides are present on a protein tail, the rate of GTP hydrolysis will be slow even when some COPI has been recruited to the membrane. Vesicles that capture preferred cargo will retain their ARF1 protein long enough to complete coat assembly, whereas vesicles that capture other proteins will not. This proposal suggests that cargo selection is kinetically regulated: it depends not on different affinities of cargo for the coat but instead on a dynamic regulation of the rate of coat release. Although there is substantial confidence that most of the major components of this pathway have been identified, there is still some way to go before we understand the functions of each component. The structural characterization of the components of the pathway is just beginning — the crystal structures of ARF1 and ARF1 complexed to a domain of GAP have been determined but no part of the COPI complex has yet been visualized. Structures for these components are likely to emerge before long. Clathrin

Clathrin-coated vesicles are the most prominent of the carrier vesicles and were the first to be discovered and

www.nature.com/reviews/molcellbio

REVIEWS

Activation (TGN)

GTP GDP

Activation (plasma membrane)

Cargo capture

ATP ADP+Pi ?

Hsc70 Coat assembly

Scission

Uncoating Auxilin

ATP ADP+Pi

GTP GDP+Pi

ARF1GEF

Dynamin

ARF1–GTP

Receptor (Yppφ)

ARF1–GDP

Amphiphysin

Receptor (LL)

SNARE

Endophilin

Docking complex β-arrestin

AP-2

Clathrin

AP-1

Figure 4 | The key steps in the formation of clathrin-coated vesicles. At the trans-Golgi network, coat assembly is activated by the recruitment of ARF1 to the membrane. It is not clear how coat assembly is activated at the plasma membrane. One end of adaptor proteins bind to cargo molecules and the other end to other coat components, including clathrin. Clathrin triskelions polymerize into hexagons and pentagons, forming a cage, which leads to membrane deformation. When the coat is almost complete, dynamin (together with accessory proteins) pinches off the vesicle. Uncoating requires ATP hydrolysis by Hsc70 and auxilin. (AP, adaptor protein; ARF1, ADP-ribosylation factor 1; ARF1GEF, ADP-ribosylation factor 1 guanine exchange factor.)

studied5,29–31. The low-density lipoprotein (LDL) receptor and many other plasma membrane proteins and their ligands are internalized by the clathrin pathway for traffic to endosomes, and clathrin-coated vesicles bud from the trans-Golgi network (TGN) to fuse with endosomes.Yeast cells also use a clathrin-based pathway for membrane traffic, although its use in the secretory pathway seems to be more prominent than its role in endocytosis32–35. Recruitment of the clathrin coat. Clathrin is the most abundant protein in the coat of these vesicles and it provides the scaffold that orchestrates protein sorting, membrane deformation and budding5,36,37. By contrast to the COPI and COPII vesicles, clathrin-coated vesicles have a large variety of associated proteins (TABLE 1). So far, more than 25 proteins have been identified as partners in the endocytic pathway alone5,38. The functions of most of these proteins still have to be elucidated (TABLE 2). Vesicles that form at the TGN carry heterotetrameric AP-1 adaptor protein complexes, whereas vesicles that form at the plasma membrane carry related AP-2 complexes. AP-1 and AP-2 adaptors bind to certain sorting signals (FDNPVY, tyrosine-based YppØ NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

motifs and dileucine motifs) found in the cytosolic tails of a large number of membrane proteins. The nonvisual arrestins are a second type of adaptor, which recruits seven-transmembrane-helix G-protein-coupled receptors at the plasma membrane to clathrincoated vesicles. This occurs by direct contacts with the cytosolic side of the G-protein-coupled receptors, with clathrin and with AP-2 (REFS 39–41). Adaptors also interact with a number of other proteins involved in clathrin-coated-vesicle function4, and they recruit clathrin to the membrane, initiating coat formation. As with the COPI and COPII systems, the recruitment of clathrin coat components to the appropriate target membranes is nucleotide dependent (FIG. 4). The role of ARF1 in the clathrin pathway is similar to, but not the same as, its role in the COPI pathway. In both cases,ARF1 helps to recruit key coat components but, in the clathrin pathway, it is involved only in the recruitment of AP-1 to Golgi membranes42–45; ARF1 does not seem to act as a timer for uncoating as it does for COPI. In vitro recruitment of AP-2 to the plasma membrane, endosomes and lysosomes also requires the activation of cytosolic components by ATP and GTP, but the identity of the

VOLUME 1 | DECEMBER 2000 | 1 9 3

REVIEWS

Actin

Clathrin

Syndapin

Intersectin

Eps15

Epsin Phosphoinositide

Synaptojanin

AP-180 R

Endophilin

N-WASP

Amphiphysin

AP-2

Dynamin

β-arrestin

Synaptotagmin

Figure 5 | Network of protein–protein and protein–lipid interactions involved in clathrin coat formation at the plasma membrane. Proteins that bind to phosphoinositides are represented here in direct contact with the plasma membrane, although this interaction might be transient. Often, two or more interactions have been detected, leading to the idea that these proteins might operate as coincidence detectors. (AP, adaptor protein; Eps15, EGF receptor pathway substrate clone 15; N-WASP, neuronal Wiskott–Aldrich syndrome protein; R, transmembrane receptor; R7, seven-transmembrane-helix G-protein-coupled receptor.)

J DOMAIN

Approximately 73 amino-acid region found in DnaJ-like heat shock proteins, which catalytically activates proteins of the Hsc70 family.

194

nucleotide-binding or -hydrolysing protein(s) is still obscure46,47. It has been proposed that the membrane protein synaptotagmin, a Ca2+ sensor, is required for AP-2 recruitment, perhaps as part of a set of postulated docking proteins required for efficient recruitment48. The interaction between AP-2 and synaptotagmin is nucleotide independent but it is enhanced by peptides bearing the YQRL tyrosine-based endocytic signal48. As synaptotagmin is primarily found in brain tissues, its role in clathrin-coat formation might be more as a coincidence detector than as a general docking protein. A simple model for clathrin-coat formation, which covered most of the evidence until very recently, was that the AP complexes have two essential functions: they bind to certain sorting motifs in receptor tails and they recruit clathrin to the membrane to initiate coat formation. Although this model is still logical and plausible, two new reports that genetic knockouts of all of the AP proteins in yeast do not show the expected profound defects in vesicle formation or membrane traffic force a re-examination of its details49,50. If the AP proteins are performing the functions of cargo selection and coat initiation at all, there must be a second protein or set of proteins available to take over these functions when required. Genetic knockouts of µ- or α- subunits of the AP proteins in multicellular organisms such as Drosophila melanogaster51 and Caenorhabditis elegans52,53 are not lethal, whereas, in mice, knockouts of the µ1 or γ- subunits of AP-1 are embryonically lethal54,55. These data indicate that the severity of the deletion is compounded by the cellular context in which AP function is hindered. The unexpected outcome of some of these genetic experiments

| DECEMBER 2000 | VOLUME 1

should be taken as a cautionary note that the biochemical studies used to propose models are still not sufficient to take into account all of the functions exerted by these coat proteins. Regulation of vesicle formation. Amphyphysin56, epsin57,58, synaptojanin59 and Eps15 (EGF receptor pathway substrate clone 15)60,61 are some of the many accessory proteins found in association with clathrin coats at the plasma membrane. They not only interact with clathrin but also have binding sites for AP adaptors, for proteins such as the large GTPase dynamin (itself involved in the budding step) and even for specialized lipids such as phosphoinositides. Although the functions of many of these proteins are still not known, it is becoming increasingly clear that they are part of a network of complex molecular switches and contacts that regulate various aspects of clathrin-mediated traffic (FIG. 5). On the basis of biochemical characteristics, many of these proteins seem to function as coincidence detectors, needing two or more simultaneous relatively weak interactions to exert their function. For example, AP-2 and β-arrestin have binding sites for clathrin, for cargo recruitment and for each other41,62–66. Simultaneous interactions probably occur between these molecules when they are recruited to the clathrin coat, providing a control point for synchronizing the concentration of cargo into endocytic vesicles. Uncoating. The heat shock protein Hsc70 and auxilin, a JDOMAIN-containing protein, are responsible for clathrin

disassembly67–70 through an ATP-dependent reaction, which is presumably subject to regulation. The uncoating mechanism is therefore fundamentally different in the COPI or COPII and clathrin pathways. In both COP pathways, uncoating results from a change in the properties of coat components in response to GTP hydrolysis in Sar1p or ARF1. In the clathrin pathway, Hsc70 and auxilin, proteins that do not participate in the process of coat assembly, are required to achieve the same effect. The most plausible reason for this difference is a requirement for independent control over various steps in the clathrin pathway and over various distinct uses of clathrin vesicles. Vesicle fission. The mechanism of vesicle fission also seems to be more complex in the clathrin pathway. In the COPI and COPII pathways, vesicle fission is intrinsic to the completion of coat assembly. Deformation and fission of the membrane requires energy and, for the COPs, this energy comes primarily from the energy of association of the coat proteins as the coat forms. In the case of clathrin, the energy of deformation probably also comes from coat assembly, but the fission step requires enzymatic activity from a GTPase, dynamin. Dynamin is recruited to coated pits and, under conditions that interfere with its GTPase activity, dynamin forms a collar or ring around the neck of the budding vesicle. Whether this protein acts as a mechanochemical transducer to generate fission (the ‘boa constrictor’, ‘blue collar’ or ‘pinchase’ model)71–73, as a recruiter to attach other proteins that are directly responsible for the fission step (the ‘rattlesnake’ or ‘white collar’

www.nature.com/reviews/molcellbio

REVIEWS

Figure 6 | Clathrin coats, a collage of medium and high-resolution views. The model of a clathrin cage at 21-Å resolution was obtained by electron microscopy92. The clathrin triskelion is a puckered and relatively rigid molecule93. The proximal84 and distal81 leg domains of the clathrin heavy chain have similar α-zigzag atomic structures, and the globular terminal domain of the clathrin heavy chain is a β-propeller81.

model)72,74,75 or as a combination of both is still a matter that needs to be resolved. Lipid-modifying enzymes such as endophilin, synaptojanin and phospholipase D are also involved in vesicle formation59,74,76–78. Endophilin is an acyltransferase that interacts with dynamin and that generates lysophosphatidic acid by fatty acyl transfer from arachidonic acid or palmitic acid74. The current view is that this reaction produces a negative curvature at the neck of the vesicle, which is postulated to facilitate pinching-off of the membrane in the fission step79,80. Synaptojanin acts in the endocytic pathway to remove the 5-phosphate from phosphatidylinositol-4,5-bisphosphate, probably modulating the recruitment of phosphoinositide-binding proteins such as AP-2, β-arrestin and dynamin to the plasma membrane. Phospholipase D produces phosphatidic acid, modulating membrane curvature and increasing the concentration of the second messenger diacyl glycerol79. The relative importance of these different enzymatic activities is not yet clear but, once again, there is obviously a greater complexity in the clathrin pathway than in the COP pathways and greater potential for regulation.

SH3 DOMAINS

(Src homology region 3 domains.) Protein sequences of about 50 amino acids that recognize and bind sequences rich in proline.

Visualization at medium and high resolutions. Considerable progress has been made in the visualization of some elements of the clathrin coat using a combination of X-ray crystallography and electron microscopy41,81–93. This information has allowed us to understand the molecular basis of some of the interactions that are known to occur within the coat, including sorting and coat formation. Because of the size and complexity of many of the coat components, the first atomic views are of recombinant fragments from adaptors, amphiphysin, dynamin, epsin, Eps15 and clathrin41,81–91.

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

The µ subunit of AP complexes is the portion of the adaptor that recognizes short peptides bearing YppØ sorting signals94. Detection of this interaction led to further experiments based on combinatorial approaches95,96 that established rules of engagement between YppØ sorting signals and µ-subunits found in different APs. These observations were rapidly followed by the determination of the atomic structure of a fragment of the µ2 subunit of AP-2, either alone or in association with short peptides bearing the endocytic signals YQRL from TGN38 or YRAL from the EGF receptor82. For several reasons, these atomic models have had a large effect on our understanding of the mechanism of sorting. First, they show that the structure of µ2 does not change on interaction with peptides bearing sorting signals, arguing that this portion of the adaptor is relatively rigid. Second, they show that the sorting signals adopt an extended conformation on binding to µ2, and they therefore help to explain the basis of specificity in the recognition of sorting signals by different µ-subunits. Third, they will facilitate the implementation of mutagenesis studies by aiding in the design of µ-variants for in vivo and in vitro studies. Structures have also been determined of fragments corresponding to the globular carboxy-terminal ‘ears’ of the α- and βadaptins of AP-283,87,91, the EH domain of Eps15 (REFS 85,86), the SH3 DOMAIN of amphiphysin 2 (REF. 90), part of the ENTH domain of epsin89 and the pleckstrin-homology domain of dynamin88. As with µ2, these structures will probably facilitate the design and interpretation of functional experiments. The structural studies of clathrin and its partners have had a clear and important effect on the way that we now think about coat formation, and provide a basis for understanding the mechanism of coat formation and the interaction between clathrin and adaptors. The crystal structures of two large fragments of the clathrin heavy chain, comprising about half of its total mass, are now known81,84 (FIG. 6). One fragment contains the globular amino-terminal domain of the heavy chain linked to a segment, about 100 Å long, that joins this domain to the distal portion of the clathrin leg81. The atomic structure of the amino-terminal domain was shown to be a seven-bladed β-propeller81 with a groove between blades 1 and 2 that accommodates the clathrin box41, a short motif found in proteins that interact with clathrin (for example, β-arrestin and APs)5,62. The structure of the amino-terminal domain bound to a clathrin-box peptide is consistent with the inward orientation of the amino-terminal domains within a coat revealed by fitting the atomic structure of the amino-terminal domain and linker to the lower-resolution model of a clathrin coat obtained by electron microscopy93. The second crystallized fragment maps to the opposite end of the leg, and corresponds to a linear stretch of about 100 Å derived from the proximal leg, the region involved in the interaction with light chains84. Comparison of the two structures reveals that the extended segments are similar. They contain a polypeptide chain folded into an ‘α-zigzag’, a series of short, apposed α-helices that run back and forth,

VOLUME 1 | DECEMBER 2000 | 1 9 5

REVIEWS roughly perpendicular to the overall direction of the leg. The parts of the leg not contained in the crystal structures are known to have a high α-helical content and they are also believed to contain α-zigzags. The αzigzag is not a completely rigid fold, as shown by a modest variability of curvature in the linker segment seen in the crystals of the fragment containing the amino-terminal domain and the linker. The extent of this flexibility is sufficient to accommodate the changes required for the formation of pentagons and hexagons in the facets of the coat, particularly at the joints between the proximal and distal legs. Electron microscopy studies of clathrin triskelions and coats have revealed how the legs interact within a surface lattice92,93,97,98 (see online animation). This has led to a model that explains how a coat can form by the sequential addition of soluble clathrin molecules to the assembling lattice93. The model highlights the idea that the addition of triskelions can happen at various locations on the lattice but that rearrangement of elements in a preformed coat (from hexagons to pentagons or vice versa) is topologically impossible36. That is, a budding coated pit must form by the sequential incorporation of coat elements (clathrin, adaptors, and so on) and not by the direct transformation of a pre-existing flat array into a curved coat. Flat arrays99, which abound at the plasma membrane and in the TGN, could act as dynamic reservoirs, similar to the relatively stable COPII reservoirs on ER membranes discussed above, disassembling at their margins to provide triskelions for de novo coat formation in the immediate vicinity. Studies in live cells. Time-lapse fluorescence microscopy has been used to observe GFP-tagged clathrin in Dictyostelium100 and mammalian101 cells. In Dictyostelium, the endogenous clathrin heavy chain was replaced with heavy chain tagged with GFP at its carboxyl terminus. Bright spots appeared at the plasma membrane, the perinuclear region and the cytosol that were highly mobile and seemed to move in synchrony, like a swirl following the internal motions of the cell. The highly mobile spots persisted for 30 seconds or less and their disappearance could reflect uncoating or movement away from the plane of focus. Thus, the results from this experiment do not resolve the question of whether the observed dynamic behaviour reflects the budding cycle of a clathrin-coated vesicle. In mammalian cells, transient expression of GFPtagged clathrin light chain A resulted in bright spots associated with the plasma membrane and bright patches located in the perinuclear region101. The bright spots were stable most often, very much as described above for Sec13p–GFP. Occasionally, it was possible to observe smaller and weaker spots emanating from locations at the plasma membrane close to the brighter ones. The proposed interpretation of the data was that all the spots represent various stages in the process of budding of coated pits and formation of coated vesicles. A word of caution is needed, however. The expected level of replacement of the endogenous light chains by LCa–GFP by transient expression is limited. It is possi-

196

| DECEMBER 2000 | VOLUME 1

ble that the bright, stable spots represented the large, flat, hexagonal flat arrays of clathrin acting as potential reservoirs for coated-vesicle formation, rather than vesicles themselves. The smaller coated pits and vesicles (with <60–100 triskelions in the coat) would be harder to detect and, in general, the intensity of the fluorescence signal would probably have been close to the detection limits (5–10 GFP molecules per location). Where are we heading and what can we expect?

Current research is attempting to define the remaining components that participate in the COPI, COPII and clathrin pathways. Structural and functional comparisons between the elements of each of these systems have proved to be very useful. Efforts to understand other routes of traffic are just starting to bear fruit, using a combination of genetics and biochemistry. Soon, there should be a better understanding of how molecules move between all membrane-bound intracellular compartments, whether by coated vesicles or by vesiculo–tubular structures. With the complete list in hand, it should be possible, for any given pathway, to do the appropriate biochemical characterization and to obtain a mechanistic description, at atomic resolution, of the interactions that regulate the traffic. These pathways are strikingly complex and have a large number of components and an even larger number of interactions and regulating steps that are used to control proper membrane flow. Until recently, it was thought that relatively simple genetic manipulations such as gene disruptions would provide direct clues to the function of any given protein component. However, the multicomponent character of the coated-vesicle-based pathways frequently allows compensation, sometimes to the extent that only a weak cellular phenotype is manifest, even in a multiple knockout. In some cases, it will be possible to obtain important and useful information by overexpressing defective proteins or molecules modified to a fixed state of their normal cycle. However, other forms and techniques for selective dissection are needed, including the development of improved methods for visualizing small vesicles and molecules as they form and move inside live cells, and the discovery and use of small molecules that can enter cells and act acutely and specifically on given steps of a selected pathway. Links DATABASE LINKS Sec23p | Sec24p | Sec13p | Sec31p |

[Contents page] ENCYCLOPEDIA OF LIFE SCIENCES Clathrin-coated

vesicles and receptor-mediated endocytosis

www.nature.com/reviews/molcellbio

REVIEWS 1.

2. 3. 4. 5. 6. 7. 8. 9. 10.

11.

12.

13. 14.

15.

16.

17.

18.

19.

20.

21.

22.

23.

24.

25.

26.

27.

28.

Wieland, F. & Harter, C. Mechanisms of vesicle formation: insights from the COP system. Curr. Opin. Cell Biol. 4, 440–446 (1999). Lewin, D. A. & Mellman, I. Sorting out adaptors. Biochem. Biophys. Acta - Mol. Cell Res. 1401, 129–145 (1998). Barlowe, C. Traffic COPs of the early secretory pathway. Traffic 1, 371–377 (2000). Kirchhausen, T. Adaptors for clathrin mediated traffic. Annu. Rev. Cell. Dev. Biol. 15, 705–732 (1999). Kirchhausen, T. Clathrin. Annu. Rev. Biochem. 69, 699–727 (2000). Rothman, J. E. & Wieland, F. T. Protein sorting by transport vesicles. Science 272, 227–234 (1996). Schekman, R. & Orci, L. Coat proteins and vesicle budding. Science 271, 1526–1533 (1996). Springer, S., Spang, A. & Schekman, R. A primer on vesicle budding. Cell 97, 145–148 (1999). Anderson, R. G. The caveolae membrane system. Annu. Rev. Biochem. 67, 199–225 (1998). Hirschberg, K. et al. Kinetic analysis of secretory protein traffic and characterization of Golgi to plasma membrane transport intermediates in living cells. J. Cell Biol. 143, 1485–1503 (1998). Toomre, D., Keller, P., White, J., Olivo, J. C. & Simons, K. Dual-color visualization of trans-Golgi network to plasma membrane traffic along microtubules in living cells. J. Cell Sci. 112, 21–33 (1999). Polischchuk, R. S. et al. Correlative light–electron microscopy reveals the tubular–saccular ultrastructure of carriers operating between the Golgi apparatus and plasma membrane. J. Cell Biol. 148, 45–58 (2000). Presley, J. F. et al. Golgi membrane dynamics. Mol. Biol. Cell 9, 1617–1626 (1998). Gruenberg, J. & Maxfield, F. R. Membrane transport in the endocytic pathway. Curr. Opin. Cell Biol. 7, 552–563 (1995). Volchuk, A. et al. Megavesicles implicated in the rapid transport of intracisternal aggregates across the Golgi stack. Cell 102, 335–348 (2000). Explores the consequences of acutely changing the size of cargo on the dynamics and morphological characteristics of membrane traffic along the secretory pathway. 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). First demonstration that coat components are used to drive vesicular traffic from the endoplasmic reticulum towards the Golgi. Barlowe, C. & Schekman, R. SEC12 encodes a guanine nucleotide exchange factor essential for transport vesicle formation from the ER. Nature 365, 347–349 (1993). Springer, S. & Schekman, R. Nucleation of COPII vesicular coat complex by endoplasmic reticulum to Golgi vesicle SNAREs. Science 281, 698–700 (1998). Peng, R. W., De Antoni, A. & Gallwitz, D. Evidence for overlapping and distinct functions in protein transport of coat protein Sec24p family members. J. Biol. Chem. 275, 11521–11528 (2000). Schimmoller, F. et al. The absence of Em24p, a component of ER-derived COPII-coated vesicles, causes a defect in transport of selected proteins to the Golgi. EMBO J. 14, 1329–1339 (1995). Kaiser, C. Thinking about p24 proteins and how transport vesicles select their cargo. Proc. Natl Acad. Sci. USA 97, 3783–3785 (2000). Stephens, D. J., Lin-Marq, N., Pagano, A., Pepperkok, R. & Paccaud, J. P. COPI-coated ER-to-Golgi transport complexes segregate from COPII in close proximity to ER exit sites. J. Cell Sci. 113, 2177–2185 (2000). Hammond, A. T. & Glick, B. S. Dynamics of transitional endoplasmic reticulum sites in vertebrate cells. Mol. Biol. Cell 11, 3013–3030 (2000). Harter, C. et al. Nonclathrin coat protein γ, a subunit of coatomer, binds to the cytoplasmic dilysine motif of membrane proteins of the early secretory pathway. Proc. Natl Acad. Sci. USA 93, 1902–1906 (1996). Dominguez, M. et al. gp25L/emp24/p24 protein family members of the cis-Golgi network bind both COPI and COPII coatomer. J. Cell Biol. 140, 751–765 (1998). Franco, M. et al. Arno3, a Sec7-domain guanine nucleotide exchange factor for Adp ribosylation factor 1, is involved in the control of Golgi structure and function. Proc. Natl Acad. Sci. USA 95, 9926–9931 (1998). Goldberg, J. Structural and functional analysis of the ARF1ARFGAP complex reveals a role for coatomer in GTP hydrolysis. Cell 96, 893–902 (1999). Goldberg, J. Decoding of sorting signals by coatomer through a GTPase switch in the COPI coat complex. Cell 100, 671–679 (2000).

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

Kinetic model for the regulation of COPI coats that couples cargo with coat formation. 29. Roth, T. F. & Porter, K. R. Yolk protein uptake in the oocyte of the mosquito Aedes aegypti. L. J. Cell Biol. 20, 313–331 (1964). 30. Kanaseki, T. & Kadota, K. The ‘vesicle in a basket’. A morphological study of the coated vesicle isolated from the nerve endings of the guinea pig brain, with special reference to the mechanism of membrane movements. J. Cell Biol. 42, 202–220 (1969). 31. Pearse, B. M. F. Coated vesicles from pig brain: purification and biochemical characterization. J. Mol. Biol. 97, 93–98 (1975). 32. Payne, G. S. & Schekman, R. A test of clathrin function in protein secretion and cell growth. Science 230, 1009–1014 (1985). 33. Seeger, M. & Payne, G. S. A role for clathrin in the sorting of vacuolar proteins in the Golgi complex of yeast. EMBO J. 11, 2811–2818 (1992). 34. Lemmon, S. K. & Jones, E. W. Clathrin requirement for normal growth of yeast. Science 238, 504–509 (1987). 35. Tan, P. K., Davis, N. G., Sprague, G. F. & Payne, G. S. Clathrin facilitates the internalization of seven transmembrane segment receptors for mating pheromones in yeast. J. Cell Biol. 123, 1707–1716 (1993). 36. Kirchhausen, T. Coated pits and coated vesicles — sorting it all out. Curr. Opin. Struct. Biol. 3, 182–188 (1993). 37. Smith, C. J. & Pearse, B. M. F. Clathrin: anatomy of a coat protein. Trends Cell Biol. 9, 335–338 (1999). 38. Jarousse, N. & Kelly, R. B. Selective inhibition of adptor complex-mediated vesiculation. Traffic 1, 378–384 (2000). 39. Ferguson, S. S. G. et al. Role of β-arrestin in mediating agonist-promoted G protein-coupled receptor internalization. Science 271, 363–366 (1996). 40. Goodman, O. B. Jr et al. β-arrestin acts as a clathrin adaptor in endocytosis of the β2-adrenergic receptor. Nature 383, 447–450 (1996). First demonstration of the direct coupling between clathrin and β-arrestin: it opens up the concept that other adaptors, besides the canonical AP-1 or AP-2, can mediate sorting of membrane proteins into clathrin-coated pits. 41. Ter Haar, E., Harrison, S. C. & Kirchhausen, T. Peptide-ingroove interactions link target proteins to the β-propeller of clathrin. Proc. Natl Acad. Sci. USA 97, 1096–1100 (2000). Atomic model obtained by X-ray crystallography of complexes formed between clathrin and peptides of β-arrestin or of AP-2 adaptor that contain sequences known to be involved in the interaction with clathrin. 42. Chen, C. Y. & Graham, T. R. An arf1 Delta synthetic lethal screen identifies a new clathrin heavy chain conditional allele that perturbs vacuolar protein transport in Saccharomyces cerevisiae. Genetics 150, 577–589 (1998). 43. Seaman, M. N. J., Sowerby, P. J. & Robinson, M. S. Cytosolic and membrane-associated proteins involved in the recruitment of AP-1 adaptors onto the trans-Golgi network. J. Biol. Chem. 271, 25446–25451 (1996). 44. Stamnes, M. A. & Rothman, J. E. The binding of AP-1 clathrin adaptor particles to Golgi membranes requires ADP-ribosylation factor, a small GTP-binding protein. Cell 73, 999–1005 (1993). 45. Zhu, Y., Traub, L. M. & Kornfeld, S. ADP-ribosylation factor 1 transiently activates high-affinity adaptor protein complex AP-1 binding sites on Golgi membranes. Mol. Biol. Cell 9, 1323–1337 (1998). 46. Seaman, M. N. J., Ball, C. L. & Robinson, M. S. Targeting and mistargeting of plasma membrane adaptors in vitro. J. Cell Biol. 123, 1093–1105 (1993). 47. Traub, L. M. et al. AP-2 containing clathrin coats assemble on mature lysosomes. J. Cell Biol. 135, 1801–1814 (1996). 48. Haucke, V. & De Camilli, P. AP-2 recruitment to synaptotagmin stimulated by tyrosine-based endocytic motifs. Science 285, 1268–1271 (1999). 49. Yeung, B. G., Phan, H. L. & Payne, G. S. Adaptor complex-independent clathrin function in yeast. Mol. Biol. Cell 10, 3643–3659 (1999). 50. Huang, K. M., D’Hondt, K., Riezman, H. & Lemmon, S. K. Clathrin functions in the absence of heterotetrameric adaptors and AP180-related proteins in yeast. EMBO J. 18, 3897–3908 (1999). This unexpected observation that removal of all clathrin adaptors does not affect clathrin-mediated pathways challenges the view that this type of adaptor is a required link between the clathrin coat and the membrane components in a coated vesicle. 51. Gonzalez-Gaitan, M. & Jackle, H. Role of Drosophila αadaptin presynaptic vesicle recycling. Cell 88, 767–776 (1997). 52. Lee, J., Jongeward, G. D. & Sternberg, P. W. unc-101, a

53.

54.

55. 56.

57.

58.

59.

60.

61.

62.

63.

64.

65.

66.

67.

68.

69.

70.

71.

72. 73. 74.

75.

gene required for many aspects of Caenorhabditis elegans development and behavior, encodes a clathrin-associated protein. Genes Dev. 8, 60–73 (1994). Shim, J. & Lee, J. Molecular genetic analysis of apm-2 and aps-2, genes encoding the medium and small chains of the AP-2 clathrin-associated protein complex in the nematode Caenorhabditis elegans. Mol. Cell 10, 309–316 (2000). Meyer, C. et al. µ 1A-adaptin-deficient mice: lethality, loss of AP-1 binding and rerouting of mannose 6-phosphate receptors. EMBO J. 19, 2193–2203 (2000). Zizioli, D. et al. Early embryonic death of mice deficient in γ-adaptin. J. Biol. Chem. 274, 5385–5390 (1999). Wigge, P. et al. Amphiphysin heterodimers: potential role in clathrin-mediated endocytosis. Mol. Biol. Cell 8, 2003–2015 (1997). Chen, H. et al. Epsin is an EH-domain-binding protein implicated in clathrin-mediated endocytosis. Nature 394, 793–797 (1998). Wendland, B., Steece, K. E. & Emr, S. D. Yeast epsins contain an essential N-terminal ENTH domain, bind clathrin and are required for endocytosis. EMBO J. 18, 4383–4393 (1999). Haffner, C. et al. Synaptojanin 1 — localization on coated endocytic intermediates in nerve terminals and interaction of its 170 kDa isoform with Eps15. FEBS Lett. 419, 175–180 (1997). Benmerah, A. et al. The tyrosine kinase substrate Eps15 is constitutively associated with the plasma membrane adaptor AP-2. J. Cell Biol. 131, 1831–1838 (1995). Tebar, F., Sorkina, T., Sorkin, A., Ericsson, M. & Kirchhausen, T. Eps15 is a component of clathrin-coated pits and vesicles and is located at the rim of coated pits. J. Biol. Chem. 271, 28727–28730 (1996). Dell’Angelica, E. C., Klumperman, J., Stoorvogel, W. & Bonifacino, J. S. Association of the AP-3 adaptor complex with clathrin. Science 280, 431–434 (1998). Goodman, O. B. Jr, Krupnick, J. G., Gurevich, V. V., Benovic, J. L. & Keen, J. H. Arrestin/clathrin interaction. Localization of the arrestin binding locus to the clathrin terminal domain. J. Biol. Chem. 272, 15017–15022 (1997). Krupnick, J. G., Goodman, O. B. Jr, Keen, J. H. & Benovic, J. L. Arrestin/clathrin interaction. Localization of the clathrin binding domain of nonvisual arrestins to the carboxy terminus. J. Biol. Chem. 272, 15011–15016 (1997). Laporte, S. A., Oakley, R. H., Holt, J. A., Barak, L. S. & Caron, M. G. The interaction of β-arrestin with the AP-2 adaptor is required for the clustering of β2-adrenergic receptor into clathrin-coated pits. 275, 23120–23126 (2000). Laporte, S. A. et al. The β2-adrenergic receptor/β-arrestin complex recruits the clathrin adaptor AP-2 during endocytosis. Proc. Natl Acad. Sci. USA 96, 3712–3717 (1999). Ungewickell, E. et al. Role of auxilin in uncoating clathrincoated vesicles. Nature 378, 632–635 (1995). The discovery that auxilin, a clathrin coat protein containing a J domain, is required as a cofactor for the ATP-dependent activity of Hsc70 validates the role of Hsc70 as the uncoating enzyme. Schlossman, D. M., Schmid, S. L., Braell, W. A. & Rothman, J. E. An enzyme that removes clathrin coats: purification of an uncoating ATPase. J. Cell Biol. 99, 723–733 (1984). Greener, T., Zhao, X. H., Nojima, H., Eisenberg, E. & Greene, L. E. Role of cyclin G-associated kinase in uncoating clathrin-coated vesicles from non-neuronal cells. J. Biol. Chem. 275, 1365–1370 (2000). Umeda, A., Meyerholz, A. & Ungewickell, E. Identification of the universal cofactor (auxilin 2) in clathrin coat dissociation. Eur. J. Cell Biol. 79, 336–342 (2000). Sweitzer, S. M. & Hinshaw, J. E. Dynamin undergoes a GTP-dependent conformational change causing vesiculation. Cell 93, 1021–1029 (1998). First in vitro demonstration that dynamin can lead to membrane vesiculation, a required step in the scission process required to transform a clathrincoated pit into a coated vesicle. Kirchhausen, T. Cell biology — boa constrictor or rattlesnake? Nature 398, 470–471 (1999). Yang, W. & Cerione, R. A. Endocytosis: is dynamin a ‘blue collar’ or ‘white collar’? Curr. Biol. 9, R511–R514 (1999). Schmidt, A. et al. Endophilin I mediates synaptic vesicle formation by transfer of arachidonate to lysophosphatidic acid. Nature 401, 133–141 (1999). Sever, S., Muhlberg, A. B. & Schmid, S. L. Impairment of dynamin’s GAP domain stimulates receptor-mediated endocytosis. Nature 398, 481–486 (1999).

VOLUME 1 | DECEMBER 2000 | 1 9 7

REVIEWS 76. Woscholski, R. et al. Synaptojanin is the major constitutively active phosphatidylinositol-3,4,5trisphosphate 5-phosphatase in rodent brain. J. Biol. Chem. 272, 9625–9628 (1997). 77. Ringstad, N. et al. Endophilin/SH3p4 is required for the transition from early to late stages in clathrin-mediated synaptic vesicle endocytosis. Neuron 24, 143–154 (1999). 78. Bi, K., Roth, M. G. & Ktistakis, N. T. Phosphatidic acid formation by phospholipase D is required for transport from the endoplasmic reticulum to the Golgi complex. Curr. Biol. 7, 301–307 (1997). 79. Zimmerberg, J. Are the curves in the right places? Traffic 1, 366–368 (2000). 80. Barr, F. A. & Shorter, J. Do cones mark sites of fission? Curr. Biol. 10, R141–R144 (2000). 81. Ter Haar, E., Musacchio, A., Harrison, S. C. & Kirchhausen, T. Atomic structure of clathrin — a β-propeller terminal domain joins an α-zigzag linker. Cell 95, 563–573 (1998). First atomic structure of a portion of clathrin. The structure shows how the globular domain at the most distal end of the leg is folded (a β-propeller with seven blades) and leads to a prediction for the organization of the remaining linear portion of the leg (an α-zigzag made of alternating short helices extending along the main axis of the leg). 82. Owen, D. J. & Evans, P. R. A structural explanation for the recognition of tyrosine-based endocytotic signals. Science 282, 1327–1332 (1998). First atomic structure of a complex between a peptide containing a tyrosine-based endocytic sorting signal of the type YppØ and the µ2 subunit of the clathrin adaptor complex AP-2. 83. Traub, L. M., Downs, M. A., Wistrich, J. L. & Fremont, D. H. Crystal structure of the α appendage of AP-2 reveals a

198

| DECEMBER 2000 | VOLUME 1

recruitment platform for clathrin coat assembly. Proc. Natl Acad. Sci. USA 96, 8907–8912 (1999). 84. Ybe, J. A. et al. Clathrin self-assembly is mediated by a tandemly repeated superhelix. Nature 399, 371–375 (1999). 85. De Beer, T., Carter, R. E., Loel-Rice, K. E., Sorkin, A. & Overduin, M. Structure and Asn–Pro–Phe binding pocket of the Eps15 homology domain. Science 281, 1357–1360 (1998). 86. Whitehead, B., Tessari, M., Carotenuto, A., van Bergen en Henegouwen, P. M. & Vuister, G. W. The EH1 domain of Eps15 is structurally classified as a member of the S100 subclass of EF-hand containing proteins. Biochemistry 38, 11271–11277 (1999). 87. Owen, D. J. et al. A structural explanation for the binding of multiple ligands by the α-adaptin appendage domain. Cell 97, 805–815 (1999). 88. Timm, D. et al. Crystal structure of the pleckstrin homology domain from dynamin. Nature Struct. Biol. 1, 782–788 (1994). 89. Hyman, J., Chen, H., Di Fiore, P. P., De Camilli, P. & Brunger, A. T. Epsin1 undergoes nucleocytosolic shuttling and its ENTH domain, structurally similar to armadillo and HEAT repeats, interacts with the transcription factor PLZF. J. Cell Biol. 149, 537–546 (2000). 90. Owen, D. J. et al. Crystal structure of the amphiphysin-2 SH3 domain and its role in the prevention of dynamin ring formation. EMBO J. 17, 5273–5285 (1998). 91. Owen, D. J., Vallis, Y., Pearse, B. M. F., McMahon, H. T. & Evans, P. R. The structure and function of the β2-adaptin appendage domain. EMBO J. 19, 4216–4227 (2000). 92. Smith, C. J., Grigorieff, N. & Pearse, B. M. Clathrin coats at 21 Å resolution: a cellular assembly designed to recycle multiple membrane receptors. EMBO J. 17, 4943–4953 (1998).

First model obtained by electron cryomicroscopy of any vesicle coat. 93. Musacchio, A. et al. Functional organization of clathrin in coats: combining electron cryomicroscopy and X-ray crystallography. Mol. Cell 3, 761–770 (1999). Determination of the tri-dimensional model of a clathrin triskelion and the analysis of how triskelions are packaged in the coat lead to a simple proposal of how triskelions can be added (assembly) or removed (uncoating) from the coat. 94. Ohno, H. et al. Interaction of tyrosine-based sorting signals with clathrin-associated proteins. Science 269, 1872–1875 (1995). 95. Boll, W. et al. Sequence requirements for the recognition of tyrosine-based endocytic signals by clathrin AP-2 complexes. EMBO J. 15, 5789–5795 (1996). 96. Ohno, H. et al. The medium subunits of adaptor complexes recognize distinct but overlapping sets of tyrosine-based sorting signals. J. Biol. Chem. 273, 25915–25921 (1998). 97. Crowther, R. A. & Pearse, B. M. Assembly and packing of clathrin into coats. J. Cell Biol. 91, 790–797 (1981). 98. Kirchhausen, T., Harrison, S. C. & Heuser, J. Configuration of clathrin trimers: evidence from electron microscopy. J. Ultrastruct. Mol. Struct. Res. 94, 199–208 (1986). 99. Heuser, J. Three-dimensional visualization of coated vesicle formation in fibroblasts. J. Cell Biol. 84, 560–583 (1980). 100. Damer, C. K. & O’Halloran, J. Spatially regulated recruitment of clathrin to the plasma membrane during capping and cell translocation. Mol. Biol. Cell 11, 2151–2159 (2000). 101. Gaidarov, I., Santini, F., Warren, R. A. & Keen, J. H. Spatial control of coated-pit dynamics in living cells. Nature Cell Biol. 1, 1–7 (1999).

www.nature.com/reviews/molcellbio

REVIEWS

P63 AND P73: P53 MIMICS, MENACES AND MORE Annie Yang and Frank McKeon Inactivation of the tumour suppressor p53 is the most common defect in cancer cells. The discovery of its two close relatives, p63 and p73, was therefore both provocative and confounding. Were these new genes tumour suppressors, p53 regulators, or evolutionary spin-offs? Both oncogenic and tumour-suppressor properties have now been attributed to the p53 homologues, perhaps reflecting the complex, often contradictory, protein products encoded by these genes. p63 and p73 are further implicated in many p53-independent pathways, including stem-cell regeneration, neurogenesis and sensory processes.

Department of Cell Biology, Harvard Medical School, 240 Longwood Avenue, Boston, Massachusetts 02115, USA. e-mail: fmckeon@hms.harvard.edu

The p53 protein was discovered in 1979 through its physical interactions with transforming proteins of DNA tumour viruses1,2. Since that time, p53 has evolved, conceptually at least, from a potential oncogene to the principal tumour suppressor in mammals: its inactivation is a precondition to most human cancers. Further confirmation of the role of p53 in tumour suppression has come from animal models that show increased tumorigenesis in p53-null mice3,4. Now, p53 is positioned at the vertex of cellular signals warning of threats of genomic damage, chromosome missegregation and oxidative stress. In response to such signals, p53 induces cell-cycle arrest to allow repair processes or, failing that, promotes cellular senescence or cell death1,5. Despite its long tenure, however, p53 continues to surprise the biomedical community. The recent discovery of genes related to TP53 (REFS 6–10) therefore triggered great excitement that other tumour suppressors, with the previously unmatched power of p53, might exist. Indeed, the TP73 and TP63 (KET, TP51, AIS) genes — named largely to reflect their TP53 kinship — encode products with remarkable similarity to the famed tumour suppressor. The hallmark features of the p53 protein — an acidic, amino-terminal transactivation (TA) domain, a core domain for DNAbinding (DBD) and a carboxy-terminal oligomerization domain — are shared by both p63 and p73 (FIG. 1). Notably, p63 and p73 sequences in the highly conserved DNA-binding domain, wherein nearly all cancer-asso-

ciated p53 mutations are found, show over 60% identity with that of p53. In experimental systems, p63 and p73 showed many p53-like properties: they could bind to p53 DNA target sites, transactivate p53-responsive genes and induce apoptosis when exogenously expressed in cells7,9,11–16. Furthermore, TP73, at least, was localized to human chromosome 1p36, a tumour-suppressor hot spot for cancers including neuroblastoma, and breast and colorectal carcinoma6. The stage was set for p63 and p73 to emerge triumphant — the answer to p53-independent cancers, and dutiful replacements for situations when p53 goes awry. The search for mutations in tumours began, interactions with p53-binding oncoproteins and p53 itself were assessed, and mouse knockouts were made to give a definitive verdict on tumour suppression by p63 and p73. It is already apparent, however, that the TP63 and TP73 genes are immensely complex, giving rise to proteins that can functionally resemble but also counteract the actions of p53. This duality, in turn, continues to fuel arguments concerning tumour suppression and oncogenesis. The polemics of cancer biology aside, the flood of data on p63 and p73 has shed new light onto emerging areas of stem-cell biology, neurogenesis and evolution of the p53 family. Two genes in one

Despite the sequence similarity among p53 family

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 1 | DECEMBER 2000 | 1 9 9

REVIEWS TP63/TP73

∆N

TA/TA*

p63/p73 ∆Nα ∆Nβ ∆Nγ

TAα TAβ TAγ ~25%

~65%

~35%

% Identity

p53

Transactivation Steric α-motif (SAM)

DNA binding

Oligomerization

Post-SAM domain

Figure 1 | Gene structure of TP63/TP73. The TP63 and TP73 genes share a common genomic organization: two separate promoters give rise to the transactivating (TA) and ∆N classes of product. (TA* is a 39-amino-acid amino-terminal extension encoded in TP63 transcripts but not found in TP53 or TP73.) Alternative splicing at the carboxyl terminus yields further p63/p73 isotypes (for example, α, β and γ). Exons encoding the various functional domains in p63 are colour coded (p73 isotypes are derived in a similar, but not identical manner and are not shown here). Sequence identity and domain homology to the tumour suppressor p53 are indicated.

STERIC α-MOTIF

(SAM). Domain of ~70 amino acids roughly conserved in many proteins and thought to participate in protein–protein interactions.

200

members, there is a sharp contrast in the gene structure of TP53 compared with that of TP63 and TP73. The TP53 gene is simple: a single promoter directs the synthesis of a single transcript that encodes a single protein of 393 amino acids, which itself can act as a transcription factor for a set of target genes (FIG. 1). The TP63 and TP73 genes, in contrast, each contain two independent promoters and make heavy use of differential splicing, rendering a dazzling array of at least six unique proteins7–9,17. As detailed in FIG. 1 and TABLE 1, the two separate promoters in both TP63 and TP73 yield essentially two classes of protein: those containing an acidic amino terminus (TA-p63, TA-p73) analogous to the transactivation domain of p53; and those with a truncated amino terminus (∆N-p63, ∆N-p73) that lacks this region. Inherent to this unusual gene structure is the idea of ‘two genes in one’, whereby TA-p63/TA-p73 products embody the necessary pieces for p53-like function, whereas ∆N-p63/∆N-p73 products are entirely the opposite. An important issue, therefore, is whether p63 and p73 act primarily as repressors of p53dependent gene expression, or control separate legions of genes for effects quite unrelated to p53. And what are the functions of the highly conserved TA isoforms of p63/p73? Do they cooperate with p53, or act alone to control or respond to unique cellular events?

products at the carboxyl terminus by the differential splicing of exons. The result is at least three ‘tails’ (α, β, γ) of varying lengths for each ‘head’ (TA or ∆N). In all six isotypes of p63 and p73, however, the core DNAbinding and oligomerization domains remain the same, a fact that will probably have implications for the functional consequences of isotype expression. The existence of several carboxy-terminal splice variants for p63 and p73 is an intriguing feature of these two genes. The conserved tetramerization and basic domains at the end of p53 are present in all p63/p73 isotypes, but other sequence elements are found in the three different carboxyl termini, α, β and γ. The shortest of these, the γ-isoform, looks most like p53, with a slight extension containing a small polyglutamate stretch found in all p63/p73 proteins. The β- and α-isoforms are considerably longer, with the latter containing a STERIC α-MOTIF (SAM)-like sequence, thought to mediate protein–protein interactions18. The solution structure of this SAM region for p73 has been described recently19, and the domain is highly homologous in p63. Cellular assays have indicated that the α-isoform carboxyl terminus may exert an inhibitory effect on the p53-like activities of p63 and p73. Indeed, TA-p63γ and TA-p63β transactivate p53-responsive reporter genes at levels comparable to wild-type p53, whereas TA-p63α completely lacks this ability. In addition, TA-p63γ and TAp63β are potent inducers of apoptosis when exogenously expressed in mammalian cells, but TA-p63α is well tolerated in identical transfections (REFS 7,9; and our unpublished observations; TABLE 1). A lower level of transactivation by p73α compared with p73β has also been observed20; however, there are conflicting data on the relative transactivation potentials of the p73 isoforms13, possibly related to the target gene in question. Nonetheless, a pattern is emerging that clearly points to functional consequences of carboxy-terminal variability in p63 and p73 proteins. The presence of a SAM structure hints at the existence of binding proteins that modulate p63 and p73 activities. Despite the attractiveness of the SAM domain as the key regulatory element in the α-isoform, however, there is convincing evidence that a post-SAM region, namely the last ~70 amino acids (FIG. 1), is necessary for inhibiting the transactivation function of TA-p63α and TA-p73α. Deletion of this carboxy-terminal end region restores transactivation potential in both TA-p73α (REF. 21) and TA-p63α (V. Doetsch, personal communication). Doetsch and colleagues further show that this post-SAM domain binds to the amino-terminal transactivation domain, evoking an intramolecular autoinhibitory mechanism. The SAM domain, in turn, may alleviate or enhance this inhibition, in cooperation with p63-/p73-binding proteins. Biochemical or genetic screens should prove useful in identifying such molecules, and provide insights into p63 and p73 pathways in vivo. TA isotypes: mimicking p53

Swapping tails

Further diversity has been engineered into p63/p73

Numerous groups have reported p53-like activities for both p63 and p73. The p63 and p73 proteins can transwww.nature.com/reviews/molcellbio

| DECEMBER 2000 | VOLUME 1

REVIEWS

Table 1 | Physical and physiological parameters of p53, p63 and p73 isotypes

MDM2

Mouse double minute 2. Overexpressed in certain cancers and implicated in the ubiquitination and destabilization of p53.

Isotype

No. of amino acids

Predicted molecular mass (kDa)

Apparent Transactivation molecular mass (kDa) domain

Causes apoptosis

p53

393

Yes

TA*-p63α

680

—

A 19-kDa protein expressed from the INKA4A locus in response to oncogenic stimuli. It functions by stabilizing p53.

TA*-p63β

555

—

TA*-p63γ

483

—

TA-p63α

641

—

WILMS TUMOUR 1 PROTEIN

TA-p63β

516

—

Yes

TA-p63γ

444

—

Yes

∆N-p63α

586

~75

∆N-p63β

461

—

∆N-p63γ

389

—

TA-p73α

637

~85

Yes

TA-p73β

499

~70

Yes

TA-p73γ

403

—

∆N-p73α

588

—

∆N-p73β

450

—

∆N-p73γ

354

—

p19ARF

Zinc-finger transcription factor implicated in genitourinary development, and in tumorigenesis in these tissues.

Human p53, along with various human isotypes of p63 and p73, are listed according to the designations in FIG. 1, and are accompanied by amino-acid sequence length, predicted mass, apparent mass of endogenous proteins by electrophoresis, and ability to transactivate p53 reporter genes and induce apoptosis on exogenous expression in cells. (TA* refers to the conserved 39amino-acid amino-terminal extension encoded by the TA-p63 transcripts but not found in either p53 or p73.)

activate reporter genes that contain consensus p53binding sites, as well as traditional p53 target genes such as the cyclin-dependent kinase inhibitor p21 and its binding protein GADD45; the proapoptotic protein Bax; and MDM2, a negative regulator of p53 (REFS 7,9,11,13–16). Moreover, like p53, p63 and p73 are each capable of inducing apoptosis when expressed exogeActive

Target gene

Competition

Sequestration

Target gene

nously in mammalian cells7,9,11. Biochemical experiments have revealed further similarities: quite intriguingly, p73 is phosphorylated in response to ionizing radiation, DNA-damaging drugs and the microtubuledisrupting anti-cancer drug taxol through pathways linked to the activation of the c-Abl tyrosine kinase22–25. Similarly, we have observed the induction of TA-p63 expression, following ultraviolet irradiation, as well as upon differentiation, of human keratinocytes (our own unpublished data). These stimuli are akin to those that induce p53 stabilization, and are consistent with circumstances requiring the activity of a negative regulator of cell growth. MDM2, a protein that binds to the transactivation domain of p53 and promotes its ubiquitin-mediated degradation, associates with a similar site on TA-p73 (REFS 26–28). Although this interaction does not destabilize p73, the binding of MDM2 suppresses p73’s association with the transcriptional co-activator p300–CBP, and therefore its ability to transactivate target genes29. Given the central role of MDM2 in regulating the abundance and activity of p53, and in mediating the p53 response to oncogenic signals through p19 (REF. 30), its collateral actions on p73 may be hinting at some role for p73 in the oncogenic response. It has also been reported that p63 and p73 both bind the WILMS TUMOUR 1 PROTEIN, perhaps even better than p53 does (REF. 31). Finally, recent genetic studies have fuelled the idea that p73 activity is important in p53 or damageresponse pathways. In particular, a provocative result from the analysis of certain human cancers indicates that p73 inactivation may be an important step in oncoARF

p53/transactivating isoform

∆N isoform

Figure 2 | Dual mechanism of inhibition by ∆N p63/p73 isotypes. ∆N isotypes, which contain the same DNA-binding and oligomerization domains as the TA proteins, may inhibit p53 and TA-p63/73 activities by two mechanisms. In the competition model (a), highly stable ∆N molecules compete for DNA sites, preventing active p53 or TA-p63/p73 tetramers from binding and consequently transactivating the target gene. In the sequestration model (b), ∆N molecules may physically interact with and sequester p53, and even more so p63/p73, to form hetero-oligomers that are transactivation incompetent.

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 1 | DECEMBER 2000 | 2 0 1

REVIEWS

LOSS OF HETEROZYGOSITY

Deletion or disruption of one of two copies of a gene. KNUDSON’S TWO-HIT HYPOTHESIS

Loss of gene function through successive deletion or mutation of both alleles. IMPRINTING

A genetic mechanism by which genes are selectively expressed from the maternal or paternal chromosomes. DOMINANT-NEGATIVE

A defective protein that retains interaction capabilities and so distorts or competes with normal proteins.

genesis32. In squamous cell carcinomas of the head and neck with LOSS OF HETEROZYGOSITY at the TP53 locus, the remaining mutant TP53 allele showed a striking bias for an arginine codon at the otherwise polymorphic position 72. So, the Arg 72 polymorphism seems to confer an added virulence to the mutant TP53 gene. Interestingly, mutant p53 with the Arg 72 polymorphism binds to and inactivates p73, whereas mutant p53 with a proline residue at position 72 does not. By extension, these data indicate that the inactivation of p73 by certain mutant forms of p53 might provide some selective advantage, and that p73 is therefore acting in some capacity to control proliferation or cell death. In mice, TP73 has been mapped to a region of chromosome 4 lost in radiation-induced T-cell lymphomas33, again suggesting some selection for the loss of p73 in tumours provoked by ionizing radiation. It was therefore tempting to cast the homologues in the p53 mould, with roles in suppressing the initiation or progression of tumours. But even in the earliest stages of their discovery, p63 and p73 showed hints of rebellion against that stereotype. For example, the initial cloning report of p73 noted that, in neuroblastoma cell lines that show loss of heterozygosity at 1p36, the remaining allele was not mutated6. This violation of KNUDSON’S TWO-HIT HYPOTHESIS was made especially blatant, Staining against p63

Epidermis

Cervix

because many established tumour cell lines, regardless of whether they showed 1p36 deletions, expressed abundant amounts of the wild-type p73 transcript6. Subsequent mutation searches in other tumour types, such as breast, colorectal and prostate carcinoma, oligodendrogliomas and small-cell lung carcinoma, all yielded a similar absence of p73 mutations34,35. IMPRINTING at the TP73 locus might represent one means of suppressing activity of the remaining allele in these tumours6, but the tissue and tumour specificity of this phenomenon remains unclear. Recent data, however, indicate that there may be a strong link between haematopoietic cancers and the silencing of TP73 loci through hypermethylation36,37. Very recent studies showing a role for p73 in activation-induced cell death of T cells may further implicate this gene in cancers of haematopoietic origin38,39. As with the TP73 gene, very few mutations have been discovered in the TP63 gene, and those that have been are of unknown functional significance9. An important insight gained from the recent mouse model of p73 deficiency was that these animals show little propensity for increased tumorigenesis17. These mice can survive up to two years without developing any tumours, whereas 100% of TP53–/– mice die of cancer between 2 and 12 months3,4. Similar analyses could not be done on TP63–/– mice, as these die within a day of birth40,41. Celli et al.42 further identified human patients with point mutations in the DNA-binding domain of p63, at positions analogous to tumour-associated p53 mutations. These p63 mutations, however, did not lead to a propensity for tumours in affected patients, despite causing profound developmental defects (see below). So, at this point, there is little convincing evidence to indicate that these p53 homologues act as tumour suppressors in the manner of p53. Future studies will no doubt fuel the debate concerning whether p63 and p73 are mere impostors of the archetypal tumour suppressor. ∆N isotypes: a menace to p53?

Urothelium

b p73 core

Prostate TA-p73

∆N-p73

Figure 3 | ∆N isotypes are predominantly expressed in vivo. a | Immunohistochemical staining with a monoclonal antibody against p63 reveals high levels of p63 protein in human epithelial tissues, as indicated. Protein-size and RNA analysis further indicate that ∆N-p63 isotypes are the predominant proteins in these cells7. b | In situ hybridization on embryonic day 12 mouse brain sections using various p73 probes to distinguish between TA and ∆N expression. The patterns of expression with the core and ∆N probes are nearly identical, whereas TA shows signal in only a small percentage of p73-positive cells.

202

If the p63/p73 carboxy-terminal variants are deviations from the p53 ideal, then the ∆N class of proteins presents an even greater challenge. As anticipated, these truncated isotypes cannot carry out the p53-associated tasks that require an intact transactivation domain. Not only that, ∆N-p63 and ∆N-p73 proteins have proved to be powerful antagonists of p53 activity. In assays for transcriptional activation, ∆N isotypes effectively block the function of p53, as well as the function of the transactivating TA-p63 and TA-p73 isotypes7,17. Dual mechanisms may be at play in this inhibition process. In the first, as ∆N proteins retain the core DNA-binding domain, simple competition for DNA sites might prevent p53 or TA-p63/TA-p73 from binding target gene promoters. In the second, ∆N-p63 and ∆N-p73 proteins may bind to p53, as well as to their own and each other’s TA isoforms. This can occur through the conserved oligomerization domain1 or, as recent studies indicate, through interactions independent of this region (D. Roop and D. Kaplan, personal communication). In either case, the sequestration of p53 or TA-p63/TA-p73 molecules, through the formation of sub- or inactive www.nature.com/reviews/molcellbio

| DECEMBER 2000 | VOLUME 1

REVIEWS a

Wild type

Knockout

Wild type

Knockout

Wild type

Knockout

Differentiation

p63? p63

TAC

Stem cell

Figure 4 | Epithelial defects in p63-deficient mice: hints for a role in stem-cell regeneration? a | Late stage TP63-deficient embryos show an unusual sloughing (arrows) of the epidermis, as compared with the normal, intact skin of wild-type littermates. b | Skin sections of late-stage TP63 knockout and wild-type embryos were stained with antibodies to keratin 5, a marker of basal, proliferative cells of the epidermis. Wild-type sections show strong staining, whereas the tattered epidermis lacking TP63 shows no keratin 5 expression, indicating a lack of basal cells. c | Residual epithelial cells along the denuded surface of TP63 knockout embryos show positive staining for loricrin, a marker of differentiated, mature cells in cornified layers of a stratified epithelium. Both wild-type and knockout cells stain for loricrin, indicating that TP63 knockout epidermis undergoes differentiation processes. d | Model for p63 function in maintaining regenerative capacity of epithelial stem cells. Left, representation of epidermal differentiation, showing a percentage of stem cells in the basal layer that undergo asymmetric division to yield another, identical stem cell, and a transient amplifying cell (TAC) that is committed to differentiation. TACs are also capable of active proliferation (amplification), but eventually reach terminal differentiation. Right, role of p63 in preserving the regenerative and proliferative capacity of stem cells and TACs, respectively.

hetero-oligomers, may underlie the DOMINANT-NEGATIVE action of the ∆N isotopes. Evidence that ∆N molecules are key players in vivo came from the initial characterization of TP63 expression7. Antibodies to p63 showed intense staining of the basal, or progenitor, cell layers of a wide range of epithelial tissues, including the epidermis, cervix, urogenital tract, prostate, myoepithelium of the breast and other glandular tissues7 (FIG. 3). Together with analyses of protein size and RNA expression, these findings showed that ∆N products account for virtually all p63 protein expression in the epithelial tissues examined. Likewise, it seems that p73 expression is also biased towards the amino-terminally truncated products. In both the developing and adult brain, evidence is strong that ∆Np73 isotypes are the main expression products, yielding protein levels 20-fold greater that those of the TA-p73 isotypes17,43. So, the principal forms of expressed p63 and p73 have activities contrary to those we attribute to p53 as a transcriptional activator. The two decades of work on p53 have therefore not prepared us for the situation presented by the TP63 and TP73 genes. What, for instance, would a high concentration of either ∆N-p63 or ∆N-p73 mean for p53 activity in the same cell, and are these isotypes repressing a wider set of genes than those regulated by p53? A study published before the discovery of p63 and p73 provided some hints, by examining p53 activity in immature and maturing keratinocytes44. p53-dependent transactivation was low in immature keratinocytes and high in differentiating keratinocytes, even though p53 protein and transcript levels showed the opposite distribution. This paradox might now be explained by the possibility that the abundant ∆N-p63 observed in the immature keratinocytes inactivates p53, and this inhibition is alleviated by the loss of ∆N-p63 during differentiation7,45. Exciting evidence that ∆N proteins modulate p53 activity in vivo has come from two recent studies. First, data from TP73 knockout mice indicate that ∆N-p73 may suppress p53 activity in certain cell populations such as sympathetic neurons43. Sympathetic neurons undergo p53-dependent apoptosis on nerve growth factor withdrawal in culture, a process that is markedly suppressed by the expression of exogenous ∆N-p73 in these cells. These neurons, which normally express ∆Np73 proteins, would therefore be expected to have enhanced p53 activity in the TP73–/– mouse and consequently higher rates of p53-dependent apoptosis. This is indeed the case: TP73 knockout mice have significantly increased sympathetic neuronal cell death43. In another study, immature keratinocytes exposed to ultraviolet radiation showed a rapid loss of ∆N-p63 coincident with the stabilization of p53 (REF. 46). Whether this loss of p63 is provoked by stabilized p53 or a separate but coordinate response to DNA damage remains an unanswered but intriguing question. It is perhaps ironic, then, that the p53 homologues may turn out to be the tumour suppressor’s worst nightmare. Dysregulation of naturally occurring p63/p73 isotypes may provide a proliferative advantage

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 1 | DECEMBER 2000 | 2 0 3

REVIEWS

Table 2 | Comparison of TP53/TP73/TP63 knockout phenotypes Defect TP53

Potential cause

Refs

Loss of genome checkpoints Unclear; strain- and sex-dependent

3,4

–/–

Early onset of thymic lymphoma, fibrosarcoma, other tumours Exencephaly, other developmental defects (small percentage) TP73 –/– Somatic growth retardation Sinus, middle ear inflammation/infections Hydrocephalus Gastrointestinal haemorrhages Defects in reproductive and social behaviour Hippocampal dysgenesis

Generalized stress 17 Hyper-reactive epithelial secretion Hypersecretion of cerebrospinal fluid ? (Pan mucositis) Vomeronasal organ malfunction; loss of pheromone receptors Loss of reelin-secreting Cajal–Retzius neurons

TP63 –/– Neonatal death Absence of limbs Absence of epidermis and other squamous epithelia Absence of urothelium Absence of secretory epithelia (prostate, breast, etc)

by countering p53 or p53-like activities in the cell. The spectre of oncogenic potential is indeed raised by evidence that 3q27-ter locus, where the TP63 gene is located, is amplified in advanced cervical carcinoma and other squamous cell carcinomas7,10,47. Many groups are now observing alterations in p63 and p73 levels in tumours compared with normal tissue34. It will be important to understand the clinical and physiological significance of these variations, particularly in the context of isotype specificity. To that end, mouse knockouts are already providing several leads. p63-deficient mice: a stem cell phenotype?

MESENCHYME

Undifferentiated connective tissue present in the early embryo. PYKNOTIC NUCLEI

Hypercondensed chromatin typical of cells undergoing apoptosis.

204

Data from TP63 knockout mice point to a pivotal function for p63 in epithelial, craniofacial and limb development. Apparently a key regulator of proliferation and differentiation programmes in these structures, p63 may act to maintain the regenerative or ‘immortal’ quality of epithelial stem cells. It is a mystery how p63 carries out this remarkable task, but understanding its mechanisms of action may offer valuable insights into stemcell biology, as well as links to cellular immortalization events that are a prerequisite to oncogenesis. The TP63 knockout mouse was generated by two independent groups, and both yielded a similar, marked phenotype, albeit with different interpretations40,41. Mice deficient in p63 show striking epithelial defects, including a complete absence of skin, hair, mammary tissue, prostate, lachrymal glands and salivary glands, as well as pronounced alterations in epithelia of the urogenital tract, tongue and stomach40,41. Severe limb truncations and craniofacial malformations were also seen in TP63–/– mice. Seemingly complex at first, the phenotype of p63 deficiency turned out to have a common theme. Many of the affected sites are tissues that, in normal animals, show epithelial stratification, with immature, progenitor cells along the basal layer and more differentiated cells above. The epidermis, or skin, for example, is the classic example of a squamous stratified

Dehydration 40,41 Failure to maintain apical ectodermal ridge Failure to maintain regenerative cell population Failure to maintain regenerative cell population Failure to maintain regenerative cell population

epithelium. Perhaps less obviously, developing limb buds in the embryo are also covered by an epithelium that later stratifies to form the apical ectodermal ridge, a multi-layered structure essential for growth and patterning of the underlying MESENCHYME48. The affected tissues in the TP63 knockout mouse are generally characterized by absence of cells: the epidermis, tongue and limb-bud epithelium all show denuded surfaces. Mills et al.41, who analysed newborn TP63–/– mice, interpreted these phenomena as aborted epithelial differentiation; but we arrived at different conclusions40. Looking at younger mice, during the late stages of embryogenesis (embryonic days 16–18), a notable sloughing of cells could be seen on the skin of TP63–/– embryos40 (FIG. 4). Remnant cells on the skin showed staining for epithelial differentiation markers (FIG. 4), and PYKNOTIC NUCLEI, which are indicative of normal cell death in differentiated cells of the epidermis. These data indicated that TP63–/– epithelial cells might progress through the end stages of differentiation, and that p63 is not essential for this process of epithelial commitment. The function of p63, then, seems to be in maintaining the basal, progenitor cell populations that underlie the capacity of such tissues to develop and regenerate49,50 (FIG. 4). Importantly, high levels of p63 protein are seen in these basal cells7 (FIG. 3), indicating that p63 might be actively required in progenitor populations. The chief distinguishing feature of stem cells at any stage of development is that they undergo ‘asymmetric’ divisions, such that one daughter cell proceeds onto differentiation pathways while the other retains stem-cell identity and regenerative potential51 (FIG. 4). Indeed, the TP63 knockout phenotype is consistent with a failure of such asymmetric cell division to occur in epithelial progenitors. Once triggered, say, by stratification cues, all TP63–/– cells undergo differentiation, thereby experiencing a massive depletion of stem cells and regenerative capacity. Celli et al.42 showed that heterozygous TP63 mutawww.nature.com/reviews/molcellbio

| DECEMBER 2000 | VOLUME 1

REVIEWS

Box 1 | p53: chicken or egg? An interesting debate is simmering on the evolutionary a p63 relationships among p53 family members, the outcome of which will probably affect our thinking on their p73 mechanisms of action, regulation and functional Dmp53 Mammalian p53 interactions. The issue is: which member of the vertebrate OR p53 family came first? The mouse knockout phenotypes might argue that TP63 or TP73 was the original gene Invertebrates Vertebrates b ?? — the developmental and physiological consequences of p73 their disruption are severe and affect basic physiological processes, whereas TP53–/– mice, apart from their p53 susceptibility to tumours, appear normal. Alternatively, (loss of ∆N p63/p73-like Dmp53 protection against genomic damage, rather than tumour promoter) suppression per se, may be a more ancient concern of an organism, and the antecedent gene might behave more like p63 TP53 as we know it in vertebrates. So far, comparative genomics has not provided any conclusive answers: the Drosophila melanogaster genome encodes a single p53-like protein (Dmp53) for which no genetic data are yet available. However, Dmp53 can transactivate genes and is pro-apoptotic. Moreover, dominant-negative experiments and sequence homology have been used to argue that Dmp53 and human p53 lie on a direct evolutionary line55–57 (a). This would argue that p63 and p73 arose by gene duplication from the original ‘tumour suppressor’ p53 and later assumed their present functions in epithelial stem-cell maintenance, secretion and neurogenesis. We cannot be more definitive here because p63 and p73 can transactivate p53 target genes, and several p63 and p73 isotypes induce apoptosis in mammalian cells6,7,11. Strictly at the level of amino-acid sequence, Dmp53 is more closely related to the p63 and p73 homologues in molluscs58 and human p63 than it is to human p53. Obviously we are dealing with incomplete information here: other than Dmp53, the only invertebrate p53 homologues characterized so far are from squid, whose p53 homologue is most related to p73, and clam, which has one most similar to human p63. Bona fide p53 orthologues are not apparent until the vertebrate branch of the evolutionary tree,whereas zebrafish, Xenopus and birds all have p63/p73 orthologues. Human p63 is much more similar to p73 than either is to p53. At first glance, then, it can be argued that p63 was the original gene that evolved in invertebrates, and p73 arose by a gene-duplication event before vertebrate evolution. As p53 is closer in sequence to p73 than to p63, we could surmise a further gene-duplication event during chordate evolution, which gave rise to the TP53 gene (b). Curiously, the p63 protein shows almost no deviation at the aminoacid level between mouse and humans (99% identity), p73 shows more (96% identity), whereas p53 is surprisingly divergent (86% identity). Whether the conservation of p63 and p73, in contrast to the TP53 gene, reflects constraints imposed by co-evolving interactions, dispensability or ‘evolvability’ of p53 function is unclear at present.

HYDROCEPHALUS

Condition marked by expansion of cerebral ventricles and compression of neural structures owing to block in flow or overproduction of cerebral spinal fluid. CHOROID PLEXUS

Capillary bed containing the cerebral ventricles responsible for producing cerebral spinal fluid. VOMERONASAL ORGAN

Cluster of sensory neurons in the nasal arch that detects pheromones and transmits this information to higher cortical centres.

tions were an underlying cause of the human disorder ectodermal dysplasia, ectrodactaly and cleft palate (EEC) syndrome. In a manner markedly similar to the TP63 knockout mouse, EEC patients have skin defects and severe limb and craniofacial abnormalities. Taken together, these findings establish a critical function for p63 in controlling the delicate balance of proliferation and differentiation in epithelial cells. Given that epithelial tissues are the target of over 80% of all human malignancies, the ability of p63 to maintain the immortalized quality of stem cells may further hint at its roles in tumorigenesis. p73-deficient mice

Unlike those lacking TP63, TP73–/– mice survive postnatally and some live well into adulthood, despite having multiple defects17. Remarkably, given the similarity of the genes, the TP73 knockout phenotype shows no obvious overlap with that of TP63-deficient mice. Instead, TP73–/– mice have malfunctions in fluid dynamics in the central nervous system and respiratory airways, defective neurogenesis, and abnormal reproductive and social behaviour (TABLE 2). The loss of secretory control in TP73–/– mice is provocative because it

suggests a role for p73 in responding to basic physiological stimuli that is arguably more critical to life than tumour suppression. Specifically, TP73–/– mice develop HYDROCEPHALUS, most probably owing to hypersecretion of cerebral spinal fluid by the CHOROID PLEXUS, and massive inflammation of nasal mucus membranes, primarily owing to hypersecretion by respiratory epithelia. So, two fundamental and highly regulated secretory processes, one controlling intraventricular pressure and the other innate immune responses to microorganisms, seem constitutive in the absence of p73. In addition, the intestinal haemorrhages frequently observed in TP73–/– mice may be the consequence of similar secretory defects in the gastrointestinal tract. Perhaps one of the more unusual aspects of the TP73–/– phenotype lies in their abnormal reproductive and social behaviour, which, for rodents in general, depends on pheromone signalling52. TP73–/– mice are uncharacteristically uninterested in mating and most other social interactions governed by pheromone signals. The origin of this malaise seems to be a frank absence of pheromone receptor expression on the neurosensory cells of the VOMERONASAL ORGAN, which express very high levels of TP73 in wild-type mice17. So p73 has

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 1 | DECEMBER 2000 | 2 0 5

REVIEWS

MARGINAL ZONE

Superficial region of the cerebral cortex generally devoid of cells. PIONEER NEURONS

One of several sets of early migrating neurons, which act as developmental and positional cues for local neurogenesis. DENTATE GYRUS

Subdomain of hippocampal formation comprising granule cell neurons.

a central, albeit poorly defined, function in primitive sensory and signal transduction pathways. Finally, TP73–/– mice show a severe distortion of the hippocampal formation, a region of the brain considered to be the centre of learning and memory in higher mammals17. The defect here could be attributed to the loss of TP73-expressing Cajal–Retzius cells, which are bipolar neurons found in the MARGINAL ZONE of the cortex and molecular layers of the hippocampus. Cajal–Retzius neurons have attracted considerable interest because they are one of several populations of PIONEER NEURONS that express reelin, a secreted glycoprotein implicated in development of the cerebral cortex53. Although TP73–/– mice lack Cajal–Retzius neurons along the marginal zone of the cortex, they retain other reelin-secreting neurons such as those of the cortical layer, which do not normally express TP73. Interestingly, cortical layering is normal in the TP73–/– mice, indicating that Cajal–Retzius neurons of the marginal zone may function exclusively in regulating hippocampal morphogenesis. Development of the hippocampus, particularly of the DENTATE GYRUS, is especially important, as recent studies have shown that this process persists throughout adulthood54. Stem cells of the granule neurons produce new progeny that require positional cues. So p73 may mediate the survival or proper migration of neurons essential for the constant remodelling of the hippocampus throughout life.

The few short years since p63 and p73 were brought into the p53 family have yielded impressive progress, but also important questions that need to be answered. What are the unique functions of the TA and ∆N isotypes for these genes? So far, the TP63 and TP73 knockout models in mice reveal only the consequences of losing both these apparently contradictory isotope classes. An important goal, then, is to generate isotype-specific and conditional knockouts that affect one isotype while preserving the other. More difficult questions surround the diverse genetic programmes that depend on p63 or p73 to control epithelial stem-cell identity, pioneer-neuron survival, and various secretory systems. The evolutionary relationships in the p53 family are also uncertain, leaving open the possibility that TP63 and TP73 represent antecedent genes to TP53 (BOX 1). Could p73’s ability to respond to extracellular stimuli be a prelude to p53’s capacity to sense cellular stresses and damage? And could p63’s role in proliferation and differentiation have preceded p53’s dominance in cell-cycle control? Perhaps the reverse is true, and p53 was the ancestor that bestowed on p63 and p73 their pivotal roles in stem-cell regeneration, neurogenesis and sensory pathways. Whichever the answer, it is increasingly clear that the homologues can hardly be considered poor relations of the p53 family. Intriguing molecules in their own right, they have potential to mimic, menace and eclipse one of the most intriguing genes of our time.

Back to p53

In spite of the striking and unique phenotypes presented by the TP63 and TP73 knockout mice, the findings may leave some wanting for more in the way of links to p53. Certainly the discovery of p53 homologues complicates much of what was presumed about p53. Any p53 target gene must now contend with three potential regulators, and even p53 itself is vulnerable to the actions of the highly complex p63 and p73 proteins.

1. 2.

4. 5. 6.

206

Levine, A. p53, the cellular gatekeeper for growth and division. Cell 88, 323–331 (1997). Steele, R. J., Thompson, A. M., Hall, P. A. & Lane, D. P. The p53 tumour suppressor gene. Br. J. Surg. 85, 1460–1467 (1998). Donehower, L. A. et al. Mice deficient for p53 are developmentally normal but susceptible to spontaneous tumours. Nature 356, 215–221 (1992). Jacks, T. et al. Tumour spectrum analysis in p53-mutant mice. Curr. Biol. 4, 1–7 (1994). Lowe, S. W. Activation of p53 by oncogenes. Endocr. Relat. Cancer 6, 45–48 (1999). Kaghad, M. et al. Monoallelically expressed gene related to p53 at 1p36, a region frequently deleted in neuroblastoma and other human cancers. Cell 90, 809–819 (1997). Reports the discovery of the first p53 homologue, p73, and shows that p73 can transactivate p53 reporter genes and inhibit colony formation on transfection. p73 is shown to be lost in many neuroblastoma and other tumour cell lines, whereas the remaining allele is shown to be wild type and frequently overexpressed. Yang, A. et al. p63, a p53 homologue at 3q27-29, encodes multiple products with transactivating, deathinducing, and dominant-negative activities. Mol. Cell 2, 305–316 (1998). Reports the discovery of second p53 homologue, p63, and defines the six principal transcripts derived from this gene. The dominant-negative action of the

10.

11.

12.

13.

14.

Links DATABASE LINKS

p53 | p73 | p63 | Human chromosome 1p36 | neuroblastoma | SAM domain | p21 | GADD45 | Bax | MDM2 | cAbl | ubiquitin | p300 | EEC syndrome | Dmp53 ENCYCLOPEDIA OF LIFE SCIENCES

Cerebral cortex development

∆N isoforms is also described. Augustin, M., Bamberger, C., Paul, D. & Schmale, H. Cloning and chromosomal mapping of the human p53related KET gene to chromosome 3q27 and its murine homologue Ket to mouse chromosome 16. Mamm. Genome 9, 899–902 (1998). Osada, M. et al. Cloning and functional analysis of human p51, which structurally and functionally resembles p53. Nature Med. 4, 839–843 (1998). Details the cloning of one of the isotypes of p63, and describes the general absence of mutations in this gene in a large number of tumour-derived cell lines. Hibi, K. et al. AIS is an oncogene amplified in squamous cell carcinoma. Proc. Natl Acad. Sci. USA 97, 5462–5467 (2000). Jost, C. A., Marin, M. C. & Kaelin, W. G. Jr p73 is a simian [correction of human] p53-related protein that can induce apoptosis. Nature 389, 191–194 (1997). Describes the ability of p73 to induce apoptosis when expressed in established cell lines. Zhu, J., Jiang, J., Zhou, W. & Chen, X. The potential tumor suppressor p73 differentially regulates cellular p53 target genes. Cancer Res. 58, 5061–5065 (1998). Di Como, C. J., Gaiddon, C. & Prives, C. p73 function is inhibited by tumour-derived p53 mutants in mammalian cells. Mol. Cell. Biol. 19, 1438–1449 (1999). Shimada, A. et al. The transcriptional activities of p53 and its homologue p51/p63: similarities and differences. Cancer Res. 59, 2781–2786 (1999).

15. Ishida, S., Yamashita, T., Nakaya, U. & Tokino, T. Adenovirus-mediated transfer of p53-related genes induces apoptosis of human cancer cells. Jpn. J. Cancer Res. 91, 174–180 (2000). 16. Levrero, M. et al. The p53/p63/p73 family of transcription factors: overlapping and distinct functions. J. Cell Sci. 113, 1661–1670 (2000). 17. Yang, A. et al. p73-deficient mice have neurological, pheromonal and inflammatory defects but lack spontaneous tumours. Nature 404, 99–103 (2000). The phenotype of the p73 knockout mouse is described. The TP73 gene is also shown to resemble the overall structure of the TP63 gene, and to produce both ∆N and TA isotypes. 18. Ponting, C. P. SAM: A novel motif in yeast sterile and Drosophila polyhomeotic proteins. Protein Sci. 4, 1924–1930 (1995). 19. Chi, S. W., Ayed, A. & Arrowsmith, C. H. Solution structure of a conserved carboxy–terminal domain of p73 with structural homology to the SAM domain. EMBO J. 18, 4438–4445 (1999). 20. De Laurenzi, V. et al. Two new p73 splice variants, γ and δ, with different transcriptional activity. J. Exp. Med. 188, 1763–1768 (1998). 21. Ozaki, T. et al. Deletion of the COOH-terminal region of p73α enhances both its transactivation function and DNAbinding activity but inhibits induction of apoptosis in mammalian cells. Cancer Res. 59, 5902–5907 (1999). 22. Gong, J. G. et al. The tyrosine kinase c-Abl regulates p73

www.nature.com/reviews/molcellbio

| DECEMBER 2000 | VOLUME 1

REVIEWS

23.

24.

25. 26.

27.

28.

29.

30. 31.

32.

33.

34.

35.

36.

37.

38.

in apoptotic response to cisplatin-induced DNA damage. Nature 399, 806–809 (1999). Agami, R., Blandino, G., Oren, M. & Shaul, Y. Interaction of c-Abl and p73α and their collaboration to induce apoptosis. Nature 399, 809–813 (1999). Yuan, Z. M. et al. p73 is regulated by tyrosine kinase c-Abl in the apoptotic response to DNA damage. Nature 399, 814–817 (1999). White, E. & Prives, C. DNA damage enables p73. Nature 399, 734–735 (1999). Ongkeko, W. M. et al. MDM2 and MDMX bind and stabilize the p53-related protein p73. Curr. Biol. 9, 829–832 (1999). Balint, E., Bates, S. & Vousden, K. H. Mdm2 binds p73α without targeting degradation. Oncogene 18, 3923–3929 (1999). Zeng, X. et al. MDM2 suppresses p73 function without promoting p73 degradation. Mol. Cell. Biol. 19, 3257–3266 (1999). Zeng, X. et al. The amino-terminal domain of p73 interacts with the CH1 domain of p300/CREB binding protein and mediates transcriptional activation and apoptosis. Mol. Cell. Biol. 20, 1299–1310 (2000). Sherr, C. J. & Weber, J. D. The ARF/p53 pathway. Curr. Opin. Genet. Dev. 10, 94–99 (2000). Scharnhorst, V., Dekker, P., van der Eb, A. J. & Jochemsen, A. G. Physical interaction between Wilms tumour 1 and p73 proteins modulates their functions. J. Biol. Chem. 275, 10202–10211 (2000). Marin, M. C. et al. A common polymorphism acts as an intragenic modifier of mutant p53 behaviour. Nature Genet. 25, 47–54 (2000). An intriguing paper that proposes a mechanism of p73 inactivation by mutations in p53. Herranz, M., Santos, J., Salido, E., Fernandez-Piqueras, J. & Serrano, M. Mouse p73 gene maps to the distal part of chromosome 4 and might be involved in the progression of γ-radiation-induced T-cell lymphomas. Cancer Res. 59, 2068–2071 (1999). Ikawa, S., Nakagawara, A. & Ikawa, Y. p53 family genes: structural comparison, expression and mutation. Cell Death Differ. 6, 1154–1161 (1999). Han, S. et al. Infrequent somatic mutations of the p73 gene in various human cancers. Eur. J. Surg. Oncol. 25, 194–198 (1999). Corn, P. G. Transcriptional silencing of the p73 gene in acute lymphoblastic leukemia and Burkitt’s lymphoma is associated with 5′ CpG island methylation. Cancer Res. 59, 3352–3356 (1999). Kawano, S. et al. Loss of p73 gene expression in leukemias/lymphomas due to hypermethylation. Blood 94, 1113–1120 (1999). Lissy, N. A., Davis, P. K., Irwin, M., Kaelin, W. G. & Dowdy, S. F. A common E2F-1 and p73 pathway mediates cell death induced by TCR activation Nature 407, 642–645 (2000).

39.

40.

41.

42.

43.

44.

45.

An interesting report that presents a role for p73 in the p53-independent process of activation-induced T-cell death. Irwin, M. et al. Role for the p53 homologue p73 in E2F-1induced apoptosis. Nature 407, 645–648 (2000). The mechanistic details of the involvement of p73 in E2F-1 induced cell death, with particular relevance to REF. 38, are reported. Yang, A. et al. p63 is essential for regenerative proliferation in limb, craniofacial and epithelial development. Nature 398, 714–718 (1999). Presents the phenotype of the TP63 knockout mouse. The authors conclude that these defects are not due to defects in commitment or differentiation of these tissues and structures, but rather a failure to maintain the regenerative stem cells. Mills, A. A. et al. p63 is a p53 homologue required for limb and epidermal morphogenesis. Nature 398, 708–713 (1999). Details the phenotype of the TP63 knockout mouse, and concludes that the underlying defect is a generalized failure in epithelial differentiation. Celli, J. et al. Heterozygous germline mutations in the p53 homologue p63 are the cause of EEC syndrome. Cell 99, 143–153 (1999). The positional cloning of mutations responsible for EEC syndrome, a dominant disease affecting limb and craniofacial development and associated with ectodermal dysplasia. These mutations are in the TP63 gene, at residues corresponding to those mutated in TP53 in tumours. This report also describes how such mutant proteins might act in a dominant manner to inhibit wild-type p63 function. Pozniak, C. D. et al. An anti-apoptotic role for the p53 family member, p73, during developmental neuron death. Science 289, 304–306 (2000). Analysis of sympathetic ganglia of TP73 knockout mice reveals a 50% reduction of neurons, which die in the absence of nerve growth factor in a p53dependent manner. ∆N-p73, the main p73 product in sympathetic neurons of wild-type animals, is shown to have the capacity to prevent cell death after nerve growth factor withdrawal, and therefore to function potentially in a dominant-negative manner towards p53. Weinberg, W. C., Azzoli, C. G., Chapman, K., Levine, A. J. & Yuspa, S. H. p53-mediated transcriptional activity increases in differentiating epidermal keratinocytes in association with decreased p53 protein. Oncogene 10, 2271–2279 (1995). This paper details unusual activities of p53 in keratinocyte development that now may be relevant to p63 function. Parsa, R., Yang, A., McKeon, F. & Green, H. Association of p63 with proliferative potential in normal and neoplastic human keratinocytes. J. Invest. Dermatol. 113,

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

1099–1105 (1999). 46. Liefer, K. M. et al. Downregulation of p63 is required for epidermal UV-B-induced apoptosis. Cancer Res. 60, 4016–4020 (2000). An interesting paper showing that the high levels of ∆N-p63 in immature keratinocytes must be eradicated for p53 to function. 47. Heselmeyer, K. et al. Gain of chromosome 3q defines the transition from severe dysplasia to invasive carcinoma of the uterine cervix. Proc. Natl Acad. Sci. USA 93, 479–484 (1996). 48. Johnson, R. L. & Tabin, C. J. Molecular models for vertebrate limb development. Cell 90, 979–990 (1997). 49. Barrandon, Y. & Green, H. Three clonal types of keratinocyte with different capacities for multiplication. Proc. Natl Acad. Sci. USA 84, 2302–2306 (1987). 50. Watt, F. M. Epidermal stem cells: markers, patterning and the control of stem cell fate. Phil. Trans. R. Soc. Lond. B 353, 831–837 (1998). 51. Lu. B., Jan, L. Y. & Jan, Y. N. Asymmetric cell division: lessons from flies and worms. Curr. Opin. Genet. Dev. 8, 392–399 (1998). 52. Buck, L. B. The molecular architecture of odor and pheromone sensing in mammals. Cell 100, 611–618 (2000). 53. Frotscher, M. Cajal–Retzius cells, Reelin, and the formation of layers. Curr. Opin. Neurobiol. 8, 570–575 (1998). 54. Eriksson, P. S. et al. Neurogenesis in the adult human hippocampus. Nature Med. 4, 1313–1317 (1998). 55. Ollmann, M. et al. Drosophila p53 is a structural and functional homologue of the tumour suppressor p53. Cell 101, 91–101 (2000). References 55–57 describe the discovery of a single p53 homologue in flies with functional ties to the DNA damage response and apoptosis. They offer the promise of a genetic analysis of the prototypic p53 molecule. 56. Brodsky, M. H. et al. Drosophila p53 binds a damage response element at the reaper locus. Cell 101, 103–113 (2000). 57. Jin, S. et al. Identification and characterization of a p53 homologue in Drosophila melanogaster. Proc. Natl Acad. Sci. USA 97, 7301–7306 (2000). 58. Ishioka, C. et al. Mutational analysis of the carboxyterminal portion of p53 using both yeast and mammalian cell assays in vivo. Oncogene 10, 1485–1492 (1995).

Acknowledgements We are grateful to V. Doetsch, D. Caput, C. Crum, H. van Bokhoven, P. Duijf, A. Sharpe, D. Roop, D. Kaplan and H. Green for enjoyable collaborations and continued support. F.M. is supported by grants from the National Institutes of Health and the American Cancer Society.

VOLUME 1 | DECEMBER 2000 | 2 0 7

REVIEWS

ARE DESMOSOMES MORE THAN TETHERS FOR INTERMEDIATE FILAMENTS? Kathleen J. Green and Claire A. Gaudry Desmosomes are intercellular adhesive junctions that anchor intermediate filaments at membrane-associated plaques in adjoining cells, thereby forming a three-dimensional supracellular scaffolding that provides tissues with mechanical strength. But desmosomes have also recently been recognized as sensors that respond to environmental and cellular cues by modulating their assembly state and, possibly, their signalling functions.

METAZOANS

Refers to the kingdom Animalia (animals) that comprises roughly 35 phyla of multicellular organisms. MICROFILAMENT

Cytoskeletal filament typically 6 nm in diameter, consisting of polymerized actin. Microfilaments form the main component of the cellular contractile machinery. INTERMEDIATE FILAMENT

Cytoskeletal filament, typically 10 nm in diameter, occurring in higher eukaryotic cells.

Departments of Pathology and Dermatology and the Robert H. Lurie Cancer Center, Northwestern University Medical School, Chicago, Illinois 60611, USA. Correspondence to K.J.G. e-mail: kgreen@nwu.edu

208

Of the common vertebrate junctions, desmosomes are perhaps the most specialized, arising later in METAZOAN evolution than their relatives, adherens junctions (FIG. 1). Both of these ‘anchoring junctions’ have, at their core, transmembrane glycoproteins that belong to the cadherin class of calcium-dependent adhesion molecules1. Cadherins are linked indirectly to the MICROFILAMENT or INTERMEDIATE FILAMENT cytoskeleton through a series of adaptor molecules that do more than tether the cadherin tails to the cytoskeleton — they also regulate junction assembly state and adhesive strength2,3. The other main types of vertebrate intercellular junctions can be broadly categorized as ‘communicating’ junctions4 and ‘occluding’ junctions5. The latter, also known as tight junctions, seal adjacent membranes through multiply spanning transmembrane proteins, such as occludins and claudins, to corral APICAL and BASOLATERAL MEMBRANE proteins into their correct pens. Tight junctions also provide a gating mechanism to regulate paracellular diffusion of solutes. Communicating (or gap) junctions are assembled from hexameric protein arrays known as connexons. Each connexon has a pore at its centre that provides a window to the neighbouring cell, allowing passage of small ions and molecules. Gap junctions therefore facilitate electrical coupling of cells within a sheet, as opposed to the occluding and anchoring junctions, which couple cells mechanically. Attention has recently focused on cell junctions as nodes at the intersection of mechanical and chemical

signalling pathways6. Intercellular junctions process instructions that dictate assembly state and function. They also regulate the availability of signalling molecules and may themselves propagate intracellular signals that control cell motility, growth and differentiation. Adherens junctions and their resident adaptor protein β-catenin have received the lion’s share of attention. But desmosomes are emerging now as even more complex structures with the potential to fulfil numerous tissuespecific structural and signalling functions. Here we attempt to unravel some of the structural complexity of desmosomes and raise issues about their functional diversity. We pull together disparate views of the mechanism of assembly of these highly organized structures, and tackle the puzzle of the desmosomal adhesive interface. Finally, the potential of the desmosome as both a receiver and transducer of signals is explored. Structure and function of desmosomes

Desmosomes are abundant in tissues that experience mechanical stress, and have been proposed to have a primarily structural function. This idea is supported by the existence of diseases in which tissue integrity is breached by gene defects or by autoimmune antibodies targeting desmosome components7–15. The composition of desmosomes changes both in different cell types, and as cells travel to the surface of a STRATIFIED EPITHELIAL 3,16 SHEET . Desmosomes are common in epithelial tissues where they anchor keratin intermediate filaments to the www.nature.com/reviews/molcellbio

| DECEMBER 2000 | VOLUME 1

REVIEWS

Table 1 | Classes of intermediate filaments* Homology class

Name

Tissue distribution

I‡

Acidic cytokeratins§

All epithelia

‡

Basic cytokeratins§

All epithelia

III

Vimentin§, desmin§, GFAP (glial fibrillary acidic protein), peripherin

Mesenchyme, muscle, diverse neuronal cells

Neurofilament triplet proteins, α-internexin

Neurons

Nuclear lamins

All cell types

*Adapted from Herrmann and Aebi109. ‡Type I and II keratins form obligate heterodimers. § Intermediate filaments associated with desmosomes.

membrane, but they are also found in myocardial and Purkinje fibre cells of the heart, where they anchor desmin intermediate filaments, as well as in meningeal cells and the follicular dendritic cells of lymph nodes, where they associate with vimentin intermediate filaments (TABLE 1). So desmosomes seem to be tailor-made for cell type-specific functions. The true extent of the structural and possible functional complexity of desmosomes is only now being appreciated. The main building blocks of desmosomes come from three gene superfamilies (FIG. 2). These include the desmosomal cadherins, which are further divided into desmogleins (DSG1–3)17–19 and desmocollins (DSC1–3)20–24; the armadillo family of nuclear and junctional proteins, including plakoglobin25 and plakophilins 1–3 (REFS 26–30); and finally, the plakins, which include desmoplakin31, plectin32,33 and the cell envelope proteins envoplakin and periplakin34,35. Components of desmosomes that fall outside these families may have other structural and/or regulatory functions36,37. The combinations of protein–protein interactions in which these components engage in a given desmosome are unknown; however, the structure of a given desmosome will depend on its localization within a stratified epithelial sheet as desmosomal com-

Adjacent epithelial cells

ponents are expressed in a stratification-specific pattern38,39 (FIG. 3). Because desmosomes are highly insoluble structures, our current understanding of their structure comes from in vitro and cell-culture domain mapping, and reconstitution studies (FIG. 4). The model driving these studies was based on work on adherens junctions. In adherens junctions, microfilaments are tethered to the membrane through a tripartite complex: a classic cadherin is coupled directly through its cytoplasmic tail to β-catenin, which in turn is linked to α-catenin, which binds to actin40. Consequently, it was predicted that intermediate filaments are also tethered to the membrane by a linear complex: desmogleins and desmocollins associate with plakoglobin, which in turn binds to desmoplakin, which links intermediate filaments to the membrane. Several studies have provided data consistent with this model. Plakoglobin was shown to associate directly with Dsg1 and other desmosomal cadherins through sequences in the cadherin intracellular catenin-binding site region41,42 (FIG. 2). Further studies defined domains in plakoglobin important for this interaction43–45, as well as for regulatory functions such as the control of desmosome diameter46 (FIG. 2). Although the importance of the desmosomal cadherin tails and plakoglobin in recruiting desmoplakin to the plasma membrane47–49 is well established, little work has been done to directly compare different desmosomal cadherin tails. The function of the unique downstream sequences in the desmoglein subfamily remains a mystery. At the other end of the complex, early studies focused on interactions between intermediate filaments and the plakin family member desmoplakin, a protein shown to associate with intermediate filaments50,51 and to be required for anchoring intermediate filaments to the plaque7,52. These studies support the basic plan for cadherin–armadillo–plakin interactions and are consistent with a high-resolution map of the desmosome obtained by immunoelectron microscopy53.

Light microscopy

Electron microscopy

Tight junction Adherens junction Gap junction

Desmosome APICAL MEMBRANE

The surface of an epithelial cell that faces the lumen. Hemidesmosome BASOLATERAL MEMBRANE

The surface of an epithelial cell that adjoins underlying tissue. STRATIFIED EPITHELIAL SHEET

Multilayered epithelial cell sheet.

Figure 1 | The junctional complex of epithelial cells. In simple epithelial cells, tight junctions, adherens junctions and desmosomes form the ‘junctional complex’, which functions together with gap junctions to conduct the physiological functions of a polarized epithelial-cell sheet. The light microscopy image shows that keratin tonofilaments (in red) link the cell sheet through connecting desmosomes, which are highlighted by the desmoplakin (in green) staining. The electron micrograph further illustrates the highly organized ultrastructure of a desmosome in which mirror-image, tripartite electron-dense plaques sandwich a central core consisting of adjacent plasma membranes bisected by an intercellular zipper-like midline.

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 1 | DECEMBER 2000 | 2 0 9

REVIEWS

REVIEWS Sorting out the repertoire of protein–protein interactions in the plaque has become more challenging as further family members such as plakophilins have emerged. In addition, lateral interactions of a more complex nature than the ‘linear’ chain originally predicted have been proposed. The plaque is, in fact, probably more akin to a three-dimensional chain-link fence, with each protein node making several points of contact with neighbouring proteins, which would greatly strengthen the junction. For instance, desmosomal cadherin tails can interact directly in an armadillo-independent manner with desmoplakin; plakophilins can interact directly with intermediate filaments; and desmoplakin can interact with itself 54–56. As one example of how lateral interactions might be used in a celltype specific manner, plakophilin 1, which is concentrated primarily in the desmosomes of the superficial

a Desmosomal cadherins

PKP1 head Pg

EII

EIII

EIV

ICS

Desmoglein

IPL

RUD

DTD

PKP1/DP Pg a

ICS EI

EII

EIII

EIV

EA TM

Desmocollin

Dsc1a

b Armadillo family Dsg Plakoglobin

Dsg

1 2 3 4 5 6 7 8 9 10

11 12 13

Desmoplakin Plakophilins 1 2 3 4

7 8 9 10

Desmoplakin/IFs/Dsg1

c Plakin family IFs ABD

Plectin N

NN Z Y

ROD

NN Z Y

ROD

A B

NN Z Y

ROD

NN Z Y

ROD

B B B B

Desmoplakin Dsc1a/Pg/PKP1 N Envoplakin N

IFs C

Periplakin N

210

layers of stratified epithelia, enhances the recruitment of desmoplakin to the desmosomal plaque. A model has been suggested whereby plakoglobin links desmoplakin to the cadherin tails through linear interactions, whereas plakophilin 1 extends the submembranous plaque laterally through plakophilin 1–desmoplakin interactions (FIG. 4). In this way, numerous intermediate filament-binding sites would be available in the upper layers of the epidermis where they are most needed57. Do members of a particular gene family have similar functions within these macromolecular complexes, and can they compensate for other family members — in other words, are they interchangeable? So far, all knockouts of desmosomal molecules have resulted in readily observable phenotypes (TABLE 2). Nevertheless, one recent study58 indicates that forced expression of Dsg3 in the superficial epidermal layers of transgenic mice Figure 2 | Schematic structure of principal desmosomal proteins. a | Desmosomal cadherin superfamily. Like the classic cadherins, desmosomal cadherins are type I membrane molecules with extracellular consensus calciumbinding sites. The three members of the desmoglein (~160 kDa) subfamily are unique in having extended tails beyond the ICS (intracellular catenin-binding site segment), that could function as a scaffolding for further protein interactions. The three members of the desmocollin (110–115 kDa) subfamily each have two spliced forms. The ‘b’ form lacks the ICS and is therefore unable to bind plakoglobin. b | Desmosomal armadillo family members. The desmosomal armadillo family members include two subclasses, one containing plakoglobin, and the other the plakophilins. Plakoglobin is β-catenin’s close relative whereas the plakophilins are a subclass more related to p120. Plakoglobin links the desmosomal cadherin tails to desmoplakin, through binding sites indicated by the arrows, but is probably engaged in lateral interactions as well. Binding partners for the plakophilins are just now being defined, with plakophilin 1 showing a more restricted repertoire than plakoglobin. c | Desmosomal plakin family members. Plakins are dumb-bell-shaped molecules comprising three domains, a central α-helical coiled-coil rod, flanked by globular carboxy (C)- and amino (N)-terminal domains, that in desmoplakin interact with intermediate filaments and armadillo/cadherin family members, respectively, as illustrated by the black arrows. The plakin family started out as a small group of intermediate filamentbinding proteins, the domain structure for which was initially characterized for desmoplakin and observed in the hemidesmosomal protein bullous pemphigoid antigen 1, and then plectin. Envoplakin and periplakin, which are also keratinocyte cell envelope proteins, have been added to the group. The term ‘plakin’ has now been coined to describe the entire group107, which has expanded to include members harbouring actin- and microtubule-binding domains, some of which lack intermediate-filament-binding domains. The amino terminus contains a series of predicted α-helical bundles designated NN, Z, Y, X, W and Z, whereas the carboxy-terminal intermediate-filament-binding domain contains homology units A, B and C108. | The regions for interactions with other proteins are indicated by arrows throughout. (ABD, actin-binding protein; Dsc1a, desmocollin 1a; DTD, desmoglein terminal domain; EI–IV, cadherin repeats I–IV; EA, extracellular anchor; IA, intracellular anchor; IF, intermediate filament; IPL, proline-rich linker; Pg, plakoglobin; PKP1, plakophilin 1; RUD, repeating unit domain; TM, transmembrane domain.)

www.nature.com/reviews/molcellbio

| DECEMBER 2000 | VOLUME 1

REVIEWS

Table 2 | Mouse models and human diseases related to desmosomal genes Mutation/target antigen

Phenotype

Mouse models Plakoglobin knockout9,10

• Early embryonic lethality caused by heart defects due to loss of intercalated disc integrity. • Later survivors show epidermal defects.

Desmoglein-3 knockout8

• Hair loss and loss of epithelial integrity.

Epidermally targeted truncated desmoglein-3 transgenic11

• Blackened tail tip, flakiness of back skin and paw swelling are visible two days after birth. • Desmosomes are reduced in number. • Perturbed epidermal cell–cell adhesion leads to thickened epidermis with areas of parakeratosis. • Some areas of the skin show hyperproliferation and inflammation.

Desmoplakin knockout7

• Early embryonic lethality during the expansion of the egg cylinder, presumably due to loss of integrity of the embryonic endoderm.

Human diseases — gene mutation Plakoglobin carboxy-terminal truncation due to frameshift and premature termination15

Naxos disease • Autosomal recessive arrythmogenic right ventricular cardiomyopathy (ARVC) combined with striate palmoplantar keratoderma (SPPK) and woolly hair.

Plakophilin-1 null due to premature • Ectodermal dysplasia resulting in a skin fragility syndrome. termination on both alleles12 Desmoglein-1 mutation (possible haploinsufficiency*)13

Striate palmoplantar keratoderma (SPPK) • Lesions of the palms and soles exacerbated by mechanical trauma.

Desmoplakin haploinsufficiency14

Striate palmoplantar keratoderma (SPPK) • Lesions of the palms and soles exacerbated by mechanical trauma.

Human diseases — autoimmune Desmoglein-3 autoantibodies110

Pemphigus vulgaris • Blistering disease of the oral cavity caused by circulating antibodies directed against desmoglein 3. • Presence of antibodies against both desmoglein 3 and 1 cause mucocutaneous form with blisters also in deep epidermis.

Desmoglein-1 autoantibodies110

Pemphigus foliaceus • Blistering disease of the superficial epidermis caused by circulating antibodies directed against desmoglein 1.

*Whether haploinsufficiency or dominant-negative interference underlies this disease awaits further analysis.

prevents the blistering that normally occurs after injection of antibodies from PEMPHIGUS FOLIACEUS patients directed against Dsg1, which normally predominates in these epidermal layers. This indicates that Dsg3 can compensate for loss of Dsg1 adhesive function. Whether other functions are also compensated for remains to be determined. Desmosome assembly and maintenance

PEMPHIGUS FOLIACEUS

A rare, blistering autoimmune disease that affects the skin and mucosal membranes.

Like other intercellular junctions, desmosomes in cultured cells assemble in response to cell–cell contact and raised levels of extracellular calcium59,60. Perturbing intracellular calcium stores with the Ca2+-ATPase inhibitor thapsigargin interferes with intercellular junction assembly even when cells are in contact61. Recently, Darier’s and Hailey–Hailey’s disease have been shown to arise from mutations in calcium ATPases, ATP2A2 (SERCA2b)62 and ATP2C1 (REF. 63), respectively. These disorders were long hypothesized to arise from defects in desmosomes, based on the loss of desmosomal structure and adhesion observed in patients. The finding that mutations in these molecular pumps, which are responsible for sequestering calcium in intracellular organelles,

cause these diseases supports the importance of intracellular calcium homeostasis for desmosomal adhesion. When calcium levels are low, newly synthesized desmosomal plaque and membrane proteins are unstable64,65. The desmosomal cadherins move to the membrane together with plakoglobin molecules, apparently in a separate compartment from desmoplakin and intermediate filaments. After desmosomal assembly is triggered, cadherins and desmoplakin are stabilized, presumably as they become associated with the cytoskeleton. Cadherin and desmoplakin complexes seem to remain separate until positioned at the membrane, when they become insoluble and more difficult to analyse biochemically. This final stage may involve dephosphorylation as the phosphatase inhibitor okadaic acid does not inhibit trafficking of components to the membrane, but prevents ultrastructurally recognizable plaques from forming66. Desmosome assembly in response to calcium is reversible during the early stages, but ultimately desmosomes mature and can no longer be dissociated by calcium depletion59. However, desmosomes rapidly re-acquire ‘calcium dependence’ at the free edge of wounded epithelial cell sheets, and this

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 1 | DECEMBER 2000 | 2 1 1

REVIEWS

Dsg1 Dsc1 PKP1

Stratum corneum Granular layer

Spinous layer

Dsg3 Dsc3 PKP2

Basal layer

Figure 3 | Differentiation-specific expression of desmosomal components. This schematic of the epidermis shows that proteins are expressed in a differentiation-dependent manner. For example, in the case of desmosomal cadherins, desmoglein (Dsg) 1 and desmocollin (Dsc) 1 are patterned in a gradient, with the highest level in the superficial layers and tapering off in the basal layers. Desmoglein 3 and desmocollin 3 show an opposite pattern. Likewise, the armadillo family member plakophilin (PKP) 1 is concentrated in junctions of the superficial layers, whereas PKP2 is found in desmosomes deeper in the epidermis.

ENDOCYTOSIS

The uptake of extracellular materials by cells. The plasma membrane invaginates and vesicles pinch off containing endocytosed molecules and plasma membrane components. HELA CELLS

An established tissue-culture strain of human epidermoid carcinoma cells, containing 70–80 chromosomes per cell. These cells were originally derived from tissue taken from a patient named Henrietta Lacks in 1951.

process depends on activation of protein kinase C-α (PKC-α)67. Throughout calcium-induced desmosome assembly, a pool of desmoplakin is associated with the cytoskeleton as ‘dots’ or packets of non-membrane bound desmoplakin decorating intermediate filament networks60. These packets have been proposed to be desmosome precursors that are delivered along intermediate filaments to build the plaque. Others have raised the possibility that these cytoplasmic dots are membrane bound and represent either preassembled structures or ENDOCYTOSED remnants of previously formed desmosomes68,69. High-resolution analysis using fluorescently tagged molecules in living cells should clarify this issue by determining the fate of these cytoplasmic structures during junction assembly in real time. Intermediate filaments do not seem to be absolutely required for desmoplakin’s assembly into desmosomes, as carboxy-terminally truncated desmoplakin that lacks the intermediate filament-binding site can still incorporate into junctions52,70. Furthermore, embryonic stem cells lacking Intermediate filament

Desmoplakin

Desmoglein

Desmocollin

Plakophilin

Plakoglobin

Figure 4 | Molecular model of the desmosome. This simplified model shows representative protein–protein interactions in which principal desmosomal components participate. These interactions are based on yeast two hybrid, co-immunoprecipitation and recruitment assays in cultured cells.

212

keratin 8 can still make desmosomes71, even though they are unable to assemble intermediate filament networks with the complementary partners of keratin 8 — keratin 18 and 19. Developmental studies of desmosome assembly also indicate that intermediate filaments grow out of nascent membrane-bound desmosomes, in contrast to the idea that intermediate filaments deliver plaque components to desmosomal cadherins at the membrane72. Adherens junctions arose first in evolution, and also assemble first both during development and in individual cells that are stimulated to make junctions. When adherens junction formation is blocked through inhibitory antibodies, dominant-negative constructs or as a result of naturally occurring mutations, desmosome assembly is inhibited or delayed73,74. Adherens-junction assembly is an active process that requires the engagement of E-cadherin and the active participation of the actin cytoskeleton75. It has been proposed that desmosomes form passively, assembling in the gaps between adherens junctions. The purpose of adherens junctions, then, could be to bring the membranes into close proximity, thereby allowing desmosomal molecules to engage and cluster. The fact that expression of a non-cadherin adhesion molecule, protein zero, led to the formation of desmosomes in HELA CELLS is consistent with this idea76. It has also been reported that desmosomal plaque proteins can cluster, forming half-desmosomal plaques in the absence of desmosomal cadherin engagement69. This is in contrast to the requirement for classic cadherin ligation during adherens junction assembly. Understanding this difference could be key to understanding desmosome assembly. Classic cadherins may have a more direct role in desmosome assembly. Cells that cannot assemble Ecadherin–plakoglobin complexes do not make desmosomes, even though they express desmosomal components77. It has been proposed that such a complex instructs the cell to make desmosomes through a structural or chemical signal. Disrupting desmosomes through the loss of plakoglobin9 or expression of a dominant-negative desmoplakin52 results in mixing of adherens junction and desmosome components, providing further support for a close structural connection between these intercellular junctions. One way to interpret these studies is that an intermediate is required for nucleating desmosome assembly that contains a classic cadherin, plakoglobin and some other desmosomal protein, perhaps a plakin. Segregation could subsequently occur through further desmosome-specific protein–protein interactions. It is interesting to note that certain cells have naturally mixed junctions that anchor both intermediate filaments and microfilaments78. It has been suggested that, in endothelial cell junctions, VE-cadherin links both actin and vimentin through β-and αcatenin or plakoglobin and desmoplakin, respectively79. It seems possible that this intermingling of junctional components is due to the fact that endothelial junctions are missing desmosomal cadherins, which www.nature.com/reviews/molcellbio

| DECEMBER 2000 | VOLUME 1

REVIEWS

L CELL FIBROBLASTS

A mouse fibroblast line derived from connective tissue that does not express adhesion molecules.

could therefore theoretically segregate these components into separate domains in E-cadherin-containing epithelial cells. Alternatively, rather than acting as a structural intermediate, E-cadherin might provide a chemical signal required for desmosome assembly. Such an idea is supported by studies showing that activation of PKC leads to desmosome formation, bypassing the normal requirement for extracellular calcium80 and assembly of the classic cadherin complex81. Supporting the close relationship between adherens junctions and desmosomes are reports that, under certain circumstances, both desmocollin 1 (REF. 82) and desmoglein 1 can bind β-catenin83,84. Moreover, lateral formation of desmosomal and classic cadherins have been observed to occur in low calcium concentrations85. Reconciliation of all these observations into a single model for adherens junction-dependent desmosome assembly is one of the important challenges facing the field. The puzzle of desmosomal adhesion

Initial structural studies of classic cadherins led to the proposal that their extracellular domains form a ‘zipper’ at the adhesive interface, in which laterally associated cadherin ‘strand’ dimers homophilically engage similarly orientated dimers on the neighbouring cell86. Although details of this model have since been questioned, this idea was appealing in light of ultrastructural images in which an electron-dense zig–zag midline seems to join the two halves of the desmosome in the centre of the intercellular space (FIG. 1). However, determining the organization of desmosomal cadherins in situ is a challenge that has not been met. Several labs have attempted to reconstitute the adhesive interface by introducing desmosomal cadherins into non-adherent mouse L CELL FIBROBLASTS. Individual desmosomal cadherins, with or without plakoglobin, are unable to confer robust adhesive properties on these cells87–90. Expression of desmocollins and desmogleins together in L cells has yielded mixed results — in some cases, cell–cell adhesion was observed and in others it was not88,89,91. What underlies these conflicting findings? First, the desmoglein and desmocollin expressed in these experiments may not be preferred partners. However, two studies with differing results used the same pairs, Dsg1 and Dsc2. Therefore, although a possible contributing factor, this was probably not the only reason underlying the discrepancies. Second, the shorter spliced variant, or ‘b’ form, of desmocollin may be required; however, this is again unlikely to be the sole difference in studies as this form was used in one study in which adhesion was reconstituted but not in another88,91. Last, a likely possibility is that the correct stoichiometry of desmocollins and desmogleins is critical. This is an idea that has not been tested systematically by varying the level of one cadherin while keeping the other constant. Are both desmocollins and desmogleins required for desmosomal cadherin-mediated adhesion because they function through heterophilic, rather than homophilic,

interactions? Dsg2 and Dsc1 have been shown to form a complex mediated by their extracellular domains92, providing compelling evidence for such heterophilic interactions. Such studies have not been carried out for any other desmosomal cadherin pair, and no information is available on the preferred pairing of desmosomal cadherins in vivo. Finally, the functional significance and potential differences that particular pairs might confer on cells are unknown. Desmosomes as signalling centres

Desmosomes are not simply ‘spot welds’, as they have so often been dubbed by histology textbooks. Like adherens junctions, desmosomes come and go in response to growth factors93 and desmosomal molecules have also been reported to associate with kinases and phosphatases94. In NBTII cells, activation of the tyrosine kinase receptor by FGF-1 treatment leads to junction dissolution. Although it was shown that a transcription factor, slug, is involved in this process93, the downstream desmosomal targets and molecular basis of junction disassembly are unknown. In fact, although the vast majority of desmosomal proteins studied are phosphoproteins, the consequences of phosphorylation are only understood in very few cases. Plakoglobin has often been cited as a tyrosine kinase substrate95, and preliminary work indicates that phosphorylation by the epidermal growth factor receptor (EGFR) and Src may compromise its interaction with desmoplakin, possibly inhibiting its association with the cytoskeleton (C.A.G. and M.F. Denning, unpublished observations). Phosphorylation of a serine in the carboxyl terminus of desmoplakin has been shown to inhibit its interactions with intermediate filaments, therefore providing a possible mechanism for modulating its availability for assembly and/or the dynamic state of the junctions96. The epithelial blistering disease pemphigus vulgaris has provided an important cell-biological window on desmosomes as sensors. It has been suggested that binding of pemphigus vulgaris antibodies to Dsg3 in a cultured-keratinocyte line leads to the phosphorylation of Dsg3 on serines and its dissociation from plakoglobin97. The resulting loss of the cadherin from desmosomes has been suggested to contribute to epithelial blistering in patients with these autoimmune antibodies. By using plakoglobin-null cells, Müller et al. have shown that plakoglobin is required for keratin retraction that typifies the in vitro response to pemphigus vulgaris IgG (E. Müller, personal communication). This sensory machinery may normally regulate the turnover of junctional complexes and remodelling of desmosomes during differentiation or wound healing. Desmosomes provide a home for at least four members of the armadillo family of nuclear and junctional proteins. Plakoglobin is the closest relative of β-catenin and is thought to function primarily in desmosomal integrity. It has been known for some time that plakoglobin can partner with adenomatosis polyposis coli (APC) and lymphoid enhancer-binding factor (LEF/TCF), and that, like β-catenin, plakoglobin is subject to degradation by the APC/proteasome pathway98.

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 1 | DECEMBER 2000 | 2 1 3

REVIEWS

AXIS DUPLICATION

Duplication of body parts about an axis (for example the anterior–posterior axis) as a result of mutation. NEURITE

Process extended by a nerve cell that can give rise to an axon or a dendrite. MESENCHYME

Undifferentiated connective tissue present in the early embryo.

Furthermore, when overexpressed in Xenopus laevis oocytes, both plakoglobin and β-catenin cause AXIS 99 DUPLICATION . Several investigators have proposed that plakoglobin exerts these effects by interfering with the degradation of β-catenin which, in essence, activates the β-catenin pathway100. Others have questioned this idea and have reported signalling directly due to plakoglobin. In one study, plakoglobin reduces growth of hair follicles in transgenic mice101. In contrast, two reports indicate that plakoglobin can transform cells by specifically activating c-myc in a β-catenin-independent pathway102 and by inhibiting apoptosis concomitant with induction of the anti-apoptotic protein Bcl-2 (REF. 103). These findings not only raise the possibility that desmosomal armadillo proteins signal on their own, but they also highlight that signalling may depend on cellular context. The fact that plakoglobin and the plakophilins, like desmosomes, arose relatively late in evolution, raises the possibility that their signalling functions may contribute to later stages of differentiation in complex tissues. The associated desmosomal cadherins also arose late in evolution and are expressed in a cell type-dependent manner. Therefore, the observed regulation of plakoglobin stability by desmogleins and desmocollins104, and possible differences in binding affinity or stoichiometry between armadillo proteins and different desmosomal cadherins, provide further mechanisms for regulating signalling in a differentiation-dependent manner44. The function of plakophilins in the nucleus is so far unknown, as are the factors that regulate their subcellular localization. Preliminary studies have hinted at the possibility that plakophilin 1 may be involved in regulating cell shape or organization of the actin cytoskeleton 56. Ectopic expression of the related armadillo protein p120 has been reported to trigger NEURITE-like outgrowth in cells. The ability of plakophilin to alter actin-dependent structures would be the first instance of such a function for a protein more commonly associated with the intermediate filament cytoskeleton105.

Links DATABASE LINKS desmogleins | desmocollins | plakophilins | junctional plakoglobin |

214

Koch, P. J. & Franke, W. W. Desmosomal cadherins: another growing multigene family of adhesion molecules. Curr. Opin. Cell Biol. 6, 682–687 (1994). Adams, C. L. & Nelson, W. J. The cytomechanics of cadherin-mediated cell–cell adhesion. Curr. Opin. Cell Biol. 10, 572–577 (1998). Kowalczyk, A. P., Bornslaeger, E. A., Norvell, S. M., Palka, H. L. & Green, K. J. Desmosomes: intercellular adhesive junctions specialized for attachment of intermediate filaments. Int. Rev. Cytol. 185, 237–302 (1999). Simon, A. M. & Goodenough, D. A. Diverse functions of vertebrate gap junctions. Trends Cell Biol. 8, 477–483

Do desmosomes or, for that matter, anchoring junctions in general signal directly? In the case of classic cadherins, there are several instances where cadherins have been shown to activate signalling pathways (reviewed in REF. 106). N-cadherin is thought to activate FGFRdependent neurite outgrowth, and possibly motility in tumour cells. N- and E-cadherin also drive divergent pathways of differentiation when introduced into embryonic stem cells, with N-cadherin promoting MESENCHYMAL structures, and E-cadherin promoting epithelial structures. E-cadherin and VE-cadherin regulate cell survival by activating the phosphatidylinositol-3-(OH)kinase-dependent serine kinase PKB/Akt, and contact inhibition by increasing levels of the cell-cycle-dependent kinase inhibitor p27. So far, there is little in the published literature to implicate desmosomal cadherins in cell growth or differentiation. However, this area of active investigation will probably uncover differentiation-specific functions for desmosomal cadherin isoforms in the near future. Looking to the future

We still have much to learn about desmosome assembly, regulation and functions. In the future, high-resolution structural and kinetic approaches will be required to bolster our current knowledge with respect to protein–protein interactions and the affinity of these interactions. The advent of microscopical approaches for imaging living cells promises to aid in establishing temporal and spatial patterns during desmosome assembly, and to address several unanswered questions. For instance, how dynamic are desmosomes? Do individual components exchange in and out of desmosomes once established? What is the half-life of an entire desmosome? A second set of questions relates to desmosomal armadillo proteins. How are plakophilins chaperoned into and out of the nucleus, and what are they doing there? Are nuclear and junctional forms of the plakophilins exchangeable and what regulates their subcellular compartmentalization? Finally, in this review, we have not even touched on the regulation of desmosomal gene expression, which of course determines the patterns of molecules involved in the assembly of these specialized structures. Beyond that, do desmosomal cadherins direct gene expression in complex tissues and affect the characteristics of the tissue or layer in which they are expressed? The advent of microarray techniques and increasing sophistication of inducible animal models promise to help meet these challenges in the coming years.

(1998). Fanning, A. S., Mitic, L. L. & Anderson, J. M. Transmembrane proteins in the tight junction barrier. J. Am. Soc. Nephrol. 10, 1337–1345 (1999). Steinberg, M. S. & McNutt, P. M. Cadherins and their connections: adhesion junctions have broader functions. Curr. Opin. Cell Biol. 11, 554–560 (1999). Gallicano, G. I. et al. Desmoplakin is required early in development for assembly of desmosomes and cytoskeletal linkage. J. Cell Biol. 143, 2009–2022 (1998). This paper, describing the phenotype in desmoplakin-null mice, is notable owing to the

extremely early stage at which the embryos show defects, and shows that desmoplakin is required both for the assembly of desmosomes and integrity of the early embryonic endoderm. 8. Koch, P. J. et al. Targeted disruption of the pemphigus vulgaris antigen (desmoglein 3) gene in mice causes loss of keratinocyte cell adhesion with a phenotype similar to pemphigus vulgaris. J. Cell Biol. 137, 1091–1102 (1997). 9. Ruiz, P. et al. Targeted mutation of plakoglobin in mice reveals essential functions of desmosomes in the embryonic heart. J. Cell Biol. 135, 215–225 (1996). 10. Bierkamp, C., McLaughlin, K. J., Schwarz, H., Huber, O. &

www.nature.com/reviews/molcellbio

| DECEMBER 2000 | VOLUME 1

REVIEWS

11.

12.

13.

14.

15.

16.

17.

18.

19.

20.

21.

22.

23.

24.

25.

26.

27.

28.

Kemler, R. Embryonic heart and skin defects in mice lacking plakoglobin. Dev. Biol. 180, 780–785 (1996). References 9 and 10 report the phenotype of plakoglobin-null mice; both show that plakoglobin function is particularly important for junction structure and integrity of cardiac muscle, as embryonic lethality occurs when the heart starts to beat. Desmosomes in the epidermis are less perturbed, indicating that other molecules compensate for its function in this tissue. Allen, E., Yu, Q.-C. & Fuchs, E. Mice expressing a mutant desmosomal cadherin exhibit abnormalities in desmosomes, proliferation, and epidermal differentiation. J. Cell Biol. 133, 1367–1382 (1996). McGrath, J. A. et al. Mutations in the plakophilin 1 gene result in ectodermal dysplasia/skin fragility syndrome. Nature Genet. 17, 240–244 (1997). This is the first paper reporting that patients with mutations in a desmosomal molecule, plakophilin 1, show several defects of the epidermis and appendages. Rickman, L. et al. N-terminal deletion in a desmosomal cadherin causes the autosomal dominant skin disease striate palmoplantar keratoderma. Hum. Mol. Genet. 8, 971–976 (1999). Armstrong, D. K. B. et al. Haploinsufficiency of desmoplakin causes a striate subtype of palmoplantar keratoderma. Hum. Mol. Genet. 8, 143–148 (1999). McKoy, G. et al. Identification of a deletion in plakoglobin in arrhythmogenic right ventricular cardiomyopathy with palmoplantar keratoderma and woolly hair (Naxos disease). Lancet 355, 2119–2124 (2000). Schmidt, A. et al. Desmosomes and cytoskeletal architecture in epithelial differentiation: cell type-specific plaque components and intermediate filament anchorage. Eur. J. Cell Biol. 65, 229–245 (1994). Koch, P. J. et al. Identification of desmoglein, a constitutive desmosomal glycoprotein, as a member of the cadherin family of cell adhesion molecules. Eur. J. Cell Biol. 53, 1–12 (1990). Amagai, M., Klaus-Kovtun, V. & Stanley, J. R. Autoantibodies against a novel epithelial cadherin in pemphigus vulgaris, a disease of cell adhesion. Cell 67, 869–877 (1991). This paper reports the identification of the antigen targeted by auto-immune antibodies from patients with pemphigus vulgaris — desmoglein 3, which is a cadherin-like molecule. Schafer, S., Koch, P. J. & Franke, W. W. Identification of the ubiquitous human desmoglein, Dsg2, and the expression catalogue of the desmoglein subfamily of desmosomal cadherins. Exp. Cell Res. 211, 391–399 (1994). Collins, J. E. et al. Cloning and sequence analysis of desmosomal glycoproteins 2 and 3 (desmocollins): cadherin-like desmosomal adhesion molecules with heterogeneous cytoplasmic domains. J. Cell Biol. 113, 381–391 (1991). Mechanic, S., Raynor, K., Hill, J. E. & Cowin, P. Desmocollins form a distinct subset of the cadherin family of cell adhesion molecules. Proc. Natl Acad. Sci. USA 88, 4476–4480 (1991). Parker, A. E. et al. Desmosomal glycoproteins II and III: Cadherin-like junctional molecules generated by alternative splicing. J. Biol. Chem. 266, 10438–10445 (1991). Kawamura, K. et al. cDNA cloning and expression of a novel human desmocollin. J. Biol. Chem. 269, 26295–26302 (1994). Theis, D. G., Koch, P. J. & Franke, W. W. Differential synthesis of type 1 and type 2 desmocollin mRNAs in human stratified epithelia. Int. J. Dev. Biol. 37, 101–110 (1993). Cowin, P., Kapprell, H.-P., Franke, W. W., Tamkun, J. & Hynes, R. O. Plakoglobin: a protein common to different kinds of intercellular adhering junctions. Cell 46, 1063–1073 (1986). This paper describes plakoglobin as the first component common to microfilament- and intermediate filament-associated cell–cell junctions. It provides a foundation for much of the recent work on the protein now known to be the closest relative of the armadillo protein β-catenin. Hatzfeld, M., Kristjansson, G. I., Plessmann, U. & Weber, K. Band 6 protein, a major constituent of desmosomes from stratified epithelia, is a novel member of the armadillo multigene family. J. Cell Sci. 107, 2259–2270 (1994). Heid, H. W. et al. Cell type-specific desmosomal plaque proteins of the plakoglobin family: plakophilin 1 (band 6 protein). Differentiation 58, 113–131 (1994). Mertens, C., Kuhn, C. & Franke, W. W. Plakophilins 2a and 2b: constitutive proteins of dual location in the karyoplasm

29.

30.

31.

32.

33.

34.

35.

36.

37.

38.

39.

40.

41.

42.

43.

44.

45.

46.

47.

48.

and the desmosomal plaque. J. Cell Biol. 135, 1009–1025 (1996). Bonne, S., Hengel, J. v., Nollet, F., Kools, P. & Roy, F. v. Plakophilin-3, a novel armadillo-like protein present in nuclei and desmosomes of epithelial cells. J. Cell Sci. 112, 2265–2276 (1999). Schmidt, A. et al. Plakophilin 3 — a novel cell-type-specific desmosomal plaque protein. Differentiation 64, 291–306 (1999). Green, K. J. et al. Structure of the human desmoplakins: implications for function in the desmosomal plaque. J. Biol. Chem. 265, 2603–2612 (1990). This paper is the first to describe much of the domain structure of what is now known as the ‘plakin’ family, and the first to uncover that desmoplakin and the hemidesmosomal bullous pemphigoid antigen 1 (BPAG1) belong to what was to emerge as an important family of cytoskeleton-linking proteins. Wiche, G. et al. Cloning and sequencing of rat plectin indicates a 466-kD polypeptide chain with a three-domain structure based on a central α-helical coiled coil. J. Cell Biol. 114, 83–99 (1991). Skalli, O., Jones, J. C. R., Gagescu, R. & Goldman, R. D. IFAP 300 is common to desmosomes and hemidesmosomes and is a possible linker of intermediate filaments to these junctions. J. Cell Biol. 125, 159–170 (1994). Ruhrberg, C., Hajibagheri, M. A. N., Simon, M., Dooley, T. P. & Watt, F. M. Envoplakin, a novel precursor of the cornified envelope that has homology to desmoplakin. J. Cell Biol. 134, 715–729 (1996). Ruhrberg, C., Hajibagheri, M. A. N., Parry, D. A. D. & Watt, F. M. Periplakin, a novel component of cornified envelopes and desmosomes that belongs to the plakin family and forms complexes with envoplakin. J. Cell Biol. 139, 1835–1849 (1997). Ouyang, P. & Sugrue, S. P. Characterization of pinin, a novel protein associated with the desmosome–intermediate filament complex. J. Cell Biol. 135, 1027–1042 (1996). Tsukita, S. & Tsukita, S. Desmocalmin: a calmodulinbinding high molecular weight protein isolated from desmosomes. J. Cell Biol. 101, 2070–2080 (1985). Koch, P. J., Goldschmidt, M. D., Zimbelmann, R., Troyanovsky, R. & Franke, W. W. Complexity and expression patterns of the desmosomal cadherins. Proc. Natl Acad. Sci. USA 89, 353–357 (1992). North, A. J., Chidgey, M. A. J., Clarke, J. P., Bardsley, W. G. & Garrod, D. R. Distinct desmocollin isoforms occur in the same desmosomes and show reciprocally graded distributions in bovine nasal epidermis. Proc. Natl Acad. Sci. USA 93, 7701–7705 (1996). Jou, T. S., Stewart, D. B., Stappert, J., Nelson, W. J. & Marrs, J. A. Genetic and biochemical dissection of protein linkages in the cadherin–catenin complex. Proc. Natl Acad. Sci. USA 92, 5067–5071 (1995). Korman, N. J., Eyre, R. W., Klaus-Kovtun, V. & Stanley, J. R. Demonstration of an adhering-junction molecule (plakoglobin) in the autoantigens of pemphigus foliaceus and pemphigus vulgaris. N. Engl. J. Med. 321, 631–635 (1989). Mathur, M., Goodwin, L. & Cowin, P. Interactions of the cytoplasmic domain of the desmosomal cadherin Dsg1 with plakoglobin. J. Biol. Chem. 269, 14075–14080 (1994). Wahl, J. K. et al. Plakoglobin domains that define its association with the desmosomal cadherins and the classical cadherins: identification of unique and shared domains. J. Cell Sci. 109, 1143–1154 (1996). Witcher, L. L. et al. Desmosomal cadherin binding domains of plakoglobin. J. Biol. Chem. 271, 10904–10909 (1996). Chitaev, N. A. et al. The binding of plakoglobin to desmosomal cadherins: patterns of binding sites and topogenic potential. J. Cell Biol. 133, 359–369 (1996). Palka, H. L. & Green, K. J. Roles of plakoglobin end domains in desmosome assembly. J. Cell Sci. 110, 2359–2371 (1997). Troyanovsky, S. M., Eshkind, L. G., Troyanovsky, R. B., Leube, R. E. & Franke, W. W. Contributions of cytoplasmic domains of desmosomal cadherins to desmosome assembly and intermediate filament anchorage. Cell 72, 561–574 (1993). This paper was the first to analyse the plaqueforming potential of the desmosomal cadherin tails. Troyanovsky, S. M., Troyanovsky, R. B., Eshkind, L. G., Leube, R. E. & Franke, W. W. Identification of amino acid sequence motifs in desmocollin, a desmosomal glycoprotein, that are required for plakoglobin binding and plaque formation. Proc. Natl Acad. Sci. USA 91,

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

10790–10794 (1994). 49. Kowalczyk, A. P. et al. The amino-terminal domain of desmoplakin binds to plakoglobin and clusters desmosomal cadherin–plakoglobin complexes. J. Cell Biol. 139, 773–784 (1997). 50. Stappenbeck, T. S. & Green, K. J. The desmoplakin carboxyl terminus coaligns with and specifically disrupts intermediate filament networks when expressed in cultured cells. J. Cell Biol. 116, 1197–1209 (1992). This was the first paper to show that the carboxyl terminus of a plakin family member, desmoplakin, associates with intermediate filaments. Subsequently, this domain of bullous pemphigoid antigen and plectin were also shown to interact with intermediate filaments. 51. Kouklis, P. D., Hutton, E. & Fuchs, E. Making a connection: direct binding between keratin intermediate filaments and desmosomal proteins. J. Cell Biol. 127, 1049–1060 (1994). This paper showed that the desmoplakin carboxyl terminus interacts directly with intermediate filament polypeptides, specifically type II epidermal keratins. 52. Bornslaeger, E. B., Corcoran, C. M., Stappenbeck, T. S. & Green, K. J. Breaking the connection: Displacement of the desmosomal plaque protein desmoplakin from cell–cell interfaces disrupts anchorage of intermediate filament bundles and alters intercellular junction assembly. J. Cell Biol. 134, 985–1002 (1996). This paper used a dominant-negative approach to show for the first time that desmoplakin is required for anchoring intermediate filaments to the desmosomal plaque and that it may be involved in segregating adherens junctions and desmosomes. 53. North, A. J. et al. Molecular map of the desmosomal plaque. J. Cell Sci. 112, 4325–4336 (1999). 54. Smith, E. A. & Fuchs, E. Defining the interactions between intermediate filaments and desmosomes. J. Cell Biol. 141, 1229–1241 (1998). 55. Hofmann, I. et al. Interaction of plakophilins with desmoplakin and intermediate filament proteins: an in vitro analysis. J. Cell Sci. 113, 2471–2483 (2000). 56. Hatzfeld, M., Haffner, C., Schulze, K. & Vinzens, U. The function of plakophilin 1 in desmosome assembly and actin filament organization. J. Cell Biol. 149, 209–222 (2000). 57. Kowalczyk, A. P. et al. The head domain of plakophilin-1 binds to and enhances its recruitment to desmosomes: implications for cutaneous disease. J. Biol. Chem. 274, 18145–18148 (1999). 58. Wu, H. et al. Protection against pemphigus foliaceus by desmoglein 3 in neonates. N. Engl. J. Med. 343, 31–35 (2000). 59. Watt, F. M., Mattey, D. L. & Garrod, D. R. Calcium-induced reorganization of desmosomal components in cultured human keratinocytes. J. Cell Biol. 99, 2211–2215 (1984). 60. Jones, J. C. R. & Goldman, R. D. Intermediate filaments and the initiation of desmosome assembly. J. Cell Biol. 101, 506–517 (1985). 61. Stuart, R. O., Sun, A., Bush, K. T. & Nigam, S. K. Dependence of epithelial intercellular junction biogenesis on thapsigargin-sensitive intracellular calcium stores. J. Biol. Chem. 271, 13636–13641 (1996). 62. Sakuntabhai, A. et al. Mutations in ATP2A2, encoding a Ca2+ pump, cause Darier disease. Nature Genet. 21, 271–277 (1999). 63. Hu, Z. et al. Mutations in ATP2C1, encoding a calcium pump, cause Hailey–Hailey disease. Nature Genet. 24, 61–65 (2000). 64. Penn, E. J., Hobson, C., Rees, D. A. & Magee, A. I. Structure and assembly of desmosome junctions: biosynthesis, processing, and transport of the major protein and glycoprotein components in cultured epithelial cells. J. Cell Biol. 105, 57–68 (1987). References 64 and 65 were the first of a series of important papers from these two groups that analysed the biosynthesis and processing of desmosomal components. 65. Pasdar, M. & Nelson, W. J. Kinetics of desmosome assembly in Madin–Darby canine kidney epithelial cells: temporal and spatial regulation of desmoplakin organization and stabilization upon cell–cell contact. I. Biochemical analysis. J. Cell Biol. 106, 677–685 (1988). 66. Pasdar, M., Li, Z. & Chan, H. Desmosome assembly and disassembly are regulated by reversible protein phosphorylation in cultured epithelial cells. Cell. Motil. Cytoskeleton 30, 108–121 (1995). 67. Wallis, S. et al. The α isoform of protein kinase C is involved in signaling the response of desmosomes to wounding in cultured epithelial cells. Mol. Biol. Cell 11,

VOLUME 1 | DECEMBER 2000 | 2 1 5

REVIEWS 1077–1092 (2000). 68. Burdett, I. D. J. Internalisation of desmosomes and their entry into the endocytic pathway via late endosomes in MDCK cells. J. Cell Sci. 106, 1115–1130 (1993). 69. Demlehner, M. P., Schafer, S., Grund, C. & Franke, W. W. Continual assembly of half-desmosomal structures in the absence of cell contacts and their frustrated endocytosis: a coordinated Sisyphus cycle. J. Cell Biol. 131, 745–760 (1995). 70. Stappenbeck, T. S. et al. Functional analysis of desmoplakin domains: specification of the interaction with keratin versus vimentin intermediate filament networks. J. Cell Biol. 123, 691–705 (1993). 71. Baribault, H. & Oshima, R. G. Polarized and functional epithelia can form after the targeted inactivation of both mouse keratin 8 alleles. J. Cell Biol. 115, 1675–1684 (1991). 72. Overton, J. Desmosome development in normal and reassociating cells in the early chick blastoderm. Dev. Biol. 4, 532–548 (1962). 73. Gumbiner, B., Stevenson, B. & Grimaldi, A. The role of the cell adhesion molecule uvomorulin in the formation and maintenance of the epithelial junctional complex. J. Cell Biol. 107, 1575–1587 (1988). 74. Amagai, M. et al. Delayed assembly of desmosomes in keratinocytes with disrupted classic-cadherin-mediated cell adhesion by a dominant negative mutant. J. Invest. Derm. 104, 27–32 (1995). 75. Vasioukhin, V., Bauer, C., Yin, M. & Fuchs, E. Directed actin polymerization is the driving force for epithelial cell–cell adhesion. Cell 100, 209–219 (2000). 76. Doyle, J. P., Stempak, J. G., Cowin, P., Colman, D. R. & D’Urso, D. Protein zero, a nervous system adhesion molecule, triggers epithelial reversion in host carcinoma cells. J. Cell Biol. 131, 465–482 (1995). 77. Lewis, J. E. et al. Cross-talk between adherens junctions and desmosomes depends on plakoglobin. J. Cell Biol. 136, 919–934 (1997). This paper provides some of the first evidence that desmosome assembly is dependent on adherens junctions because of the common junction component, plakoglobin, and its required association with E-cadherin. 78. Schmelz, M. & Franke, W. W. Complexus adhaerentes, a new group of desmoplakin-containing junctions in endothelial cells: The syndesmos connecting retothelial cells of lymph nodes. Eur. J. Cell Biol. 61, 274–289 (1993). 79. Kowalczyk, A. P. et al. VE-cadherin and desmoplakin are assembled into dermal microvascular endothelial intercellular junctions: a pivotal role for plakoglobin in the recruitment of desmoplakin to intercellular junctions. J. Cell Sci. 111, 3045–3057 (1998). 80. Sheu, H.-M., Kitajima, Y. & Yaoita, H. Involvement of protein kinase C in translocation of desmoplakins from cytosol to plasma membrane during desmosome formation in human squamous cell carcinoma cells grown in low to normal calcium concentration. Exp. Cell Res. 185, 176–190 (1989). 81. Hengel, J. v. et al. Protein kinase C activation upregulates intercellular adhesion of a-catenin-negative human colon cancer cell variants via induction of desmosomes. J. Cell Biol. 137, 1103–1116 (1997).

216

82. Hanakawa, Y., Amagai, M., Shirakata, Y., Sayama, K. & Hashimoto, K. Different effects of dominant negative mutants of desmocollin and desmoglein on the cell–cell adhesion of keratinocytes. J. Cell Sci. 113, 1803–1811 (2000). 83. Norvell, S. M. & Green, K. J. Contributions of extracellular and intracellular domains of full length and chimeric cadherin molecules to junction assembly in epithelial cells. J. Cell Sci. 111, 1305–1318 (1998). 84. Bierkamp, C., Schwarz, H., Huber, O. & Kemler, R. Desmosomal localization of β-catenin in the skin of plakoglobin null-mutant mice. Development 126, 371–381 (1998). 85. Troyanovsky, R. B., Klingelhofer, J. & Troyanovsky, S. Removal of calcium ions triggers a novel type of intercadherin interaction. J. Cell Sci. 112, 4379–4387 (1999). 86. Weis, W. I. Cadherin structure: a revealing zipper. Structure 3, 425–427 (1995). 87. Amagai, M., Karpati, S., Klaus-Kovtun, V., Udey, M. C. & Stanley, J. R. The extracellular domain of pemphigus vulgaris antigen (desmoglein 3) mediates weak homophilic adhesion. J. Invest. Derm. 102, 402–408 (1994). 88. Marcozzi, C., Burdett, I. D. J., Buxton, R. S. & Magee, A. I. Coexpression of both types of desmosomal cadherin and plakoglobin confers strong intercellular adhesion. J. Cell Sci. 111, 495–509 (1998). 89. Kowalczyk, A. P., Borgwardt, J. E. & Green, K. J. Analysis of desmosomal cadherin-adhesive function and stoichiometry of desmosomal cadherin–plakoglobin complexes. J. Invest. Derm. 107, 293–300 (1996). 90. Chidgey, M. A. J., Clarke, J. P. & Garrod, D. R. Expression of full-length desmosomal glycoproteins (desmocollins) is not sufficient to confer strong adhesion on transfected L929 cells. J. Invest. Derm. 106, 689–695 (1996). 91. Tselepis, C., Chidgey, M., North, A. & Garrod, D. Desmosomal adhesion inhibits invasive behavior. Proc. Natl Acad. Sci. USA 95, 8064–8069 (1998). 92. Chitaev, N. A. & Troyanovsky, S. M. Direct Ca2+-dependent heterophilic interaction between desmosomal cadherins, desmoglein and desmocollin, contributes to cell–cell adhesion. J. Cell Biol. 138, 193–201 (1997). The first paper to provide direct evidence for an interaction between desmoglein and desmocollin extracellular domains, consistent with the idea that heterophilic cadherin interactions may be involved in desmosomal adhesion. 93. Savagner, P., Yamada, K. M. & Thiery, J. P. The zinc-finger protein slug causes desmosome dissociation, an initial and necessary step for growth factor-induced epithelial–mesenchymal transition. J. Cell Biol. 137, 1403–1419 (1997). 94. Fuchs, M., Muller, T., Lerch, M. M. & Ullrich, A. Association of human protein-tyrosine phosphatase k with members of the armadillo family. J. Biol. Chem. 271, 16712–16719 (1996). 95. Shibamoto, S. et al. Tyrosine phosphorylation of β-catenin and plakoglobin enhanced by hepatocyte growth factor and epidermal growth factor in human carcinoma cells. Cell Adh. Commun. 1, 295–305 (1994). 96. Stappenbeck, T. S., Lamb, J. A., Corcoran, C. M. & Green, K. J. Phosphorylation of the desmoplakin COOH terminus

negatively regulates its interaction with keratin intermediate filament networks. J. Biol. Chem. 269, 29351–29354 (1994). 97. Aoyama, Y., Owada, M. K. & Kitajima, Y. A pathogenic autoantibody, pemphigus vulgaris-IgG, induces phosphorylation of desmoglein 3, and its dissociation from plakoglobin in cultured keratinocytes. Eur. J. Immunol. 29, 2233–2240 (1999). 98. Sadot, E. et al. Differential interaction of plakoglobin and βcatenin with the ubiquitin-proteasome system. Oncogene 19, 1992–2001 (2000). 99. Merriam, J. M., Rubenstein, A. B. & Klymkowsky, M. W. Cytoplasmically anchored plakoglobin induces a WNT-like phenotype in Xenopus. Dev. Biol. 185, 67–81 (1997). 100. Miller, J. R. & Moon, R. T. Analysis of the signaling activities of localization mutants of β-catenin during axis specification in Xenopus. J. Cell Biol. 139, 229–243 (1997). 101. Charpentier, E., Lavker, R. M., Acquista, E. & Cowin, P. Plakoglobin suppresses epithelial proliferation and hair growth in vivo. J. Cell Biol. 149, 503–520 (2000). 102. Kolligs, F. T. et al. γ-Catenin is regulated by the APC tumor suppressor and its oncogenic activity is distinct from that of β-catenin. Genes Dev. 14, 1319–1331 (2000). 103. Hakimelahi, S. et al. Plakoglobin regulates the expression of the anti-apoptotic protein BCL-2. J. Biol. Chem. 275, 10905–10911 (2000). 104. Kowalczyk, A. P. et al. Posttranslational regulation of plakoglobin expression: Influence of the desmosomal cadherins on plakoglobin metabolic stability. J. Biol. Chem. 269, 31214–31223 (1994). 105. Anastasiadis, P. Z. & Reynolds, A. B. The p120 catenin family: complex roles in adhesion, signaling and cancer. J. Cell Sci. 113, 1319–1334 (2000). 106. Vleminckx, K. & Kemler, R. Cadherins and tissue formation: integrating adhesion and signaling. BioEssays 21, 211–220 (1999). 107. Ruhrberg, C. & Watt, F. M. The plakin family: versatile organisers of cytoskeletal architecture. Curr. Opin. Genet. Dev. 7, 392–397 (1997). 108. Green, K. J., Virata, M. L. A., Elgart, G. W., Stanley, J. R. & Parry, D. A. D. Comparative structural analysis of desmoplakin, bullous pemphigoid antigen and plectin: members of a new gene family involved in organization of intermediate filaments. Int. J. Biol. Macromol. 14, 145–153 (1992). 109. Herrmann, H. & Aebi, U. Intermediate filaments and their associates: multi-talented structural elements specifying cytoarchitecture and cytodynamics. Curr. Opin. Cell Biol. 12, 79–90 (2000). 110. Amagai, M. Autoimmunity against desmosomal cadherins in pemphigus. J. Dermatol. Sci. 20, 92–102 (1999).

Acknowledgements The authors would like to thank all of our colleagues who provided input and information before publication. Thanks go also to A. Kowalczyk, J. Stanley and members of the Green lab for critical reading of the manuscript, and to T. Mauro for helpful discussion. The authors are supported by grants from the NIH to K.G. and a Warner Lambert Consumer Healthcare Research Fellowship from the Dermatology Foundation to C.G..

www.nature.com/reviews/molcellbio

| DECEMBER 2000 | VOLUME 1

REVIEWS

INTRAMEMBRANE PROTEOLYSIS BY PRESENILINS Harald Steiner and Christian Haass Many neurodegenerative diseases involve the deposition of insoluble amyloid molecules. In Alzheimer’s disease, for example, the amyloid β-peptide (Aβ) is the main component of the characteristic senile plaques. Proteolytic enzymes called secretases are involved in generating Aβ, and one of these may have been identified as presenilin — a discovery that paves the way for a more complete understanding of presenilin structure and function. ENDOPROTEOLYSIS

Hydrolysis of internal peptide bonds by proteases. Several classes of protease are discriminated by their characteristic active-site domains. These include serine proteases, cysteine proteases, aspartyl proteases and metalloproteases.

Adolf Butenandt-Institute, Department of Biochemistry, Laboratory for Alzheimer’s Disease Research, LudwigMaximilians University, 80336 Munich, Germany. Correspondence to C.H. e-mail: chaass@pbm.med. uni-muenchen.de

Owing to spectacular advances in biomedical research, life expectancy in developed countries has increased markedly. But the main risk factor for diseases such as Alzheimer’s disease is ageing, so the number of people affected by this devastating disorder is constantly increasing. Progress in understanding the molecular mechanisms underlying Alzheimer’s disease has led to unexpected insights into the process of cellular differentiation, and has also provided the basis for developing possible treatments for those with this disease. Alzheimer’s disease is not the only neurodegenerative disorder to be characterized by the deposition of insoluble amyloid molecules — others include Parkinson’s disease, prion diseases and frontotemporal dementia. These amyloid molecules undergo structural changes that initiate and facilitate their rapid aggregation and precipitation into β-sheeted fibrils1,2, which are believed to be neurotoxic3 and responsible for the disease-specific pathology. Here we focus on the presenilins, proteins that have not only helped us to understand the molecular mechanisms underlying Alzheimer’s disease, but have also taught us about fundamental biological processes, such as the connection of cell-fate decisions with proteolysisdependent signal transduction. Secretases process βAPP

The main insoluble molecule involved in Alzheimer’s disease is amyloid β-peptide (Aβ), which is deposited in the brains of affected people as senile plaques (FIG. 1). Most Aβ peptides terminate after amino acid 40, but a small portion is elongated by two extra car-

boxy-terminal residues. The longer form, Aβ42, is believed to be the main toxic component in Alzheimer’s disease. Aβ is generated from a large protein, the β-amyloid precursor protein (βAPP), by ENDOPROTEOLYSIS4. The βAPP protein is a ubiquitously expressed type I transmembrane domain protein, which is encoded by a gene located on chromosome 21 (REF. 5). Familial Alzheimer’s disease (FAD) has been observed in a small fraction of patients, and autosomal-dominant mutations are located within the βAPP gene (BOX 1). Interestingly, patients with trisomy 21 (Down’s syndrome) develop an early Alzheimer’s-disease-like pathology, with senile plaques that are indistinguishable from those of patients with Alzheimer’s disease. This early deposition may be due simply to the extra copy of the βAPP gene, which could allow higher levels of Aβ to be produced. The proteases involved in generating Aβ are called secretases. Three distinct secretase activities have been identified (FIG. 2), and evidence for their identity is accumulating. The first is the β-secretase, which generates the amino terminus of the Aβ domain (FIG. 2). Two highly homologous β-secretase enzymes — BACE (βsite APP cleaving enzyme) and BACE2 — have been identified (for a review, see REF. 6). Both are membranebound aspartyl proteases, which cleave βAPP in a sequence-specific manner. BACE is thought to be the main β-secretase activity in the brain, whereas BACE2 seems to be located predominantly in peripheral tissues. BACE is produced as a precursor (probably representing the inactive ZYMOGEN). It contains a propeptide at its

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 1 | DECEMBER 2000 | 2 1 7

REVIEWS Finally we have the α-secretase, which cleaves in the centre of the Aβ domain (FIG. 2) and inhibits the generation of Aβ15. Two related proteases seem to exert an αsecretase activity16,17, both of which are metalloproteases of the ADAM (a disintegrin and metalloprotease) family18. Substrate recognition by these proteases depends on the distance of the α-helical structures within the target substrates from the plasma membrane19. Presenilins: key to understanding Alzheimer’s

Figure 1 | Deposition of amyloid β-peptide in the brains of patients with Alzheimer’s disease. Many neurodegenerative disorders — such as prion diseases, Parkinson’s disease, Huntington’s disease, Alzheimer’s disease, frontotemporal dementia — are characterized by the aggregation of amyloidogenic polypeptides. Each disease is characterized by its own characteristic amyloid molecule: prion protein, α-synuclein, huntingtin, amyloid β-peptide (Aβ) and tau, respectively. These molecules have similar biochemical properties, which can force them to precipitate into highly insoluble fibrillar structures. Either the resulting plaques or the protofibrils (oligomeric fibrillar aggregates) are believed to cause the neuronal loss observed in all these diseases. The invariant pathogenic feature of Alzheimer’s disease is the accumulation of senile plaques. These plaques consist mainly of Aβ. The figure shows Aβ depositions in the brain of an Alzheimer´s patient by immunohistochemical staining using an anti-Aβ antibody.

ZYMOGEN

A proteolytically inactive precursor of a protease. Most of these proteases contain a prodomain at the amino terminus, which keeps the corresponding enzyme inactive. The prodomain is removed by endoproteolysis. This can be mediated by other proteases (so zymogens and their activating proteases are often members of a proteolytic cascade), or by autoproteolysis.

amino terminus, which is removed early during maturation — probably by a furin-like protease. BACE has an acidic pH optimum, consistent with the idea that βsecretase processing may occur, at least to some extent, within acidic endosomes7,8. The second secretase is the γ-secretase, which mediates the carboxy-terminal cleavage of βAPP and finally liberates Aβ (FIG. 2). The γ-secretase is an unusual aspartyl protease9. Contrary to all textbook knowledge, this protease seems to cleave βAPP almost in the centre of the transmembrane domain (FIG. 2). It also has loose sequence specificity — at least for the recognition of βAPP10. Aβ is produced under physiological conditions11–14, and the combined activities of the β- and γsecretases lead to its secretion into biological fluids4 (FIG. 2). In fact, the time-bomb ticking in all of us depends on that physiological mechanism, which enables our neurons to produce amyloidogenic Aβ constitutively throughout our lives.

Box 1 | Mutations in the β-amyloid precursor protein gene Mendelian inheritance of Alzheimer’s disease has been observed in a small fraction of patients, and autosomal-dominant mutations are located within the β-amyloid precursor protein ( βAPP) gene. All βAPP mutations are clustered within and around the Aβ domain at three main locations — each mutation is close to one of the three secretase cleavage sites (FIG. 2). When these mutations were investigated in tissue culture, as well as in the human brain, most were found to increase production of the highly amyloidogenic 42-amino-acid form of Aβ (Aβ42)95. This peptide not only aggregates much faster than its shorter counterpart1,2, but is also specifically deposited in senile plaques36. Moreover, Aβ42 also accumulates in senile plaques formed in sporadic cases of Alzheimer’s disease96. So Aβ42 is probably the main pathological component in all cases of Alzheimer’s disease36.

218

Many cases of early-onset FAD are associated with two homologous genes, presenilin (PS)1 and PS2 (REFS 20–22). Some PS1 mutations cause the most aggressive known form of Alzheimer’s disease, with an age of onset under 30 (REF. 23). So, shortly after they were identified by St George Hyslop and colleagues20 in 1995, the presenilin genes were proposed to hold the key to understanding the molecular mechanisms that underlie Alzheimer’s disease — and this turned out to be true. A channel-like structure? Presenilins are polytopic membrane proteins that undergo constitutive endoproteolysis24 in their large cytoplasmic loop (FIG. 3). This cleavage results in the generation of stable amino- and carboxy-terminal fragments (the NTF and CTF, respectively), which bind to each other and form an enzymatically active heterodimeric complex25–28. However, no mixed complexes between the fragments of PS1 and PS2 have been observed29. Work from many laboratories has led to the widely accepted eight-transmembrane domain model30–32, in which the amino terminus, the large cytoplasmic loop and the carboxyl terminus are all located in the cytoplasm. Although the topology of presenilins seems clear between transmembrane domains 1 and 6, their structure downstream of the large cytoplasmic loop is not entirely solved. There are different models33, which may indicate a variable positioning of the cytoplasmic tail in the cytoplasm or the lumen34. At the moment, though, the eight-transmembrane domain model seems the most likely structure. So we use this model throughout the review, although more evidence is required to be sure about the topology and structure of the carboxy-terminal domain. It is tempting to assemble the eight putative transmembrane domains in a way that allows a pore- or channel-forming unit to be predicted. Such a channel may be required to insert substrates into the enzymatically active complex. This model may also help to explain why mutations are spread over the entire PS1 molecule, and do not cluster around a certain hot spot — that is, an active centre of the molecule. The poreforming model would predict that mutations in the various transmembrane domains of PS1 could all affect a putative active site (that is, a catalytic centre) within the inner part of the pore. Mutations in the short loops between the transmembrane domains may interfere with the structure of transmembrane domains, thereby indirectly affecting the active site. Indeed, such an activesite domain has been identified in the middle of transmembrane domains 6 and 7. Assuming that presenilin mutations affect an active www.nature.com/reviews/molcellbio

| DECEMBER 2000 | VOLUME 1

REVIEWS

42 N'

C' Aβ

Aβ

CTFβ

γ-secretase cleavage CTFα

p3 Lumen

Cytoplasm

Figure 2 | Generation of amyloid β-peptide. Amyloid β-peptide (Aβ) is derived by proteolytic processing from its precursor, the β-amyloid precursor protein (βAPP), which is initially cleaved by either β- or α-secretase. This cleavage generates carboxy-terminal membrane-bound fragments, termed CTFβ and CTFα, respectively. These are the immediate substrates for cleavage by the γ-secretase to liberate Aβ (β- and γ-secretase cleavage) or a smaller peptide termed p3 (α- and γ-secretase cleavage). The Aβ40 and Aβ42 forms are probably both generated by the same γ-secretase activity. Carboxy-terminal fragments of βAPP produced by β- or α-secretase cleavage accumulate in cells with reduced or eliminated presenilin function.

site, all presenilin mutations — regardless of where they are located — would be expected to cause the same pathological dysfunction. Indeed, all mutations responsible for FAD that have been analysed so far increase the production of the highly amyloidogenic Aβ42. This was shown originally in primary cells derived from carriers of presenilin mutations35, and subsequently confirmed in transfected cell lines (summarized by Selkoe36) and transgenic mice (summarized by Price et al.37). In such mouse models, expression of both FAD-linked presenilin and βAPP mutants notably increased the tendency of Aβ to precipitate into amyloid plaques37. Moreover, the fact that mutations in both the βAPP (BOX 1) and presenilin genes affect identical pathological mechanisms provides strong evidence for the pivotal role of Aβ in Alzheimer’s disease. Although a direct effect of βAPP mutations on Aβ production was obvious — owing to the position of the mutations close to the sites of secretase cleavage (BOX 1) — it has been much more difficult to understand the role of the presenilins in generating aberrant Aβ. But the answer has turned out to be surprisingly simple.

SOMITE

A series of paired blocks of cells that form during early vertebrate development and give rise to the backbone and body muscle.

Proteolysis-dependent signal transduction. The knockouts of presenilin genes provided spectacular and unexpected insights into their biological function. Mice with a PS1 deletion had at least some phenotypic similarities to mice lacking Notch1 (REF. 38), a key molecule in developmental signal transduction39. In these mice, problems in SOMITE and rib formation were observed, as well as cerebral haemorrhage40,41. Furthermore, a double knockout of PS1 and PS2 was almost indistinguishable from the developmental deficits that result from the deletion of the Notch1 gene42,43. Genetic evidence from a simple model system, the worm Caenorhabditis elegans, also supported a function for the presenilins in cell-fate decisions44. The phenotype of sel-12 mutations, which seem to be associated with a failure in Notch signalling, could be

rescued by the transgenic expression of human PS1 and PS2 genes45–49. Similar results — again showing a pivotal function for the presenilins in the Notch signalling cascade — were also observed in another model system, the fruitfly Drosophila melanogaster50,51. But how can a function in Notch signalling, and therefore cellular differentiation and embryogenesis, be related to the malfunction of mutant presenilins in Aβ production? The answer came again from the PS1 knockout. A ground-breaking experiment by De Strooper and colleagues52 showed that primary neurons derived from the brains of PS1 knockout animals could not efficiently produce Aβ or a smaller peptide termed p3. Moreover, these cells accumulated high levels of βAPP carboxyterminal fragments generated by α- or β-secretase. These fragments are the immediate precursors of Aβ and p3 production (FIG. 2). So in the absence of PS1, a defect in γ-secretase activity apparently leads to reduced generation of Aβ and p3 and a build-up of γ-secretase substrates. A logical conclusion is that the presenilins are directly required for γ-secretase function. But how can that be connected with a failure in Notch signalling in the same animals? For a long time, the signal transduction initiated after the binding of Notch to its natural ligands, such as members of the DSL (Delta/Serrate/Lag-2) family, was not understood. Coincidentally, work in Drosophila, as well as in transfected cells, revealed that Notch undergoes endoproteolysis after binding to its ligands53,54. The Notch intracellular domain (NICD) is released by a proteolytic cleavage, which occurs at, or close to, the cytoplasmic side of the plasma membrane (FIG. 4). The NICD then translocates to the nucleus, where it binds to transcription factors of the CSL (CPB/SuH/LAG-1) family and regulates the transcription of target genes. The final evidence that endoproteolytic release of

NTF

CTF

Figure 3 | Presenilins are endoproteolytically processed. Endoproteolytic cleavage of the presenilins (PS) occurs within the large hydrophilic loop between transmembrane domains 6 and 7. The blue box represents the cleavage-site domain. Cleavage of PS1 occurs at, or close to, methionine 292 (REF. 76). Cleavage of PS2 occurs at, or close to, Met 298 (REF. 77). The resulting amino- and carboxy-terminal fragments (NTF and CTF) accumulate as stable heterodimers.

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 1 | DECEMBER 2000 | 2 1 9

REVIEWS bers of the ADAM family16,17. Ectodomain cleavage of Notch and βAPP may cause structural changes in the remaining carboxy-terminal fragment, facilitating the third (S3) cut by an intramembrane processing step mediated by a γ-secretase-like protease activity61. Notch and βAPP are not the only substrates for presenilindependent endoproteolysis. The amyloid-precursor-like proteins (APLPs), which are homologous to βAPP, may also be cleaved in a PS1-dependent manner63 — again, probably within their transmembrane domains.

Delta/Serrate/Lag-2 Cytoplasm

Extracellular space

S1 cleavage

S2 cleavage

S1 (furin family)

Presenilins: unusual aspartyl proteases?

S2 (ADAM family) S3 (γ-secretase-like) S3 cleavage Notch1 Cytoplasm

NICD

CSL

No signalling Transcription of target genes

Figure 4 | Presenilins and Notch signalling. Notch signalling depends on three endoproteolytic cleavages. Notch is first cleaved at site 1 (S1) during maturation in the Golgi39. The cleaved heterodimer binds at the cell surface to its ligands (members of the DSL (Delta/Serrate/Lag-2) family) on a neighbouring cell. Cleavage at S2 then generates a small, membrane-bound fragment (NEXT, for Notch extracellular truncation)61, which can now be cleaved at S3 (REF. 53). This cleavage is reminiscent of the γ-secretase cleavage of βAPP, and liberates the Notch intracellular domain (NICD), which is translocated to the nucleus. There it regulates transcription of target genes by binding to transcription factors of the CSL (for CPB/SuH/LAG-1) family39.

NICD is absolutely required for Notch signalling was provided by mutagenesis of the intramembrane cleavage site of Notch1 (REF. 55). Mice homozygous for a Notch1 gene variant that carries the V1744G mutation, which blocks its endoproteolysis, developed a phenotype resembling that caused by deletion of the Notch1 gene. Proteolytic processing of Notch is remarkably similar to that of βAPP because both occur within, or close to, the membrane. But cleavage of Notch is not only similar to that of βAPP — it also depends on the presenilins. Neurons that lack PS1 show a pronounced reduction in generation of the NICD56,57. Moreover, a deletion of both PS1 and PS2 completely abolished Aβ production and generation of the NICD58,59. So both presenilins seem to facilitate similar intramembrane endoproteolytic cleavages: they support the liberation of membrane-associated proteins, which are subsequently secreted or transported to the nucleus. This explains why the PS1 knockout affects both Notch signalling and Aβ production. However, there is a further — and striking — similarity between βAPP and Notch processing. Both molecules require a cleavage in their ectodomain before the presenilin-dependent final cut (FIG. 4). In the case of βAPP, this is the α- or β-secretase cleavage. For Notch, it is a cleavage by a furin-like protease at site 1 (S1)60, followed by a second cut (S2) mediated by the tumour necrosis factor-α (TNF-α)-converting enzyme TACE, a member of the ADAM family61,62. Similarly to S2 cleavage of Notch, the α-secretase cleavage of βAPP is also mediated by mem-

220

All the results discussed above indicate that the presenilins directly affect endoproteolysis of selected membrane proteins. So a logical conclusion is that presenilins are the γ-secretase. There is evidence that γ-secretase activity can be specifically blocked by aspartyl protease inhibitors9,64. On the basis of these findings, Wolfe and colleagues65 searched for critical aspartate residues in the presenilin proteins that could represent an active centre of a ‘presenilin-aspartyl’ protease. Although they found no protease active-site domains containing the expected consensus sequence D(T/S)G(T/S), which is typical for all aspartyl proteases known so far, Wolfe and co-workers identified two interesting aspartyl residues within transmembrane domains 6 and 7 (FIG. 5). Mutation of critical aspartate residues in conventional aspartyl proteases is known to inhibit their proteolytic function, and, when either of the two critical aspartates in PS1 was mutated, a striking result was observed65: presenilin endoproteolysis was inhibited and carboxy-terminal fragments of βAPP generated by α- or β-secretase accumulated65 (FIG. 2), which immunoprecipitated together with PS1 and PS2 (REF. 66). Furthermore, Aβ production was markedly inhibited65. In other words, inhibition of the γ-secretase activity seems to be associated with the lack of presenilin-dependent endoproteolysis of βAPP. This also suggests that the full-length protein is a zymogen, which can be activated by autoproteolysis (FIG. 5). Similar results were obtained when the corresponding aspartate residues of PS2 were mutated47,67. Moreover, the aspartates are functionally conserved during evolution68 and all of the aspartate mutant derivatives of PS1 and PS2 inhibited Notch endoproteolysis47,48,69 and did not rescue the sel-12 mutant phenotype in C. elegans47,48,70. These data are also consistent with the finding that the presenilins form a complex composed of their amino- and carboxy-terminal fragments — both fragments would be required to form the active centre of a bilobed aspartyl protease. For example, an artificial presenilin molecule corresponding to the amino-terminal fragment alone does not rescue the sel-12 mutant phenotype46 and does not induce the generation of Aβ42 when it contains a mutation associated with FAD71–73. Wolfe and Selkoe therefore concluded65 that the presenilins and the γ-secretase are one and the same. In this model, substrates may be cleaved in a channel-like pore of the presenilins (FIG. 5). If the enzyme’s catalytic sites faced the channel, this may also explain the identical effects of the many mutations on γ-secretase cleavage of βAPP. Although alternative interpretations may be possible www.nature.com/reviews/molcellbio

| DECEMBER 2000 | VOLUME 1

REVIEWS

UNFOLDED PROTEIN RESPONSE

An intracellular signalling pathway that connects the endoplasmic reticulum (ER) with the nucleus. Under stress conditions (when unfolded proteins accumulate in the ER), cells react by the increased transcription of chaperone genes. These chaperones are required for maintenance of protein folding. CAPACITATIVE CALCIUM ENTRY

A cellular mechanism required for refilling of intracellular calcium stores.

(BOX 2), this intriguing hypothesis is now supported by

elegant experiments. Inhibitors of the γ-secretase were covalently crosslinked to the amino- and carboxy-terminal fragments of the presenilins74,75. Interestingly, these inhibitors bind exclusively to the presenilin fragments — not to the full-length proteins — indicating that the heterodimeric presenilin complex is indeed the biologically active version of presenilins. But it is still not clear whether the full-length presenilin protein is a zymogen activated by autoproteolysis, because several artificial cleavage-site mutations block endoproteolysis without affecting Aβ production76,77 and Notch signalling76,77. Attempts to reconstitute the γ-secretase activity in vitro also implicate the presenilins in endoproteolysis of βAPP. Early experiments25–28 indicated that presenilin fragments not only form a heterodimer, but that they are also part of a high-molecular-weight complex. Using the extraction methods described by

Box 2 | Alternative functions of presenilins Although an aspartyl protease activity of presenilins seems likely, there is evidence for other functions and alternative explanations of the current data.

The spatial paradox. The cellular distribution of presenilins does not reflect that of

their substrates, such as βAPP and Notch. These substrates undergo presenilindependent cleavage at, or close to, the cell surface, whereas presenilins are restricted to the endoplasmic reticulum and early Golgi97. This paradox may have been solved by the identification of small amounts of presenilins on the cell surface69, and perhaps in endosomes98; a localization consistent with the site of γ-secretase cleavage.

Differential support of βAPP and Notch endoproteolysis. Selective artificial mutations of PS1 block Notch endoproteolysis and Notch signalling, but still allow Aβ production48,99. This may be due to differential effects of the substrate-binding site99; to the participation of endogenous wild-type PS1 in formation of a presenilin complex48; or to the involvement of two proteases with γ-secretase activity in production of the Notch intracellular domain and Aβ48. Subcellular transport. The reduced γ-secretase cleavage in neurons from PS1-knockout animals may be due to subtle effects on the subcellular transport of presenilin substrates63, and may indicate a function of presenilins in selective protein transport. The unfolded protein response. Presenilins have been implicated in the UNFOLDED 100 PROTEIN RESPONSE (UPR) pathway (although other evidence indicates that they are not101 ), which may be associated with neuronal loss in Alzheimer’s disease102. The UPR is downregulated both by a PS1 knockout102 and FAD-associated PS1 mutations103. Downregulation of the UPR is suggested to be due to failure in a γsecretase-like cleavage of the ER stress sensor Ire1102, a type I transmembrane protein required for expression of several UPR chaperone genes100.

Regulators of apoptosis? Many studies indicate that presenilins may affect cellular survival. Mutations associated with familial Alzheimer’s disease may facilitate induction of apoptosis104,105, causing cell death, although it is not understood how such a function could be related to a proteolytic activity of presenilins. However, it is interesting to speculate that presenilins could activate the apoptotic cascade by liberating a cytoplasmic fragment from the Fas death receptor. Memory defects. Presenilins rescue a memory deficit in sel-12 mutant worms49. However, the function of presenilins in memory and axonal guidance has been attributed to a facilitation of Notch signalling49, which is consistent with their proteolytic function.

Calcium entry. Presenilins affect CAPACITATIVE CALCIUM ENTRY (CCE)106. A PS1 knockout, as well as mutations in the active-site aspartate, potentiates CCE, whereas FADassociated presenilin mutations attenuate CCE. It is not clear whether presenilins modulate calcium entry through their proteolytic function, because γ-secretase inhibitors seem to have no effect on CCE.

Capell and colleagues26, an in vitro assay for γ-secretase activity has now been established78. This assay strictly requires a high-molecular-weight presenilin complex78, which may indeed indicate that the γ-secretase activity is a large, multi-subunit protease complex similar to the proteasome. It is tempting to speculate that such a ‘secretosome’79/‘secretasome’ complex may contain components that determine the specificity of γ-secretase cleavage, as in βAPP or Notch cleavage, for instance. In fact, a type I transmembrane protein called nicastrin has been identifed79 as a component of the presenilin complex that seems to modulate γ-secretase activity. Nicastrin binds to βAPP CTFs and presenilins, and may regulate the access or positioning of γ-secretase to its substrates. Moreover, FAD-associated presenilin mutations alter the interaction of nicastrin with βAPP CTFs79. So it is possible that FAD mutations affect not only the active site of presenilin (see above) but also the nicastrin/βAPP CTF interaction and the subsequent γsecretase cut. Interestingly, suppression of nicastrin expression also blocks Notch signalling in C. elegans 79. A new active-site motif within presenilins? If the presenilins are indeed identical with the γ-secretase activity, why do they share no sequence homology with typical aspartyl proteases? Perhaps they are unprecedented proteases — owing to their intramembrane cleavage — and so are completely unique. However, there is evidence that presenilins may use a protease active site shared with certain bacterial proteases. This work was initiated by the finding that a naturally occurring mutation associated with FAD is located immediately adjacent to the critical aspartate 385 (REFS 80,81). This mutation, G384A, causes a sixfold increase in production of Aβ42 (REFS 82,83) — the most pronounced increase ever recorded for a PS1 mutation. (The PS1 ∆exon9 mutation, which previously had the strongest effect on generation of Aβ42, increased this highly amyloidogenic variant by only a factor of about three84.) So amino acid 384 must be of particular importance for PS1 function. This is indeed the case, as artificial mutations inserted at this position inhibit the generation of both Aβ and the NICD83. The effect of amino-acid exchanges at position 384 therefore differs fundamentally from that of exchanges at the neighbouring critical aspartate. Whereas all aspartate mutations at position 385 inhibit the function of presenilins in the generation of Aβ and Notch endoproteolysis/signalling, mutations of residue 384 can either markedly increase the pathological effects of the presenilins or decrease their biological activities. Glycine 384 seems to be important for presenilin function because it is conserved in all other members of the presenilin family, including distant members such as spe-4 of C. elegans83 (FIG. 6). Moreover, sequence comparisons also revealed a highly conserved motif around aspartate 385 and, when this motif was used to screen databases, considerable homology to a family of recently identified bacterial aspartyl proteases was found83 (FIG. 6). These proteases belong to the family of type 4 prepilin peptidases (TFPPs). The TFPPs are polytopic proteins containing

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 1 | DECEMBER 2000 | 2 2 1

REVIEWS eight transmembrane domains, and they are found in many prokaryotes85. In TFPPs, the critical aspartates are located close to transmembrane domains (or, in three cases, probably within the membrane83; see FIG. 6). These aspartates are not only fully conserved in all members of the TFPP family, but their mutation completely blocks the proteolytic activity of these proteases85. What is the function of the TFPPs, and is there any similarity with presenilin function? The TFPPs are required for removal of the hydrophobic leader sequence of type 4 prepilins, precursors of a subtype of bacterial pili. The cleavage occurs at the cytosolic site of the plasma membrane, and results in immediate secretion of the mature substrate. The secreted prepilins then polymerize in the extracellular space and form the pili. So this process is remarkably similar to cleavage mediated by the eukaryotic γ-secretases. Moreover, the TFPPs, as well as the presenilins, lack the typical D(T/S)G(T/S) motif found in all conventional aspartyl proteases.

Autoproteolysis?

Instead, they seem to share a G(A)X′GDX′′ motif (where X′ is variable, and X′′ is preferably a hydrophobic amino acid; FIG. 6) around the carboxy-terminal aspartyl residue. Although a slightly similar motif may also be found in K+-channels, the aspartyl residue, which is functionally required in all presenilins and TFPPs, is not conserved in several of these channels. There may also be homology around the amino-terminal aspartyl residue. In that case, a DXXXXLXP motif (where X represents variable amino acids; leucine and proline are preferred residues in all presenilins and TFPPs) may be conserved (FIG. 6). Although these motifs seem to be conserved, there is no other homology between the TFPPs and the presenilins. However, this is similar to the case of the site 2 protease (S2P)86, a polytopic metalloprotease that carries out an intramembrane endoproteolyic cleavage similar to that of the γ-secretase. S2P contains a classical metalloprotease active-site HEXXH motif (where X is typically a non-charged amino acid) within a hydrophobic domain. Although S2P uses a typical protease active site, similar to those in presenilins and TFPPs, S2P shares no sequence homology outside its catalytic centre with other metalloproteases that do not belong to the S2P family. However, more evidence is needed before we can conclude that the two conserved-sequence motifs around the critical aspartates of the presenilins and the TFPPs are part of a novel protease active site. Nevertheless, the remarkable sequence conservation indicates possible convergent evolution of the protease active sites in TFPPs and presenilins. It would therefore not be too surprising if more polytopic eukaryotic proteases with a conserved glycine/aspartate motif were found in the near future.

NTF

A vision for the near future?

CTF Aβ β-secretase

Aβ

p3 α-secretase

Aβ

Figure 5 | Are presenilins aspartyl proteases with γ-secretase function? Critical aspartate residues in transmembrane domains 6 and 7 indicate that presenilins may be new aspartyl proteases with γ-secretase function. According to this model65, βAPP carboxyterminal fragments generated by either β- or α-secretase are internally cleaved by the intrinsic aspartyl protease activity of presenilin, which itself is activated by autoproteolysis. ( NTF, amino-terminal fragment; CTF, carboxy-terminal fragment.)

222

Who would have believed, only two years ago, that we would have all the secretases to hand by the turn of the century? Obvious possibilities for treating Alzheimer’s disease now emerge from our detailed knowledge of the molecular mechanisms that underlie this devastating condition. For example, we now have highly specific β-87 and γ-secretase-specific inhibitors9,74,75,88 and, using wellcharacterized transgenic mouse models for Alzheimer’s disease (for a review see Price et al.37), they can be tested for their ability to inhibit the formation of senile plaques. Indeed, the first generation of γ-secretase inhibitors is in Phase I trials. However, inhibitors (as well as inhibitor concentrations) must be selected that reduce Aβ production but allow enough production of the NICD to maintain Notch signalling, even in adults. If that is not possible, other intriguing approaches to slowing Alzheimer’s disease are now available. These include the stimulation of proteases responsible for Aβ clearance89–91 and immunization with anti-Aβ antibodies92,93, which protects transgenic mouse models against the symptoms of Alzheimer’s disease. A general function of presenilins

The substrate specificity of the presenilin-mediated intramembrane proteolysis seems to be loose; indeed, www.nature.com/reviews/molcellbio

| DECEMBER 2000 | VOLUME 1

REVIEWS a Presenilin family

YDLVAVLCP

GLGDF

TFPP family

3 D

IDADTQLLP

GYGDF

G(A)X'GDX"

Figure 6 | Sequence comparison of the peptidase family with the type 4 prepilin peptidase family. a | Type 4 prepilin peptidases (TFPPs) and presenilins share sequence homology around their critical aspartate residues. Arrowheads indicate the critical aspartate residues of presenilins and TFPPs. Blue arrowheads denote familial Alzheimer’s disease (FAD) mutations in the vicinity of the critical aspartate residues of PS1. b | Presenilins and TFPPs appear to share a common G(A)X′GDX′′ motif (dark grey box; for details see text) around the carboxy-terminal active-site aspartate residue.

there is now evidence that presenilin-mediated endoproteolysis may not depend on the primary sequence of the substrate at all94. Rather, substrate recognition seems to require prior truncation of the ectodomain by ‘shedding’ enzymes. In fact, the known substrates of presenilins, such as βAPP, APLP and Notch, all undergo an ectodomain cleavage before the intramembrane proteolysis. Struhl and Adachi94 now show that the size of the ectodomain that remains after the initial cleavage determines whether or not presenilin-mediated cleavage will occur: the smaller the remaining ectodomain, the higher

Teplow, D. B. Structural and kinetic features of amyloid βprotein fibrillogenesis. Amyloid 5, 121–142 (1998). Lansbury, P. T. Jr Structural neurology: are seeds at the root of neuronal degeneration? Neuron 19, 1151–1154 (1997). 3. Yankner, B. A. Mechanisms of neuronal degeneration in Alzheimer’s disease. Neuron 16, 921–932 (1996). 4. Haass, C. & Selkoe, D. J. Cellular processing of β-amyloid precursor protein and the genesis of amyloid β-peptide. Cell 75, 1039–1042 (1993). 5. Kang, J. et al. The precursor of Alzheimer’s disease amyloid A4 protein resembles a cell-surface receptor. Nature 325, 733–736 (1987). 6. Vassar, R. & Citron, M. Aβ-generating enyzmes: recent advances in β- and γ-secretase research. Neuron 27, 419–422 (2000). 7. Haass, C., Koo, E. H., Mellon, A., Hung, A. Y. & Selkoe, D. J. Targeting of cell-surface β-amyloid precursor protein to lysosomes: alternative processing into amyloid-bearing fragments. Nature 357, 500–503 (1992). 8. Perez, R. G., Squazzo, S. L. & Koo, E. H. Enhanced release of amyloid β-protein from codon 670/671 ‘Swedish’ mutant β-amyloid precursor protein occurs in both secretory and endocytic pathways. J. Biol. Chem. 271, 9100–9107 (1996). 9. Wolfe, M. S. et al. Peptidomimetic probes and molecular modeling suggest that Alzheimer’s γ-secretase is an intramembrane-cleaving aspartyl protease. Biochemistry 38, 4720–4727 (1999). 10. Lichtenthaler, S. F. et al. Mechanism of the cleavage specificity of Alzheimer’s disease γ-secretase identified by 1.

11.

12.

13.

14.

15.

16.

17.

18.

the probability for the subsequent intramembrane cut. These findings indicate that presenilins may have a function in the general degradation of transmembrane domains. So the presenilins may have evolved as special proteases required for the removal of transmembrane domains. Later during evolution, an extra function in signal transduction in the Notch pathway may have been obtained. Clearly, this cleavage can occur constitutively in an unregulated manner53,54. However, because generation of the NICD also depends on earlier truncation of the ectodomain (FIG. 4), presenilin-mediated cleavage (and, consequently, Notch signalling itself) is highly regulated. Such a general function of presenilins in transmembrane domain degradation may, in fact, be the cause of the Alzheimer’s disease problem. βAPP just happens to be a suitable substrate for PS-mediated endoproteolysis, and unfortunately this cleavage results in the production of Aβ. There was no evolutionary pressure on this cleavage, since we simply used to die long before enough Aβ had accumulated to allow aggregation and plaque formation. However, thanks to the tremendous medical advances the situation changed dramatically and we now frequently reach an age that allows sufficient Aβ accumulation to cause Alzheimer’s diease pathology. Thus modern medical research produced a never-ending story: we prevent many health problems but in the same moment we create new challenges. At least this will keep the many Alzheimer’s disease researchers busy even after the Alzheimer’s disease problem has finally been solved. Links DATABASE LINKS Alzheimer’s disease | Parkinson’s

phenylalanine-scanning mutagenesis of the transmembrane domain of the amyloid precursor protein. Proc. Natl Acad. Sci. USA 96, 3053–3058 (1999). Haass, C. et al. Amyloid β-peptide is produced by cultured cells during normal metabolism. Nature 359, 322–325 (1992). Busciglio, J., Gabuzda, D. H., Matsudaira, P. & Yankner, B. A. Generation of β-amyloid in the secretory pathway in neuronal and nonneuronal cells. Proc. Natl Acad. Sci. USA 90, 2092–2096 (1993). Shoji, M. et al. Production of the Alzheimer amyloid βprotein by normal proteolytic processing. Science 258, 126–129 (1992). Seubert, P. et al. Isolation and quantification of soluble Alzheimer’s β-peptide from biological fluids. Nature 359, 325–327 (1992). Esch, F. S. et al. Cleavage of amyloid β-peptide during constitutive processing of its precursor. Science 248, 1122–1124 (1990). Lammich, S. et al. Constitutive and regulated α-secretase cleavage of Alzheimer’s amyloid precursor protein by a disintegrin metalloprotease. Proc. Natl Acad. Sci. USA 96, 3922–3927 (1999). Buxbaum, J. D. et al. Evidence that tumor necrosis factor-α converting enzyme is involved in regulated α-secretase cleavage of the Alzheimer amyloid protein precursor. J. Biol. Chem. 273, 27765–27767 (1998). Primakoff, P. & Myles, D. G. The ADAM gene family: surface proteins with adhesion and protease activity. Trends Genet. 16, 83–87 (2000).

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

19. Sisodia, S. S. β-Amyloid precursor protein cleavage by a membrane-bound protease. Proc. Natl Acad. Sci. USA 89, 6075–6079 (1992). 20. Sherrington, R. et al. Cloning of a gene bearing missense mutations in early-onset familial Alzheimer’s disease. Nature 375, 754–760 (1995). 21. Rogaev, E. I. et al. Familial Alzheimer’s disease in kindreds with missense mutations in a gene on chromosome 1 related to the Alzheimer’s disease type 3 gene. Nature 376, 775–778 (1995). 22. Levy-Lahad, E. et al. Candidate gene for the chromosome 1 familial Alzheimer’s disease locus. Science 269, 973–977 (1995). 23. Wisniewski, T. et al. A novel Polish presenilin-1 mutation (P117L) is associated with familial Alzheimer’s disease and leads to death as early as the age of 28 years. Neuroreport 9, 217–221 (1998). 24. Thinakaran, G. et al. Endoproteolysis of presenilin 1 and accumulation of processed derivatives in vivo. Neuron 17, 181–190 (1996). First demonstration that presenilins undergo endoproteolysis and occur as stable amino- and carboxy-terminal fragments in vivo. 25. Seeger, M. et al. Evidence for phosphorylation and oligomeric assembly of presenilin 1. Proc. Natl Acad. Sci. USA 94, 5090–5094 (1997). 26. Capell, A. et al. The proteolytic fragments of the Alzheimer’s disease-associated presenilin-1 form heterodimers and occur as a 100–150-kDa molecular mass complex. J. Biol. Chem. 273, 3205–3211 (1998).

VOLUME 1 | DECEMBER 2000 | 2 2 3

REVIEWS 27. Yu, G. et al. The presenilin 1 protein is a component of a high molecular weight intracellular complex that contains βcatenin. J. Biol. Chem. 273, 16470–16475 (1998). 28. Thinakaran, G. et al. Stable association of presenilin derivatives and absence of presenilin interactions with APP. Neurobiol. Dis. 4, 438–453 (1998). 29. Saura, C. A. et al. Evidence that intramolecular associations between presenilin domains are obligatory for endoproteolytic processing. J. Biol. Chem. 274, 13818–13823 (1999). 30. Doan, A. et al. Protein topology of presenilin 1. Neuron 17, 1023–1030 (1996). 31. Li, X. & Greenwald, I. Membrane topology of the C. elegans SEL-12 presenilin. Neuron 17, 1015–1021 (1996). 32. Li, X. & Greenwald, I. Additional evidence for an eighttransmembrane-domain topology for Caenorhabditis elegans and human presenilins. Proc. Natl Acad. Sci. USA 95, 7109–7114 (1998). 33. Lehmann, S., Chiesa, R. & Harris, D. A. Evidence for a sixtransmembrane domain structure of presenilin 1. J. Biol. Chem. 272, 12047–12051 (1997). 34. Nakai, T. et al. Membrane topology of Alzheimer’s diseaserelated presenilin 1. Evidence for the existence of a molecular species with a seven membrane-spanning and one membrane-embedded structure. J. Biol. Chem. 274, 23647–23658 (1999). 35. Scheuner, D. et al. Secreted amyloid β-protein similar to that in the senile plaques of Alzheimer’s disease is increased in vivo by the presenilin 1 and 2 and APP mutations linked to familial Alzheimer’s disease. Nature Med. 2, 864–870 (1996). Familial Alzheimer’s disease-linked mutations of PS1 and PS2 enhance Aβ42 production in vivo. 36. Selkoe, D. J. Translating cell biology into therapeutic advances in Alzheimer’s disease. Nature 399, A23–A31 (1999). 37. Price, D. L., Tanzi, R. E., Borchelt, D. R. & Sisodia, S. S. Alzheimer’s disease: genetic studies and transgenic models. Annu. Rev. Genet. 32, 461–493 (1998). 38. Conlon, R. A., Reaume, A. G. & Rossant, J. Notch1 is required for the coordinate segmentation of somites. Development 121, 1533–1545 (1995). 39. Mumm, J. S. & Kopan, R. Notch signaling: From the outside in. Dev. Biol. (in the press). 40. Wong, P. C. et al. Presenilin 1 is required for Notch1 and DII1 expression in the paraxial mesoderm. Nature 387, 288–292 (1997). 41. Shen, J. et al. Skeletal and CNS defects in Presenilin-1deficient mice. Cell 89, 629–639 (1997). 42. Herreman, A. et al. Presenilin 2 deficiency causes a mild pulmonary phenotype and no changes in amyloid precursor protein processing but enhances the embryonic lethal phenotype of presenilin 1 deficiency. Proc. Natl Acad. Sci. USA 96, 11872–11877 (1999). 43. Donoviel, D. B. et al. Mice lacking both presenilin genes exhibit early embryonic patterning defects. Genes Dev. 13, 2801–2810 (1999). 44. Levitan, D. & Greenwald, I. Facilitation of lin-12-mediated signalling by sel-12, a Caenorhabditis elegans S182 Alzheimer’s disease gene. Nature 377, 351–354 (1995). 45. Levitan, D. et al. Assessment of normal and mutant human presenilin function in Caenorhabditis elegans. Proc. Natl Acad. Sci. USA 93, 14940–14944 (1996). 46. Baumeister, R. et al. Human presenilin-1, but not familial Alzheimer’s disease (FAD) mutants, facilitate Caenorhabditis elegans Notch signalling independently of proteolytic processing. Genes Funct. 1, 149–159 (1997). 47. Steiner, H. et al. A loss of function mutation of presenilin-2 interferes with amyloid β-peptide production and Notch signaling. J. Biol. Chem. 274, 28669–28673 (1999). 48. Capell, A. et al. Presenilin-1 differentially facilitates endoproteolysis of the β-amyloid precursor protein and Notch. Nature Cell Biol. 2, 205–211 (2000). 49. Wittenburg, N. et al. Presenilin is required for proper morphology and function of neurons in C. elegans. Nature 406, 306–309 (2000). 50. Ye, Y., Lukinova, N. & Fortini, M. E. Neurogenic phenotypes and altered Notch processing in Drosophila Presenilin mutants. Nature 398, 525–529 (1999). 51. Struhl, G. & Greenwald, I. Presenilin is required for activity and nuclear access of Notch in Drosophila. Nature 398, 522–525 (1999). 52. De Strooper, B. et al. Deficiency of presenilin-1 inhibits the normal cleavage of amyloid precursor protein. Nature 391, 387–390 (1998). Pioneering work showing that PS1-knockout cells are deficient in γ-secretase cleavage of βAPP. 53. Schroeter, E. H., Kisslinger, J. A. & Kopan, R. Notch-1 signalling requires ligand-induced proteolytic release of intracellular domain. Nature 393, 382–386 (1998). 54. Struhl, G. & Adachi, A. Nuclear access and action of Notch in vivo. Cell 93, 649–660 (1998). 55. Huppert, S. S. et al. Embryonic lethality in mice homozygous for a processing-deficient allele of Notch1. Nature 405, 966–970 (2000).

224

56. De Strooper, B. et al. A presenilin-1-dependent γsecretase-like protease mediates release of Notch intracellular domain. Nature 398, 518–522 (1999). PS1 deficiency as well as γ-secretase inhibitors block the γ-secretase-like cleavage of Notch. 57. Song, W. et al. Proteolytic release and nuclear translocation of Notch-1 are induced by presenilin-1 and impaired by pathogenic presenilin-1 mutations. Proc. Natl Acad. Sci. USA 96, 6959–6963 (1999). 58. Zhang, Z. et al. Presenilins are required for γ-secretase cleavage of βAPP and transmembrane cleavage of Notch1. Nature Cell Biol. 2, 463–465 (2000). 59. Herreman, A. et al. Total inactivation of γ-secretase activity in presenilin-deficient embryonic stem cells. Nature Cell Biol. 2, 461–462 (2000). 60. Logeat, F. et al. The Notch1 receptor is cleaved constitutively by a furin-like convertase. Proc. Natl Acad. Sci. USA 95, 8108–8112 (1998). 61. Mumm, J. S. et al. A ligand-induced extracellular cleavage regulates γ-secretase-like proteolytic activation of Notch1. Mol. Cell 5, 197–206 (2000). 62. Brou, C. et al. A novel proteolytic cleavage involved in Notch signaling: the role of the disintegrin-metalloprotease TACE. Mol. Cell 5, 207–216 (2000). 63. Naruse, S. et al. Effects of PS1 deficiency on membrane protein trafficking in neurons. Neuron 21, 1213–1221 (1998). 64. Shearman, M. S. et al. L-685,458, an aspartyl protease transition state mimic, is a potent inhibitor of amyloid βprotein precursor γ-secretase activity. Biochemistry 39, 8698–8704 (2000). 65. Wolfe, M. S. et al. Two transmembrane aspartates in presenilin-1 required for presenilin endoproteolysis and γsecretase activity. Nature 398, 513–517 (1999). Mutational analysis of two conserved aspartates within TM6 and TM7 indicates that presenilins may be unusual aspartyl proteases identical with γsecretase. 66. Xia, W. et al. Presenilin complexes with the C-terminal fragments of amyloid precursor protein at the sites of amyloid β-protein generation. Proc. Natl Acad. Sci. USA 97, 9299–9304 (2000). 67. Kimberly, W. T., Xia, W., Rahmati, T., Wolfe, M. S. & Selkoe, D. J. The transmembrane aspartates in presenilin 1 and 2 are obligatory for γ-secretase activity and amyloid β-protein generation. J. Biol. Chem. 275, 3173–3178 (2000). 68. Leimer, U. et al. Zebrafish (Danio rerio) presenilin promotes aberrant amyloid β-peptide production and requires a critical aspartate residue for its function in amyloidogenesis. Biochemistry 38, 13602–13609 (1999). 69. Ray, W. J. et al. Cell surface presenilin-1 participates in the γ-secretase-like proteolysis of Notch. J. Biol. Chem. 274, 36801–36807 (1999). 70. Brockhaus, M. et al. Caspase-mediated cleavage is not required for the activity of presenilins in amyloidogenesis and Notch signaling. Neuroreport 9, 1481–1486 (1998). 71. Steiner, H. et al. Expression of Alzheimer’s diseaseassociated presenilin-1 is controlled by proteolytic degradation and complex formation. J. Biol. Chem. 273, 32322–32331 (1998). 72. Tomita, T. et al. Molecular dissection of domains in mutant presenilin 2 that mediate overproduction of amyloidogenic forms of amyloid β-peptides. Inability of truncated forms of PS2 with familial Alzheimer’s disease mutation to increase secretion of Aβ42. J. Biol. Chem. 273, 21153–21160 (1998). 73. Citron, M. et al. Additive effects of PS1 and APP mutations on secretion of the 42-residue amyloid β-protein. Neurobiol. Disord. 5, 107–116 (1998). 74. Li, Y.-M. et al. Photoactivated γ-secretase inhibitors directed to the active site covalently label presenilin 1. Nature 405, 689–694 (2000). 75. Esler, W. P. et al. Transition-state analogue inhibitors of γsecretase bind directly to presenilin-1. Nature Cell Biol. 2, 428–433 (2000). References 74 and 75 show that transition-state analogue inhibitors bind to presenilins, indicating that presenilins may contain the active site of γsecretase. 76. Steiner, H. et al. Amyloidogenic function of the Alzheimer’s disease-associated presenilin 1 in the absence of endoproteolysis. Biochemistry 38, 14600–14605 (1999). 77. Jacobsen, H. et al. The influence of endoproteolytic processing of familial Alzheimer’s disease presenilin 2 on Aβ42 amyloid peptide formation. J. Biol. Chem. 274, 35233–35239 (1999). 78. Li, Y.-M. et al. Presenilin 1 is linked with γ-secretase activity in the detergent solubilized state. Proc. Natl Acad. Sci. USA 97, 6138–6143 (2000). Identification of nicastrin, a component of the presenilin complex that modulates γ-secretase activity. 79. Yu, G. et al. Nicastrin modulates presenilin-mediated notch/glp-1 signal transduction and βAPP processing. Nature 407, 48–54 (2000).

80. Cruts, M. et al. Molecular genetic analysis of familial earlyonset Alzheimer’s disease linked to chromosome 14q24.3. Hum. Mol. Genet. 4, 2363–2371 (1995). 81. Tanahashi, H. et al. Sequence analysis of presenilin-1 gene mutation in Japanese Alzheimer’s disease patients. Neurosci. Lett. 218, 139–141 (1996). 82. De Jonghe, C. et al. Evidence that Aβ42 plasma levels in presenilin-1 mutation carriers do not allow for prediction of their clinical phenotype. Neurobiol. Disord. 6, 280–287 (1999). 83. Steiner, H. et al. Glycine 384 is required for presenilin–1 function and is conserved in polytopic bacterial aspartyl proteases. Nature Cell Biol. 2, 848–851 (2000). Presenilins share sequence homology around an active site aspartate with polytopic bacterial aspartyl proteases. 84. Steiner, H. et al. The biological and pathological function of the presenilin-1 ∆exon 9 mutation is independent of its defect to undergo proteolytic processing. J. Biol. Chem. 274, 7615–7618 (1999). 85. LaPointe, C. F. & Taylor, R. K. The type 4 prepilin peptidases comprise a novel family of aspartic acid proteases. J. Biol. Chem. 275, 1502–1510 (2000). 86. Brown, M. S., Ye, J., Rawson, R. B. & Goldstein, J. L. Regulated intramembrane proteolysis: a control mechanism conserved from bacteria to humans. Cell 100, 391–398 (2000). 87. Sinha, S. et al. Purification and cloning of amyloid precursor protein β-secretase from human brain. Nature 402, 537–540 (1999). 88. Wolfe, M. S. et al. A substrate-based difluoroketone selectively inhibits Alzheimer’s γ-secretase activity. J. Med. Chem. 41, 6–9 (1998). 89. Qiu, W. Q. et al. Insulin-degrading enzyme regulates extracellular levels of amyloid β-protein by degradation. J. Biol. Chem. 273, 32730–32738 (1998). 90. Vekrellis, K. et al. Neurons regulate extracellular levels of amyloid β-protein via proteolysis by insulin-degrading enzyme. J. Neurosci. 20, 1657–1665 (2000). 91. Iwata, N. et al. Identification of the major Aβ1-42-degrading catabolic pathway in brain parenchyma: suppression leads to biochemical and pathological deposition. Nature Med. 6, 143–150 (2000). 92. Schenk, D. et al. Immunization with amyloid-β attenuates Alzheimer-disease-like pathology in the PDAPP mouse. Nature 400, 173–177 (1999). 93. Bard, F. et al. Peripherally administered antibodies against amyloid β-peptide enter the central nervous system and reduce pathology in a mouse model of Alzheimer disease. Nature Med. 6, 916–919 (2000). 94. Struhl, G. & Adachi, A. Requirements for presenilindependent cleavage of Notch and other transmembrane proteins. Mol. Cell 6, 625–636 (2000). Presenilins may have a general role in endoproteolysis of transmembrane domains of type I transmembrane proteins, which undergo ectodomain shedding. 95. Suzuki, N. et al. An increased percentage of long amyloid β-protein secreted by familial amyloid β-protein precursor (β-APP717) mutants. Science 264, 1336–1340 (1994). 96. Iwatsubo, T. et al. Visualization of Aβ-42(43) and Aβ-40 in senile plaques with end-specific Aβ-monoclonals: evidence that an initially deposited species is Aβ-42(43). Neuron 13, 45–53 (1994). 97. Annaert, W. G. et al. Presenilin 1 controls γ-secretase processing of amyloid precursor protein in pre-Golgi compartments of hippocampal neurons. J. Cell Biol. 147, 277–294 (1999). 98. Lah, J. J. & Levey, A. I. Endogenous presenilin-1 targets to endocytic rather than biosynthetic compartments. Mol. Cell. Neurosci. 16, 111–126 (2000). 99. Kulic, L. et al. Separation of presenilin function in amyloid βpeptide generation and endoproteolysis of Notch. Proc. Natl Acad. Sci. USA 97, 5913–5918 (2000). 100. Mori, K. Tripartite management of unfolded proteins in the endoplasmic reticulum. Cell 101, 451–454 (2000). 101. Sato, N. et al. Upregulation of BiP and CHOP by the unfolded protein response is independent of presenilin expression. Nature Cell Biol. 2, 863–870 (2000). 102. Niwa, M., Sidrauski, C., Kaufman, R. J. & Walter, P. A role for presenilin-1 in nuclear accumulation of Ire1 fragments and induction of the mammalian unfolded protein response. Cell 99, 691–702 (1999). 103. Katayama, T. et al. Presenilin-1 mutations downregulate the signalling pathway of the unfolded-protein response. Nature Cell Biol. 1, 479–485 (1999). 104. Wolozin, B. et al. Participation of presenilin 2 in apoptosis: enhanced basal activity conferred by an Alzheimer mutation. Science 274, 1710–1713 (1996). 105. Kim, T. W., Pettingell, W. H., Jung, Y. K., Kovacs, D. M. & Tanzi, R. E. Alternative cleavage of Alzheimer-associated presenilins during apoptosis by a caspase-3 family protease. Science 277, 373–376 (1997). 106. Yoo, A. S. et al. Presenilin-mediated modulation of capacitative calcium entry. Neuron 27, 561–572 (2000).

www.nature.com/reviews/molcellbio

| DECEMBER 2000 | VOLUME 1

PERSPECTIVES TIMELINE

Krebs and his trinity of cycles Hans Kornberg Everyone who has ever taken biology at school has heard of the Krebs cycle, but few realize that Hans Krebs also discovered two other cycles. It is appropriate, at the centenary of his birth, to consider the circumstances and experiments that led Krebs to establish these metabolic pathways.

Of all the biochemists whose work has given us insight into the molecular mechanisms by which cells synthesize their components from food and obtain biologically useful energy, Sir Hans Krebs FRS (FIG. 1) was undoubtedly among the greatest. He was born on 25 August 1900 and died at the age of 81, laden with honours and regarded almost with reverence by biologists all over the world. How did this unassuming, gentle, self-effacing man overcome so many difficulties and achieve so much? Hans Krebs, the son of a prominent otolaryngologist, was born in the little town of Hildesheim in Germany and attended the local grammar school — the ‘Andreanum’ — from 1910 to 1918. In the economic and political chaos that followed the First World War, he attended a variety of universities (as was the German custom) and graduated in medicine in 1925. However, although he practised internal medicine, Krebs had been drawn to scientific research even as an undergraduate. His chance to concentrate his efforts in that direction came in 1926, when he was offered a research assistantship (his first paid post) in the laboratory of the doyen of biochemistry, Otto Warburg, at the Kaiser Wilhelm Institut in Berlin–Dahlem. It was there that Krebs learned the tech-

niques developed by Warburg for investigating the processes by which sugars are broken down in slices of muscle and other tissues — processes that accurately reflect those in intact organisms. He also learned how to plan a fruitful attack on a scientific problem: as he once said to me in correcting an experimental protocol I had submitted for his approval, “A competent biochemist knows what experiments to do; a good one knows what experiments not to do”. However, Warburg also made it clear that he would not help Krebs obtain a university post after he left in 1930 and, indeed, he sent a letter to Krebs’s future employer urging him to keep Krebs busy with clinical duties, as he would never make a good biochemist! Cycle 1: The ornithine cycle

A year after leaving Warburg’s laboratory, Krebs obtained a clinical post at the hospital of his old medical school in Freiburg. He was expected to look after a ward of 22 patients; after a while, this was increased to 44. He was also expected to assist his professor in various time-consuming tasks. Despite that, he devoted every spare moment to tackling a problem that more experienced workers felt could not be solved — and, in less than a year, he solved it. It was known that urea, the main nitrogenous end product of metabolism, is formed exclusively in the liver, presumably from carbon dioxide and ammonia: CO2 + 2 NH3 → CO(NH2)2 + H2O Would liver slices be able to effect this synthesis, under conditions where Krebs (and a

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

Figure 1 | Sir Hans Krebs on a motorized cycle. (Photograph courtesy of Gil Hardy, Oxford Brookes University, UK.)

medical student, Kurt Henseleit, who assisted him) could measure the amounts of ammonia taken up and of urea formed? Krebs had first to design a salt solution that mimicked the saline composition of blood. Using the experimental techniques he had learned in Warburg’s laboratory, he showed that liver slices did indeed make some urea from added ammonium salts, the energy for this synthesis being supplied by the breakdown of stored carbohydrates. Addition of different amino acids stimulated the production of urea from ammonia to various degrees. However, one amino acid — ornithine — stimulated it to a much greater extent than any other tested. It had been known since 1904 that the amino acid arginine could be hydrolysed

VOLUME 1 | DECEMBER 2000 | 2 2 5

PERSPECTIVES

Urea

Ornithine

+H2O

α-Keto acid

+CO2

–H2O

Arginine

CO2 + NH4+

Carbamoyl phosphate

+NH3

Citrulline –H2O

Citrulline

Aspartate

Argininosuccinate

Ornithine

α-Amino acid Oxaloacetate Malate

Urea Arginine

Fumarate

+NH3

Figure 2 | The ornithine cycle. a | The cycle as formulated by Krebs and Henseleit4 in 1932. b | Modern version of the ornithine cycle.

(split by water) by the enzyme arginase to give ornithine and urea, and that the liver is rich in arginase1. So was the ornithine perhaps contaminated with arginine? Careful purification of the ornithine, and measurements of the amounts of ornithine added and of urea formed, not only removed this anxiety but revealed an unexpected effect: the addition of one molecule of ornithine brought about the extra formation of more than 20 molecules of urea, provided that ammonia was also present. This established that ornithine acted as a catalyst. Krebs was guided by the idea that a catalyst must take part in the reaction to yield some intermediate that could then give rise to urea and also regenerate the ornithine: clearly, arginine fitted the latter part of this hypothesis. But could liver slices be shown to form arginine from ornithine and ammonia? That is: Ornithine + CO2 + 2 NH3 → arginine Because this process was unlikely to proceed in one step, Krebs postulated that citrulline might be an intermediate, and that the sequence of reactions might be:

England, Sir Frederick Gowland Hopkins (Professor of Biochemistry at the University of Cambridge and President of The Royal Society) used Krebs’s work as the main scientific topic of his presidential address to the society in November 1932. Of more immediate significance, the Dean of the Medical Faculty of Freiburg recommended in glowing terms that Krebs be granted tenure. All was set fair for a brilliant future. But then, in January 1933, Hitler was elected Reich Chancellor, and three months later the Government issued an order placing on indefinite leave of absence “… all members of the Jewish race (irrespective of their religion) who are employed in ... teaching establishments ...”. Shortly thereafter, Krebs was dismissed from his post and forbidden to enter university premises. It was indeed fortunate that Krebs had attracted the notice of Sir Frederick: he readily offered to house Krebs in the biochemistry department of the University of Cambridge, and helped him to obtain a grant of £300 (for one year only) from the Rockefeller Foundation. Krebs gratefully seized the lifeline that had been thrown to him and, in June 1933, accompanied by 16

Ornithine + CO2 + NH3 → citrulline +H2O Citrulline + NH3 → arginine + H2O Arginine + H2O → urea + ornithine By a happy coincidence, two scientists2,3 had just independently isolated citrulline. Krebs wrote to both of them, obtained a few milligrams from each, and showed that citrulline indeed acted catalytically to promote the formation of urea from ammonia and carbon dioxide. And so the ornithine cycle4 was born (FIG. 2). Triumph and disaster

The discovery of the ornithine cycle was immediately recognized by the scientific world as an important achievement. Krebs was showered with invitations to lecture, his name was suggested for a Chair, and, in

226

Citric acid

Isocitric acid

Oxalosuccinic acid α-Ketoglutaric acid Triose Succinic acid

Fumaric acid

l-malic acid

Oxaloacetic acid

Figure 3 | The citric acid cycle, formulated by Krebs and Johnson6 in 1937.

wooden boxes and several suitcases containing his scientific equipment, left Freiburg to begin a new life in freedom. Cycle 2: The citric acid cycle

After three happy and scientifically fruitful but financially straitened years, Krebs accepted a lectureship in pharmacology at the University of Sheffield. This enabled him, and William A. (‘Johnny’) Johnson, a graduate student, to study how foodstuffs — proteins, carbohydrates and fats — were oxidized to carbon dioxide and water to yield the energy for the many energyrequiring reactions characteristic of living organisms. Previously, the Hungarian biochemist Albert Szent–Györgyi had studied this process using the breast muscle of pigeons. This tissue has to power the birds’ flight and is therefore well adapted to extracting biologically useful energy from food. When minced and suspended in a saline solution, such muscle cells took up oxygen rapidly, and the rate of oxygen uptake was greatly increased when carbohydrates or their fermentation products (such as pyruvate or lactate) were added. Clearly, the minced muscle could carry out the total combustion of these C 3 compounds. For example, for lactate: CH3CH(OH)CO2– + H+ + 3 O2 → 3 CO2 + 3 H2O Krebs argued that this process could not occur in just one step and that any intermediates between the starting materials and the end products must be oxidized at least as rapidly as lactate or pyruvate. Szent–Györgyi had already shown that relatively few compounds obeyed this criterion, but that succinate, fumarate, malate and oxaloacetate — all salts of C4 acids — did: Succinate → fumarate + 2 H Fumarate + H2O → malate Malate → oxaloacetate + 2 H Moreover, Szent–Györgyi and his co-workers had also shown that these materials acted catalytically — they promoted the uptake of more oxygen than was required for oxidation of the quantities of substrates added. But the significance of these observations had not been realized; indeed, Szent–Györgyi and colleagues believed that these C4-acid salts did not participate in the oxidative process as food materials, but that they acted as shuttles for hydrogen from such foodstuffs to oxygen, being alternately oxidized and reduced5. In 1937, Krebs and Johnson showed6 that succinate could be synthesized by animal tiswww.nature.com/reviews/molcellbio

| DECEMBER 2000 | VOLUME 1

PERSPECTIVES sues if pyruvate was available. They speculated that the C4-acid salt might have arisen from the oxidation of citrate by a sequence of reactions that had already been shown to occur in liver tissue7,8: Citrate → cis-aconitate + H2O cis-Aconitate + H2O → isocitrate Isocitrate → α-ketoglutarate + CO2 + 2 H α-Ketoglutarate → succinate + CO2 + 2 H Would citrate also be readily oxidized in minced muscle? Would it act catalytically? And, most importantly, would citrate be formed from oxaloacetate if pyruvate were also added? In an epoch-making paper9 (which had initially been rejected by Nature), Krebs and Johnson demonstrated conclusively that the answer to all these questions was yes. This conclusion was based on the observations that citrate was not only oxidized rapidly, but that it did not disappear during that process — clearly, it was being continuously re-formed. Furthermore, in the absence of oxygen, large amounts of citrate were formed by minced muscle if both oxaloacetate and pyruvate were added, but not if either component was omitted. In the presence of arsenite, which was known to inhibit the oxidation of α-ketoacids, α-ketoglutarate accumulated from citrate. In the presence of malonate, an inhibitor of succinate oxidation10,11, succinate accumulated when either citrate or oxaloacetate were added — a particularly striking point. As is apparent from the formulae of these two C4-acid salts, succinate (–O2CCH2CH2CO2–) is more reduced than oxaloacetate (–O2C(CO)CH2CO2–) — its formation from oxaloacetate with the uptake of oxygen must imply the occurrence of a cyclic, oxidative sequence of reactions. What Krebs and Johnson termed the “citric acid cycle” was born (FIG. 3). Cycle 3: The glyoxylate cycle

As often happens in science, a new idea — even when well supported by hard evidence — is not immediately accepted. It took a number of confirmatory experiments by Krebs and colleagues over several years to establish the citric acid cycle as the main pathway of terminal respiration in nearly all aerobic organisms. These included decisive experiments with isotopically labelled substrates, and the identification of acetyl-coenzyme A (REFS 12,13) as the ‘triose’ of Krebs’s and Johnson’s original cycle. But this left one thorny question unanswered. The cycle explained how microorganisms oxidized acetate to carbon dioxide and water: CH3CO2– + H+ + 2 O2 → 2 CO2 + 2 H2O

Acetate

Oxaloacetate Oxaloacetate + 12 O2

Malate

Citrate

+ 12

cis-Aconitate

Malate

Isocitrate + 12 O2 CO2 α-Ketoglutarate

Fumarate + 12 O2 Succinate

+ 12

Citrate cis-Aconitate

Isocitrate Acetate Glyoxylate

Succinate

CO2

Figure 4 | The tricarboxylic acid (TCA) and glyoxylate cycles, formulated by Kornberg and Krebs20 in 1957. a | The main stages of the TCA cycle. The net effect of one turn of the cycle is the conversion of acetate and two O2 molecules into two CO2 molecules. b | Main stages of the glyoxylate cycle — a variant of the TCA cycle. The net effect of one turn is the conversion of two acetate molecules and half an O2 molecule into succinate. Further metabolism of succinate can lead to the synthesis of other metabolites.

But it failed to explain how such organisms could grow on acetate — that is, how they could make the carbon skeletons of all their cell constituents from acetate as the sole source of carbon. This puzzle nagged Krebs for many years, but he felt that the time was ripe to tackle it only when new experimental approaches could be brought to bear on it. I was fortunate to join Krebs’ laboratory — which had been based in Oxford since 1953 — after gaining some experience in

“A competent biochemist knows what experiments to do; a good one knows what experiments not to do.” the techniques that had enabled Melvin Calvin and his colleagues to elucidate the path of carbon in photosynthesis14. Incubating suspensions of acetate-grown bacteria with 14C-labelled acetate for very brief periods, I noticed15 that, although the intermediates of the citric acid cycle — now generally known as the tricarboxylic acid (TCA) cycle — and amino acids directly derived from it were initially the only labelled products formed, their labelling patterns were not consistent with the operation of that cycle. This observation was confirmed by my colleague J. Rodney Quayle16, who measured the position and amount of isotopic carbon that appeared in these products. It seemed that intermediates of the TCA cycle acquired labelled carbon not only from the turning of that cycle, but also from an ancillary pathway that fed acetate into it through a second point of entry. Such

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

an ancillary sequence had been shown using cell-free extracts of acetate-grown bacteria17, and resulted from the combined action of two previously identified enzymes. One enzyme condensed acetate (in the form of its coenzyme A derivative) with glyoxylate (–O2CCHO) to form malate18, whereas the other provided the glyoxylate needed for this reaction by cleaving isocitrate19. The sequence was thus: Isocitrate → succinate + glyoxylate Glyoxylate + acetyl-coenzyme A → malate + coenzyme A My co-worker Neil Madsen and I showed17 that acetyl-coenzyme A indeed reacted with one intermediate of the TCA cycle to yield two: Acetyl-coenzyme A + isocitrate → malate + succinate + coenzyme A This bypassed the two reactions of the TCA cycle in which carbon was lost as carbon dioxide. It was because of this that we termed this sequence the “glyoxylate bypass” of the TCA cycle. Krebs and I subsequently put the two cycles together and termed the combination the “glyoxylate cycle”20 (FIG. 4). Although Krebs (by now Sir Hans and a Nobel laureate) did not himself carry out any of the experimental work, it is indisputable that identification of the problem, the impetus to tackle it and the means to investigate it, were his. This cycle may thus be rightly regarded as the third important metabolic route to which his name will forever be attached. Hans Kornberg is in the Department of Biology, Boston University, 2 Cummington Street, Boston, Massachusetts 02215, USA. e-mail: hlk@bu.edu

VOLUME 1 | DECEMBER 2000 | 2 2 7

PERSPECTIVES Links ENCYCLOPEDIA OF LIFE SCIENCES Krebs,

Hans Adolf Kossel, A. & Dakin, H. D. Über die Arginase. Z. Physiol. Chem. 41, 321–331 (1904). 2. Wada, M. Über Citrullin, eine neue Aminosaure im presssaft der Wassermelone, Citrullis vulgaris schrad. Biochem. Z. 224, 420–429 (1930). 3. Ackermann, D. Über den biologischen Abbau des Arginins zu Citrullin. Biochem. Z. 203, 66–69 (1931). 4. Krebs, H. A. & Henseleit, K. Untersuchungen über die Harnstoffbildung im tierkorper. Z. Physiol. Chem. 210, 33–66 (1932). 5. Annau, E. et al. Über die Bedeutung der Fumarsaure fur die tierische Gewabsatmung. Z. Physiol. Chem. 235, 1–68 (1935). 6. Krebs, H. A. & Johnson, W. A. Metabolism of ketonic acids in animal tissues. Biochem. J. 31, 645–660 (1937). 7. Martius, C. & Knoop, F. Der physiologische Abbau der Citronensaure. Vorläufige mitteilung. Z. Physiol. Chem. 246, 1–11 (1936). 8. Martius, C. Über den Abbau der Citronensaure. Z. Physiol. Chem. 247, 104–110 (1937). 9. Krebs, H. A. & Johnson, W. A. The role of citric acid in intermediate metabolism in animal tissues. Enzymologia 4, 148–156 (1937). 10. Quastel, J. H. & Whetham, M. D. The equilibria existing between succinic, fumaric and malic acids in the presence of resting bacteria. Biochem. J. 18, 519–534 (1924). 11. Quastel, J. H. & Whetham, M. D. Dehydrogenations

12.

13.

14.

15.

16.

17.

18.

19.

20.

produced by resting bacteria. Biochem. J. 19, 520–531 (1925). Stern, J. R. & Ochoa, S. Enzymatic synthesis of citric acid by condensation of acetate and oxaloacetate. J. Biol. Chem. 179, 491–492 (1949). Stern, J. R. et al. Enzymatic synthesis of citric acid. V. Reaction of acetyl coenzyme A. J. Biol. Chem. 198, 313–321 (1952). Bassham, J. A. & Calvin, M. The Path of Carbon in Photosynthesis (Prentice–Hall, Englewood Cliffs, New Jersey, 1957). Kornberg, H. L. The metabolism of C2-compounds in micro-organisms. 1. The incorporation of [2-14C]-acetate by Pseudomonas fluorescens, and by a Corynebacterium, grown on ammonium acetate. Biochem. J. 68, 535–542 (1958). Kornberg, H. L. & Quayle, J. R. The metabolism of C2compounds in micro-organisms. 2. The effect of carbon dioxide in the incorporation of [14C]-acetate by acetategrown Pseudomonas KB 1. Biochem. J. 68, 542–549 (1958). Kornberg, H. L. & Madsen, N. B. Synthesis of C4dicarboxylic acids from acetate by a glyoxylate bypass of the tricarboxylic acid cycle. Biochim. Biophys. Acta 24, 651–653 (1957). Wong, D. T. O. & Ajl, S. J. Conversion of acetate and glyoxylate to malate. J. Am. Chem. Soc. 78, 3230–3231 (1956). Smith, R. A. & Gunsalus, I. C. Isocitratase: A new tricarboxylic acid cleavage system. J. Am. Chem. Soc. 76, 5002–5003 (1954). Kornberg, H. L. & Krebs, H. A. Synthesis of cell constituents from C2-units by a modified tricarboxylic acid cycle. Nature 179, 988–991 (1957).

TIMELINE

The elusive cytostatic factor in the animal egg Yoshio Masui While animal eggs await fertilization, their cell cycle needs to be halted. The molecule responsible for this arrest — the cytostatic factor — was first described in 1971. But its identity was not revealed until 1989, and even now questions remain about this elusive factor.

Most animals develop from a fertilized egg — the zygote. The question of why eggs require fertilization to begin development is an old one: how are they prevented from spontaneous development, or parthenogenesis? In 1911, Frank Lillie1 asked the question as follows: “The nature of the inhibition that causes the need for fertilization is a most fundamental problem”. A year later, his answer2 was that a “lack of interchange between the egg nucleus and the egg cytoplasm” in the unfertilized egg causes the inhibition. However, according to Theodor Boveri3, the inhibition was caused by a lack of the “organ” for cell divisions, such as the centrosome. And in 1913, Jacques Loeb 4 proposed that the obstacle to cell division was stability of the cortex in the unfertil-

228

ized egg, which could be removed by partial cytolysis. The answer given by later researchers was rather more simple — that the egg is inhibited by a cell-division inhibitor that accumulates during oogenesis5. Candidates for this inhibitor ranged from carbon dioxide to complex metabolic inhibitors (TIMELINE). But the search for a cell-division inhibitor required great caution, because cell divisions may easily be inhibited by nonspecific toxicity. For the inhibitor to be authentic, it needed to be tested on the criteria indicated in BOX 1 (REF. 5). The cytostatic factor, which I described along with the late Clement Markert6 in 1971, was the first inhibitor that met all of these criteria.

Discovery of the cytostatic factor

During the 1960s, embryologists were losing confidence in studies of cell differentiation, particularly embryonic induction. There was suspicion about the specificity of the inducer and, influenced by immunologists, embryologists were inclined to think that cell differentiation resulted from clonal selection rather than from cell transformation triggered by the inducer. To escape such scepticism, I wanted to study the role of nucleo-cytoplasmic interactions in a simple differentiation process that could be triggered by a welldefined inducer in a single cell. In 1964, after reading a paper by Tatiana Dettlaff and her associates7, who analysed frog oocyte maturation by enucleation and nuclear transplantation, I decided that ‘oocyte maturation’ was a suitable system for this study. Oocyte maturation is the final process of female germ-cell differentiation, in which oocytes — fully grown and arrested at prophase of the first meiosis (prophase I) — resume meiosis to become fertilizable eggs or mature oocytes. In vertebrates, maturing oocytes complete the first meiosis, but eggs are arrested at metaphase of the second meiosis (metaphase II). Fertilization activates the eggs to release them from this arrest (FIG. 1). I began my study of oocyte maturation using the leopard frog Rana pipiens in late 1966, during my first sabbatical year in Markert’s lab at Yale University. In 1967, Allen Schuetz8 and I9 independently found that frog-oocyte maturation could easily be induced in vitro by progesterone. However, as Markert and I reported6, it was not the hormone but maturation-promoting factor (MPF) that directly induced the oocyte nucleus to resume meiosis when injected into immature oocytes. MPF is activated in the cytoplasm of the oocyte in response to progesterone, and remains active in the egg. But MPF activity markedly decreases after the egg is activated by fertilization or by pricking with a glass needle (FIG. 1). We thought that, because MPF could reinitiate meiosis in immature oocytes, it might accelerate mitosis when injected into zygotes. On the contrary, as we reported6, when the same cytoplasm was injected into one of the

Box 1 | The search for the cytostatic factor The criteria for putative cytostatic factors5 is that they should: • Appear during oocyte maturation. • Disappear during fertilization (egg activation). • Be destroyed under the same physico-chemical conditions as those that cause egg activation. • Provide the arrested zygote with same properties as those of the unfertilized egg. • Inhibit mitosis of the zygote reversibly.

www.nature.com/reviews/molcellbio

| DECEMBER 2000 | VOLUME 1

PERSPECTIVES Links ENCYCLOPEDIA OF LIFE SCIENCES Krebs,

12.

13.

14.

15.

16.

17.

18.

19.

20.

TIMELINE

228

Discovery of the cytostatic factor

www.nature.com/reviews/molcellbio

| DECEMBER 2000 | VOLUME 1

PERSPECTIVES

Timeline | Progress of research into the cytostatic factor Discovery of Ca2+-sensitivity and Mg2+-dependency of the cytostatic factor and its stabilization by a Ca2+-chelating agent in frog-egg extracts12.

Polynucleotides are proposed to be the cytostatic factor (Menkin5). CO2 is proposed as the cytostatic factor (Battaillon and Tchou-Su5).

1930

1954

The polysaccaride heparin is proposed to be the cytostatic factor (Heilbrunn et al.5).

Discovery of the cytostatic factor in unfertilized egg cytoplasm that causes the metaphase arrest of meiosis and mitosis6.

1959

1967

1971

Metabolic inhibitors are proposed to be the cytostatic factor (Monroy and Tyler5).

Partial purification and molecular characterization of the cytostatic factor as a protease-sensitive and RNase-resistant molecule with sedimentation coefficient 3S13,14.

1974

1977

Extraction of the cytostatic factor in a soluble form from frog eggs11.

1988

1989

Biochemical characterization.

It was no easy task to extract the cytostatic factor from unfertilized eggs, because activation of the egg by slight injuries to the cortex inac-

Identification of the cytostatic factor as p90Rsk protein kinase37,38.

1993

1999

Identification of the cytostatic factor as a MAPK33. Discovery of CaMKII as the key enzyme in destruction of Mos21.

tivates cytostatic factor and MPF. But in 1974, I succeeded11 in preparing extracts without losing activities of these factors by compressing eggs using centrifugal force. Peter Meyerhof tested the effects of ions on activity of the cytostatic factor in the egg extracts prepared by this method and, in 1977, he reported12 that addition of Ca2+ to, or removal of Mg2+ from, the extracts quickly causes the loss of cytostatic factor activity. Conversely, removal of Ca2+ from, or addition of Mg2+ to, the extracts stabilizes the activity. Progesterone

1994

2000

Discovery of spontaneous parthenogenetic activation of eggs from c-mosdeficient mice35,36.

Identification of the cytostatic factor as Mos15.

Further stabilization and enhancement of cytostatic factor activity by ATP and phosphatase inhibitors13.

cells (blastomeres) of the two-cell embryo, “the injected blastomere frequently stopped cleaving. This result was unexpected” (FIG. 2). However, when the zygote’s cytoplasm was injected into blastomeres, cleavage was not inhibited, indicating that initiation of cleavage might result from the removal of the inhibitory factor from the egg by fertilization. So we concluded6 that “a specific cytoplasmic factor or factors [in unfertilized eggs] is responsible for the inhibition of mitosis and cleavage. This hypothetical factor for the cytoplasm of maturing oocytes can be tentatively labelled cytostatic factor”. The cytostatic-factor-arrested blastomere and the unfertilized egg show the same cytological features5 — both have a mitotic apparatus containing condensed chromosomes, but lacking asters (FIG. 2)6,10, and their cytoplasm shows chromosome-condensation activity as well as MPF activity10. Cytostaticfactor-arrested blastomeres resumed DNA replication, mitotic chromosome condensation and formation of cleavage furrows when the blastomeres were injected with a large number of sperm nuclei, or with Ca2+ ions, which destroy the cytostatic factor and activate the egg10. Therefore, the cytostatic-factormediated inhibition of cell division was reversible, indicating that the inhibition is not due to nonspecific toxicity of this factor.

Discovery that p90Rsk inhibits APC to prevent degradation of cyclin B39.

Discovery of MAPK activation by Mos (Posada et al.; Nebreda and Hunt; Shibuya and Ruderman32).

Fertilization

Discovery of the c-mos gene in starfish oocytes16.

The extreme instability of the cytostatic factor and the inefficiency of the assay method for its activity (microinjection into blastomeres) hampered purification and further characterization for the next ten years. But in 1988, Ellen Shibuya13 succeeded in not only stabilizing, but also enhancing activity of the cytostatic factor using phosphatase inhibitors and ATP. In 1989, she14 partially purified the cytostatic factor eight times from frog-egg extracts by ammonium sulphate precipitation followed by ultracentrifugation First cleavage

Blastula

MPF

CSF

Time

Figure 1 | Changes in activity of maturation-promoting factor and the cytostatic factor during oocyte maturation, fertilization and cleavage. The relative maturation-promoting factor (MPF) and cytostatic factor (CSF) activities are expressed in arbitrary units. MPF was assayed for its activity to induce oocyte maturation by injecting MPF into fully grown oocytes, or for its protein-kinase activity by its ability to phosphorylate histone H1. The CSF was assayed for its activity to arrest mitosis of blastomeres by injecting it into two-cell stage frog embryos.

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 1 | DECEMBER 2000 | 2 2 9

PERSPECTIVES a

Figure 2 | Blastomere of a Rana pipiens embryo injected with the cytostatic factor. a | The arrested blastomere, which is injected with extracts from unfertilized eggs. b | Arrested mitotic spindle of the blastomere. The photograph of the blastomere (a) was taken 24 hours after injection of the egg extract, and fixed and sectioned for histological examination, then the photograph of the arrested mitotic spindle was taken. Note the absence of the aster structures at the spindle poles — the characteristic feature of cytostatic-factor-arrested mitosis. (Photographs by Ellen Shibuya.)

through a sucrose-density gradient, and did more tests on its activity. Thus the cytostatic factor was found to be a protease-sensitive — but RNase-resistant — molecule with a sedimentation coefficient of 3S. These results indicated that the cytostatic factor is a small protein activated by its Mg2+-dependent phosphorylation, but inactivated by Ca2+ ions (TIMELINE, BOX 2). The identity of the cytostatic factor was revealed from a different angle by Noriyuki Sagata and his associates15 in 1989. They noted a striking resemblance between the cytostatic factor and Mos — the protein kinase coded by the proto-oncogene c-mos — in their molecular characteristics and behaviour during oocyte maturation and fertilization (BOX 2). The conclusion from this was that “the c-mos proto-oncogene product, pp39mos (Mos), is either the longknown endogenous meiotic inhibitor of amphibian eggs, the cytostatic factor, or at least a catalytic component of it”15. Mos is now known to exist not only in the frog, but also in other vertebrates, and even in starfish16, whose oocytes are not arrested at metaphase II (TIMELINE). Destruction of the cytostatic factor.

When eggs are activated, their cytoplasm loses its cytostatic factor activity. Calcium has

been known to be pivotal in egg activation since Daniel Mazia’s discovery17, in 1937, that Ca2+ is released from fertilized sea-urchin eggs. In 1974, the Ca2+ ionophore A23187, which causes a release of intracellular Ca2+, was found to be a universal egg activator18. Then, in 1977, Lionel Jaffe and colleagues19, using a chemiluminescent protein aequorin, showed an explosive surge of Ca2+ in the fertilized fish egg. That same year, Meyerhof and I found12 inactivation of the cytostatic factor after injection into the fertilized frog egg at early periods of fertilization as well as after treatment with Ca2+ ions. Destruction of Mos in frog-egg extracts by Ca2+, and in A23187treated Xenopus eggs, was also reported20 in 1989. Finally, four years later, destruction of cytostatic factor activity was found to be mediated by Ca2+/calmodulin-dependent protein kinase II (CaMKII)21 (TIMELINE). The activation, by protein-synthesis inhibitors, of eggs from various animals has been known since the demonstration in 1973 that eggs of a polychaete Chaetopterus are activated by cycloheximide22. In 1988, mollusc eggs were shown to be activated by 6-dimethyl-amino-purine (6-DMAP), a protein-kinase inhibitor23. Among vertebrates, mouse24 and newt25 eggs can be activated by protein-synthesis inhibitors, but frog26 eggs cannot. Newt25 and frog26 eggs are

Box 2 | Criteria for identifying c-Mos as the cytostatic factor • Both arrest meiosis and mitosis at metaphase. • Both appear in unfertilized eggs, but disappear in fertilized eggs. • Both are destroyed in vivo by egg activation and in vitro by Ca2+ ions. • Both are neutralized and immunodepleted with the same antibodies against Mos.

activated by 6-DMAP in the presence of a Ca2+-chelating agent, BAPTA, although mouse27 eggs are not activated under these conditions. Last year, destruction of the cytostatic factor was found to be accelerated in 6-DMAP-activated frog eggs28. These results indicate that stability of the cytostatic

“Although nobody would argue today that Mos is part of the cytostatic factor, the ultimate identity of the cytostatic factor is still unclear.” factor depends on continuous protein synthesis and/or phosphorylation. The cytostatic factor activity also seems to depend on the state of microtubules in the egg. Heavy water (deuterium oxide, D2O), which enhances microtubule assembly, was found to activate eggs of sea urchins29, newt25 and mouse (Y.M., unpublished observations). A concomitant loss of the cytostatic factor by D2O has been shown in the newt egg25. Conversely, mouse eggs treated with microtubule-disassembling agents such as nocodazole and colcemid cannot be activated by the ionophore A23187, 6-DMAP or a protein-synthesis inhibitor alone30,31. However, if these microtubule-depleted eggs are treated with the ionophore A23187 and then exposed to either 6-DMAP or a protein synthesis inhibitor, the eggs are activated31.

• Both depend on phosphorylation for their activities.

Action of the cytostatic factor

• Both are sedimented in egg extracts in the same manner (150,000 g for 6 h).

In 1993, Mos activation was shown to lead to

230

www.nature.com/reviews/molcellbio

| DECEMBER 2000 | VOLUME 1

PERSPECTIVES the mitogen-activated protein kinase (MAPK) cascade32. In 1995, Jim Maller and his colleagues showed33 that “microinjection of thiophosphorylated MAPK into one blastomere of a two-cell embryo induced metaphase arrest similar to that induced by c-Mosxe”. Furthermore, oocytes from mice lacking the c-mos gene were found to lack active MAPK and to be spontaneously activated34–36. Finally, one of the MAPK-activated proteins, the 90 kDa ribosomal protein S6 kinase (p90Rsk), was found to have cytostatic factor activity37,38. These results indicate that the MAPK cascade, initiated by Mos, mediates the action of the cytostatic factor (TIMELINE). Furthermore, it has been shown39 that p90Rsk hyperphosphorylates cdc27, a component of the anaphase-promoting complex (APC), to prevent degradation of cyclin B. Therefore, one way in which the MAPK cascade mediates cytostatic-factor-driven mitotic arrest might be by inhibiting the APC to maintain MPF activity. However, this cytostatic-factor-mediated arrest may be brought about independently of MPF activity as other studies40,41 showed that, once arrested at metaphase, mitosis can be sustained by MAPK activity alone, even after MPF activity has dropped to the low levels seen during interphase. Also, this arrest must occur independently of regulation of the spindle-assembly checkpoint, because the cytostatic factor arrests blastomeres of early frog embryos lacking the checkpoint mechanism. Therefore, we may consider another possibility that the cytostatic factor arrests mitosis by directly inhibiting chromosome movement during segregation to opposite poles of the cell. This is indicated by observations that Mos, MAPK and the centromere-binding kinesin-like protein (CENPE) are localized together in the kinetochore of chromosome42,43, and that “CENPE is modified or masked during the natural, Mos-dependent, cell-cycle arrest that occurs at metaphase II”43. Future directions

Many questions remain about the cytostatic factor — even 30 years after its discovery. Although nobody would argue today that Mos is part of the cytostatic factor, the ultimate identity of the cytostatic factor is still unclear, as is the question of how it, or its downstream kinases, arrest mitosis in the middle of metaphase. The answers will provide us with indispensable knowledge for controlling mitosis and cell division. So one direction of future study is to identify the cytostatic factor and clarify how it works. The most convincing evidence for the cytostatic factor would be to purify it to a high degree

Diplotene

GVBD

Metaphase I

Growth

First division Hormone Spisula Urechis

Insect Tunicate

Metaphase II

Zygote

First PB

Two-cell

Second PB

Pronucleus

Second division Vertebrate

Echinoderm Coelenterate

Figure 3 | Progression of meiosis during oocyte maturation and the timing of fertilization. The diagram represents information published by Wilson1 in 1925 and Masui5 in 1985. Animals are indicated whose oocytes are arrested and await fertilization at the stages of meiosis shown. (GVBD, germinal vesicle breakdown; PB, polar body.)

from frog eggs, as was done to identify MPF44. However, this task will require new methods to stabilize the cytostatic factor in egg extracts and to assay it rapidly and accurately in vitro. Another area of future study is to investigate the cytostatic factor in the entire animal kingdom. As is well known1,5 (except in some species such as the surf clam Spisula solidissima, whose oocyte maturation is initiated by fertilization), maturing oocytes arrest again at metaphase I in insects, molluscs and ascidians (for example, the sea squirt) and at metaphase II in vertebrates. Only in echinoderms (such as starfish and sea urchins) and coelenterates (for instance, jellyfish) do oocytes complete meiosis and arrest at the pronuclear stage until fertilization (FIG. 3). Activation of MAPK during oocyte maturation, and its inactivation upon fertilization, have been observed in echinoderms45 and ascidians46. In starfish, eggs arrested at the pronuclear stage can be activated to begin DNA synthesis when MAPK is inactivated, but they cannot be activated by fertilization if MAPK has been constitutively activated47,48. We also know that activation of MAPK during entry to mitosis arrests the cell cycle at metaphase, but that activation of this same protein during interphase inhibits entry to mitosis49. If so, does this mean that the MAPK cascade may have the same role as the cytostatic factor in both invertebrate and vertebrate oocytes arrested at various stages of meiosis? Finally, earlier this year, Takeo Kishimoto and his associates16 succeeded in cloning the c-mos gene from oocytes of the starfish Asterina pectinifera. They found that ablation

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

of c-mos messenger RNA in starfish oocytes prevented them from entering meiosis II, but that it caused them to initiate DNA replication immediately after meiosis I and to develop parthenogenetically. Will Mos, then, be found as a ubiquitous factor throughout the animal kingdom? If so, the molecular regulation of meiosis is highly conserved throughout evolution. Or will we discover another protein kinase that does the job of the cytostatic factor in invertebrate eggs? If this is the case, studies into the relationship between this newly found protein and Mos will shed light on the problem of evolution of meiotic regulation in the animal kingdom. Whichever, it is important from a zoological point of view to know what molecules the invertebrate cytostatic factors are. And it seems to me that the technical expertise to solve the problems stated above is now mature. Yoshio Masui is in the Department of Zoology, University of Toronto, 25 Harbord Street, Toronto, Ontario, Canada M5S 3G5. e-mail: masui@zoo.utoronto.ca

Links DATABASE LINKS Mos | CaMKII | MAPKs |

cdc27 | CENPE 1.

Lillie, F. R. Studies of fertilization in Nereis. I. The cortical changes in the egg: II. Partial fertilization. J. Morphol. 22, 361–393 (1911). Lillie, F. R. Studies of fertilization in Nereis. VI. The fertilizing power of porations of the spermatozoon. J. Exp. Zool. 12, 427–476 (1912). Boveri, T. Zellenstudien VI. Die Entwicklung dispermer Seeigeleier. Ein Beitrag zur Befruchtungslehre und zur Theorie des Kernes. Jena Zeitschri. Naturw. 43, 1–292 (1907).

VOLUME 1 | DECEMBER 2000 | 2 3 1

PERSPECTIVES

OPINION

Human cancer cell lines: fact and fantasy John R. W. Masters Cancer cell lines are used in many biomedical research laboratories. Why, then, are they often described as unrepresentative of the cells from which they were derived? Here, I argue that they have been unjustly accused. Under the right conditions, and with appropriate controls, properly authenticated cancer cell lines retain the properties of the cancers of origin.

Cell lines provide an almost unlimited supply of cells with similar genotypes and phenotypes. Their use avoids variation between individuals and bypasses ethical issues associated with animal and human experiments. However, many scientists question whether they retain the characteristics of the cells from which they were derived. How have cancer cell lines attained this dubious reputation, and what can cancer researchers do to ensure their appropriate use? Culture of human cancer cells

There is a misconception that because cancers seem to have unlimited growth potential in patients, the cells are easy to culture and have limitless growth potential in the laboratory (BOX 1). Nothing could be further from the truth — for many types of cancer, it is far easier to grow the normal cells than the cancer cells1. Even for cancers that are relatively easy to grow, such as melanomas, only the metastatic cancers can be established as continuous cell lines in most cases2. So are the human cancer cell lines that have been produced representative of the cancers from which they were derived? There are two aspects to this question and, as discussed below, two opposite answers. Representing the cancer of origin

Individual cancer cell lines “provide a snapshot of the tumor at the time the biopsy was taken”3. Evidence to support this statement includes the following: Histopathology. When human cancer cell lines are transplanted subcutaneously into immunodeficient mice (such as the nude

strain), most can form tumours. Of 127 human cancer cell lines that produced tumours in nude mice after subcutaneous injection, the histopathology correlated with the tumour of origin in every case4. Molecular genetics and receptor expression. Few direct comparisons have been made of

the genotypic and phenotypic characteristics of cancer cell lines with those of the tumours from which they were derived. Two exceptional studies compared a series of breast cancer cell lines5 and lung cancer cell lines6 with the cancers from which they were derived, and showed that the cultures retain many of the phenotypic (such as oestrogen receptor expression) and genotypic properties of their corresponding tumours for long periods of time. Similarly, mutations in cyclin-dependent kinase genes and p53 were almost always identical in cell lines and the lymphomas or leukaemias from which they were derived7,8. Gene expression. Complementary DNA microarray studies of over 8,000 genes in 60 human cancer cell lines revealed consistent similarities between cell lines from the same

Box 1 | Cell line models There are a multitude of definitions for each tissue culture term. This perspective follows the definitions of the terminology committee of the Society for In Vitro Biology (formerly the American Tissue Culture Association)33. Primary culture Produced by growing cells from tissue taken directly from an individual. Cell line A primary culture becomes a cell line when it is transferred into the next culture vessel. For adherent cultures, the cells are detached using a protease, such as trypsin, and/or a chelating agent, such as EDTA, and subdivided — this process is known as passaging. For cells that grow in suspension, the culture is split into new culture vessels. Unless specialized culture conditions are used, within a few passages a relatively uniform population of proliferative cells is selected. This population is probably representative of the cells that divide when the tissue of origin is wounded, and will carry on growing until the end of the natural proliferative lifespan is reached and senescence occurs. As long as the cells proliferate, they show little or no evidence of tissue-specific differentiation. However, given the appropriate signals, they can regenerate a functional tissue. Immortal cell line Normal human cells have a limited lifespan in culture and almost never spontaneously immortalize (in contrast to rodent cells). Consequently cell lines can only be used over a limited period until they senesce. Most human cancers express telomerase, but either cannot be cultured or undergo senescence. To delay senescence, the lifespan can be extended by transfection with viral genes. The products of the viral genes sequester proteins such as p53 and Rb, allowing the cells to continue dividing for more passages. The cultures still senesce (this period is described as ‘crisis’), but if one is patient, in some cultures the occasional cell will acquire the mutation(s) that make it immortal and sometimes tumorigenic. Cell immortalization and carcinogenesis have much in common. Conditionally immortalized cell lines The advantages of immortal cell lines (a constant supply of almost identical cells) can be achieved, without the disadvantage of transforming the cells into the equivalent of cancer cells, by using conditional immortalization with a temperature-sensitive mutant of the viral gene. For example, one mutant of the SV40 T-antigen is functional at 33 oC, but conformationally inactive at 39 ºC (REF. 34). Cells conditionally immortalized with this construct grow exponentially at the permissive temperature (33 ºC), but stop dividing and can express tissue-specific features at the non-permissive temperature (39 ºC)35,36. However, there is often a degree of ‘leakiness’, where dividing cells escape and grow at the non-permissive temperature. Continuous cancer cell lines Generally, it is only highly aggressive cancers that have accumulated the genetic changes necessary for unlimited growth in vitro that spontaneously become continuous cell lines. Cancer cell lines tend to be grown in commercial tissue culture medium that contains fetal calf serum, under which the main selection pressure is for proliferative cells.

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 1 | DECEMBER 2000 | 2 3 3

PERSPECTIVES tissue of origin, and consistent differences between cell lines of different origins9. Drug sensitivity. In contrast to most other solid cancers in adult humans, testicular germ cell tumours are cured in over 80% of cases using cisplatin-based combination chemotherapy. Cell lines derived from testis tumours retain their hypersensitivity to cisplatin and other DNA-damaging agents10. Not representing cancer type

Are cancer cell lines derived from a particular type of cancer representative of the clinical spectrum of cancers at that site? This is a completely different question, and the answer this time is a resounding no. In the clinic, cancer is classified by stage and grade (BOX 2). The cancers that yield continuous cell lines tend to be fast growing, high stage and poorly differentiated tumours that have accumulated the mutations required for indefinite growth in vitro. There are few cell lines derived from primary well-differentiated cancers, which form the majority of some

types of cancer, such as bladder tumours11. The genetic changes required to immortalize cells (BOX 1) are mostly late events in cancer progression, and therefore it is not surprising that most primary cancers are not immortal. Also, few scientists have had the interest, the patience or the funding to develop cell lines from slow-growing cancers. Consequently, the population doubling times of cancer cell lines are mostly short. This deficit reflects the greater difficulty of establishing slow-growing tumour cell lines and the preference of the scientific community for fast-growing cells that are easy to handle. Quality control

Careful, regular quality control is a vital, but sadly often neglected, part of cell culture. This neglect has played an important part in tarnishing the reputation of in vitro cancer models. What are the pitfalls, and how can we avoid them? Genomic instability

The problem. Why do phenotypic and geno-

Box 2 | Cancer stage and grade Cancer is classified in three ways — the site of origin (such as breast or prostate cancer), the stage of the cancer (how far it has spread) and the grade of the cancer (how similar to the normal cells it appears under the microscope). Stage Specific staging classifications are available for each type of cancer and there is a choice of classifications. The stage of the cancer is important for the patient, because the treatment options are mainly dictated by the stage. One of the most widely used staging systems is the tumour, node, metastasis (TNM) system. The details differ for each type of cancer, but the following definitions provide a rough guide. T1: Small localized cancer. T2: Larger cancer, but confined within the organ. T3: Cancer at the limits of the organ. T4: Cancer that has spread locally into other organs. N0: No cancer cells detected in those regional nodes that have been examined. N1: Cancer cells detected in one or more of the regional lymph nodes examined. M0: No metastases. M1: Metastases present. The classification can be refined for cancers where there are reliable serum markers (such as prostate and testis cancer). A more accurate stage is obtained once the surgical specimen has been examined under the microscope, giving the pathological stage (pT). In general, T1 and T2 cancers can be treated by local means (surgery or radiotherapy), but most T3 cancers and virtually all T4, N1 and M1 cancers need systemic treatment. Most cell lines are derived from high-stage cancers that are beyond local treatment. Tumour grading The histopathologist examines sections of cancers under the microscope to determine the pT stage (as described above) and to decide how aggressive the tumour is (grade). Various features are used to determine the grade, including the similarity of the morphology of the cancer to that of the normal tissue from which it is derived, the extent of morphological changes in the nucleus and the frequency of mitotic figures (dividing cells). The grade is also important for the patient, because it gives an idea of the prognosis (likelihood of the cancer progressing). There are many grading systems for each type of cancer, and the following definitions provide only a rough guide. G1: Well-differentiated cancer with good prognosis. G2: Moderately differentiated cancer with intermediate prognosis. G3: Poorly differentiated cancer with bad prognosis.

234

typic differences arise between sublines of the same cancer cell line? The first reason is that cancer cell lines constantly generate variants with phenotypic and/or genotypic differences from the predominant population. If the cells are grown continuously over many generations, faster growing and less representative clones may be selected. Second, if the cells are sent to other laboratories and exposed to different environments (such as media, sera, trypsin, carbon dioxide levels, humidity and temperature), variants that are better adapted to the new conditions are likely to be selected. The uncontrolled passing of cells between laboratories is the cause of much unreliable data12. The solution. As long as cells are not grown indefinitely and passed between laboratories, they retain most of the features of the cancers from which they were derived.Adequate frozen stocks of each cell line should be produced, and users should return to frozen stocks at regular intervals (for example, for adherent cell lines, every 10 passages or at 3-month intervals, whichever is shorter; and for cell lines growing in suspension, such as those derived from leukaemias and lymphomas, every 4–6 weeks). Treated like this, if the cells are kept under identical culture conditions, they are relatively stable phenotypically and genotypically5–8. Cross-contamination

Cross-contamination of cell lines happens in two ways. It can result from poor culture technique when, for various reasons, two cell lines accidentally get into the same culture. After a few passages, there is no trace of the slower-growing cell line, and it has been completely displaced by the faster-growing cell line. The second reason is clerical error — mislabelling of growing cells or frozen stocks. Such accidents can and do occur frequently in any laboratory, and have devastating consequences unless simple quality control measures are adopted. The two most notorious examples of human cell line cross-contamination are ‘KB’ cells and ‘ECV304’ cells.‘KB’ cells are widely used as a model for keratinocytes, despite being HeLa cells, derived from a glandular cancer of the cervix.‘ECV304’ cells are widely used as a model of endothelium, despite being T24 cells, derived from a bladder cancer13. It was first shown over 30 years ago that ‘KB’ cells are in fact HeLa cells14, but this false cell line continues to appear in hundreds of publications every year. A common misconception is that crosscontamination leads to hybrids of the crosscontaminating cell line and the original cell line. But consider what happens during crosswww.nature.com/reviews/molcellbio

| DECEMBER 2000 | VOLUME 1

PERSPECTIVES

Unique bar code for each cell line Cancer cell line Extract DNA Use software to determine number of repeats of each allele (plus manual check)

Multiplex PCR

Electrophoresis

Fluorescently labelled primer pairs

Figure 1 | DNA profiling of short tandem repeats. Every genome, whether it belongs to a crime suspect or a cancer cell line, has a characteristic pattern of repetitive sequences, including short tandem repeats (STRs). These are loci (specific sites in the DNA) at which highly variable numbers of a short repeated sequence (2–6 nucleotides in length) are found. For STR profiling, the number of repeats at 5–10 different loci is calculated by extracting the DNA, adding different primers (fluorescently tagged with coloured dyes — blue, green or yellow) for each locus and amplifying the DNA using the polymerase chain reaction (PCR). The PCR products are then separated by gel electrophoresis and compared with standard size markers (fluorescently tagged with a red dye)37. Cells derived from the same clone will have an identical pattern of markers that can be converted into a series of numbers corresponding to the number of repeats in each allele at each locus. This provides a ‘bar code’ or international reference standard for that cell line. Although small differences can occur between cells that originated from the same culture but have subsequently been cultured separately for long periods of time, they can still be accurately matched to a consensus bar code using appropriate cut-off values

contamination: within a few passages, the faster-growing cell line outgrows the slowergrowing cell line, and all trace of the slowergrowing cell line is gone. There is no evidence that HeLa(KB) or T24(ECV-304) are somatic cell hybrids or have acquired any genetic information from the cells they contaminated. Claims of mixed parentage have little, if any, foundation. Any genotypic or phenotypic differences between the various sublines of HeLa probably reflect the different selection pressures under which they have been placed in different laboratories. The problem. In the 1970s, Walter NelsonRees fought a bitter campaign to expose the

use of cross-contaminated cell lines15 — a thankless task for which he became notorious. He showed that a large proportion of the cell lines being used worldwide and being distributed by cell banks were HeLa cells16,17 — the first human cancer cell line developed18. The campaign was so successful that by the time he retired in 1981, some people thought that all human cancer cell lines were HeLa cells. Sadly, Nelson-Rees’s campaign was rapidly forgotten after his retirement, and the use of cross-contaminated cell lines is now greater than ever. Of the cancer cell lines submitted for cell banking, 17–36% are either from a different individual or

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

from a different species to that claimed19,20. The implication of these figures — that a similar proportion of the publications describing work with human cell lines contain misleading data — is impossible to avoid. It is bizarre that this problem continues to be ignored, particularly as it so easy to detect. The solution. Cross-contamination can be detected by karyotyping21, isozyme analysis22, HLA (human leukocyte antigen)typing23 or DNA fingerprinting24, but none of these methods provides a simple numerical output that can be compared between laboratories. However, DNA profiling (FIG. 1), the technique used by forensic services for the identification of individuals, is now available for cell lines. DNA profiling provides a simple, cheap and universal solution applicable to all human cell lines. If it is adopted routinely and incorporated into standard operating protocols, it will provide an international reference standard for every cell line and prevent all but the most deliberate fraud. Problems such as those outlined above could be tackled by the careful training of all users of cancer cell lines. Information on how best to use cancer cell lines is readily available12,25. Journal editors and referees could also play a part, by requesting evidence that all the cell lines discussed in a research paper have been authenticated. Microbial contamination

The problem. Just as serious is the widespread contamination of cell lines with microorganisms, especially Mycoplasma. On the basis of submissions to cell banks, it is estimated that 15–35% of cell lines are contaminated with Mycoplasma26,27. Mycoplasma infection can have marked effects on gene expression and cell behaviour26 and work done with Mycoplasma-infected cells cannot be regarded as valid. Infection is often at a low level that is undetectable with microscopic techniques, and Mycoplasma is highly infectious and can rapidly spread through all cell stocks. The Mycoplasma problem, like the HeLa problem, is probably only the tip of the iceberg. There are many other insidious microorganisms lurking in cell cultures. For example, Mycobacterium avium has been found in cell lines in London and Berkeley, California (J.R.W.M. and G. Buehring, unpublished observations), and is probably far more widespread. Even more worrying, screening for viruses is completely non-existent in almost all research laboratories. Expression of viral products could influence experiments and biotechnology products, as

VOLUME 1 | DECEMBER 2000 | 2 3 5

PERSPECTIVES well as transcriptome and proteome analyses. Another bad — but extremely common — practice is the inclusion of antibiotics in tissue culture medium. For routine management of cell lines, antibiotics are unnecessary and only provide a cover for inadequate technique12. The antibiotic could reduce an infection to a level where it is not noticed, despite the fact that microorganisms are present. Their products will contribute to DNA, RNA and protein analyses and could alter the behaviour of the cells. Also, antibiotics might influence the behaviour and characteristics of the cells. The solution. Sensitive tests for Mycoplasma contamination (including polymerase chain reaction (PCR)-based assays, indicator cells, and broth and agar culture) are available12,26. All laboratories that use cell lines should test their cell stocks for Mycoplasma. Laboratories that do not test can reasonably make the assumption that all their cells are contaminated with Mycoplasma. Screening for viruses, in addition to Mycoplasma, could be carried out by cell-line suppliers. For example, the German Collection of Microorganisms and Cell Cultures screens all human cell lines for some of the more frequently occurring viruses. In common with cross-contamination, there is a lack of awareness of the magnitude and seriousness of the problem. Scientists using cell lines must be thoroughly trained and educated. Tests for Mycoplasma infection are readily available, and their use should be built into routine laboratory practice. Future needs

Most cancer cell lines have already acquired all the changes needed for the cells to grow as metastatic deposits in distant sites. Consequently, they are of questionable value for studying the changes associated with cancer progression, except perhaps as models with which to reverse these changes. Cell lines are needed from early stage and well-differentiated cancers, with matching cell lines from the corresponding normal tissues from the same patients. We also need cell lines from inherited cancers. Cell lines are essential gene discovery tools in human cancer and more are needed because most molecular genetic changes are much easier to detect using these pure populations of cancer cells. For example, it is difficult to detect homozygous deletions in tumour tissue because of the presence of contaminating non-malignant cells. Identification of such mutations in cell lines is relatively easy28 and can result in the detection of new tumour suppressor genes. Methylation and loss of heterozygosity (often

236

the first evidence of a new tumour suppressor gene) are usually first detected in cell lines29,30. We need to know why, in many types of tissue, the normal cells grow much more readily than the cancer cells — have the cancer cells lost some of the properties needed for them to grow in isolation? One of the main goals for cancer research is to routinely culture every human cancer from every patient. For the patient, this advance will facilitate individualized drug therapy, autologous immunotherapy, and more accurate molecular staging and prognosis using transcriptome and proteome analyses31,32. Most of the criticisms concerning phenotypic and genotypic drift in cell lines are due to lack of quality control. All cell lines need to be authenticated by DNA profiling, and contamination by Mycoplasma and other microorganisms excluded. Once these simple quality control measures are taken by every laboratory working with cancer cell lines, we will be able to rebuff most of the criticisms levelled against these cells. John R. W. Masters is at the Institute of Urology, University College London, 67 Riding House Street, London W1W 7EY, UK. e-mail: j.masters@ucl.ac.uk

Links FURTHER INFORMATION

Laboratory of the Government Chemist | German Collection of Microorganisms and Cell Cultures | Society for In Vitro Biology | Time-lapse movies of five cancer cell lines 1.

O’Hare, M. J. in Human Cancer in Primary Culture (ed. Masters, J. R. W.) 271–286 (Kluwer Academic, Dordrecht, 1991). 2. Hsu, M.-Y., Elder, D. A. & Herlyn, M. in Human Cell Culture Volume 1, Cancer Cell Lines Part 1 (eds Masters, J. R. W. & Palsson, B.) 259–274 (Kluwer Academic, Dordrecht, 1999). 3. Lansford, C. D. et al. in Human Cell Culture Volume 2, Cancer Cell Lines Part 2 (eds Masters, J. R. W. & Palsson, B.) 185–255 (Kluwer Academic, Dordrecht, 1999). 4. Fogh, J., Fogh, J. M. & Orfeo, T. One hundred and twenty-seven cultured human tumour cell lines producing tumours in nude mice. J. Natl Cancer Inst. 59, 221–226 (1977). 5. Wistuba, I. I. et al. Comparison of features of human breast cancer cell lines and their corresponding tumours. Clin. Cancer Res. 4, 2931–2938 (1998). 6. Wistuba, I. I. et al. Comparison of features of human lung cancer cell lines and their corresponding tumours. Clin. Cancer Res. 5, 991–1000 (1999). 7. Drexler, H. G. Review of alterations of the cyclindependent kinase inhibitor INK4 family genes p15, p16, p18 and p19 in human leukemia-lymphoma cells. Leukemia 12, 845–859 (1998). 8. Drexler, H. G. et al. p53 alterations in human leukemialymphoma cell lines: in vitro artefact or prerequisite for cell immortalization? Leukemia 14, 198–206 (2000). 9. Ross, D. T. et al. Systematic variation in gene expression patterns in human cancer cell lines. Nature Genet. 24, 227–235 (2000). 10. Walker, M. C., Parris, C. N. & Masters, J. R. W. Differential sensitivities to chemotherapeutic drugs between testicular and bladder cancer cells. J. Natl Cancer Inst 79, 213–216 (1987). 11. Knuechel, R. & Masters, J. R. W. in Human Cell Culture Volume 1, Cancer Cell Lines Part 1 (eds Masters, J. R. W. & Palsson, B.) 213–230 (Kluwer Academic, Dordrecht, 1999).

12. UKCCCR guidelines for the use of cell lines in cancer research. Br. J. Cancer 82, 1495–1509 (2000). 13. Dirks, W. G., MacLeod, R. A. F. & Drexler, H. G. ECV304 (endothelial) is really T24 (bladder carcinoma): cell line cross-contamination at source. In Vitro Cell Dev. Biol. Anim. 35, 558–559 (1999). 14. Gartler, S. M. Genetic markers as tracers in cell culture. Natl Cancer Inst. Monogr. 26, 167–195 (1967). 15. Gold, M. A Conspiracy of Cells. One Woman’s Immortal Legacy and the Scandal It Caused (State University of New York, Albany,1986). 16. Nelson-Rees, W. A., Flandermeyer, R. R. & Hawthorne, P. K. Banded marker chromosomes as indicators of intraspecies cellular contamination. Science 184, 1093 (1974). 17. Nelson-Rees, W. A., Daniels, D. W. & Flandermeyer, R. R. Cross-contamination of cells in culture. Science 212, 446–452 (1981). 18. Gey, G. O., Coffman, W. D. & Kubicek, M. T. Tissue culture studies of the proliferative capacity of cervical carcinoma and normal epithelium. Cancer Res. 12, 264–265 (1952). 19. Markovic, O. & Markovic, N. Cell cross-contamination in cell cultures: the silent and neglected danger. In Vitro Cell Dev. Biol. Anim. 34, 1–8 (1998). 20. MacLeod, R. A. F. et al. Widespread intraspecies crosscontamination of human tumour cell lines. Int. J. Cancer 83, 555–563 (1999). 21. MacLeod, R. A. F. & Drexler, H. G. in Human Cell Culture Volume 3, Leukemias and Lymphomas (eds Masters, J. R. W. & Palsson, B.) 373–399 (Kluwer Academic, Dordrecht, 2000). 22. Povey, S., Hopkinson, D. A., Harris, H. & Franks, L. M. Characterisation of human cell lines and differentiation from HeLa by enzyme typing. Nature 264, 60–63 (1976). 23. O’Toole, C. M., Povey, S., Hepburn, P. & Franks, L. M. Identity of some human bladder cancer cell lines. Nature 301, 429–430 (1981). 24. Gilbert, D. A. et al. Application of DNA fingerprints for cell line individualization. Am. J. Hum. Genet. 47, 499–514 (1990). 25. Freshney, R. I. Culture of Animal Cells. A Manual of Basic Technique 4th edn (Wiley-Liss, New York, 2000). 26. Drexler, H. G. & Uphoff, C. C. in The Encyclopedia of Cell Technology (eds Spier, R. E., Griffiths E. & Scragg, A. H.) 609–627 (Wiley, New York, 2000). 27. Hay, R. J., Macy, M. L. & Chen, T. R. Mycoplasma infection of cultured cells. Nature 339, 487–499 (1989). 28. Virmani, A. K. et al. Promoter methylation and silencing of the retinoic acid receptor-β gene in lung carcinomas. J. Natl Cancer Inst. 92, 1303–1307 (2000). 29. Virmani, A. K. et al. Allelotyping demonstrates common and distinct patterns of chromosomal loss in human lung cancer types. Genes Chrom. Cancer 21, 308–319 (1998). 30. Shivapurkar, N. et al. Multiple regions of chromosome 4 demonstrating allelic losses in breast carcinomas. Cancer Res. 59, 3576–3580 (1999). 31. Masters, J. R. W. & Lakhani, S. How microarrays can help cancer patients. Nature 404, 921 (2000). 32. Alizadeh, A. A. et al. Distinct types of diffuse large B-cell lymphoma identified by gene expression profiling. Nature 403, 503–511 (2000). 33 Schaeffer, W. I. Usage of vertebrate, invertebrate and plant cell, tissue and organ culture terminology. In vitro 20, 19–24 (1984). 34 Jat, P. S. & Sharp, P. A. Cell lines established by a temperature-sensitive simian virus 40 large-T-antigen are growth restricted at the nonpermissive temperature. Mol. Cell. Biol. 9, 1672–1681 (1989). 35. Stamps, A. C., Davies, S. C., Burman, J. & O’Hare, M. J. Analysis of proviral integration in human mammary epithelial cell lines immortalized by retroviral infection with a temperature-sensitive SV40 T-antigen construct. Int. J. Cancer 57, 865–874 (1994). 36. Simon, L. V., Beauchamp, J. R., O’Hare, M. & Olsen, I. Establishment of long-term myogenic cultures from patients with Duchenne muscular dystrophy by retroviral transduction of a temperature-sensitive SV40 large T antigen. Exp. Cell Res. 224, 264–271 (1996). 37. Thomson, J. A., Pilotti, V., Stevens, P., Ayres, K. L. & Debenham, P. G. Validation of short tandem repeat analysis for the investigation of cases of disputed paternity. Forensic Sci. Int. 100, 1–16 (1999).

Acknowledgements I thank Alan Entwistle, Ludwig Institute for Cancer Research, University College London for help preparing the time-lapse movies and Jim Thomson, Laboratory of the Government Chemist, London, UK for providing images for Figure 1.

www.nature.com/reviews/molcellbio

| DECEMBER 2000 | VOLUME 1