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

toc_r

Page 1 of 2

CONTENTS

October 2000 Vol 1 No 1

1 | In this issue doi:10.1038/35036000

Highlights PDF

[1332K]

3 | CELL GROWTH A cell cycle controller rewrites its CV doi:10.1038/35036002

4 | WEB WATCH Networking with proteins doi:10.1038/35036005

4 | PROTEOMICS This way up and handle with care doi:10.1038/35036007

5 | MEMBRANE TRAFFIC SNARE hypothesis 2000 doi:10.1038/35036010

5 | CELL DIVISION Another star on drugs

11 | THE VERSATILITY AND UNIVERSALITY OF CALCIUM SIGNALLING Michael J. Berridge, Peter Lipp & Martin D. Bootman doi:10.1038/35036035 [1276K]

22 | A TALE OF TOROIDS IN DNA METABOLISM Manju M. Hingorani & Mike O'Donnell doi:10.1038/35036044 [1075K]

31 | LIPID RAFTS AND SIGNAL TRANSDUCTION Kai Simons & Derek Toomre doi:10.1038/35036052 [604K]

40 | NEW TARGETS FOR INHIBITORS OF HIV-1 REPLICATION John P. Moore & Mario Stevenson doi:10.1038/35036060 [1139K]

doi:10.1038/35036012

6 | IN BRIEF PRIONS | CELL ADHESION | ALZHEIMER'S DISEASE | CELL DIVISION doi:10.1038/35036025

6 | INOSITOL PHOSPHATES From duckweed to DNA... doi:10.1038/35036017

7 | DNA METABOLISM ... and from repair to remodelling doi:10.1038/35036020

7 | WEB WATCH Untangling inositol doi:10.1038/35036023

8 | IN BRIEF NUCLEAR TRANSPORT | CELL POLARITY | ANGIOGENESIS doi:10.1038/35036015

8 | APOPTOSIS DIABLO is double trouble doi:10.1038/35036027

50 | WALKING ON TWO HEADS: THE MANY TALENTS OF KINESIN G端nther Woehlke & Manfred Schliwa doi:10.1038/35036069 [823K]

59 | MULTILEVEL REGULATION OF THE CIRCADIAN CLOCK Nicolas Cermakian & Paolo Sassone-Corsi doi:10.1038/35036078 [985K]

69 | Timeline THE PAST, PRESENT AND FUTURE OF MOLECULAR COMPUTING Adam J. Ruben & Laura F. Landweber doi:10.1038/35036086 [1290K]

72 | TIMELINE HAYFLICK, HIS LIMIT, AND CELLULAR AGEING

http://www.nature.com.ezproxy.library.wisc.edu/nrm/journal/v1/n1/toc_r.html

12/3/2006

toc_r

Page 2 of 2

8 | TELOMERES Caps and cancer doi:10.1038/35036029

9 | STRUCTURAL BIOLOGY Pocket the difference doi:10.1038/35036032

Jerry W. Shay & Woodring E. Wright doi:10.1038/35036093 [352K]

76 | OPINION CANCER: LOOKING OUTSIDE THE GENOME Judah Folkman, Philip Hahnfeldt & Lynn Hlatky doi:10.1038/35036100 [3281K]

80 | NatureView doi:10.1038/35036104

NATURE REVIEWS | MOLECULAR CELL BIOLOGY

http://www.nature.com.ezproxy.library.wisc.edu/nrm/journal/v1/n1/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

STEPHEN P. JACKSON WELLCOME/CRC INSTITUTE, CAMBRIDGE, UK ROBERT JENSEN JOHNS HOPKINS UNIVERSITY, BALTIMORE, MD, USA VICTORIA 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

A cell cycle controller rewrites its CV Any cell biology textbook will tell you that the main function of D-type cyclins is to get cells through the G1 phase of the cell cycle. They do this by activating one of two cyclin-dependent kinases, CDK4 or CDK6, which hyperphosphorylate the retinoblastoma protein RB, preventing it from co-repressing the transcription factor E2F. A quick glance at the first of the two EMBO Journal papers summarized here might leave you thinking that cyclin D–CDK4 is out of a job, but closer examination reveals that this is far from true: it may be dispensable for G1 progression, but it’s added a new skill to its portfolio — as a key regulator of cell growth (increase in cell mass). One of the main problems with studying the functions of cyclin–CDK complexes in vertebrates is that there are so many of them: redundancy confounds knockout studies. These papers side-step the problem by turning to Drosophila, which has only one D-type cyclin (CycD), one Cdk4 and two RB-family members, Rbf and Rbf2. Claas Meyer and colleagues isolated a deletion of the Cdk4 gene that blocks its kinase activity. They expected it to be lethal, but instead they got small flies with fewer, slowly growing cells. How do these cells get through G1 phase? Heterozygosity for CycE or Cdk2 (but not other cyclins and CDKs) was lethal in the Cdk4 homozygous mutants, indicating that CycE–Cdk2 can ‘temp’ for CycD–Cdk4. This is surprising: in mammals Cdk4 and Cdk2 are

thought to work sequentially and Cdk2 can’t access its phosphorylation sites on RB without prior phosphorylation by Cdk4. Clearly this isn’t the case in flies. The small phenotype in Cdk4mutant flies implies that CycD–Cdk4 promotes cell growth. Sanjeev Datar and co-workers explored this by overexpressing CycD and/or Cdk4 with green fluorescent protein in different cell types. In proliferating cells from the developing wing, overexpression of both proteins had no effect on cellcycle phasing, whereas overexpres-

sion of CycE shortened G1 phase, leading to smaller cells. Studies of clones in the fly wing showed that the cells’ doubling time was increased, but these cells were no smaller than their wild-type counterparts, indicating that CycD–Cdk4 causes cell mass to increase. More marked were the results in the postmitotic cells of the adult eye: CycD–Cdk4 overexpression caused the eyes to bulge because the ommatidia (single facets of the fly’s compound eye) contained more, larger cells (see picture). CycD–Cdk4 can, therefore, cause growth even in

Drosophila eye showing CycD–Cdk4-overexpressing clones (green). Courtesy of Bruce A. Edgar and Sanjeev A. Datar, Fred Hutchinson Cancer Research Center, Seattle, Washington, USA.

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 1 | OCTOBER 2000 | 3

▲

NOBUTAKA HIROKAWA UNIVERSITY OF TOKYO, JAPAN

CE LL GROWTH

HIGHLIGHTS

Networking with proteins Creating user-friendly databases of protein–protein interactions is a tough challenge. Pronet — a curated database of published protein–protein interactions produced by Myriad Genetics — is a brave attempt to address the problem. Searching the database couldn’t be simpler: you just type in a keyword and hit ‘go’. This returns a list of proteins whose database entries contain that word. Clicking on any entry in this list will take you to that protein’s homepage, which contains a graphical representation of the protein’s domain structure, links to other relevant databases, and sequence information. But Pronet’s killer application is the addictive ‘view interactions graphically’ feature, which takes you to a page containing your protein in a box. Clicking on the box creates a spider diagram showing all the proteins it interacts with, and you can expand the network as far as you like. A ‘mouse mode’ menu allows you to delete proteins, move them, or squeeze the network to make room for more interactions. What’s more, if you come across a protein you’re unfamiliar with, choosing ‘info’ from the mouse mode menu will link you back to that protein’s data entry page. Pronet does have limitations: it contains only human sequences, although there are plans to include other species. It also records only interactions found using the yeast two-hybrid system, which creates some idiosyncrasies. For example, the c-Src page states that there are ‘no recorded interactions for this protein’. Links to papers describing the interactions would also be useful. That said, as more data are added, Pronet will evolve into an invaluable tool for anyone wanting to track protein–protein interactions.

▲

WEB WATCH

cells that aren’t cycling. Similar results were obtained in the salivary gland, an organ that undegoes endoreduplication (cell division without cytokinesis). Datar and colleagues turned back to the wing to study whether CycD–Cdk4 exerts its effects through Rbf. As expected, overexpression of Rbf alone slowed cell division, whereas cells expressing all three proteins had near normal cell divison rates but were larger, indicating that cell growth was promoted even while the cell cycle was being slowed by Rbf. In the eye, by contrast, Rbf overexpression didn’t influence postmitotic growth, and insertion of a null Rbf allele had

Cath Brooksbank

References and links ORIGINAL RESEARCH PAPERS Meyer, C. A. et

al. Drosophila Cdk4 is required for normal growth and is dispensable for cell cycle progression. EMBO J. 19, 4533–4542 (2000) | Datar, S. A. et al. The Drosophila cyclin D–Cdk4 complex promotes cellular growth. EMBO J. 19, 4543–4554 (2000) REVIEW Sherr, C. J. & Roberts, J. M. CDK inhibitors: positive and negative regulators of G1phase progression. Genes Dev. 13, 1501–1512 (1999) FURTHER READING Cockcroft, C. E. et al. Cyclin D control of growth rate in plants. Nature 405, 575–579 (2000) | Rane, S. G. et al. Loss of Cdk4 expression causes insulin-deficient diabetes and Cdk4 activation results in β-islet cell hyperplasia. Nature Genet. 22, 44–52 (1999) | Tsutsui, T. et al. Targeted disruption of CDK4 delays cell cycle entry with enhanced p27Kip1 activity. Mol. Cell Biol. 19, 7011–7019 (1999) FURTHER INFORMATION Cyclins and E2F: a Kohn interaction map | The interactive fly: cell cycle genes | Mitosis world

P R OT E O M I C S

This way up and handle with care Proteins are beastly to work with: they denature at the drop of a hat and have an annoying tendency to regulate their social interactions by post-translational modifications. Small wonder, then, that researchers wanting a highthroughput readout of cellular behaviour use DNA microarrays to look at messenger RNA levels instead, even though they don’t necessarily correlate with protein activity. All that might be about to change: in the 8 September issue of Science, Gavin MacBeath and Stuart Schreiber report that they can make microarrays of functionally active proteins, and can use them to measure interactions with other proteins and small molecules. Two hurdles had to be leapt: keeping the proteins active and getting them in the right orientation. A third goal was to make the technology compatible with existing microarray analysis tools. With hindsight, the solutions to these problems turned out to be laughably simple: use the gear that prints commercially available DNA microarrays, put 40% glycerol in your buffers to prevent dehydration of the nanolitre volumes applied, and coat your slides with a reagent that reacts with primary amines. This

Cath Brooksbank

no effect on cell growth in either the wing or the eye. CycD–Cdk4 must, therefore, be promoting cell growth by phosphorylating targets other than Rbf. So, rather than being dedicated to getting cells through G1, CycD–Cdk4 promotes hyperplasia (increased numbers of cell divisions) in dividing cells, hypertrophy (increased cell size) in endoreduplicating cells and both in postmitotic cells — and it doesn’t need Rbf to carry out any of these functions. The solution to the next challenge — determining the growthpromoting target of CycD–Cdk4 — might well have all our eyes bulging.

captures proteins by their amino termini or by surface-exposed lysine residues, so each protein gets stuck to the glass in a range of different orientations, one of which is almost bound to be the right way up. The slides are then quenched with bovine serum albumin (BSA) which, as well as blocking any unreacted groups, lowers background noise when the slides are probed with other proteins. These simple tricks have allowed MacBeath and Schreiber to print proteins at densities of 1,600 spots per square centimetre. Now pick your favourite protein function. Do you want to find new protein–protein interactions? Or hunt for new substrates for your pet protein kinase? Or are you more interested in finding out what proteins your library of drug

candidates binds to? The researchers did proof-of-principle experiments to show that all of these applications are feasible by flooding the slides with fluorophore-tagged proteins, kinase substrates in the presence of radiolabelled ATP, or synthetic ligands coupled to fluorescently labelled BSA. Although most of these experiments were done using a small number of arrayed protein spots, they also work in the context of a chip containing over 10,000 spots: a single spot of the FKBP12–rapamycin binding protein (FRB) can easily be located in a sea of protein-G spots when probed with a mixture of two fluorophore-tagged proteins — one binding to FRB, the other binding to protein G. The greatest barrier to commercial availability of these protein microarrays will be purification of the proteins to put on them. Let’s hope that the current trend in automation obviates the need for arrays of protein biochemists, cursing in cold rooms over jammed fraction collectors. Cath Brooksbank References and links ORIGINAL RESEARCH PAPER MacBeath,

G. & Schreiber, S. L. Printing proteins as microarrays for high-throughput function determination. Science 289, 1760–1763 (2000)

www.nature.com/reviews/molcellbio

| OCTOBER 2000 | VOLUME 1

HIGHLIGHTS

MEMBRANE TRAFFIC

SNARE hypothesis 2000 A question that has kept scientists busy for some time is how specificity is achieved during membrane traffic. In other words, how does a vesicle distinguish, say, the plasma membrane from a lysosome? In 1993, Jim Rothman proposed a working model — the ‘SNARE hypothesis’ — and variations on this theme have dominated membrane fusion research ever since. The model has been seriously challenged over the past few years, but in the 15 September issue of Nature, the Rothman and Söllner laboratories present new evidence to support it. According to the original hypothesis, transmembrane SNARE proteins present on the vesicle and target membranes pair up in antiparallel fashion to dock the membranes. The trans-SNARE complex consists of a syntaxin and a SNAP25 family member on the target membrane and a VAMP family member on the vesicle. The binding of two soluble proteins, NSF and SNAP, to the SNARE complex then drives membrane fusion. The specificity of SNARE pairing dictates the specificity of membrane recognition. Many mechanistic predictions of the SNARE hypothesis have since been disproved. First, the function of NSF and SNAP is not to drive fusion but to open SNARE complexes to allow their recycling after fusion has occurred. Then, membrane docking can occur in the absence of SNAREs. Next, SNAREs are promiscuous in their interactions, at least in vitro. Last, the genome of the yeast Saccharomyces cerevisiae does not contain enough SNAREs to specify all the transport steps in the cell. The Rothman and Söllner laboratories have now reconstituted yeast SNAREs into liposomes in all possible combinations and measured membrane mixing as a readout for correct pairing. They found remarkable specificity in SNARE pairing, as out of 33 possible combinations, only three known, one suspected and one novel SNARE combination occurred.

So, whereas SNAREs pair up almost randomly in solution, they are not at all promiscuous in the presence of lipid bilayers. Rothman and colleagues also showed that SNAREs can only interact productively with their partners if they are in the correct orientation. So if a SNARE ends up on the wrong membrane after fusion has occurred, it won’t be able to take part in a functional complex with its normal partners because it will be positioned upside down. Only after it has found its way back to its residence membrane can it form a fusogenic complex with its partners. Using a similar approach, Söllner and colleagues found that SNARE complexes are made up of four SNAREs and not three. Indeed, the two α-helices contributed by SNAP25 to the neuronal SNARE complex probably stem from two separate SNAREs in most other complexes. By using four SNAREs instead of three and by applying a combinatorial approach to SNARE pairing, there may be enough SNAREs after all to allow each SNARE complex to act in only one transport step. So, although parts of the SNARE hypothesis have been disproved, its core postulate seems to have survived the test of time — formation of SNARE complexes confers specificity to membrane fusion. The regulation of SNARE-mediated membrane fusion by Rab proteins, tethering factors and proteins of the Munc family should now keep us busy for the next few years. Raluca Gagescu References and links ORIGINAL RESEARCH PAPERS McNew, J. A. et al. Compartmental specificity of cellular membrane fusion encoded in SNARE proteins Nature 407, 153–159 (2000) | Parlati, F. et al. Topological restriction of SNARE-dependent membrane fusion. Nature 407, 194–198 (2000) | Fukuda, R. et al. Functional architecture of an intracellular membrane t-SNARE. Nature 407, 198–202 (2000) NEWS AND VIEWS Scales, S. J. et al. The specifics of membrane fusion. Nature 407, 144–146 (2000) REVIEW Jahn, R. & Südhof, T. C. Membrane fusion and exocytosis. Annu. Rev. Biochem. 68, 863–911 (1999)

Courtesy of T. Kapoor and T. Mayer, Harvard Medical School, Massachusetts, USA.

CELL DIVISION

Another star on drugs The function of the microtubule spindle during mitosis is to distribute replicated DNA equally between daughter cells. Last year Mitchison and colleagues discovered a new drug that interferes with spindle formation and they now show that this drug can be useful for the study of spindle dynamics. Monastrol is a small, membrane-permeant compound whose action is rapidly reversible. This makes it a perfect tool for the study of mitotic spindles. Instead of targeting microtubules like many other drugs, monastrol specifically inhibits the motility of the mitotic kinesin Eg5, resulting in the formation of monopolar spindles, also called monoasters. Kapoor et al. found that, in the presence of monastrol, centrosomes duplicate but do not separate. If the drug is added to preformed spindles in a cell-free assay, the two poles move towards each other until they merge. This confirms that Eg5 functions in centrosome separation and in the maintenance of microtubule crosslinking at the spindle midzone. Moreover, Kapoor et al. made two interesting observations using monastrol. First, they discovered forces orthogonal to the spindle axis that had previously not been recognized. They also found that the spindle checkpoint protein Mad2 is probably sensitive to the exact number of microtubules at each kinetochore, tension at each kinetochore or the dynamic status of the kinetochore. These observations now need to be followed up. So monastrol has been validated as a tool to study spindle dynamics, and has opened two new avenues for research. Raluca Gagescu References and links ORIGINAL RESEARCH PAPER Kapoor, T. M. et al. Probing spindle assembly mechanisms

with monastrol, a small molecule inhibitor of the mitotic kinesin Eg5. J. Cell Biol. 150, 975–988 (2000) FURTHER READING Mayer, T. U. et al. Small molecule inhibitor of mitotic spindle bipolarity identified in a phenotype-based screen. Science 286, 971–974 (1999)

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 1 | OCTOBER 2000 | 5

HIGHLIGHTS

IN BRIEF

I N O S I T O L P H O S P H AT E S

From duckweed to DNA…

PRIONS

Species-barrier-independent prion replication in apparently resistant species. Hill, A. F. et al. Proc. Natl Acad. Sci. USA 97, 10248–10253 (2000)

Until now, it was believed that a species barrier limits infection by prions from different species. For example, mice cannot be infected with hamster prions but humans can be infected with cow prions. Hill et al. now show that hamster prions can replicate in mice — they just do not cause prion disease. But if either mice or hamsters are inoculated with brain extracts from these sub-clinically infected mice, they do develop the disease. CELL ADHESION

Two cell adhesion molecules, nectin and cadherin, interact through their cytoplasmic domain-associated proteins. Tachibana, K. et al. J. Cell Biol. 150, 1161–1175 (2000)

Nectin is a calcium-independent immunoglobulin-like adhesion molecule that can trans-interact with itself to form a cell–cell adhesion system, similar to the well-known cadherin system. The nectin and cadherin adhesion complexes interact with each other through their cytoplasmic domain-associated proteins l-afadin and catenins, respectively. Although the two systems cooperate in organizing adherens junctions, their hierarchy is not known. ALZHEIMER’S DISEASE

Nicastrin modulates presenilin-mediated notch/glp-1 signal transduction and βAPP processing. Yu, G. et al. Nature 407, 48–54 (2000)

Processing of the β-amyloid precursor protein (βAPP) to the amyloid-β peptide is central in the path to Alzheimer’s disease, and this paper reports the identification of a new player in the process. Called nicastrin, this protein interacts with βAPP, and also with presenilins (PS) 1 and 2. The authors believe it to be a functional component of the PS1 and PS2 complexes, and show that, like them, nicastrin is involved in processing both βAPP and Notch. CELL DIVISION

Analysis of mid1p, a protein required for placement of the cell division site, reveals a link between the nucleus and the cell surface in fission yeast. Paoletti, A. & Chang, F.

What does a tiny aquatic flowering plant have in common with a process for repairing broken DNA molecules? The answer is that both require a phosphorylated cyclic alcohol called inositol hexakisphosphate (InsP6), and, for the DNA-repair process at least, a report by Hanakahi and colleagues in the 15 September issue of Cell explains why. The story revolves around the catalytic subunit of the DNA-dependent protein kinase (DNA-PKcs), which helps to repair double-stranded DNA breaks using the non-homologous end-joining (NHEJ) pathway. The DNA-PKcs is first targeted to each of the severed ends by a complex of the Ku70 and Ku80 proteins (see figure). DNA-PKcs then recruits a complex of DNA ligase IV and XRCC4, resulting in rejoining of the broken ends. In vitro assays have shown that although all of these players — Ku70, Ku80, DNA-PKcs, XRCC4 and DNA ligase IV — are needed for NHEJ, they are not sufficient. So something is missing, and to discover what, Hanakahi et al. used an in vitro complementation assay. They took cell-free extracts that could promote NHEJ, fractionated them on a phosphocellulose column, then added back various combinations of the fractions to restore the NHEJ activity. In so doing, the authors identified a fraction containing the elusive factor, which they named stimulatory factor A (SFA). The task of identifying SFA involved eight purification steps followed by a barrage of tests, including protease digestion, nuclease digestion and even boiling. But the activity survived all of these insults intact, indicating — unexpectedly — that SFA is neither a protein nor a nucleic acid. So what is it? The authors concluded that the active ingredient is, in

fact, InsP6, and showed that the component of the NHEJ machinery to which it binds is probably DNA-PKcs. They suggest several possible functions for InsP6 in NHEJ, including the idea that it converts DNA-PKcs from an inactive to an active form. Interestingly, DNA-PKcs, as well as several other DNA-repair proteins such as ataxia telangiectasia mutated (ATM) and ATR, contains motifs characteristic of phosphatidylinositol3-OH kinases (PI(3)K). As none of these proteins has lipid kinase activity, why have they retained such motifs? Hanakahi et al. speculate that such proteins may have evolved from a common ancestor with both protein and lipid kinase functions, and that mutation of the PI(3)K domain caused these proteins to lose their lipid kinase activity but retain the ability to bind headgroups containing inositol polyphosphates. However, there is no evidence that this is where the InsP6 binds DNA-PKcs — indeed, there is a big charge difference between the phosphatidylinositol-3,4,5-trisphosphate headgroup and InsP6 — and it may turn out that InsP6 binds to an allosteric activation site on DNA-PKcs. Either way, InsP6 is a molecule of the moment. As well as making up an enormous 60% of the volume of the aquatic plant duckweed, where it acts as a kind of ballast, InsP6 is thought to act as an antioxidant and phosphate storage source in plant seeds. In mammalian cells it has been implicated in regulating growth, inflammation and nervous transmission, and it is even thought to be involved in export of messenger RNA from the nucleus. Nonetheless, the link to DNA repair comes as a surprise, and should trigger a search for InsP6 -binding sites in other members of the PI(3)K family.

XRCC4

ORIGINAL RESEARCH PAPER Hanakahi, L. A.,

Ku 80

Ku 70

Ligase IV

Ku 70 ?

InsP6

Ku 80 ?

DNA-PKcs

In fission yeast, a contractile ring forms in the middle of the cell in early mitosis, and its position depends on the position of the nucleus. The position of the plane of cell division with respect to the nucleus is probably defined by mid1p. This protein shuttles between the nucleus and a broad band at the medial cell surface, but how exactly does it convey the positional information?

DNA-PKcs

Mol. Biol. Cell 11, 2757–2771 (2000)

Alison Mitchell References and links Bartlet-Jones, M., Chappell, C., Pappin, D. & West, S. C. Binding of inositol phosphate to DNAPK and stimulation of double-strand break repair. Cell 102, 721–729 (2000) REVIEW Smith, G. C. M. & Jackson, S. P. The DNA-dependent protein kinase. Genes Dev. 13, 916–934 (1999)

www.nature.com/reviews/molcellbio

| OCTOBER 2000 | VOLUME 1

HIGHLIGHTS

WEB WATCH

D N A M E TA B O L I S M

… and from repair to remodelling The paststarts ten years bodyfirst hereor so have seen many global

boundaries come down — from economic and geographical borders to the Berlin wall. This is the case in science too, where the lines that have traditionally divided many areas of molecular and cell biology are beginning to blur. Molecules identified in one process are increasingly being implicated in other — often unexpected — pathways. Take the TIP60 protein, for example. Originally isolated as a protein that interacts with HIV-1 Tat, it was subsequently shown to be a histone acetyltransferase. And now, according to a paper by Yoshihiro Nakatani and collaborators in Cell, TIP60 may also be involved in DNA repair and apoptosis. The first step was purification of the TIP60 complex from HeLa cells. Initial studies of TIP60’s histone acetyltransferase activity had used TIP60 monomers, and there was a problem with that — activity was detected only when free histones were used, and not with the more physiologically relevant nucleosomes. But Nakatani and colleagues showed that the TIP60 complex, which contains at least 14 subunits, can acetylate both substrates. The authors then took a closer look at the complex. Using mass spectrometry they discovered that the two 54 kDa subunits, christened TAPs — for ‘TIP60associated proteins’ — were familiar faces, corresponding to the eukaryotic RuvB-like proteins RUVBL1 and RUVBL2, respectively. RuvB is a ring-shaped protein required for DNA recombination and repair in Escherichia coli (see the review by Hingorani and O’Donnell on page 22 for more information). Because RuvB has intrinsic ATPase and helicase activities, Nakatani and co-workers first confirmed that these are also functions of the TIP60 complex. Then, given that the TIP60 complex contains what seem to be functional counterparts of RuvB, the authors wondered whether it — and, specifically, its histone acetyltransferase activity — might have some function in DNA repair. To test their idea, they ectopically expressed a mutated TIP60 with no acetyltransferase activity in HeLa cells. They then watched how these cells behaved after γirradiation, which generates double-stranded DNA breaks, compared with controls (cells expressing wildtype TIP60 or with no ectopic TIP60 at all). The result was that, in the controls, at least 40% of the breaks were repaired within 30 minutes of γ-irradiation. But only 5% of the damage in the mutant cells was mended over the same period. Normally, in cells with irreparable DNA damage, an apoptotic death pathway is activated. So Nakatani and colleagues looked for evidence of this in the cells expressing mutated TIP60. But after 12 hours of γirradiation these cells showed no apoptosis at all,

indicating that the suicide response was also affected by the TIP60 mutation. It’s therefore likely that the TIP60 complex interacts with checkpoint proteins to activate a cellular suicide programme in response to DNA damage. These results raise plenty of questions, but they could be a step towards understanding DNA repair and signalling pathways in the real world of chromosomes and chromatin. Many in vitro studies use naked DNA as a substrate, but in living cells the enzymes have nucleosomes to contend with. Until now histone acetyltransferases have been implicated mainly in transcriptional activation, where they are thought to relax the structure of the chromatin so that transcription factors can access the DNA. But DNA repair enzymes presumably face many of the same challenges, so it would make a lot of sense to have both activities in a single complex. It turns out that other subunits of the TIP60 complex have already turned up in several different chromatin-remodelling complexes, indicating that this may be a general phenomenon — maybe not a global one, but certainly one that blurs the boundary between chromatin biology and DNA metabolism. Alison Mitchell References and links ORIGINAL RESEARCH PAPER Ikura, T. et al. Involvement of the TIP60

histone acetyltransferase complex in DNA repair and apoptosis. Cell 102, 463–473 (2000) REVIEW Sterner, D. E. & Berger, S. L. Acetylation of histones and transcription-related factors. Microbiol. Mol. Biol. Rev. 64, 435–459 (2000)

Untangling inositol Where do you turn if your research flings you into a completely new field? The story on page 6 will doubtless have experts in DNA repair grappling with myo-inositol and its extended family; but if you don’t know your phytate from your pentakisphosphates, help is at hand. Most of the researchers who study inositol phosphates view the calcium-releasing second messenger inositol-1,4,5trisphosphate as the centre of their universe (see the review by Michael J. Berridge and colleagues on page 11 of this issue), but Stephen Shears at the US National Institute of Environmental Health Sciences has a different perspective: he has put together an online tutorial on the more-highly phosphorylated inositol polyphosphates, with inositol hexakisphosphate (InsP6, also known as phytate) at its hub. The front page provides a map of inositol polyphosphate metabolism, and clicking on different sections of it takes you to reviews on the metabolism and functions of just about every inositol polyphosphate known to exist. There’s some valuable historical background information (see, for example, the section on the Ins(1,4,5)P3–Ins(1,3,4,5)P4 cycle) and fascinating insights into the variety of cellular processes that these molecules have been implicated in. Each review has an extensive reference list and is regularly updated. One thing that is missing, however, is a guide to their nomenclature. A myo-inositol guide written by the International Union of Biochemistry and Molecular Biology’s committee of nomenclature experts explains all. It may shock you to learn that this was written in 1988, but it is comforting to know that one thing in this field has stayed the same for 12 years.

Cath Brooksbank

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 1 | OCTOBER 2000 | 7

HIGHLIGHTS

IN BRIEF

A P O P TO S I S

N U CLE A R TRA N S PO RT

Binding of the Mex67p/Mtr2p heterodimer to FXFG, GLFG, and FG repeat nucleoporins is essential for nuclear mRNA export. Strässer, K., Bassler, J. & Hurt, E. J. Cell Biol. 150, 695–706 (2000)

Messenger RNA export through the nuclear pore involves a number of nucleoporins and the Mex67p/Mtr2p complex, but its mechanism is not well understood. Hurt and colleagues show that Mex67p and Mtr2p bind as a heterodimer to RNA and also to several conserved nucleoporin motifs. They propose that transport through the pore is directed by sequential interactions with first the Nup82 and then the Nup116 complex. This mechanism resembles that used by importin family members to transport proteins through the nuclear pore. CELL POLARITY

CHE-14, a protein with a sterol-sensing domain, is required for apical sorting in C. elegans ectodermal epithelial cells. Michaux, G. et al. Curr. Biol. 10, 1098–1107 (2000)

It’s a mystery how trafficking of proteins to the apical or basolateral membranes of polarized cells is controlled, but evidence implicates proteins containing sterol-sensing domains. Here, Michaux and colleagues characterize CHE-14, a new member of the Patched family of sterol-sensing proteins, in Caenorhabditis elegans. CHE-14 is most closely related to Dispatched, a Drosophila protein with 12 putative transmembrane domains that is involved in releasing the secreted signalling molecule Hedgehog from cells. CHE-14 mutants accumulate vesicles at the apical membranes of epithelial cells, and CHE-14 tagged with green fluorescent protein rescues the mutant phenotype and localizes to apical membranes. Deletion of the predicted extracellular loops and transmembrane domains, including the sterol-sensing domain, abolishes its ability to rescue the secretion phenotype, whereas the predicted cytoplasmic loops seem dispensable. The authors propose a model in which CHE-14 and its close relative Dispatched are required for exocytosis, whereas Patched and the Niemann–Pick C protein — another sterol-sensing protein — are required for endocytosis. ANGIOGENESIS

Genes expressed in human tumor endothelium. St. Croix, B. et al. Science 289, 1197–1202 (2000)

To survive and grow, tumours need their own supply of blood. They produce factors to stimulate the formation of new blood vessels, but does the endothelial lining of these vessels differ from that in vessels from healthy tissues? This paper indicates that they do — and dramatically so. The authors compare gene-expression profiles in endothelium derived from normal and tumour tissue, and find that of the 170 transcripts predominantly expressed in the endothelium, 79 are differentially expressed.

DIABLO is double trouble Scientific announcements are a bit like catching a bus — a long wait, and then two arrive at once. This was the case back in July, when the groups of Xiaodong Wang in Dallas and David Vaux in Melbourne independently described a new mammalian protein that promotes apoptosis. Its activity has now been further characterized, as Wang and colleagues report in Nature. The protein in question — named Smac by Wang’s group and DIABLO by Vaux and colleagues — promotes apoptosis by binding to and antagonizing members of the IAP (inhibitor of apoptosis protein) family. These IAPs have an anti-apoptotic activity because they bind and inhibit caspases, the key effectors of cell death. In Drosophila, the activity of IAPs is countered by the imaginatively named Reaper, Hid and Grim proteins — hence the search for mammalian functional homologues. And, as described in the two initial reports, Smac/DIABLO fits the bill. It

promotes apoptosis by binding to the IAPs, and prevents them from sequestering caspases. But how does it do this? Wang and colleagues now provide clues with the crystal structure of Smac/DIABLO at a resolution of 2.2 Å. The structure reveals that Smac/DIABLO forms a homodimer in solution, so the authors first asked whether this is the functional form. They engineered missense mutations to perturb dimer formation, and found that the interaction of monomers with an IAP (XIAP) was indeed disrupted (although not abolished). At least one function of Smac/DIABLO is to stimulate the cleavage of procaspase 3 from an inactive procaspase precursor to the mature form, caspase 3. Wang and co-workers reconstituted this function in vitro using purified recombinant components (including XIAP), then asked whether Smac/DIABLO might also promote the catalytic activity of mature caspase 3. They found that it could, and conclude that Smac/DIABLO triggers apoptosis through at least two mechanisms — it induces the proteolytic activation of procaspase 3, and also promotes the enzymatic activity of mature caspase 3.

T E LO M E R E S

Caps and cancer Telomerase is activated in many human cancers, and the possibility of targeting tumours with telomerase inhibitors is attractive. But a study in September’s Nature Genetics shows that Terc–/– mice, which lack functional telomerase, are especially sensitive to ionizing radiation. The scientific implication is that functionally intact telomeres and the response to ionizing radiation are somehow linked; the clinical one is that it may not be wise to treat cancer patients with both telomerase inhibitors and ionizing radiation. Alison Mitchell References and links ORIGINAL RESEARCH PAPER Wong, K.-K. et al. Telomere dysfunction impairs DNA repair and enhances sensitivity to ionizing radiation. Nature Genet. 26, 85–88 (2000) FURTHER READING González-Suárez, E., Samper, E., Flores, J. M. & Blasco, M. A. Telomerase-deficient mice with short telomeres are resistant to skin tumorigenesis. Nature Genet. 26, 114–117 (2000) REVIEW Lundblad, V. DNA ends: maintenance of chromosome termini versus repair of double strand breaks. Mutat. Res. 451, 227–240 (2000)

www.nature.com/reviews/molcellbio

| OCTOBER 2000 | VOLUME 1

Finally Wang and colleagues looked for other functional parallels with the Drosophila homologues. It turns out that, in both cases, the amino-terminal region is essential for function. The authors narrowed this down to just seven amino acids, which, on their own, are enough to activate procaspase 3 (albeit much less efficiently than the full-length protein). Given that the only region of sequence conservation in Grim, Reaper and Hid is the amino-terminal 14 amino acids, this work, say the authors, defines “an evolutionarily conserved structural and biochemical basis for the activation of apoptosis by Smac/DIABLO”. Alison Mitchell References and links ORIGINAL RESEARCH PAPERS Chai, J.

et al. Structural and biochemical basis of apoptotic activation by Smac/DIABLO. Nature 406, 855–862 (2000) | Du, C., Fang, M., Li, Y., Li, L. & Wang, X. Smac, a mitochondrial protein that promotes cyctochrome c-dependent caspase activation by eliminating IAP inhibition. Cell 102, 33–42 (2000) | Verhagen, A. M. et al. Identification of DIABLO, a mammalian protein that promotes apoptosis by binding and antagonizing IAP proteins. Cell 102, 43–53 (2000) FURTHER READING Green, D. R. Apoptotic pathways: Paper wraps stone blunts scissors. Cell 102, 1–4 (2000)

DAPP1-PH

Grp1-PH

Image kindly provided by Kathryn Ferguson and Mark Lemmon, University of Pennsylvania, Pennsylvania, USA.

S T R U C T U R A L B I O LO G Y

Pocket the difference There’s a molecular identity parade at the plasma membrane every time cells activate phosphatidylinositol 3-OH kinase (PI(3)K). This enzyme catalyses the production of the phospholipid messengers p h o s phatidylinositol-3,4,5-tr isphosphate (PtdIns(3,4,5)P3) and PtdIns(3,4)P2 which, in turn, recruit proteins containing pleckstrin homology (PH) domains to the membrane. These proteins choreograph a wide variety of molecular dances, from migration and membrane trafficking to adhesion. But how do they pick out tiny amounts of PtdIns(3,4,5)P3 or PtdIns(3,4)P2 from among much higher concentrations of PtdIns(4,5)P2 in the cell membrane? Some PH domains recognize only PtdIns(3,4,5)P3 but others are less discriminating and recognize both PtdIns(3,4,5)P3 and PtdIns(3,4)P2. This makes a big difference to the length of the dance because activation of PI(3)K leads to a transient blip in the amount of PtdIns(3,4,5)P3 but a much more prolonged rise in the amount of PtdIns(3,4)P2. Two papers in the August issue of Molecular Cell provide an answer, by comparing the structure of a PH domain that picks out PtdIns(3,4,5)P3 every time with one that can’t tell the difference between PtdIns(3,4,5)P3 and PtdIns(3,4)P2. Kathryn Ferguson and colleagues at the University of Pennsylvania, and Susan Lietzke and co-workers up the coast at the University of Massachusetts, both chose the PH domain from the PtdIns(3,4,5)P3-specific protein Grp1 (also known as ARNO3 or cytohesin 3) as the subject of their crystallographic analysis. Ferguson et al. also solved the structure of DAPP1 (PHISH), which binds PtdIns(3,4,5)P3 and PtdIns(3,4)P2 with similar affinity. Lietzke and colleagues compared features of the Grp1 PH domain with the promiscuous PH domain from protein kinase B (AKT). In both studies, inositol-1,3,4,5-tetrakisphosphate (Ins(1,3,4,5)P4) — the water-soluble headgroup of PtdIns(3,4,5)P3 — was used to crystallize the ligand-bound PH domains. It was no surprise that the PH domains of both Grp1 and DAPP1 are β-sandwiches, just like previ-

ously solved PH domains, but Grp1 has an extra two β-strands. These form a snug pocket accommodating the 5-phosphate (see picture), which simply isn’t there in the non-discriminating PH domains of DAPP1 and protein kinase B. This difference explains how Grp1 binds PtdIns(3,4,5)P3, but how does it stop PtdIns(3,4)P2 from binding? The answer lies in the number of hydrogen bonds necessary to stabilize the interaction. Ferguson and colleagues find that both Grp1 and DAPP1 make a total of 14 side-chain hydrogen bonds with Ins(1,3,4,5)P4, but in Grp1 these are equally distributed among the 3-, 4- and 5-phosphates, whereas in DAPP1 there are no hydrogen bonds to the 5-phosphate. So, when PtdIns(3,4)P2 enters Grp1’s PH domain, there’s nothing for this extra pocket to grab hold of, and Grp1 can’t hang on tightly enough to just the 3- and 4-phosphates. As well as explaining why some PH domains are fussier than others, these studies also have predictive value. With their new-found understanding of which residues are important for gripping each phosphate group, Ferguson and colleagues used sequence comparisons to predict the specificity of a previously uncharacterized PH domain, then tested their prediction (which turned out to be correct) using a simple gel filtration assay. Armed with this knowledge, it should now be possible to work out the length and strength of responses that different PH-domain proteins mediate. As Lietzke and colleagues point out, it also paves the way towards blocking PI(3)K responses therapeutically, by designing drugs that bind specific PH domains. Cath Brooksbank References and links ORIGINAL RESEARCH PAPERS Ferguson, K. M. et al. Structural

basis for discrimination of 3-phosphoinositides by pleckstrin homology domains. Mol. Cell 6, 373–384 (2000) | Lietzke, S. E. et al. Structural basis of 3-phosphoinositide recognition by pleckstrin homology domains. Mol. Cell 6, 385–394 (2000) REVIEWS Lemmon, M. A. & Ferguson, K. M. Signal-dependent membrane targeting by pleckstrin homology (PH) domains. Biochem. J. 350, 1–18 (2000) | Chan, T. O. et al. AKT/PKB and other D3 phosphoinositide-regulated kinases: kinase activation by phosphoinositide-dependent phosphorylation. Annu. Rev. Biochem. 68, 965–1014 (1999)

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 1 | OCTOBER 2000 | 9

REVIEWS THE VERSATILITY AND UNIVERSALITY OF CALCIUM SIGNALLING Michael J. Berridge, Peter Lipp and Martin D. Bootman The universality of calcium as an intracellular messenger depends on its enormous versatility. Cells have a calcium signalling toolkit with many components that can be mixed and matched to create a wide range of spatial and temporal signals. This versatility is exploited to control processes as diverse as fertilization, proliferation, development, learning and memory, contraction and secretion, and must be accomplished within the context of calcium being highly toxic. Exceeding its normal spatial and temporal boundaries can result in cell death through both necrosis and apoptosis.

Ca2+-INDUCED Ca2+ RELEASE

An autocatalytic mechanism by which cytoplasmic Ca2+ activates the release of Ca2+ from internal stores through channels such as inositol-1,4,5trisphosphate receptors or ryanodine receptors.

Calcium (Ca2+) is a ubiquitous intracellular signal responsible for controlling numerous cellular processes. At one level, its action is simple: cells at rest have a Ca2+concentration of 100 nM but are activated when this level rises to roughly 1000 nM (FIG. 1). The immediate question is how can this elevation of Ca2+ regulate so many processes? The answer lies in the versatility of the Ca2+ signalling mechanism in terms of speed, amplitude and spatio-temporal patterning. This versatility emerges from the use of an extensive molecular repertoire of signalling components, which comprise a Ca2+ signalling toolkit (FIG. 2 and online poster) that can be assembled in combinations to create signals with widely different spatial and temporal profiles. More variations are achieved through the interactions that Ca2+ makes (crosstalk) with other signalling pathways. This versatility is exploited to regulate diverse cellular responses. The Ca2+ signalling toolkit

The Babraham Institute, Laboratory of Molecular Signalling, Babraham Hall, Babraham, Cambridge, CB2 4AT, UK. e-mail: michael.berridge@ bbsrc.ac.uk Correspondance to: M.J.B.

The Ca2+ signalling network can be divided into four functional units (FIG. 1): • Signalling is triggered by a stimulus that generates various Ca2+-mobilizing signals. • The latter activate the ON mechanisms that feed Ca2+ into the cytoplasm. • Ca2+ functions as a messenger to stimulate numerous

Ca2+-sensitive processes. • Finally, the OFF mechanisms, composed of pumps and exchangers, remove Ca2+ from the cytoplasm to restore the resting state. The functional relationship between these units is illustrated in FIG. 2, which reveals that the signalling network is composed of many components (the Ca2+ signalling toolkit). Because many of the molecular components of this toolkit have several isoforms [online poster] with subtly different properties, each specific cell type can exploit this large repertoire to construct versatile Ca2+ signalling networks. Generation of Ca2+-mobilizing signals. Cells generate their Ca2+ signals by using both internal and external sources of Ca2+. The internal stores are held within the membrane systems of the endoplasmic reticulum (ER) or the equivalent organelle, the sarcoplasmic reticulum (SR) of muscle cells. Release from these internal stores is controlled by various channels, of which the inositol1,4,5-trisphosphate receptor (InsP3R) and ryanodine receptor (RYR) families have been studied most extensively1,2. The principal activator of these channels is Ca2+ itself and this process of Ca -INDUCED Ca RELEASE is central to the mechanism of Ca2+ signalling (see below). Ca2+mobilizing second messengers that are generated when

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 1 | OCTOBER 2000 | 1 1

REVIEWS

Figure 1 | The four units of the Ca2+ signalling network. Stimuli act by generating Ca2+-mobilizing signals that act on various ON mechanisms to trigger an increase in the intracellular concentration of Ca2+. The increased level of Ca2+ stimulates various Ca2+-sensitive processes to trigger many different cellular pathways. The response is terminated by OFF mechanisms that restore Ca2+ to its resting level. Details of these four functional units, with the same colour coding, are revealed in FIG. 2.

stimuli bind to cell surface receptors (FIG. 2) determine whether Ca2+ can activate these channels. One is Ins(1,4,5)P3 (REF. 1), which diffuses into the cell to engage the InsP3Rs and release Ca2+ from the ER (FIG. 2). The ability of Ca2+ to stimulate the RYRs is modulated by cyclic ADP ribose (cADPR)3. A related messenger, nicotinic acid dinucleotide phosphate (NAADP)4, acts on a separate, as yet uncharacterized, channel. Sphingosine 1phosphate (S1P) releases Ca2+ from the ER — possibly by binding to a sphingolipid Ca2+ release-mediating protein of the ER (SCaMPER)5. These different Ca2+-mobilizing messengers can coexist in cells, where they seem to be controlled by different receptors. For example, in the exocrine pancreas, muscarinic acetylcholine receptors act through Ins(1,4,5)P3, whereas cholecystokinin receptors use cADPR6. Similarly, human SH-SY5Y cells have acetylcholine receptors linked through Ins(1,4,5)P3 whereas lysophosphatidic acid acts through S1P7. So the versatility of the signalling network is enhanced by having different Ca2+-mobilizing messengers linked to separate input signals.

VOLTAGE-OPERATED CHANNELS

Plasma-membrane ion channels that are activated by membrane depolarization. RECEPTOR-OPERATED CHANNELS

Plasma membrane ion channels that open in response to binding of an extracellular ligand. STORE-OPERATED CHANNELS

Plasma membrane ion channels, of uncertain identity, that open in response to depletion of internal Ca2+ stores.

ON mechanisms. The ON mechanisms depend on Ca2+ channels that control the entry of external Ca2+ or the release of Ca2+ from internal stores. For the first case, there are families of Ca2+ entry channels defined by the way in which they are activated. We know most about VOLTAGE-OPERATED CHANNELS (VOCs). In addition there are many channels that open in response to receptor activation. RECEPTOR-OPERATED CHANNELS (ROCs) open on binding external stimuli, usually transmitters such as glutamate, ATP or acetylcholine. Other channels are sensitive to various signals generated following receptor activation such as store emptying8, diacylglycerol (DAG)9 and arachidonic acid10,11. Most attention has focused on capacitative Ca2+ entry (FIG. 3a)where empty stores activate STORE-OPERATED CHANNELS (SOCs) in the plasma membrane through an unknown mechanism. Recent evi-

dence12,13 lends support to a conformational-coupling mechanism14, which proposes that InsP3Rs in the ER are directly coupled to SOCS (FIG. 3a). There is considerable interest in SOCs because they provide the Ca2+ signals that control many cellular processes (see later). Signal Ca2+ is also derived from the internal stores using the channels and Ca2+-mobilizing messengers described above. As little is known about the channels opened by NAADP and S1P, we will focus on the InsP3Rs and the RYRs. These two channels are regulated by several factors, the most important of which is Ca2+ itself, which regulates Ca2+ release by acting from either the lumenal or cytoplasmic sides of the channel. Increasing the level of Ca2+ within the lumen of the ER/SR enhances the sensitivity of the RYRs and the same may apply to the InsP3Rs. The cytosolic action of Ca2+ is more complex: it can be both stimulatory and inhibitory and can vary between the different InsP3R isoforms. In general, the InsP3Rs have a bell-shaped Ca2+ dependence when treated with low concentrations of Ins(1,4,5)P3: low concentrations of Ca2+ (100–300 nM) are stimulatory but above 300 nM, Ca2+ becomes inhibitory and switches the channel off15. Emerging evidence indicates that InsP3Rs are sometimes not inhibited by high cytosolic Ca2+ concentrations. Instead of a bell-shape, the relationship between InsP3R activity and cytosolic Ca2+ is sigmoidal. This is particularly true in the presence of high Ins(1,4,5)P3 levels, indicating that Ins(1,4,5)P3 acts as a molecular switch and that once the receptor binds Ins(1,4,5)P3, it becomes sensitive to the stimulatory, but not the inhibitory, action of Ca2+ (REF. 15). The function of cADPR is not so clear but it is known to increase the Ca2+ sensitivity of RYRs. The autocatalytic process of Ca2+-induced Ca2+ release enables the InsP3Rs and RYRs to communicate with each other to establish coordinated Ca2+ signals, often organized into propagating waves1,2. The main function of the Ca2+-mobilizing messengers, therefore, is to alter the sensitivity of the InsP3Rs and RYRs to this stimulatory action of Ca2+. How do cells exploit the Ca2+ toolkit? Let us consider three tissues that generate Ca2+ signals in different ways (FIG. 4): Muscle. Perhaps the most specialized mechanism is found in skeletal muscle, which has a L-type VOC (α1S) located in the plasma membrane that interacts directly with the large cytoplasmic head of the RYR1 embedded in the SR (FIGS 3b, 4a). Membrane depolarization induces a conformational change in α1S that is transmitted directly to RYR1, causing it to release Ca2+ from the SR. By contrast, cardiac cells use a related α1C L-type channel to gate a small amount of trigger Ca2+ that then diffuses across the plasma membrane to activate RYR2 channels in the SR through Ca2+-induced Ca2+ release. Neurons. Neurons have numerous Ca2+ channels in different parts of the cell to carry out separate functions (FIG. 4b). N- and P/Q-type VOCs at synaptic endings trigger the release of neurotransmitters. The L-type VOCs on the cell body and proximal dendrites are ideally positioned to provide the Ca2+ signals that induce gene activation. They also function as ‘kinetic filters’, allowing www.nature.com/reviews/molcellbio

| OCTOBER 2000 | VOLUME 1

REVIEWS them to respond effectively to the small depolarizations that occur at synaptic spines16. The Ca2+ signals in spines, responsible for mediating the early synaptic modifications that are implicated in learning and memory, are provided by entry through such VOCs and also through ROCs (such as NMDA (N-methyl-D-aspartate) receptors) and by release from RYRs and InsP3Rs (reviewed in REF. 17). As InsP3Rs are sensitive to both Ins(1,4,5)P3 and Ca2+, they could act as coincidence detectors to correlate the activity of pre- and postsynaptic inputs, which is central to memory formation17. In hippocampal neurons, for example, electrical activity resulting in Ca2+ entry through VOCs acts together with Ins(1,4,5)P3 produced by metabotropic glutamate receptors (mGluR1) to produce a synergistic release of internal Ca2+ (REF. 18).

Pancreas. RYRs have also been described in nonexcitable cells such as the pancreas, where they collaborate with InsP3Rs to control both fluid and enzyme secretion (FIG. 4c)6. Acetylcholine and cholecystokinin act through Ins(1,4,5)P3. Cholecystokinin also acts through both NAADP and cADPR. NAADP might also initiate Ca2+ release from RYRs19. As a result of these ON mechanisms, Ca2+ flows into the cytoplasm to produce the increase in concentration that constitutes a Ca2+ signal (FIG. 1). However, the concentration that is measured in cells using various Ca2+ indicators (for example, aequorin or Ca2+-sensitive dyes such as Fura2 or Fluo3) is only the tip of the iceberg because most of the Ca2+ that enters the cytoplasm is rapidly bound to various cytosolic buffers such as

Figure 2 | Elements of the Ca2+ signalling toolkit. Cells have an extensive signalling toolkit that can be mixed and matched to create Ca2+ signals of widely different properties. Ca2+-mobilizing signals (blue) are generated by stimuli acting through a variety of cell-surface receptors (R), including G-protein (G)-linked receptors and receptor tyrosine kinases (RTK). The signals generated include: inositol-1,4,5-trisphosphate (Ins(1,4,5)P3), generated by the hydrolysis of phosphatidylinositol-4,5bisphosphate (PtdIns(4,5)P2) by a family of phospholipase C enzymes (PLCβ, PLCγ); cyclic ADP ribose (cADPR) and nicotinic acid dinucleotide phosphate (NAADP), both generated from nicotinamide-adenine dinucleotide (NAD) and its phosphorylated derivative NADP by ADP ribosyl cyclase; and sphingosine 1-phosphate (S1P), generated from sphingosine by a sphingosine kinase. ON mechanisms (green) include plasma membrane Ca2+ channels, which respond to transmitters or to membrane depolarization (∆V), and intracellular Ca2+ channels — the Ins(1,4,5)P3 receptor (InsP3R), ryanodine receptor (RYR), NAADP receptor and sphingolipid Ca2+ release-mediating protein of the ER (SCaMPER). The Ca2+ released into the cytoplasm by these ON mechanisms activates different Ca2+ sensors (purple), which augment a wide range of Ca2+-sensitive processes (purple), depending on cell type and context. OFF mechanisms (red) pump Ca2+ out of the cytoplasm: the Na+/Ca2+ exchanger and the plasma membrane Ca2+ ATPase (PMCA) pumps Ca2+ out of the cell and the sarco-endoplasmic reticulum Ca2+ ATPase (SERCA) pumps it back into the ER/SR. (TnC, troponin C; CAM, calmodulin; MLCK, myosin light chain kinase; CAMK, Ca2+/calmodulin-dependent protein kinase; cyclic AMP PDE, cyclic AMP phosphodiesterase; NOS, nitric oxide synthase; PKC, protein kinase C; PYK2, proline-rich kinase 2; PTP, permeability transition pore.) Online poster

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 1 | OCTOBER 2000 | 1 3

REVIEWS parvalbumin, calbindin-D28K and calretinin. The buffer capacity — the number of Ca2+ ions that are bound for each free ion — varies considerably between cells20. Cytosolic buffers are involved in shaping both the amplitude and duration of Ca2+ signals. During each spike, they act as a halfway house for Ca2+ by loading it up during the ON mechanisms and then unloading it during the OFF mechanisms described later (FIG. 2). Buffers also limit the spatial spreading of local Ca2+ signals. This is particularly important in neurons that contain high concentrations of buffers such as parvalbumin and calbindin, which ensure that Ca2+ signals are largely confined to synapses. Ca2+-sensitive processes. Once the ON mechanisms have generated a Ca2+ signal, various Ca2+-sensitive processes

Figure 3 | Ca2+ signalling by conformational coupling using macromolecular complexes. a | Capacitative Ca2+ entry. In response to a Ca2+-mobilizing signal such as inositol1,4,5-trisphosphate (Ins(1,4,5)P3), Ca2+ is released from the endoplasmic reticulum. Emptying of the store is detected by a protein, most probably an inositol-1,4,5-trisphosphate receptor (InsP3R) or a ryanodine receptor (RYR), which undergoes a conformational change (white arrows) that is transmitted to the store-operated channel (SOC) to induce Ca2+ entry across the plasma membrane. b | Ca2+ release in skeletal muscle. Voltage sensors (α1S subunit of an L-type Ca2+ channel) located in the plasma membrane induce a conformational change in the RYR1 channels (open arrows) that then release Ca2+ from the sarcoplasmic reticulum.

translate this into a cellular response (FIGS 2 AND 4). The Ca2+ signalling toolkit has numerous Ca2+-binding proteins, which can be divided into Ca2+ buffers (described above) and Ca2+ sensors, on the basis of their main functions [online poster]. The latter respond to an increase in Ca2+ by activating diverse processes (FIG. 2). The classical sensors are troponin C (TnC) and calmodulin (CAM), which have four EF hands that bind Ca2+ and undergo a pronounced conformational change to activate various downstream effectors. TnC has a limited function to control the interaction of actin and myosin during the contraction cycle of cardiac and skeletal muscle (FIG. 4a). CAM is used more generally to regulate many processes such as the contraction of smooth muscle, crosstalk between signalling pathways, gene transcription, ion channel modulation and metabolism. The same cell can use different sensors to regulate separate processes. In skeletal muscle, for example, TnC regulates contraction whereas CAM stimulates phosphorylase kinase to ensure a parallel increase in ATP production (FIG. 4a). In addition to the above proteins, which act generally, there are numerous Ca2+-binding proteins designed for more specific functions. For example, synaptotagmin is associated with membrane vesicles and is a Ca2+ sensor for exocytosis. The versatility of Ca2+ signalling is greatly enhanced by some of the Ca2+-sensitive processes linking into other signalling pathways (BOX 1). The ability of Ca2+ to recruit the control elements of other signalling pathways (for example, cyclic AMP and mitogen-activated protein kinase pathways) is particularly evident in the control of gene transcription in neurons (FIG. 4b). OFF mechanisms. Once Ca2+ has carried out its signalling functions, it is rapidly removed from the cytoplasm by various pumps21 and exchangers22 (FIG. 2). The plasma membrane Ca2+-ATPase (PMCA) pumps and Na+/Ca2+ exchangers extrude Ca2+ to the outside whereas the sarco-endoplasmic reticulum ATPase (SERCA) pumps return Ca2+ to the internal stores. The mitochondrion is another important component of the OFF mechanism in that it sequesters Ca2+ rapidly during the development of the Ca2+ signal and then releases it back slowly during the recovery phase (FIG. 2). This uptake of Ca2+ by the mitochondrion is important in shaping both the amplitude23 and the spatio-temporal patterns of Ca2+ signals24–26. Mitochondria extrude protons to create the electrochemical gradient that allows ATP synthesis. The same gradient is used to drive Ca2+ uptake through a uniporter that has a low sensitivity to Ca2+ (half-maximal activation around 15 µM). This low sensitivity means that mitochondria accumulate Ca2+ more effectively when they are close to Ca2+-releasing channels27. Here, they may form a ‘quasi-synapse’, allowing them to directly sense the high local Ca2+ concentration that builds up in the vicinity of open Ca2+ channels, such as the InsP3Rs and RYRs28. There seem to be reciprocal interactions between the two organelles in that the ER/SR provides the Ca2+ that enters the mitochondria, which in turn modifies the Ca2+ feedback mechanisms that regulate Ca2+ release from the ER/SR. www.nature.com/reviews/molcellbio

| OCTOBER 2000 | VOLUME 1

REVIEWS Various proteins such as presenilins and apoptosis regulatory proteins (such as Bcl-2 described later) modulate the way these two organelles handle Ca2+. The presenilins, located in the ER membrane, not only function to process the β-amyloid precursor protein but also modulate Ca2+ signalling. Mutations of presenilin result in overfilling of the ER leading to larger Ca2+ signals and a decrease in capacitative Ca2+ entry29. The mitochondrion has an enormous capacity to accumulate Ca2+ and the mitochondrial matrix contains buffers that prevent the concentration from rising too high. Once the cytosolic Ca2+ has returned to its resting level, a mitochondrial Na+/Ca2+ exchanger pumps the large load of Ca2+ back into the cytoplasm, from which it is either returned to the ER or removed from the cell (FIG. 2). Ca2+ can also leave the mitochondrion through a permeability transition pore (PTP)26,30, which has all the elements of Ca2+-induced Ca2+ release because its formation is activated by the build up of Ca2+ within the mitochondrial matrix31. This PTP may have two functional states. A low conductance state of the pore can act reversibly, allowing mitochondria to become excitable, and this may contribute to the generation of Ca2+ waves31. On the other hand, an irreversible high conductance state of the PTP has a marked effect on the mitochondrion in that it collapses the transmembrane potential and leads to the release of cyctochrome c and the initiation of apoptosis (see later). Global aspects of Ca2+ signalling

Figure 4 | Application of the Ca2+ signalling toolkit to regulate different cellular processes. a | In skeletal muscle, an L-type Ca2+ channel (α1S) senses membrane depolarization (∆V) and undergoes a conformational change that is transmitted to the ryanodine receptor 1 (RYR1) (FIG. 3b). Ca2+ released from the sarcoplasmic reticulum (SR) interacts with two sensors, troponin C (TnC), which triggers contraction, and calmodulin (CAM), which activates glycogen metabolism to synthesize ATP. b | Neurons have several Ca2+sensitive processes located in different regions. Membrane depolarization (∆V) is sensed by Nor P/Q-type channels at the synaptic endings to produce a localized pulse of Ca2+ that triggers exocytosis. In the cell body and dendrites, L-type channels sense the same depolarization and induce the entry of Ca2+ which has a number of targets: adenylyl cyclase I or III (AC I/III) leading to cyclic AMP production, proline-rich tyrosine kinase (PYK2), mitogen-activated protein kinase (MAPK), Ca2+/calmodulin-dependent protein kinase II (CAMKII) and calmodulin–calcineurin (CAM–CN). Some of these targets induce gene transcription. The neurotransmitter glutamate can also generate Ca2+ signals either by activating receptor-operated channels such as NMDA (N-methyl-D-aspartate) receptors, or by stimulating the metabotropic glutamate receptor mGluR1 to produce inositol-1,4,5-trisphosphate (Ins(1,4,5)P3) to mobilize internal Ca2+ from the endoplasmic reticulum (ER). These glutamate-induced Ca2+ signals are localized to synaptic endings, where they contribute to processes such as long-term potentiation (LTP) and longterm depression (LTD), which have been implicated in learning and memory. c | The exocrine pancreas uses two signalling systems regulated by separate receptors. Acetylcholine uses Ins(1,4,5)P3 to release internal Ca2+. As well as stimulating Ins(1,4,5)P3 formation, cholycystokinin also acts through both cyclic ADP ribose (cADPR) and nicotinic acid dinucleotide phosphate (NAADP). The latter seems to act by releasing a small amount of trigger Ca2+ through the NAADP receptor (NR) that then acts together with cADPR to release further Ca2+ through RYRs.

Elementary events. Further versatility is achieved by varying the spatial and temporal aspects of Ca2+ signalling32,33. The different types of Ca2+ signals shown in FIG. 5 result from the InsP3Rs and/or RYRs having different degrees of excitability depending on the levels of the appropriate Ca2+-mobilizing messenger. At low levels of stimulation, the degree of excitability is such that individual RYRs or InsP3Rs open and these single-channel events have been recorded as quarks34 or blips35, respectively (FIG. 5b). These may be considered as the fundamental events that are the building blocks from which more complex Ca2+ signals are constructed. These single-channel events are rare and the more usual event is larger, resulting from the coordinated opening of clusters of InsP3Rs or RYRs, known as puffs or sparks, respectively (FIG. 5c). Sparks were first described in cardiac cells36 where they represent Ca2+ signals from a group of RYR2 channels opening in concert. The puffs recorded in either Xenopus oocytes37,38 or HeLa cells39 have diverse amplitudes indicating that there are either variable numbers of InsP3Rs within each cluster or variable numbers of channels open within an individual cluster. Ca2+ waves. Sparks and puffs contribute to intracellular Ca2+ signals, such as the Ca2+ waves that sweep through cells (FIG. 5d). For waves to occur, most of the InsP3Rs and the RYRs must be sufficiently sensitive to Ca2+ to respond to each other through the process of Ca2+induced Ca2+ release. One group of channels releases Ca2+, which then diffuses to neighbouring receptors to

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 1 | OCTOBER 2000 | 1 5

REVIEWS

Box 1 | A Ca2+ nexus — crosstalk between signalling pathways The Ca2+ signalling toolkit interacts with many other signalling pathways. The interactions are reciprocal in nature in that information flows to and from the other signalling pathways. It is difficult to make any generalizations and each set of interactions has to be treated separately: Ca2+–cyclic AMP interactions (1) The function of the cyclic AMP and Ca2+ signalling systems are intimately linked. Some of the adenylyl cyclase isomers are activated by Ca2+ whereas others are inhibited. Ca2+ can also stimulate some of the cAMP phosphodiesterases (cAMP PDE). Changes in the level of cAMP can feed back to influence the level of Ca2+ by acting on both Ca2+ channels and pumps. In cardiac and skeletal muscle, the activity of the L-type Ca2+ channel is enhanced by cAMP. Ca2+–NO interactions (2) An important function of Ca2+ is to activate nitric oxide (NO) synthase to generate NO, which functions as a local hormone to regulate the activity of neighbouring cells. The NO activates guanylyl cyclase to produce cyclic GMP, which feeds back to influence the activity of Ca2+ channels and pumps. For example, smooth muscle cells relax when cGMP phosphorylates an inositol-1,4,5-trisphosphate receptor (InsP3R)-associated cGMP kinase substrate that reduces inositol-1,4,5-trisphosphate (Ins(1,4,5)P3)-induced Ca2+ release108. Ca2+–phosphatidylinositol-3-OH kinase interaction (3) The ubiquitous phosphatidylinositol-3-OH-kinase (PI(3)K) signalling pathway has many functions in cells, which are mediated by the lipid second messenger phosphatidylinositol-3,4,5-trisphosphate (PtdIns(3,4,5)P3). One function of PtdIns(3,4,5)P3 is to activate the non-receptor tyrosine kinase Btk that then phosphorylates and activates phospholipase Cγ1 (PLCγ1)81. The tumour suppressor PTEN, a 3-phosphatase that lowers the level of PtdIns(3,4,5)P3, reduces both the level of Ins(1,4,5)P3 and the influx of external Ca2+ (REF. 109). Ca2+ feedback interactions (4) There are numerous feedback interactions within the Ca2+ signalling pathway whereby Ca2+ can modulate its own activity. For example, Ca2+ can activate phospholipase Cδ1 (PLCδ1) to increase the level of Ins(1,4,5)P3. Conversely, it can lower the level of this second messenger by stimulating the Ins(1,4,5)P3 kinase to produce Ins(1,3,4,5)P4. Finally, Ca2+ can exert profound effects on the Ca2+ channels and pumps. Ca2+–mitogen-activated protein kinase interaction (5) Ca2+ can interact with the mitogen-activated protein kinase (MAPK) signalling pathway by activating a proline-rich tyrosine kinase 2 (PYK2), which then acts through the small GTPase Ras to induce the MAPK cascade110. For example, the growth of smooth muscle cells may depend on the Ca2+-dependent activation of the MAPK pathway111. A more indirect method may depend on Ca2+ stimulating a metalloproteinase to release epidermal growth factor (EGF) from a precursor, as seems to occur in prostate carcinoma cells112.

excite further release, therefore setting up the regenerative process. When gap junctions connect cells, these intracellular waves can spread to neighbouring cells, to create intercellular waves (FIG. 5e) capable of coordinating the activity of many cells40 (see Sanderson lab page). For example, intercellular Ca2+ waves in the lung epithelium stimulate the beat frequency of the cilia that expel inhaled contaminants from the airways40. Intercellular waves have also been recorded in the intact liver41 and in insect salivary glands42. Just how the wave traverses the gap junction is a matter of considerable debate. For the two examples given above, Ca2+ seems to be the signal that crosses the gap junction41,42 (FIG. 5e). In the case of the lung epithelium40, the messenger seems to be Ins(1,4,5)P3 and there is also evidence for the presence of 16

extracellular messengers such as ATP in other cell types. In addition to creating global responses, these elementary events have signalling functions within highly localized cellular domains. A classic example is the process of exocytosis at synaptic endings where N- or P/Q-type VOCs create a local pulse of Ca2+ to activate synaptotagmin and trigger vesicle release (FIG. 4b). Ca2+ released through InsP3Rs43 can stimulate exocytosis in various secretory cells. In adrenal glomerulosa cells, T-type Ca2+ channels in the plasma membrane seem to have a ‘Ca2+ pipeline’, enabling them to feed Ca2+ directly into the mitochondria to stimulate steroidogenesis44. Sparks that activate Ca2+-sensitive K+ channels to trigger membrane hyperpolarization control the excitability of neurons and smooth muscle cells. In HeLa cells, Ca2+ puffs are concenwww.nature.com/reviews/molcellbio

| OCTOBER 2000 | VOLUME 1

REVIEWS trated around the nucleus where they feed Ca2+ directly into the nucleoplasm45. Finally, as mentioned earlier, the mitochondria located near the sites of elementary events take up Ca2+ rapidly and this stimulates mitochondrial metabolism to increase ATP formation (FIG. 2). Temporal aspects of Ca2+ signalling

Ca2+ signals are usually presented as brief spikes. In some cases, individual spikes are sufficient to trigger a cellular response such as the contraction of skeletal muscle (FIG. 4a) or neurotransmitter release(FIG. 4b). When longer periods of signalling are necessary, spikes are repeated to give waves with different frequencies, ranging from 1–60 seconds (in pancreas and liver) to 24 hours (in the Ca2+ oscillator that initiates mitosis during the cell cycle). Cells often respond to changes in stimulus intensity by varying the frequency of Ca2+ waves. To use such a frequency-modulated signalling system, cells have evolved sophisticated ‘molecular machines’ for decoding frequency-encoded Ca2+ signals. The two Ca2+-sensitive proteins that seem to decode wave frequency are Ca2+/calmodulin-dependent protein kinase II (CAMKII)46 and protein kinase C47. Frequency coding is used to control processes such as liver metabolism, smooth muscle contractility and differential gene transcription, especially in developing systems. For example, Ca2+ spikes can initiate gene expression more effectively than a steadily maintained level of the same average [Ca2+]48. A low frequency of spiking activates the transcription factor NF-κB, whereas higher frequencies are necessary to switch on the transcription factor NF-AT49. Ca2+ may also be important in entraining the circadian clock in the suprachiasmatic nucleus. This can be reset by releasing Ca2+ from either the RYR-sensitive50 or the InsP3R-sensitive51 stores. The universality of Ca2+ signalling

Ca2+ signalling is used throughout the life history of an organism. Life begins with a surge of Ca2+ at fertilization and this versatile system is then used repeatedly to control many processes during development and in adult life. One of the fascinating aspects of Ca2+ is that it plays a direct role in controlling the transcriptional events that select out the types of Ca2+ signalling systems that are expressed in specific cell types. Such a role for Ca2+ in differential gene transcription is still in its infancy but is rapidly developing into an active area of research.

CDC25

A dual-specificity threonine/tyrosine phosphatase required for progression of the cell cycle. It dephosphorylates and activates cyclin–CDK complexes.

Fertilization. During fertilization, mammalian eggs generate regular Ca2+ spikes that persist for about two hours and initiate development. Each spike is a global signal that sweeps through the egg, driven by Ca2+ release from InsP3Rs52. The increase in Ins(1,4,5)P3 necessary to support such waves may be generated by a unique phospholipase C that is transferred into the egg by the sperm at fertilization53. This regular pattern of Ca2+ spiking stimulates CAMKII, which then acts through CDC25 to dephosphorylate the enzyme cyclin-dependent kinase 1 (CDK1), resulting in cyclin B activation and the completion of meiosis. The male and female nuclei now fuse, marking the end of the fertilization-induced Ca2+ spikes.

Figure 5 | The spatial organization of Ca2+ release from internal stores. a | Inositol-1,4,5-trisphosphate receptors (InsP3Rs) and ryanodine receptors (RYRs) are distributed over the surface of the endoplasmic and/or sarcoplasmic reticulum (ER/SR). b | In response to weak stimuli, individual channels open to give either blips (InsP3Rs) or quarks (RYRs). c | At higher levels of stimulation, groups of InsP3Rs or RYRs open together to produce puffs or sparks, respectively. d | When cells are fully excitable, the elementary events depicted in (c) can excite neighbouring receptors through a process of Ca2+-induced Ca2+ release to set up an intracellular wave. e | When gap junctions connect cells, waves can travel from one cell to the next to set up an intercellular wave. Animated online

The Ca2+ signalling system has completed its first task in the young embryo but it is soon called into play again to trigger the mitotic events at the end of the first cell cycle. The cell cycle consists of an orderly programme of events controlled by two-linked oscillators — a cell cycle oscillator and a Ca2+ oscillator54. The former depends on the synthesis and periodic proteolysis of various cyclins at specific points during the cell cycle. The Ca2+ oscillator, based on the periodic release of stored Ca2+, is responsible for initiating specific events associated with mitosis. As the embryo approaches mitosis, a series of spontaneous Ca2+ transients trigger specific events such as nuclear envelope breakdown55 and cell cleavage56. In the case of Xenopus oocytes, the Ca2+ oscillator persists for at least 5 hours with a periodicity of 30 minutes, which exactly coincides with the length of each cell cycle57,58. Just what drives the Ca2+ oscillator is a mystery but it seems to depend on the periodic elevation of Ins(1,4,5)P3 (REFS 59–61).

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 1 | OCTOBER 2000 | 1 7

REVIEWS

SOMITES

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

Embryonic pattern formation. During the next stage of development, the zygote proliferates rapidly to produce large groups of cells and the Ca2+ signalling system controls the specification processes responsible for pattern formation and cell differentiation. The orchestrated sequences of Ca2+ signals that occur during developmental processes (for example, gastrulation, formation of SOMITES and neural induction) have been documented in the zebrafish embryo62–64. There are indications that Ins(1,4,5)P3 and Ca2+ act during specification of the dorsoventral axis. The concentration of Ins(1,4,5)P3 increases significantly during development of the dorsoventral axis in both Xenopus65,66 and zebrafish67. Imaging studies have revealed a standing gradient with a prolonged elevation of Ca2+ in the ven-

tral region of early zebrafish embryos62. Dorsoventral gradients of Ca2+ have also been recorded in early Drosophila embryos68. The proposed gradient of Ins(1,4,5)P3/Ca2+, being high in the ventral region and low in the dorsal region, may specify pattern formation in the developing embryo. Consistent with the existence of such a gradient, the activity of CAMKII is higher in the ventral region69. Procedures designed to disrupt this Ins(1,4,5)P3/Ca2+ gradient, such as the injection of an antibody that inhibits Ins(1,4,5)P3induced Ca2+ release, can respecify the axis in Xenopus embryos70 — as can altering the CAMKII activity gradient69. The dorsoventral axis may therefore be determined by a gradient in the activity of the Ca2+ signalling pathway. Cell differentiation. Later in development, Ca2+ is involved in inducing the differentiation of individual cells. In contrast to the standing gradient of Ca2+ responsible for axis specification, Ca2+ spiking induces cell differentiation, at least in neural and muscle cells71. In Xenopus, spontaneous Ca2+ spikes produced by RYRs during a narrow developmental window72 drive the differentiation of myocytes into somites. The development of neurons is also regulated by Ca2+ spikes that control processes such as the expression of specific neurotransmitters and channels73,74, the behaviour of growth cones75 and the establishment of the specific connections within neural circuits76. Differentiation culminates with the emergence of different cell types specialized for specific functions, some of which were described earlier (FIG. 4). A key element of the differentiation process, therefore, is to install those components of the Ca2+ signalling toolkit that each specialized cell needs to fulfil its particular function.

Figure 6 | Ca2+ function during lymphocyte proliferation. Antigen interacts with the T-cell receptor (TCR) to recruit phospholipase Cγ1 (PLCγ1) to generate both diacylglycerol (DAG) and inositol-1,4,5-trisphosphate (Ins(1,4,5)P3). The production of Ins(1,4,5)P3 is maintained by the phosphatidylinositol-3-OH kinase (PI(3)K) pathway, which generates phosphatidylinositol3,4,5-trisphosphate (PtdIns(3,4,5)P3). This stimulates the non-receptor tyrosine kinase Btk which, in turn, phosphorylates and activates phospholipase Cγ1 (PLCγ1). Ins(1,4,5)P3 releases Ca2+ from the endoplasmic reticulum (ER) through the type 1 Ins(1,4,5)P3 receptor (InsP3R1). Emptying of this store activates store-operated channels (SOCs) (FIG. 3b). The latter are kept open by potassium channels, which hyperpolarize the membrane, and by mitochondria, which reduce the negative feedback effect of Ca2+ on the SOCs. Ca2+ initiates the proliferative response by stimulating various transcription factors such as NF-κB, NF-AT and CREB. The stimulatory action of Ca2+ on the calmodulin (CAM)–calcineurin (CN) complex that dephosphorylates NF-AT is inhibited by the immunosuppressants cyclosporin A (CsA) and FK506. (PtdIns(4,5)P2, phosphatidylinositol-4,5-bisphosphate; PKC, protein kinase C; IKB, inhibitor of NF-κB; P, phosphate)

Cell proliferation. Once cells have been assigned specific jobs, they usually stop proliferating. In many cases, however, such differentiated cells maintain the option of reentering the cell cycle and this usually occurs in response to growth factors. Ca2+ is one of the key regulators of cell proliferation, functioning in conjunction with other signalling pathways such as those regulated through MAPK and phosphatidylinositol-3-OH kinase (PI(3)K) (BOX 1)77,78. The function of Ca2+ is well illustrated in lymphocytes responding to antigen (FIG. 6). Here, the ‘growth factor’ is the antigen that binds to the T-cell receptor to initiate the assembly of a supramolecular activation cluster79 containing scaffolding and signal transducing elements. One of the latter is phospholipase Cγ1 (PLCγ1), which produces both DAG and Ins(1,4,5)P3 for a period of at least two hours to activate proliferation. The need for such prolonged periods of signalling to initiate proliferation is not unique to lymphocytes. For example, stimulation of Chinese hamster ovary cell proliferation by gastrin correlated with its ability to maintain an oscillatory Ca2+ signal for at least two hours. Conversely, carbachol, acting through muscarinic M3 receptors, gave a short-lived Ca2+ response and failed to stimulate proliferation80. www.nature.com/reviews/molcellbio

| OCTOBER 2000 | VOLUME 1

REVIEWS As Ca2+ stores have a limited capacity, such a prolonged period of Ins(1,4,5)P3-induced Ca2+ signalling depends on the influx of external Ca2+ through SOCs (FIG. 6), controlled by several modulatory mechanisms. The first is an example of the crosstalk between signalling pathways (BOX 1) and concerns the ability of the PI(3)K signalling pathway to stimulate PLCγ1 to maintain the supply of Ins(1,4,5)P3 (REF. 81). The second is the activation of K+ channels that serve to hyperpolarize the membrane to enhance the entry of external Ca2+ (FIG. 6)82. Finally, SOCs are prone to Ca2+-induced inhibition but this negative feedback pathway is reduced by mitochondria, which soak up the Ca2+ entering through the channels83. One possibility is that the mitochondria may then redistribute the Ca2+ by releasing it deeper within the cell83.

SPHINGOMYELIN SIGNALLING

Several metabolites of sphingomyelin affect apoptosis through poorly undertood mechanisms: ceramide and sphingomyelin are generally proapoptotic whereas sphingosine 1-phosphate is generally antiapoptotic. STRESS-ACTIVATED PROTEIN KINASES

Members of the mitogenactivated protein kinase (MAPK) family that are activated by stress, including cJun N-terminal kinase (JNK) and p38 MAPK.

Transcription factor activation. The main function of Ca2+ in controlling cell proliferation is to activate transcription factors either in the cytoplasm (NF-AT, NFκB) or within the nucleus (CREB) (FIG. 6). The function of Ca2+ during stimulation of gene transcription in lymphocytes (FIG. 6) is similar to that in neurons during learning (FIG. 4b). One action of Ca2+ is to stimulate the Ca2+-sensitive protein phosphatase calcineurin to dephosphorylate NF-AT, which then enters the nucleus84. As soon as Ca2+ signalling stops, kinases in the nucleus rapidly phosphorylate NF-AT, which then leaves the nucleus, and transcription of NF-AT-responsive genes ceases. The prolonged period of Ca2+ signalling that is required to induce proliferation is therefore necessary to maintain NF-AT in its active form. Interrupting this signalling cascade at various points decreases gene transcription and cell proliferation. Transcription is inhibited in mutants with defective SOCs that cannot sustain Ca2+ signalling85. Likewise, the immunosuppressants cyclosporin A and FK506 prevent transcription by inhibiting the action of calcineurin (FIG. 6). An increase in Ca2+ is one of the signals that can trigger the proteolysis of the inhibitory IκB subunit, allowing the active NF-κB subunit to enter the nucleus. CREB, in contrast to the factors discussed above, is a nuclear Ca2+-responsive transcription factor, which is phosphorylated by CAMKII and CAMKIV. In addition, Ca2+ acting within the nucleus is also responsible for stimulating the Ca2+sensitive transcriptional co-activator CREB-binding protein (CBP)86,87. A CAM inhibitory peptide targeted to the nucleus can block DNA synthesis and cell-cycle progression, emphasizing the importance of a nuclear Ca2+ signal for cell proliferation88. These Ca2+-sensitive transcription factors activate numerous target genes; some code for progression factors such as the interleukin 2 system that is responsible for switching on DNA synthesis, whereas others produce components such as Fas and the Fas ligand that trigger apoptosis. So Ca2+ is central in setting up the signalling systems that enable cells to decide whether to grow or to die. Ca2+ disregulation and cancer. Phospholipase C has been referred to as a malignancy-linked signal transducing enzyme89 and its overexpression will promote trans-

formation and tumorigenesis in NIH3T3 cells90. The activity of the enzyme phosphatidylinositol-4-OH kinase, which catalyses production of the precursor that is hydrolysed to form Ins(1,4,5)P3, is greatly enhanced in certain cancer cells91. Several drugs that block Ca2+ entry can retard the growth of human melanoma, lung and colon carcinoma cells92, vascular smooth muscle cells93 and human prostate cancer cells94. One inhibitor, carboxy-amidotriazole, has been used in clinical trials to control refractory cancers95. Finally, the auxiliary subunit of a voltage-dependent Ca2+ channel α2δ is a potential tumour suppressor for several cancers96. Calcium and apoptosis. The function of Ca2+ in apoptosis is an enormously complex subject involving interplay between many systems including the SPHINGOMYELIN SIGNALLING PATHWAY, the redox system, the STRESS-ACTIVATED 2+ PROTEIN KINASE cascade and the Ca signalling pathway. In the last case, one function of Ca2+ is to control the expression of the apoptotic signalling components such as the Fas system described above. In addition, Ca2+ can also induce apoptosis in response to various pathological conditions and this often depends on an interplay between the mitochondria and the ER97. As described earlier (FIG. 2), there is a continuous ebb and flow of Ca2+ between these two organelles. There are indications that pro-apoptotic stimuli such as ceramide can influence how mitochondria respond to this periodic flux of Ca2+. The Ca2+ signals produced by Ins(1,4,5)P3 are handled normally, but when superimposed on a background of ceramide they induce apoptosis through formation of the PTP98. The latter usually forms when the mitochondria become overloaded with Ca2+ and so release cytochrome c (see above). The apoptosis regulatory proteins that function either as death antagonists (Bcl-2 and Bcl-XL) or death agonists (Bax, Bak and Bad), may exert some of their actions by interfering with the Ca2+ dynamics of these two organelles. For example, Bcl-2 is located both in the ER and in mitochondria. Both Bax and Bad accelerate opening of the voltage-dependent anion channel, which is part of the permeability transition pore (PTP), and so contribute to the release of cytochrome c99.Conversely, Bcl-2 and Bcl-XL seem to block Ca2+-induced apoptosis100–101. They enable the mitochondria to cope with large loads of Ca2+ (REFS 102–104). The function of Bcl-2 on the ER is uncertain. There are reports that Bcl-2 enhances the store of Ca2+(REF. 104), perhaps by upregulating SERCA gene expression105. However, other reports indicate that it increases membrane permeability, thereby reducing the concentration of Ca2+ in the ER106,107. An important consequence of having less Ca2+ in the ER is that the amount of Ca2+ being released during signalling is reduced106,107, which also decreases the amount taken up by the mitochondria. The anti-apoptotic action of Ca2+ may therefore depend on this reduction of the amount of Ca2+ circulating within the ER/mitochondrial system. From universality to individuality

Cells have access to an extensive Ca2+ signalling toolkit,

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 1 | OCTOBER 2000 | 1 9

REVIEWS from which they can assemble signalling systems with variable spatial and temporal properties. Imaging techniques have enabled us to characterize a physiological toolkit in the form of the elementary events that are Links DATABASE LINKS

Inositol signalling | M. J. Sanderson’s lab page ELS LINKS

Calcium signalling and regulation of cell function| Calcium and neurotransmitter release | Calcium channel diversity

16.

Berridge, M. J. Inositol trisphosphate and calcium signalling. Nature 361, 315–325 (1993). Clapham, D. E. Calcium signaling. Cell 80, 259–268 (1995). Clapper, D. L., Walseth, T. F., Dargei, P. J. & Lee, H. C. Pyridine nucleotide metabolites stimulate calcium release from sea urchin egg microsomes desensitized to inositol trisphosphate. J. Biol. Chem. 262, 9561–9568 (1987). Genazzini, A. A. & Galione, A. A. Ca2+ release mechanism gated by the novel pyridine nucleotide, NAADP. Trends Pharmacol. Sci. 18, 108–110 (1997). Mao, C. G. et al. Molecular cloning and characterization of SCaMPER, a sphingolipid Ca2+ release-mediating protein from endoplasmic reticulum. Proc. Natl Acad. Sci. USA 93, 1993–1996 (1996). Cancela, J. M. & Petersen, O. H. The cyclic ADP ribose antagonist 8-NH2-cADP-ribose blocks cholecystokininevoked cytosolic Ca2+ spiking in pancreatic acinar cells. Pfluger’s Arch. 435, 746–748 (1998). Young, K. W., Challiss, R. A. J., Nahorski, S. R., & Mackrill, J. J. Lysophosphatidic acid-mediated Ca2+ mobilization in human SH-SY5Y neuroblastoma cells is independent of phosphoinositide signalling, but dependent on sphingosine kinase activation. Biochem. J. 343, 45–52 (1999). Putney, J. W. Jr. A model for receptor-regulated calcium entry. Cell Calcium 7, 1–12 (1986). Hofmann, T. et al. Direct activation of human TRP6 and TRPC3 channels by diacylglycerol. Nature 397, 259–263 (1999). The mammalian homologues of the Drosophila transient receptor potential (TRP) proteins function as Ca2+ channels but their control is still largely unknown. This paper suggests that some may be regulated by diacylglycerol. Broad, L. M., Cannon, T. R. & Taylor, C. W. A noncapacitative pathway activated by arachidonic acid is the major Ca2+ entry mechanism in rat A7r5 smooth muscle cells stimulated with low concentrations of vasopressin. J. Physiol. 517, 121–134 (1999). Mignen, O. & Shuttleworth, T. J. IARC, a novel arachidonate-regulated, noncapacitative Ca2+ entry channel. J. Biol. Chem. 275, 9114–9119 (2000). Kiselyov, K. et al. Functional interaction between InsP3 receptors and store-operated Htrp3 channels. Nature 396, 478–482 (1998). Some of the first evidence to indicate that inositol1,4,5-trisphosphate receptors might be directly linked to Ca2+ channels in the plasma membrane. Boulay, G. et al. Modulation of Ca2+ entry by polypeptides of the inositol 1,4,5-trisphosphate receptor (IP3R) that bind transient receptor potential (TRP): Evidence for roles of TRP and IP3R in store depletion-activated Ca2+ entry. Proc. Natl Acad. Sci. USA 96, 14955–14960 (1999). Berridge, M. J. Capacitative calcium entry. Biochem. J. 312, 1–11 (1995). Bootman, M. D. & Lipp, P. Calcium signalling: Ringing changes to the ‘bell-shaped curve’. Curr. Biol. 9, R876–R878 (1999). Mermelstein, P. G., Bito, H., Deisseroth, K. & Tsien, R. W.

| OCTOBER 2000 | VOLUME 1

2. 3.

8. 9.

10.

11.

12.

13.

14. 15.

17. 18.

19.

20.

21.

22.

23.

24.

25.

26.

27.

28.

29.

30.

used to create either localized signals, or the coordinated activity of these events to produce global signals in the form of intracellular and intercellular waves. Now that the molecular and physiological mechanisms have been identified, the new challenge is to determine how this versatile Ca2+ signalling system functions in specific cellular processes. The universality of this signalling system is evident in its emerging function during various developmental processes such as axis specification, pattern formation and cellular differentiation. In the developing nervous system, for example, patterns of Ca2+ spikes regulate axonal growth and neuronal connectivity. The process of differentiation is particularly interesting because Ca2+ acts directly in setting up these Ca2+ signalling pathways. An important problem for the future, therefore, is to understand how Ca2+ functions to select out those components of the molecular toolkit that are uniquely expressed in each cell type.

Critical dependence of cAMP response element-binding protein phosphorylation on L-type calcium channels support a selective response to EPSPs in preference to action potentials. J. Neurosci. 20, 266–273 (2000). Berridge, M. J. Neuronal calcium signaling. Neuron 21, 13–26 (1998). Nakamura, T., Barbara, J. G., Nakamura, K. & Ross, W. N. Synergistic release of Ca2+ from IP3-sensitive stores evoked by synaptic activation of mGluRs paired with backpropagating action potentials. Neuron 24, 727–737 (1999). Direct evidence that the inositol-1,4,5-trisphosphate receptor may act as a coincident detector, integrating a Ca2+ signal coming from an action potential and inositol-1,4,5-trisphosphate generated by a metabotropic receptor. Cancela, J. M., Churchill, G. C. & Galione, A. Coordination of agonist-induced Ca2+-signalling patterns by NAADP in pancreatic acinar cells. Nature 398, 74–76 (1999). Fierro, L. & Llano, I. High endogenous calcium buffering in Purkinje cells from rat cerebellar slices. J. Physiol. 496, 617–625 (1996). Pozzan, T., Rizzuto, R., Volpe, P. & Meldolesi, J. Molecular and cellular physiology of intracellular calcium stores. Physiol. Rev. 74, 595–636 (1994). Blaustein, M. P. & Lederer, W. J. Sodium/calcium exchange: Its physiological implications. Physiol. Rev. 79, 763–854 (1999). Budd, S. L. & Nicholls, D. G. A reevaluation of the role of mitochondria in neuronal Ca2+ homeostasis. J. Neurochem. 66, 403– 411 (1996). Jouaville, L. S., Ichas, F., Holmuhamedor, E. L., Camacho, P. & Lechleiter, J. D. Synchronization of calcium waves by mitochondrial substrates in Xenopus laevis oocytes. Nature 377, 438–441 (1995). Collins, T. J., Lipp, P., Berridge, M. J., Li, W. & Bootman, M. D. Inositol 1,4,5-trisphosphate-induced Ca2+ release is inhibited by mitochondrial depolarization. Biochem. J. 347, 593–600 (2000). Duchen, M. R. Contributions of mitochondria to animal physiology: from homeostatic sensor to calcium signalling and cell death. J. Physiol. 516, 1–17 (1999). Rizzuto, R., Brini, M., Murgia, M. & Pozzan, T. Microdomains with high Ca2+ close to IP3-sensitive channels that are sensed by neighboring mitochondria. Science 262, 744–747 (1993). The first demonstration that mitochondria sense the high concentrations of Ca2+ that build up in the vicinity of intracellular channels such as the inositol1,4,5-trisphosphate receptor. Csordas, G., Thomas, A. P. & Hajnoczky, G. Quasisynaptic calcium signal transmission between endoplasmic reticulum and mitochondria. EMBO J. 18, 96–108 (1999). Leissring, M. A. et al. Capacitative calcium entry deficits and elevated luminal calcium content in mutant presenilin-1 knockin mice. J. Cell Biol. 149, 793–797 (2000). Bernadi, P. Mitochondrial transport of cations: channels,

31.

32. 33.

34.

35.

36.

37.

38.

39.

40.

41.

42.

43.

44.

45.

exchangers, and permeability transition. Physiol. Rev. 79, 1127–1155 (1999). Ichas, F., Jouaville, L. S. & Mazat, J. P. Mitochondria are excitable organelles capable of generating and conveying electrical and calcium signals. Cell 89, 1145–1153 (1997). Berridge, M. J. Elementary and global aspects of calcium signalling. J. Physiol. 499, 291–306 (1997). Lipp, P. & Niggli, E. A hierarchical concept of cellular and subcellular Ca2+ signaling. Prog. Biophys. Mol. Biol. 65, 265–296 (1996). Lipp, P. & Niggli, E. Fundamental calcium release events revealed by two-photon excitation photolysis of caged calcium in guinea-pig cardiac myocytes. J. Physiol. 508, 801–809 (1998) Bootman, M., Niggli, E., Berridge, M. J. & Lipp, P. Imaging the hierarchical Ca2+ signalling system in HeLa cells. J. Physiol. 499, 307–314 (1997). Cheng, H., Lederer, W. J. & Cannell, M. B. Calcium sparks — elementary events underlying excitationcontraction coupling in heart-muscle. Science 262, 740–744 (1993). One of the first visualizations of the localized Ca2+ signal emerging from small groups of ryanodine receptors. Such elementary events are the basic building blocks of Ca2+ signals. Yao, Y., Coi, J. & Parker, I. Quantal puffs of intracellular Ca2+ evoked by inositol trisphosphate in Xenopus oocytes. J. Physiol. 482, 533–553 (1995). Sun, X. -P., Callamaras, N., Marchant, J. S. & Parker, I. A continuum of InsP3-mediated elementary Ca2+ signalling events in Xenopus oocytes. J. Physiol. 509, 67–80 (1998). Thomas, D. Lipp, P., Berridge, M. J. & Bootman, M. D. Hormone-evoked elementary Ca2+ signals are not stereotypic, but reflect activation of different size channel clusters and variable recruitment of channels within a cluster. J. Biol. Chem. 273, 27130–27136 (1998). Lansley, A. B. & Sanderson, M. J. Regulation of airway ciliary activity by Ca2+: Simultaneous measurement of beat frequency and intracellular Ca2+. Biophys. J. 77, 629–638 (1999). Robb-Gaspers, L. D. & Thomas, A. P. Coordination of Ca2+ signaling by intercellular propogation of Ca2+ waves in the intact liver. J. Biol. Chem. 270, 8102–8107 (1995). The first demonstration of intercellular Ca2+ waves travelling through large numbers of cells in an intact organ. Zimmermann, B. & Walz, B. The mechanism mediating regenerative intercellular Ca2+ waves in the blowfly salivary gland. EMBO J. 18, 3222–3231 (1999). Tse, F. W. & Tse, A. Regulation of exocytosis via release of Ca2+ from intracellular stores. BioEssays 21, 861–865 (1999). Maturana, A. D. et al. Angiotensin II negatively modulates L-type calcium channels through a pertussis toxinsensitive G protein in adrenal glomerulosa cells. J. Biol. Chem. 274, 19943–19948 (1999). Lipp, P., Thomas, D., Berridge, M. J. & Bootman, M. D. Nuclear calcium signalling by individual cytoplasmic

www.nature.com/reviews/molcellbio

REVIEWS

46.

47.

48.

49.

50.

51.

52.

53.

54.

55.

56.

57.

58.

59.

60.

61.

62.

63.

64.

65.

66.

calcium puffs. EMBO J. 16, 7166–7173 (1997). A demonstration that Ca2+ puffs are concentrated around the nucleus and are therefore able to feed Ca2+ directly into the nucleoplasm. DeKoninck, P. & Schulman, H. Sensitivity of CaM kinase II to the frequency of Ca2+ oscillations. Science 279, 227–230 (1998). Oancea, E. & Meyer, T. Protein kinase C as a molecular machine for decoding calcium and diacylglycerol signals. Cell 95, 307–318 (1998). Li, W. H., Llopis, J., Whitney, M., Zlokarnik, G. & Tsien, R. Y. Cell-permeant caged InsP3 ester shows that Ca2+ spike frequency can optimize gene expression. Nature 392, 936–941 (1998). Dolmetsch, R. E., Xu, K. L. & Lewis, R. S. Calcium oscillations increase the efficiency and specificity of gene expression. Nature 392, 933–936 (1998). Ding, J. M. et al. A neuronal ryanodine receptor mediates light-induced phase delays of the circadian clock. Nature 394, 381–384 (1998). Hamada, T. et al. The role of inositol trisphosphateinduced Ca2+ release from IP3-receptor in the rat suprachiasmatic nucleus on circadian entrainment mechanism. Neurosci. Lett. 263, 125–128 (1999). Miyazaki, S. et al. Essential role of the inositol 1,4,5trisphosphate receptor/Ca2+ release channel in Ca2+ waves and Ca2+ oscillations at fertilization of mammalian eggs. Dev. Biol. 158, 62–78 (1993). Jones, K. T., Matsuda, M., Parrington, J., Katan, M. & Swann, K. Different Ca2+-releasing abilities of sperm extracts compared with tissue extracts and phospholipase C isoforms in sea urchin egg homogenate, and mouse eggs. Biochem. J. 346, 743–749 (2000). Swanson, C. A., Arkin, A. P. & Ross, J. An endogenous calcium oscillator may control early embryonic division. Proc. Natl Acad. Sci. USA 94, 1194–119 (1997). Kono, T., Jones, K. T., BosMikich, A., Whittingham, D. G. & Carroll, J. A cell cycle-associated change in Ca2+ releasing activity leads to the generation of Ca2+ transients in mouse embryos during the first mitotic division. J. Cell Biol. 132, 915–923(1996). Chang, D. C. & Meng, C. L A localized elevation of cytosolic-free calcium is associated with cytokinesis in the zebrafish embryo. J. Cell Biol. 131, 1539–1545 (1995). Keating, T. J., Cork, R. J. & Robinson, K. R. Intracellular free calcium oscillations in normal and cleavage-blocked embryos and artificially activated eggs of Xenopus-laevis. J. Cell Sci. 107, 2229–2237 (1994). Kubota, H. Y., Yoshimoto, Y. & Hiramoto, Y. Oscillation of intracellular free calcium in cleaving and cleavagearrested embryos of Xenopus-laevis. Dev. Biol. 160, 512–518 (1993). Stith, B. J., Goalstone, M., Silva, S. & Jaynes, C. Inositol 1,4,5-trisphosphate mass changes from fertilization through 1st-cleavage in Xenopus laevis. Mol. Biol. Cell 4, 435–443 (1993). Han, J. K. Oscillation of inositol polyphosphates in the embryonic cleavage cycle of the Xenopus laevis. Biochem. Biophys. Res. Commun. 206, 775–780 (1995). Ciapa, B., Pesando, D., Wilding, M. & Whitaker, M. Cellcycle calcium transients driven by cyclic changes in inositol trisphosphate levels. Nature 368, 875–878 (1994). Some of the first evidence that cyclic changes in inositol-1,4,5-trisphosphate and Ca2+ are responsible for controlling certain cell-cycle events, especially those occurring at mitosis. Gilland, E., Miller, A. L., Karplus, E., Baker, R. & Webb, S. E. Imaging of multicellular large-scale rhythmic calcium waves during zebrafish gastrulation. Proc. Natl Acad. Sci. USA 96, 157–161(1999). Webb, S. E. & Miller, A. L. Calcium signalling during zebrafish embryonic development. Bioessays 22, 113–123 (2000). Creton, R., Speksnijder, J. E. & Jaffe, L. F. Patterns of free calcium in zebrafish embryos. J. Cell Sci. 111, 1613–1622 (1998). Maslanski, J. A, Leshko, L. & Busa, W. B. Lithiumsensitive production of inositol phosphates during amphibian embryonic mesoderm induction. Science 256, 243–245(1992). Kume, S., Muto, A., Okano, H. & Mikoshiba, K. Developmental expression of the inositol 1,4,5-

67. 68.

69.

70.

71.

72.

73.

74.

75.

76. 77.

78. 79.

80.

81.

82.

83.

84.

85.

86.

87.

88.

89.

trisphosphate receptor and localization of inositol 1,4,5trisphosphate during early embryogenesis in Xenopus laevis. Mech. Dev. 66, 157–168 (1997). Reinhard, E. et al. Localized calcium signals in early zebrafish development. Dev. Biol. 170, 50–71(1995). Creton, R., Kreiling, J. A. & Jaffe, L. F. Presence and roles of calcium gradients along the dorsal-ventral axis in Drosophila embryos. Dev. Biol. 217, 375–385 (2000). Kühl, M., Sheldahl, L. C., Malbon, C. C. & Moon, R. T. Ca2+/calmodulin-dependent protein kinase II is stimulated by Wnt and frizzled homologs and promotes ventral cell fates in Xenopus. J. Biol. Chem. 275, 12701–12711 (2000). Kume, S. et al. Role of inositol 1,4,5-trisphosphate receptor in ventral signaling in Xenopus embryos. Science 278, 1940–1943 (1997). A role for Ca2+ in setting up the dorsoventral axis in Xenopus oocytes was demonstrated by showing that the axis was modified by inhibiting the activity of the inositol-1,4,5-trisphosphate receptor. Buonanno, A. & Fields, R. D. Gene regulation by patterned electrical activity during neural and skeletal muscle development. Curr. Opin. Neurobiol. 9, 110–120 (1999). Ferrari, M. B., Ribbeck, K., Hagler, D. J. & Spitzer, N. C. A calcium signaling cascade essential for myosin thick filament assembly in Xenopus myocytes. J. Cell Biol. 141, 1349–1356 (1998). Gu, X. N. & Spitzer, N. C. Breaking the code: Regulation of neuronal differentiation by spontaneous calcium transients. Dev. Neurosci. 19, 33–41(1997). Carey, M. B. & Matsumoto, S. G. Spontaneous calcium transients are required for neuronal differentiation of murine neural crest. Dev. Biol. 215, 298–313 (1999). Gomez, T. M. & Spitzer, N. C. In vivo regulation of axon extension and pathfinding by growth-cone calcium transients. Nature 397, 350–355(1999). By studying Ca2+ signals in individual neurons growing in vivo, it was possible to show that brief Ca2+ transients function both in the extension of the axon and in its ability to locate its target. Wong, R. O. L. Retinal waves and visual system development. Annu. Rev. Neurosci. 22, 29–47 (1999). Lu, K. P. & Means, A. R. Regulation of the cell-cycle by calcium and calmodulin. Endocrine Rev. 14, 40–58 (1993). Berridge, M. J. Calcium signalling and cell-proliferation. Bioessays 17, 491–500 (1995). Monks, C. R. F., Freiberg, B. A., Kupfer, H., Sciaky, N. & Kupfer, A. Three-dimensional segregation of supramolecular activation clusters in T cells. Nature 395, 82–86 (1998). Akagi, K., Nagao, T. & Urushidani, T. Correlation between Ca2+ oscillation and cell proliferation via CCKB/gastrin receptor. Biochim. Biophys. Acta 1452, 243–253 (1999). Scharenberg, A. M. & Kinet, J. P. Ptdlns-3,4,5-P3: A regulatory nexus between tyrosine kinases and sustained calcium signals. Cell 94, 5–8 (1998). Lewis, R. S. & Cahalan, M. D. Potassium and calcium channels in lymphocytes. Annu. Rev. Immunol. 13, 623–653 (1995). Hoth, M., Fanger, C. M. & Lewis, R. S. Mitochondrial regulation of store-operated calcium signalling in T lymphocytes. J. Cell Biol. 137, 633–648 (1997). Crabtree, G. R. Generic signals and specific outcomes: Signaling through Ca2+, calcineurin, and NF-AT. Cell 96, 611–614 (1999). Timmerman, L. A., Clipstone, N. A., Ho, S. N., Northrop, J. P. & Crabtree, G. R. Rapid shuttling of NF-AT in discrimination of Ca2+ signals and immunosuppression. Nature 383, 837–840 (1996). Chawla, S., Hardingham, G. E., Quinn, D. R. & Bading, H. CBP: A signal-regulated transcriptional coactivator controlled by nuclear calcium and CaM kinase IV. Science 281, 1505–1509 (1998). Hardingham, G. E., Chawla, S., Cruzalegui, F. H. & Bading, H. Control of recruitment and transcription–activating function of CBP determines gene regulation by NMDA receptors and L-type calcium channels. Neuron 22, 789–798 (1999). Wang, J. H., Moreira, K. M., Campos, B., Kaetzel, M. A. & Dedman, J. R. Targeted neutralization of calmodulin in the nucleus blocks DNA synthesis and cell cycle progression. Biochim. Biophys. Acta 1313, 223–228 (1996). Yang, H., Shen, F., Herenyiova, M. & Weber, G.

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

90.

91.

92.

93.

94.

95.

96.

97.

98.

99.

100. 101. 102.

103.

104.

105.

106.

107.

108.

109.

110.

111.

112.

Phospholipase C (EC 3. 1. 4. 11): A malignancy linked signal transduction enzyme. Anticancer Res. 18, 1399–1404 (1998). Smith, M. R. et al. Overexpression of phosphoinositidespecific phospholipase C γ in NIH 3T3 cells promotes transformation and tumorigenicity. Carcinogenesis 19, 177–185 (1998). Rizzo, M. T. & Weber, G. L. Phosphatidylinositol 4-kinase — an enzyme linked with proliferation and malignancy. Cancer Res. 54, 2611–2614 (1994). Benzaquen, L. R., Brugnara, C., Byers, H. R., GattoniCelli, S. & Halperin, J. A. Clotrimazole inhibits cell proliferation in vitro and in vivo. Nature Med. 1, 534–540 (1995). Nie, L., Mogami, H., Kanzaki, M., Shibata, H. & Kojima, I. Blockade of DNA synthesis induced by platelet-derived growth factor by tranilast, an inhibitor of Ca2+ entry, in vascular smooth muscle cells. Mol. Pharm. 50, 763–769 (1996). Haverstick, D. M., Heady, T. N., Macdonald, T. L. & Gray, L. S. Inhibition of human prostate cancer proliferation in vitro and in a mouse model by a compound synthesized to block Ca2+ entry. Cancer Res. 60, 1002–1008 (2000). Kohn, E. C. et al. Clinical investigation of a cytostatic calcium influx inhibitor in patients with refractory cancers. Cancer Res. 56, 569–573 (1996). Gao, B. et al. Functional properties of a new voltagedependent calcium channel α2δ auxiliary subunit gene (CACNA2D2). J. Biol. Chem. 275, 12237–12242 (2000). One of the first indications that malignancy might be linked to an alteration in Ca2+ signalling. Kass, G. E. N. & Orrenius, S. Calcium signaling and cytotoxicity. Environ. Health Perspect. 107, 25–35 (1999). Szalai, G., Krishnamurthy, R. & Hajnoczky, G. Apoptosis driven by IP3-linked mitochondrial calcium signals. EMBO J. 18, 6349–6361 (1999). Shimizu, S., Narita, M. & Tsujimoto, Y. Bcl-2 family proteins regulate the release of apoptogenic cytochrome c by the mitochondrial channel VDAC. Nature 399, 483–487 (1999). Barr, P. J. & Tomei, L. D. Apoptosis and its role in human disease. Biotechnology 12, 487–493 (1994). Reed, J. C. Bcl-2 and the regulation of programmed celldeath. J. Cell Biol. 124, 1–6 (1994). Murphy, A., Bredesen, D. E., Cortopassi, G., Wang, E. & Fiskum, G. Bcl-2 potentiates the maximal calcium uptake capacity of neural cell mitochondria. Proc. Natl Acad. Sci. USA 93, 9893–9898 (1996). Vander Heiden, M., Chandel, N. S., Williamson, E. K., Schumacker, P. T. & Thompson, C. B. Bcl-xL regulates the membrane potential and volume homeostasis of mitochondria. Cell 91, 627–637 (1997). Zhu, L. P. et al. Modulation of mitochondrial Ca2+ homeostasis by Bcl-2. J. Biol. Chem. 274, 33267–33273 (1999). Evidence that mitochondrial metabolism can be modulated by the anti-apoptotic modulator Bcl-2. Kuo, T. H. et al. Modulation of endoplasmic reticulum calcium pump by Bcl-2. Oncogene 17, 1903–1910 (1998). Foyouzi-Youssefi, R. et al. Bcl-2 decreases the free Ca2+ concentration within the endoplasmic reticulum. Proc. Natl Acad. Sci. USA 97, 5723–5728 (2000) Pinton, P. et al. Reduced loading on intracellular Ca2+ stores and downregulation of capacitative Ca2+ influx in Bcl-2-overexpressing cells. J. Biol. Chem. 275, 857–862 (2000). Schlossmann, J. et al. Regulation of intracellular calcium by a signalling complex of IRAG, IP3 receptor and cGMP kinase Iβ. Nature 404, 197–201 (2000). Morimoto, A. M. et al. The MMAC1 tumor suppressor phosphatase inhibits phospholipase C and integrin-linked kinase-activity. Oncogene 19, 200–209 (2000). Lev, S. et al. Protein-tyrosine kinase PYK2 involved in Ca2+-induced regulation of ion-channel and map kinase functions. Nature 376, 737–745 (1995). Brinson, A. E. et al. Regulation of a calcium-dependent tyrosine kinase in vascular smooth muscle cells by angiotensin II and platelet-derived growth factor. Dependence on calcium and the actin cytoskeleton. J. Biol. Chem. 273, 1711–1718 (1998). Prenzel, N. et al. EGF receptor transactivation by Gprotein-coupled receptors requires metalloproteinases of proHB–EGF. Nature 402, 884–888 (1999).

VOLUME 1 | OCTOBER 2000 | 2 1

REVIEWS

A TALE OF TOROIDS IN DNA METABOLISM Manju M. Hingorani* and Mike O’Donnell*‡ A strikingly large number of the proteins involved in DNA metabolism adopt a toroidal — or ring-shaped — quaternary structure, even though they have completely unrelated functions. Given that these proteins all use DNA as a substrate, their convergence to one shape is probably not a coincidence. Ring-forming proteins may have been selected during evolution for advantages conferred by the toroidal shape on their interactions with DNA.

PROCESSIVITY

The ability of an enzyme to catalyse more than one turnover before releasing the substrate or product of the reaction. REPLISOME

The multi-protein assembly at the junction of the DNA replication fork.

*The Rockefeller University; ‡The Howard Hughes Medical Institute, 1230 York Avenue, New York, New York 10021, USA. e-mails: manju@mod. rockefeller.edu odonnel@rockvax. rockefeller.edu

High-resolution studies of protein structure have led to the discovery of several ring-shaped macromolecules that have essential metabolic functions. These proteins generally contain several subunits, and provide an enclosed environment for chemical reactions that may otherwise be unfavourable. Examples of such circular proteins include the F1-ATPase1, molecular chaperones that catalyse protein folding (for example, Escherichia coli GroEL)2, the proteasomes that catalyse protein degradation3, and bacterial light-harvesting complexes4. Among these diverse toroids is a subset of proteins dedicated to DNA metabolism, that physically or chemically manipulate DNA during DNA replication, repair and recombination5. In fact, a disproportionately large number of the proteins involved in DNA metabolism assume a circular shape. They range from sliding clamps, which passively diffuse on DNA, to the helicases that catalyse ATP-fuelled DNA unwinding, or to enzymes such as bacteriophage λ exonuclease and topoisomerases that chemically modify DNA. Many of these proteins seem to have unrelated evolutionary origins and catalyse very different reactions on DNA, although they share a quaternary ring-shaped structure. So why have so many proteins adopted the same overall shape? Possibly because they all work on DNA — the distinctive structure and properties of the DNA polymer could have stimulated convergence to the toroidal form. The sheer number and diversity of toroidal proteins using DNA as a substrate — in contrast to protein or RNA substrates — also implies that this form has had a marked effect on the development

of complex DNA metabolic processes. This review describes prominent ring-shaped proteins that work on DNA, and highlights the effect of their toroidal form on function, to try and understand why this form is so successful in DNA metabolism. Circular sliding clamps

Sliding-clamp proteins belong to the family of PROCESSIVITY factors, whose main function is to increase the lifetime

of other proteins on DNA. Clamps are an integral part of the REPLISOME, where they facilitate a stable interaction between DNA polymerase and the primer–template during DNA replication6,7. The speed of replication is influenced greatly by these proteins, as the polymerase alone tends to fall off the primer–template every few nucleotides, slowing down DNA synthesis. In contrast, when tethered to DNA by a sliding clamp, the polymerase becomes highly processive and extends the primer by thousands of nucleotides without falling off, resulting in efficient replication of genomic DNA8–10. Early biochemical studies showed that clamps move freely on DNA and fall off the ends of linearized DNA molecules, therefore predicting that these proteins interact with DNA using a sequence-independent topological linkage11. This hypothesis was confirmed by the first crystal structure of a clamp, E. coli β, which revealed a circular protein with a central hole large enough (35 Å diameter) to accommodate double-stranded/duplex DNA with no steric hindrance (FIG. 1a)12. Two semi-circular β monomers form a tightly sealed ring that can remain stably linked to DNA with a half-life of about www.nature.com/reviews/molcellbio

| OCTOBER 2000 | VOLUME 1

REVIEWS

Figure 1 | Circular sliding clamps. a | Top views of sliding clamps from bacteriophage to humans, showing the central channel of about 35 Å that encloses duplex DNA. b | Model structure of the phage RB69 clamp bound behind the polymerase (grey) on DNA (orange) through a short peptide connector. c | Model mechanism of ATP-driven clamp-loader-catalysed assembly of the E. coli β clamp on DNA. Clamp assembly starts when two or more γ-subunits (red) bind ATP and change the conformation of the γ-complex , so that the δ-subunit (yellow) is exposed85,86. The δ-subunit binds the dimeric β-clamp (blue) and opens it at one interface82,86. The ATP-bound γ-complex also binds primed DNA with high affinity and brings it in close proximity to the open clamp85. Next, ATP hydrolysis by the γ-subunits is coupled to closure of the clamp around DNA and release of the topologically linked clamp–DNA complex for use by DNA polymerase87.

100 minutes13. The inside of the ring is positively charged and probably contacts the DNA phosphate backbone through nonspecific, water-mediated interactions, which allow the ring to slide in a sequenceindependent fashion on DNA. Sliding clamp proteins from bacteriophage (T4 and RB69 clamps)14,15, Saccharomyces cerevisiae (the yeast proliferating cell nuclear antigeny, yPCNA)16 and humans (hPCNA)17 have similar toroidal structures despite minimal sequence homology, but these proteins are smaller, so three monomers are required to form a ring with about the same dimensions as β (FIG. 1a). A recent structure15 of the RB69 clamp in complex with a carboxy-terminal peptide of the polymerase reveals the connection between the polymerase and the clamp on DNA (FIG. 1b). This has clarified further how circular sliding clamps act as mobile tethers for DNA polymerases and increase their processivity. Hexameric helicases

Like sliding clamps, hexameric helicases are toroidal

proteins that bind DNA within the central channel and move along it. But in this case, translocation is an energy-driven process fuelled by ATP, and is coupled to unwinding of dsDNA18,19. Most complex organisms (including bacteria) have several helicases that act in DNA replication, repair and recombination. One example highlighting the importance of helicase activity in nucleic acid metabolism is the BLM helicase, mutations in which are linked to Bloom’s syndrome20 (a human genetic disorder that leads to defects in growth and fertility as well as increased susceptibility to cancer). The E. coli DnaB helicase21 and rho22 (an RNA–DNA helicase and transcription terminator) were among the first helicases found to function as hexamers, and electron microscopy revealed that the six subunits form a ring (FIG. 2a). Hexameric ringshaped helicases have now been found in various organisms including bacteriophage (for example, T7 gp4 (REFS 23,24), T4 gp41(REF. 25) and SPP1 gp40S (REF. 26)), viruses (for example, SV40 T antigen27), bacteria (for example, E. coli DnaB28, rho29 and RuvB30),

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 1 | OCTOBER 2000 | 2 3

REVIEWS eukaryotes (for example, human mini-chromosome maintenance (MCM) complex31 and BLM helicase32) and archaea (for example, Methanobacterium thermoautotrophicum MCM33) (FIG. 2a). The helicase rings range from 120–140 Å in diameter with an inner diameter of 20–40 Å, which is large enough to accommodate single- or double-stranded DNA18. Whereas helicases from the T7 (REFS 23,34) and T4 (REF. 35) bacteriophages as well as E. coli DnaB36 encircle ssDNA, the bovine papilloma virus E1 helicase37 and SV40 T antigen38 might encircle ssDNA or dsDNA. One model for DNA unwinding places the hexameric helicase at the replication fork with one ssDNA strand bound within the ring and one excluded from the ring18. FIGURE 2b depicts how the T7 helicase binds the 5′ strand inside the ring with a specific polarity23, and can unwind DNA in the 5′–3′ direction but exclude the 3′ strand. Consistent with this ‘exclusion’ model, the unwinding activity of both the T7 and T4 helicases is strongly inhibited by a large block in the 5′ ssDNA, but is insensitive to a similar block in the 3′ ssDNA39–41. So at the fork junction, just the unidirectional translocation of the helicase ring on a single DNA strand could drive duplex DNA unwinding. Other versions of this model suggest that helicases

might also manipulate the excluded ssDNA and/or dsDNA at the fork (through interactions with the outside of the ring) to destabilize the base pairs in duplex DNA. Unwinding could be aided further by interaction between the helicase and DNA polymerase. For example, both the T7 helicase42 and DnaB43,44 bind their respective DNA polymerases. The polymerase may assist passively by imposing unidirectionality on the helicase (into the duplex DNA) or by an active mechanism in which the forward driving force of the polymerase accelerates helicase activity. The translocation and DNA unwinding activity of helicases depends on their ATPase (or NTPase) activity18,19. The binding affinity of hexameric helicases to DNA is increased in the presence of NTP (nucleoside 5′triphosphate) relative to NDP (nucleoside 5′- diphosphate), indicating that nucleotide binding and hydrolysis modulate their interaction with DNA45,46. Furthermore, rapid kinetic analyses of T7 helicase47 and E. coli rho48 indicate that the hexamers may use a sequential NTPase mechanism (similar to F1-ATPase1), in which one active site after another binds and releases DNA — coupled to NTP binding, hydrolysis and NDP release — driving both stepwise movement of the helicase ring on DNA and subsequent DNA unwinding (FIG. 2c). Recombination proteins

Figure 2 | Circular helicases. a | Top views of hexameric helicases from bacteriophage to humans, showing the central channel of about 25–35 Å that encloses single- or doublestranded DNA. b | Bacteriophage T7 helicase (yellow) encircling the 5′ strand and excluding the 3′ strand to unwind DNA in the 5′–3′ direction. c | A stepwise model mechanism for helicase translocation and DNA unwinding in which active sites on the hexamer consecutively bind NTP (T, red) and DNA, hydrolyse NTP (D, yellow), and release the hydrolysis products as well as DNA.

Many proteins in DNA recombination pathways (including DNA-repair-coupled recombination) assemble into multi-subunit toroids containing a central channel. For example, the Red system in bacteriophage λ uses at least two toroidal proteins. One of these, the λ exonuclease, is a trimer of identical subunits arranged in a ring (FIG. 3a)49. It binds dsDNA within the central channel, where the exonuclease active sites are located, and degrades one DNA strand in the 5′–3′ direction to generate long 3′ ssDNA overhangs. The width of the channel varies from about 30 Å at one end of the ring to roughly 15 Å at the other, such that dsDNA fits in one end and only ssDNA can fit in the other. The structure implies that the enzyme moves with a specific orientation on DNA, taking in dsDNA at one end, degrading the 5′ strand, and expelling the 3′ strand from the other. This DNA product can be processed further for recombination by the single-strand annealing pathway and the double-strand break repair pathway. Both pathways create primed DNA intermediates, which are then replicated to complete recombination or repair. In the single-strand annealing pathway, the β protein of phage λ binds ssDNA overhangs and facilitates basepairing with homologous regions in another single strand of DNA. Electron microscopy shows that the β protein forms oligomeric rings of various sizes50. These include small (roughly 12-subunit) rings in the absence of DNA, and larger (15- or 18-subunit) rings in the presence of ssDNA and Mg2+. In the presence of dsDNA, β protein preferentially forms a long protein filament on the DNA duplex (FIG. 3b). Interestingly, when presented with a heat-denatured DNA fragment (roughly four kilobases), β protein first forms large rings on the ssDNA and then makes long helical filawww.nature.com/reviews/molcellbio

| OCTOBER 2000 | VOLUME 1

REVIEWS

STRAND EXCHANGE

The process by which a single DNA strand switches from one duplex DNA molecule to base pair with a complementary strand from a second, homologous duplex DNA molecule. HOMOLOGOUS RECOMBINATION

The process by which segments of DNA are exchanged between two DNA duplexes that share high sequence similarity. CATENATION/DECATENATION

Topoisomerases catenate (join) or decatenate (separate) two circular DNA molecules by cutting one DNA strand, passing a second strand through the break, and resealing the break in DNA.

ments under DNA annealing conditions, implying that β rings on ssDNA nucleate β–dsDNA filaments. These structural data, along with biochemical analysis of β–DNA interactions51, indicate a possible mechanism for β-catalysed annealing of homologous ssDNAs (FIG. 3b). Initially, large β rings may form on the ssDNA overhangs (created by λ exonuclease), possibly wrapping DNA on the outside of the ring such that the bases are exposed. This may accelerate annealing by removing secondary structures, and by presenting the bases for rapid recognition and pairing with a complementary strand. Once annealing begins, the rings seem to convert into the helical β–dsDNA filament shown in FIG. 3b. It is not clear whether β protein dissociates soon after the duplex DNA is formed or if the nucleoprotein filament has a further role in recombination. Structural studies of proteins analogous to β protein, such as bacteriophage P22 Erf52 and E. coli RecT53, indicate that these proteins form rings and may form filaments (RecT), and implies that they could use similar mechanisms to catalyse ssDNA annealing. During double-strand break repair, the E. coli RecA protein binds ssDNA overhangs (to form a helical RecA–DNA filament), and facilitates STRAND EXCHANGE with homologous regions in a duplex DNA strand. RecA and its bacteriophage and eukaryotic homologues (UvsX and Rad51, respectively)54,55 catalyse ATP-dependent strand exchange between homologous ss- and dsDNA substrates56. These proteins form helical filaments on ssDNA, which facilitate formation of heteroduplex DNA (between ssDNA and its dsDNA partner) during strand exchange. Recently however, RecA and Rad51

have been found to assemble into rings that could also have some function in HOMOLOGOUS RECOMBINATION57,58. The ring may be a vestigial form of the RecA protein, but its persistence in higher eukaryotes, including humans (hRad51), indicates that the ring form may be functional. Moreover, the human DMC1 protein, a meiosis-specific RecA homologue that can weakly catalyse strand exchange in vitro, has so far been detected only as an octameric ring (FIG. 3c)59. The central channel of the ring is wide enough to encircle ss- and ds-DNA regions of duplex DNA containing single-stranded overhangs. Although hDMC1 rings can stack on DNA, the protein does not seem to form helical filaments on DNA (unlike RecA). These structural data raise the possibility that RecA and its homologues function as rings in an as yet unknown manner during DNA recombination. Recently, the human RAD52 protein was found to assemble into heptameric rings with a large funnelshaped channel, 40–60 Å in diameter60 (FIG. 3d). Eukaryotic Rad52 proteins are essential for DNA recombination and promote annealing of ssDNAs61. It is not clear how Rad52 uses its toroidal shape to facilitate ssDNA annealing; it might function like the bacteriophage β protein and Erf to present ssDNA appropriately for rapid annealing, although there is no evidence that Rad52 forms filaments on DNA. Rad52 seems to work with both Rad51 and RPA (replication protein A, a eukaryotic ssDNA-binding protein) in Rad51-mediated strand-exchange reactions, and with RPA for ssDNA annealing62–64. One way Rad52 rings could stimulate strand exchange is by binding ssDNA and nucleating or accelerating formation of the Rad51–DNA filament. Likewise Rad52 could modulate the interaction between RPA and DNA to stimulate the annealing reaction. Ongoing studies of how Rad52 binds DNA and its protein partners (Rad51, RPA) will help elucidate its mechanism of action in DNA recombination. Topoisomerases

Figure 3 | Circular recombination proteins. a | Top view of the bacteriophage λ exonuclease, showing the central channel that binds DNA. b | The β protein of bacteriophage λ, as a ring with single-stranded DNA wound around it (left) and as a nucleoprotein filament with duplex DNA (right). c | Octameric human DMC1, a meiosis-specific homologue of E. coli RecA. d | Heptameric human RAD52.

DNA topoisomerases break one or both strands of the double helix, pass a single strand or duplex DNA, respectively, through the break and re-ligate the strand(s) to catalyse changes in the topological state of DNA. These enzymes change the ‘superhelicity’ of DNA and CATENATE or DECATENATE DNA molecules, and are therefore essential for solving topological problems that entangle DNA during replication, recombination, repair or transcription, as well as chromosome condensation and segregation65. Topoisomerases belong to two main families: type I enzymes that break only one DNA strand transiently (and are divided further into type IA and type IB topoisomerases), and type II enzymes that break both strands of the duplex. High-resolution structures of topoisomerases from each family have been solved, and all have a toroidal shape with a large central cavity that can accommodate DNA. Escherichia coli topoisomerases I and III (FIG. 4a) are type IA enzymes in which four domains of a single polypeptide form a toroid with a hole about 25 Å in diameter66,67. The proposed mechanism of type IA topoisomerases involves initial DNA binding to an

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 1 | OCTOBER 2000 | 2 5

REVIEWS

Figure 4 | Circular topoisomerases. a | Type IA enzymes, E. coli topoisomerase I and III. Domains I and III are thought to move to expose the active site (where one DNA strand is cleaved) and open a gate for entry of the other strand into the central cavity. This is followed by re-ligation and release of both DNA strands to change the topology of DNA. b | A type IB enzyme, human topoisomerase I. The core subdomains II and III are thought to part to allow entry of dsDNA (pink) into the central cavity, where one DNA strand is cleaved. The other strand is passed through the break, followed by re-ligation and release of the DNA. c | A type II enzyme, Saccharomyces cerevisiae topoisomerase II DNA-binding fragment. The proposed ATP-driven mechanism involves binding and cleavage of one dsDNA at the upper domain, parting of the broken strand to allow passage of another duplex into the cavity, then re-ligation and release of both DNAs.

outer site on the ring. This induces ring opening, cleavage of one strand and passage of the other strand through the break into the central cavity, re-ligation of the broken strand and release of DNA, to yield a product with changed topology. The human topoisomerase I shown in FIG. 4b is an example of a type IB enzyme, the structure and mechanisms of which are unrelated to those of type IA enzymes. In this case, a bi-lobed monomer forms a closed clamp-like structure that encircles the DNA duplex68,69. The active site is on the inside of the toroid, and strand cleavage, strand passage and re-ligation all occur in the central cavity. The third family of topoisomerases, the type II enzymes, catalyse the transport of one dsDNA strand through another in a reaction coupled to ATP70,71. FIGURE 4c shows the carboxy-terminal DNA binding or cleavage fragment of S. cerevisiae topoisomerase II, in which two monomers are arranged around a central cavity about 50–55 Å in diameter72. The enzyme uses energy from ATP to bind one dsDNA strand, cleave it, part the ends for a second DNA strand to pass through into the central cavity, and then re-ligate the first strand. The second strand is then released from the cavity to complete the reaction. As topoisomerases nick or break DNA as part of their reaction, it is critical that they hold on to the reaction intermediates until the breaks in DNA are resealed, to avoid DNA damage. The toroidal design of topoisomerase addresses this problem, with the central cavity serving as a trap for the DNA intermediates until the reaction is complete. Other toroids

Several other toroids function in nucleic-acid metabolism, including NAD+-dependent DNA ligase (Thermus filiformis)73, the trp RNA-binding attenuation protein (Bacillus subtilis TRAP)74, the bacteriophage head–tail connector (for example, φ29 connector)75 and translin76 (FIG. 5). The NAD+-dependent DNA ligase is a

monomeric protein with four domains arranged in a ring (FIG. 5a). Suh and colleagues have proposed73 that the ligase may bind DNA within the central cavity, in a manner analogous to human topoisomerase I, as it seals the nick in dsDNA. They also speculate that its clamp-like shape may allow the ligase to slide on DNA until it encounters a nicked site. The TRAP protein forms an 11-subunit ring that helps to terminate transcription of the trp operon. The crystal structure of TRAP in complex with RNA shows how each subunit interacts with a GAG triplet repeat (FIG. 5b)74. When tryptophan levels in the cell are high, TRAP binds these repeats to facilitate formation of a transcription-terminator hairpin structure in the messenger RNA, which blocks further mRNA synthesis. The head–tail connector protein has several functions in the bacteriophage life cycle. It is a circular assembly of 12 identical subunits that bind DNA77,78 and also interact with an accessory ATPase factor. Together, these proteins pump DNA into the phage head75, possibly by an ATP-dependent mechanism analogous to that of RuvB (a ring-shaped DNA pump/helicase involved in homologous recombination) (FIG. 2)79. Translin is an octameric toroid that binds specific sequences at ssDNA ends of broken DNA76. Translin is highly conserved among vertebrates80, implying that it serves essential functions in these organisms. Possibly, it recognizes chromosomal DNA breakpoints, and could also form a scaffold for assembly of other proteins that repair or otherwise process broken DNA. Ring assembly on DNA

The closed ring structure of the proteins described above, and the fact that they generally bind DNA in the central channel, raises the question of how DNA gains access to the inside of the ring. The answers are as varied as the protein rings themselves. Topoisomerases (IA and IB) and DNA ligase seem intrinsically capable of large conformational changes that open and close the ring — much like a clam shell — around DNA67,73,81. Other prowww.nature.com/reviews/molcellbio

| OCTOBER 2000 | VOLUME 1

REVIEWS

Figure 5 | More circular proteins. a | Thermus filiformis NAD+-dependent DNA ligase with dsDNA modelled within the central channel. b | B. subtilis trp RNA binding attenuation protein (TRAP) bound to 11 tryptophan molecules and 11 repeating GAG triplets; similar TRAP binding to mRNA in vivo facilitates formation of a transcription-terminator hairpin structure in the RNA, effectively blocking tryptophan biosynthesis. c | The bacteriophage φ29 head–tail connector that promotes packaging of DNA into the phage prohead. d | Human translin, an octameric protein ring that recognizes chromosomal DNA breakpoints.

MOLECULAR MATCHMAKER

A macromolecule that increases the affinity of two or more other molecules for each other, usually through a reaction using energy from ATP binding and hydrolysis.

teins require assistance to form a topological link with DNA. For example, the circular sliding-clamp proteins are assembled around DNA by MOLECULAR MATCHMAKER proteins, known as ‘clamp loaders’, in a reaction fuelled by ATP (FIG. 1c). Clamp loaders from E. coli (γ complex)82, bacteriophage T4 (gp44/62)83 and eukaryotes (replication factor C or RFC)84, are multi-protein complexes in which the ATPase activity of one or more subunits drives the other subunits to bind the circular clamp and DNA, open the clamp and close it around the DNA82–87. Hexameric DNA helicases also form stable closed rings in the presence of Mg2+ and/or NTP, so they need a mechanism for assembly around DNA18 (FIG. 2). Studies of the bacteriophage T7 helicase88 and E. coli rho (D. E. Kim and S. S. Patel, unpublished observations) indicate that the DNA may first bind outside the ring, inducing conformational changes that open the ring and allow DNA to slip inside. Several helicases also interact with accessory protein factors that may facilitate their assembly on DNA. In bacteriophage T4, for example, the gp59 protein stimulates interaction between the gp41 helicase and DNA89. In E. coli, the DnaC protein facilitates assembly of the DnaB helicase on DNA90. Likewise, the Cdc6 protein may be required for assembly of the MCM helicase complex onto DNA91,92. Both gp59 (REF. 93) and DnaC94 bind their respective helicases as well as DNA

with high affinity, and may work by simply increasing local concentration of the helicase near DNA (where it can self-assemble around DNA). Alternatively, these proteins could be molecular matchmakers that load the helicase ring onto DNA, possibly by ATP-driven mechanisms analogous to those used by clamp-loader proteins. The requirement for a mechanism of assembly around DNA introduces a measure of specificity to the interaction between DNA and toroidal proteins, and presents a method of regulating their activity on DNA. This point is highlighted by the λ exonuclease ring, which does not self-assemble around DNA and does not use an accessory exonuclease-loader. Consequently, λ exonuclease can initiate DNA degradation only at free dsDNA ends (which can thread into the ring), and its nuclease activity is limited to specific loci such as breaks in dsDNA95. Regulation of the accessory proteins that catalyse ring assembly on DNA further influences the activity of toroidal proteins, and can have far-reaching consequences on DNA metabolism. For example, Cdc6, the MCM helicase loader, is subject to cell-cycle regulatory pathways97, and helicase-catalysed unwinding of the replication origin is required for assembly of replicationinitiation proteins on DNA7,96,97. So modulation of Cdc6 activity affects MCM activity on DNA, which in turn can affect initiation of replication. Similarly, individual components of the eukaryotic RFC complex, which is responsible for loading (and unloading) PCNA clamps onto DNA, have been implicated in cell-cycle checkpoint pathways84. Therefore modulation of clamp assembly (through the clamp loader) and consequent modulation of DNA polymerase activity may be one method by which cell-cycle regulatory pathways influence DNA metabolism and vice versa. Why rings?

Within the classes of toroids discussed above, there are a few proteins that have similar functions but do not assume a toroidal shape. Most notable among these are the monomeric and dimeric DNA helicases as well as the monomeric processivity factors of DNA polymerases. E. coli Rep helicase and UvrD (helicase II) have two subunits that appear to alternately bind and release DNA (in an ATPase-coupled mechanism), for unidirectional translocation and DNA unwinding19, and a recent study indicates that the UvrD helicase is also active as a monomer98. The PcrA helicase seems to function as a monomer, with DNA binding and release alternating between two sites on the same protein, allowing the helicase to move like an ‘inch-worm’ on DNA99. So why do so many DNA helicases function as rings? The most prominent benefit of a circular shape is the ability to form a topological link with DNA. Hexameric helicases bind DNA within the central channel, ensuring that when DNA is released from one site during translocation it does not dissociate completely but remains trapped within the ring until bound by another site (FIG. 2c)18. Non-toroidal helicases have the core DNA-unwinding activity, but a topological link with DNA may confer the advantage of high processivity on toroidal helicases. Indeed, most of the hexameric helicases examined so far

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 1 | OCTOBER 2000 | 2 7

REVIEWS

Table 1 | Structure and function of protein rings involved in DNA metabolism Protein rings

Examples

Activity on DNA

Mode of action

Sliding clamps

E. coli β, S. cerevisiae and human PCNA, bacteriophage T4 gp45.

Increase the processivity of replicative DNA polymerases to several thousand nucleotides synthesized per DNA binding event.

The protein ring encircles DNA, binds DNA polymerase and diffuses passively along the duplex, acting as a mobile tether for the polymerase as it duplicates DNA.

Hexameric helicases

SV40 T antigen, bacteriophage T7 gp4, E. coli DnaB, human MCM complex.

Unwind dsDNA to generate ssDNA substrates for DNA metabolic processes such as replication, repair and recombination.

The protein ring encircles ss- (in some cases ds-) DNA, and separates the two strands of the duplex fuelled by energy from NTP binding and hydrolysis.

DNA recombination proteins

Human RAD51, RAD52, DMC1, bacteriophage λ β protein.

Facilitate homologous DNA recombination.

Present ss- or dsDNA in conformations that favour ssDNA annealing or strand exchange with dsDNA, alone or in complex with other recombination proteins.

Change the superhelicity of DNA and catenate/decatenate duplex DNA molecules.

Cleave one (ss or ds) DNA strand, pass a second strand through the break, and re-ligate the broken DNA. Activity of type II enzymes is fuelled by ATP.

Topoisomerases E. coli topoisomerase I, human topoisomerase I, S. cerevisiae topoisomerase II. Others

Bacteriophage λ exonuclease. 3′–5′ DNA exonuclease. Bacteriophage φ29 DNA transport protein. head–tail connector. Human translin. Breakpoint recognition protein.

are active in genomic DNA replication, which requires exceptionally processive helicase and DNA polymerase activity for efficient synthesis of long chromosomal DNA. In contrast, the Rep (and UvrD) and PcrA helicases function mainly in DNA repair and replication of small phage or plasmids in bacteria100, which require unwinding of relatively short lengths of DNA. Similarly, multimeric circular sliding clamps may confer higher processivity on replicative DNA polymerases than monomeric processivity factors that do not encircle DNA, such as UL42 (an accessory protein for the herpes simplex virus DNA polymerase101), the small subunit of mitochondrial polymerase γ102 or thioredoxin (an accessory protein for T7 polymerase)103. Interestingly, the polymerase γ accessory subunit has almost no effect on stability of the interaction between polymerase γ and DNA. Instead it increases polymerase processivity by increasing the rate of DNA polymerization104. So the monomeric processivity factors may differ from circular sliding-clamp proteins in their mechanism of action, which could explain the observed structural differences. The λ exonuclease, which can hydrolyse about 3,000 nucleotides per DNA-binding event, is another example of how the toroidal shape allows a protein to act with high processivity (and speed) on DNA95. Other rings, such as the β-recombination protein of bacteriophage λ50, and TRAP74, highlight how the ordered assembly of repeating nucleic-acid-binding sites (as in a ring) can be used to arrange DNA (or RNA) into specific conformations required for activity. In other proteins, such as DNA topoisomerases, the ring provides a secluded chamber where intermediates can be trapped until the reaction is complete. This property seems particularly important for type II topoisomerases, which must manipulate two dsDNA strands without letting go of either until strand transfer is complete (FIG. 4c)65.

Encircles DNA and processively degrades one strand. Pumps DNA into the phage head using ATP. Binds (encircles?) duplex DNA ends.

A ringing endorsement

The long list of ring-shaped proteins illustrates the usefulness of this shape in a variety of DNA-metabolic processes (TABLE 1). Indeed, a single toroidal protein can have different functions simply by virtue of its shape. For example, the bacteriophage T4 clamp gp45 (FIG. 1), originally discovered as a polymerase processivity factor, is also a mobile scaffold on DNA for the transcriptional co-activator T4 gp33 and the sigma factor T4 gp55, facilitating their interaction with RNA polymerase at late-gene promoters105. Similarly, the eukaryotic PCNA clamp is an accessory protein not only for replicative DNA polymerases, but also for DNA ligase I, DNA (cytosine-5) methyl transferase, Fen1 and XPG endonucleases, and several other proteins that work on DNA6. These proteins seem to have acquired (and retained during evolution) the ability to bind circular clamps, whose DNA-tracking ability may enhance their activity by allowing them to rapidly target their sites of action on DNA. The widespread use of the toroidal shape is highlighted further in a phylogenetic study by Koonin and colleagues106 describing the common roots between the RecA/Rad51/DMC1 family of ATP-dependent recombinases (FIG. 3) and the DnaB family of hexameric helicases (FIG. 2). It has been known for some time now that proteins such as RecA, F1-ATPase, the rho transcription terminator, and the PcrA and T7 helicases have a common core domain responsible for NTP binding/hydrolysis, and related conformational changes that drive their activity (REF. 107 and references therein). There are further similarities in the oligomerization domains of RecA and hexameric helicases24, and Koonin and colleagues propose that an ancestral RecA protein was recruited for helicase function as the DNA replication system evolved in bacteria. Given the advantages of forming a topological link with DNA, it is not surprising that a ring-formwww.nature.com/reviews/molcellbio

| OCTOBER 2000 | VOLUME 1

REVIEWS ing, DNA-binding, RecA-like protein was a useful progenitor for hexameric helicases. These proteins have retained their core NTPase domain and the ability to form oligomeric rings, despite having diverged in sequence and secondary structure. Links DATABASE LINKS Bloom’s syndrome | BLM helicase | DMC1 | RAD52 | translin | Cdc6 |

1. 2. 3.

5. 6. 7. 8.

10.

11.

12.

13.

14.

15.

16.

17.

18. 19.

Boyer, P. D. The ATP synthase — a splendid molecular machine. Annu. Rev. Biochem. 66, 717–749 (1997). Sigler, P. B. et al. Structure and function in GroEL-mediated protein folding. Annu. Rev. Biochem. 67, 581–608 (1998). Bochtler, M., Ditzel, L., Groll, M., Hartmann, C. & Huber, R. The proteasome. Annu. Rev. Biophys. Biomol. Struct. 28, 295–317 (1999). Isaacs, N. W., Cogdell, R. J., Freer, A. A. & Prince, S. M. Light-harvesting mechanisms in purple photosynthetic bacteria. Curr. Opin. Struct. Biol. 5, 794–797 (1995). Hingorani, M. M. & O’Donnell, M. Toroidal proteins: running rings around DNA. Curr. Biol. 8, R83–R86 (1998). Hingorani, M. M. & O’Donnell, M. Sliding clamps: a (tail)ored fit. Curr. Biol. 10, R25–R29 (2000). Baker, T. A. & Bell, S. P. Polymerases and the replisome: machines within machines. Cell 92, 295–305 (1998). Fay, P. J., Johanson, K. O., McHenry, C. S. & Bambara, R. A. Size classes of products synthesized processively by DNA polymerase III and DNA polymerase III holoenzyme of Escherichia coli. J. Biol. Chem. 256, 976–983 (1981). Tsurimoto, T. & Stillman, B. Functions of replication factor C and proliferating-cell nuclear antigen: functional similarity of DNA polymerase accessory proteins from human cells and bacteriophage T4. Proc. Natl Acad. Sci. USA 87, 1023–1027 (1990). Jarvis, T. C., Newport, J. W. & von Hippel, P. H. Stimulation of the processivity of the DNA polymerase of bacteriophage T4 by the polymerase accessory proteins. The role of ATP hydrolysis. J. Biol. Chem. 266, 1830–1840 (1991). Stukenberg, P. T., Studwell-Vaughan, P. S. & O’Donnell, M. Mechanism of the sliding beta-clamp of DNA polymerase III holoenzyme. J. Biol. Chem. 266, 11328–11334 (1991). This study predicted that the β clamp encircles DNA before the solution of a crystal structure. This prediction was made on the basis of the observation that β could fall off linear DNA, but not circular DNA. Kong, X. P., Onrust, R., O’Donnell, M. & Kuriyan, J. Threedimensional structure of the beta subunit of E. coli DNA polymerase III holoenzyme: a sliding DNA clamp. Cell 69, 425–437 (1992). The first crystal structure of a sliding clamp protein clarified greatly the mechanism by which clamps act as processivity factors for polymerase on DNA. Yao, N. et al. Clamp loading, unloading and intrinsic stability of the PCNA, beta and gp45 sliding clamps of human, E. coli and T4 replicases. Genes Cells 1, 101–113 (1996). Moarefi, I., Jeruzalmi, D., Turner, J., O’Donnell, M. & Kuriyan, J. Crystal structure of the DNA polymerase processivity factor of T4 bacteriophage. J. Mol. Biol. 296, 1215–1223 (2000). Shamoo, Y. & Steitz, T. A. Building a replisome from interacting pieces: sliding clamp complexed to a peptide from DNA polymerase and a polymerase editing complex. Cell 99, 155–166 (1999). Structure of a sliding clamp in complex with an interacting peptide from the DNA polymerase; this structure represents the first step towards determining the structure of a DNA replisome. Krishna, T. S., Kong, X. P., Gary, S., Burgers, P. M. & Kuriyan, J. Crystal structure of the eukaryotic DNA polymerase processivity factor PCNA. Cell 79, 1233–1243 (1994). Gulbis, J. M., Kelman, Z., Hurwitz, J., O’Donnell, M. & Kuriyan, J. Structure of the C-terminal region of p21(WAF1/CIP1) complexed with human PCNA. Cell 87, 297–306 (1996). Patel, S. S. & Picha, K. M. Structure and function of hexameric helicases. Annu. Rev. Biochem. (in the press). Lohman, T. M. & Bjornson, K. P. Mechanisms of helicase-

We have described a variety of proteins that work on DNA and share an overall toroidal shape. The fact that many of these proteins are evolutionarily unrelated, and have highly specialized structures and functions, makes their convergence to a common shape quite striking. Moreover, persistence of this shape through evolution in proteins with common roots, but divergent functions, highlights its advantages in protein transactions with DNA, and its effect on nucleic-acid metabolism. As more protein structures come under increasingly high-resolution scrutiny, we expect the list of toroidal proteins to expand and diversify even further.

catalyzed DNA unwinding. Annu. Rev. Biochem. 65, 169–214 (1996). 20. Karow, J. K., Wu, L. & Hickson, I. D. RecQ family helicases: roles in cancer and aging. Curr. Opin. Genet. Dev. 10, 32–38 (2000). 21. Reha-Krantz, L. J. & Hurwitz, J. The dnaB gene product of Escherichia coli. I. Purification, homogeneity, and physical properties. J. Biol. Chem. 253, 4043–4050 (1978). 22. Finger, L. R. & Richardson, J. P. Stabilization of the hexameric form of Escherichia coli. protein rho under ATP hydrolysis conditions. J. Mol. Biol. 156, 203–219 (1982). 23. Egelman, H. H., Yu, X., Wild, R., Hingorani, M. M. & Patel, S. S. Bacteriophage T7 helicase/primase proteins form rings around single-stranded DNA that suggest a general structure for hexameric helicases. Proc. Natl Acad. Sci. USA 92, 3869–3873 (1995). The electron-microscope (EM) structure of a replicative hexameric DNA helicase provided insights into how these enzymes form a ring that encircles single-stranded DNA and catalyse unwinding of duplex DNA. 24. Sawaya, M. R., Guo, S., Tabor, S., Richardson, C. C. & Ellenberger, T. Crystal structure of the helicase domain from the replicative helicase-primase of bacteriophage T7. Cell 99, 167–177 (1999). The first crystal structure of a hexameric DNA helicase. 25. Dong, F., Gogol, E. P. & von Hippel, P. H. The phage T4coded DNA replication helicase (gp41) forms a hexamer upon activation by nucleoside triphosphate. J. Biol. Chem. 270, 7462–7473 (1995). 26. Barcena, M. et al. Polymorphic quaternary organization of the Bacillus subtilis bacteriophage SPP1 replicative helicase (G40 P). J. Mol. Biol. 283, 809–819 (1998). 27. San Martin, M. C., Gruss, C. & Carazo, J. M. Six molecules of SV40 large T antigen assemble in a propeller-shaped particle around a channel. J. Mol. Biol. 268, 15–20 (1997). 28. Yu, X., Jezewska, M. J., Bujalowski, W. & Egelman, E. H. The hexameric E. coli DnaB helicase can exist in different quaternary states. J. Mol. Biol. 259, 7–14 (1996). 29. Yu, X., Horiguchi, T., Shigesada, K. & Egelman, E. H. Threedimensional reconstruction of transcription termination factor rho: Orientation of the N-terminal domain and visualization of an RNA-binding Site. J. Mol. Biol. 299, 1299–1307 (2000). 30. Stasiak, A. et al. The Escherichia coli RuvB branch migration protein forms double hexameric rings around DNA. Proc. Natl Acad. Sci. USA 91, 7618–7622 (1994). 31. Sato, M. et al. Electron microscopic observation and singlestranded DNA binding activity of the Mcm4, 6, 7 complex. J. Mol. Biol. 300, 421–431 (2000). A first view of the quaternary structure of the human MCM helicase that is required for initiation of DNA replication. 32. Karow, J. K., Newman, R. H., Freemont, P. S. & Hickson, I. D. Oligomeric ring structure of the Bloom’s syndrome helicase. Curr. Biol. 9, 597–600 (1999). 33. Chong, J. P., Hayashi, M. K., Simon, M. N., Xu, R. M. & Stillman, B. A double-hexamer archaeal minichromosome maintenance protein is an ATP-dependent DNA helicase. Proc. Natl Acad. Sci. USA 97, 1530–1535 (2000). 34. Yu, X., Hingorani, M. M., Patel, S. S. & Egelman, E. H. DNA is bound within the central hole to one or two of the six subunits of the T7 DNA helicase. Nature Struct. Biol. 3, 740–743 (1996). 35. Morris, P. D. & Raney, K. D. DNA helicases displace streptavidin from biotin-labeled oligonucleotides. Biochemistry 38, 5164–5171 (1999). 36. Bujalowski, W. & Jezewska, M. J. Interactions of

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

37.

38.

39.

40.

41.

42.

43.

44.

45.

46.

47.

48.

49. 50.

51.

52.

53.

Escherichia coli primary replicative helicase DnaB protein with single-stranded DNA. The nucleic acid does not wrap around the protein hexamer. Biochemistry 34, 8513–8519 (1995). Fouts, E. T., Yu, X., Egelman, E. H. & Botchan, M. R. Biochemical and electron microscopic image analysis of the hexameric E1 helicase. J. Biol. Chem. 274, 4447–4458 (1999). Mastrangelo, I. A. et al. ATP-dependent assembly of double hexamers of SV40 T antigen at the viral origin of DNA replication. Nature 338, 658–662 (1989). Yong, Y. & Romano, L. J. Benzo[a]pyrene-DNA adducts inhibit the DNA helicase activity of the bacteriophage T7 gene 4 protein. Chem. Res. Toxicol. 9, 179–187 (1996). Hacker, K. J. & Johnson, K. A. A hexameric helicase encircles one DNA strand and excludes the other during DNA unwinding. Biochemistry 36, 14080–14087 (1997). Raney, K. D., Carver, T. E. & Benkovic, S. J. Stoichiometry and DNA unwinding by the bacteriophage T4 41:59 helicase. J. Biol. Chem. 271, 14074–14081 (1996). Nakai, H. & Richardson, C. C. Interactions of the DNA polymerase and gene 4 protein of bacteriophage T7. Protein–protein and protein–DNA interactions involved in RNA-primed DNA synthesis. J. Biol. Chem. 261, 15208–15216 (1986). Kim, S., Dallmann, H. G., McHenry, C. S. & Marians, K. J. Coupling of a replicative polymerase and helicase: a tauDnaB interaction mediates rapid replication fork movement. Cell 84, 643–650 (1996). Yuzhakov, A., Turner, J. & O’Donnell, M. Replisome assembly reveals the basis for asymmetric function in leading and lagging strand replication. Cell 86, 877–886 (1996). Hingorani, M. M. & Patel, S. S. Interactions of bacteriophage T7 DNA primase/helicase protein with single-stranded and double-stranded DNAs. Biochemistry 32, 12478–12487 (1993). Jezewska, M. J., Kim, U. S. & Bujalowski, W. Binding of Escherichia coli primary replicative helicase DnaB protein to single-stranded DNA. Long-range allosteric conformational changes within the protein hexamer. Biochemistry 35, 2129–2145 (1996). Hingorani, M. M., Washington, M. T., Moore, K. C. & Patel, S. S. The dTTPase mechanism of T7 DNA helicase resembles the binding change mechanism of the F1ATPase. Proc. Natl Acad. Sci. USA 94, 5012–5017 (1997). Rapid kinetic analysis of the NTPase activity of a hexameric helicase yielded a model mechanism for helicase-catalysed sequential TTP hydrolysis (similar to F1-ATPase) that may be coupled to stepwise DNA unwinding. Stitt, B. L. & Xu, Y. Sequential hydrolysis of ATP molecules bound in interacting catalytic sites of Escherichia coli transcription termination protein Rho. J. Biol. Chem. 273, 26477–26486 (1998). Kovall, R. & Matthews, B. W. Toroidal structure of lambdaexonuclease. Science 277, 1824–1827 (1997). Passy, S. I., Yu, X., Li, Z., Radding, C. M. & Egelman, E. H. Rings and filaments of beta protein from bacteriophage lambda suggest a superfamily of recombination proteins. Proc. Natl Acad. Sci. USA 96, 4279–4284 (1999). Mythili, E., Kumar, K. A. & Muniyappa, K. Characterization of the DNA-binding domain of beta protein, a component of phage lambda red-pathway, by UV catalyzed crosslinking. Gene 182, 81–87 (1996). Poteete, A. R., Sauer, R. T. & Hendrix, R. W. Domain structure and quaternary organization of the bacteriophage P22 Erf protein. J. Mol. Biol. 171, 401–418 (1983). Thresher, R. J., Makhov, A. M., Hall, S. D., Kolodner, R. &

VOLUME 1 | OCTOBER 2000 | 2 9

REVIEWS

54.

55.

56.

57.

58.

59.

60. 61.

62.

63.

64.

65. 66.

67.

68.

69.

70. 71.

72.

73.

Griffith, J. D. Electron microscopic visualization of RecT protein and its complexes with DNA. J. Mol. Biol. 254, 364–371 (1995). Ando, R. A. & Morrical, S. W. Single-stranded DNA binding properties of the UvsX recombinase of bacteriophage T4: binding parameters and effects of nucleotides. J. Mol. Biol. 283, 785–796 (1998). Shinohara, A., Ogawa, H. & Ogawa, T. Rad51 protein involved in repair and recombination in S. cerevisiae is a RecA-like protein. Cell 69, 457–470 (1992); erratum in Cell 71, 180 (1992). Bianco, P. R., Tracy, R. B. & Kowalczykowski, S. C. DNA strand exchange proteins: a biochemical and physical comparison. Front. Biosci. 3, D570–D603 (1998). Yu, X. & Egelman, E. H. The RecA hexamer is a structural homologue of ring helicases. Nature Struct. Biol. 4, 101–104 (1997). Baumann, P., Benson, F. E., Hajibagheri, N. & West, S. C. Purification of human Rad51 protein by selective spermidine precipitation. Mutat. Res. 384, 65–72 (1997). Passy, S. I. et al. Human Dmc1 protein binds DNA as an octameric ring. Proc. Natl Acad. Sci. USA 96, 10684–10688 (1999). EM structure of this ring-shaped human RecA homologue opens up new possible mechanisms of action of RecA-like proteins in homologous recombination or other metabolic processes. Stasiak, A. Z. et al. The human Rad52 protein exists as a heptameric ring. Curr. Biol. 10, 337–340 (2000). Rattray, A. J. & Symington, L. S. Multiple pathways for homologous recombination in Saccharomyces cerevisiae. Genetics 139, 45–56 (1995). Baumann, P. & West, S. C. Heteroduplex formation by human Rad51 protein: effects of DNA end-structure, hRPA and hRad52. J. Mol. Biol. 291, 363–374 (1999). Sung, P. Function of yeast Rad52 protein as a mediator between replication protein A and the Rad51 recombinase. J. Biol. Chem. 272, 28194–28197 (1997). New, J. H., Sugiyama, T., Zaitseva, E. & Kowalczykowski, S. C. Rad52 protein stimulates DNA strand exchange by Rad51 and replication protein A. Nature 391, 407–410 (1998). Wang, J. C. DNA topoisomerases. Annu. Rev. Biochem. 65, 635–692 (1996). Lima, C. D., Wang, J. C. & Mondragon, A. Threedimensional structure of the 67K N-terminal fragment of E. coli DNA topoisomerase I. Nature 367, 138–146 (1994). Mondragon, A. & DiGate, R. The structure of Escherichia coli DNA topoisomerase III. Structure Fold. Des. 7, 1373–1383 (1999). Redinbo, M. R., Stewart, L., Kuhn, P., Champoux, J. J. & Hol, W. G. Crystal structures of human topoisomerase I in covalent and noncovalent complexes with DNA. Science 279, 1504–1513 (1998). Stewart, L., Redinbo, M. R., Qiu, X., Hol, W. G. & Champoux, J. J. A model for the mechanism of human topoisomerase I. Science 279, 1534–1541 (1998). The series of crystal structures of human topoisomerase I in references 68 and 69 provide detailed snapshots of the process by which this enzyme catalyses changes in the topology of DNA. Berger, J. M. Type II DNA topoisomerases. Curr. Opin. Struct. Biol. 8, 26–32 (1998). Baird, C. L., Harkins, T. T., Morris, S. K. & Lindsley, J. E. Topoisomerase II drives DNA transport by hydrolyzing one ATP. Proc. Natl Acad. Sci. USA 96, 13685–13690 (1999). A rapid kinetic study of the topoisomerase II reaction yields a model mechanism by which this enzyme couples ATP binding and hydrolysis to catenation or decatenation of duplex DNA. Berger, J. M., Gamblin, S. J., Harrison, S. C. & Wang, J. C. Structure and mechanism of DNA topoisomerase II. Nature 379, 225–232 (1996). Lee, J. Y. et al. Crystal structure of NAD+-dependent DNA ligase: modular architecture and functional implications.

EMBO J. 19, 1119–1129 (2000). 74. Antson, A. A. et al. Structure of the trp RNA-binding attenuation protein, TRAP, bound to RNA. Nature 401, 235–242 (1999). 75. Valpuesta, J. M., Fernandez, J. J., Carazo, J. M. & Carrascosa, J. L. The three-dimensional structure of a DNA translocating machine at 10 Å resolution. Structure Fold. Des. 7, 289–296 (1999). 76. Kasai, M. et al. The translin ring specifically recognizes DNA ends at recombination hot spots in the human genome. J. Biol. Chem. 272, 11402–11407 (1997). 77. Valle, M., Valpuesta, J. M., Carrascosa, J. L., Tamayo, J. & Garcia, R. The interaction of DNA with bacteriophage phi 29 connector: a study by AFM and TEM. J. Struct. Biol. 116, 390–398 (1996). 78. Turnquist, S., Simon, M., Egelman, E. & Anderson, D. Supercoiled DNA wraps around the bacteriophage phi 29 head–tail connector. Proc. Natl Acad. Sci. USA 89, 10479–10483 (1992). 79. West, S. C. Processing of recombination intermediates by the RuvABC proteins. Annu. Rev. Genet. 31, 213–244 (1997). 80. Aoki, K., Suzuki, K., Ishida, R. & Kasai, M. The DNA binding activity of Translin is mediated by a basic region in the ring-shaped structure conserved in evolution. FEBS Lett. 443, 363–366 (1999). 81. Redinbo, M. R., Stewart, L., Champoux, J. J. & Hol, W. G. Structural flexibility in human topoisomerase I revealed in multiple non-isomorphous crystal structures. J. Mol. Biol. 292, 685–696 (1999). 82. Turner, J., Hingorani, M. M., Kelman, Z. & O’Donnell, M. The internal workings of a DNA polymerase clamp-loading machine. EMBO J. 18, 771–783 (1999). 83. Kaboord, B. F. & Benkovic, S. J. Accessory proteins function as matchmakers in the assembly of the T4 DNA polymerase holoenzyme. Curr. Biol. 5, 149–157 (1995). 84. Mossi, R. & Hubscher, U. Clamping down on clamps and clamp loaders — the eukaryotic replication factor C. Eur. J. Biochem. 254, 209–216 (1998). 85. Hingorani, M. M. & O’Donnell, M. ATP binding to the Escherichia coli clamp loader powers opening of the ringshaped clamp of DNA polymerase III holoenzyme. J. Biol. Chem. 273, 24550–24563 (1998). 86. Naktinis, V., Onrust, R., Fang, L. & O’Donnell, M. Assembly of a chromosomal replication machine: two DNA polymerases, a clamp loader, and sliding clamps in one holoenzyme particle. II. Intermediate complex between the clamp loader and its clamp. J. Biol. Chem. 270, 13358–13365 (1995). 87. Hingorani, M. M., Bloom, L. B., Goodman, M. F. & O’Donnell, M. Division of labor-sequential ATP hydrolysis drives assembly of a DNA polymerase sliding clamp around DNA. EMBO J. 18, 5131–5144 (1999). 88. Ahnert, P., Moore Picha, K. & Patel, S. S. A ring-opening mechanism for single-stranded DNA binding in the central channel of T7 helicase–primase protein. EMBO J. 19, 3418–3427 (2000). 89. Tarumi, K. & Yonesaki, T. Functional interactions of gene 32, 41, and 59 proteins of bacteriophage T4. J. Biol. Chem. 270, 2614–2619 (1995). 90. Allen, G. C. Jr & Kornberg, A. Fine balance in the regulation of DnaB helicase by DnaC protein in replication in Escherichia coli. J. Biol. Chem. 266, 22096–22101 (1991). 91. Perkins, G. & Diffley, J. F. Nucleotide-dependent prereplicative complex assembly by Cdc6p, a homolog of eukaryotic and prokaryotic clamp-loaders. Mol. Cell 2, 23–32 (1998). 92. Herbig, U., Marlar, C. A. & Fanning, E. The Cdc6 nucleotidebinding site regulates its activity in DNA replication in human cells. Mol. Biol. Cell 10, 2631–2645 (1999). 93. Morrical, S. W., Hempstead, K. & Morrical, M. D. The gene 59 protein of bacteriophage T4 modulates the intrinsic and single-stranded DNA-stimulated ATPase activities of

94.

95.

96.

97. 98.

99.

100.

101.

102.

103.

104.

105.

106.

107.

gene 41 protein, the T4 replicative DNA helicase. J. Biol. Chem. 269, 33069–33081 (1994). Learn, B. A., Um, S. J., Huang, L. & McMacken, R. Cryptic single-stranded-DNA binding activities of the phage lambda P and Escherichia coli DnaC replication initiation proteins facilitate the transfer of E. coli DnaB helicase onto DNA. Proc. Natl Acad. Sci. USA 94, 1154–1159 (1997). Carter, D. M. & Radding, C. M. The role of exonuclease and beta protein of phage lambda in genetic recombination. II. Substrate specificity and the mode of action of lambda exonuclease. J. Biol. Chem. 246, 2502–2512 (1971). Fang, L., Davey, M. J. & O’Donnell, M. Replisome assembly at oriC, the replication origin of E. coli, reveals an explanation for initiation sites outside an origin. Mol. Cell 4, 541–553 (1999). Dutta, A. & Bell, S. P. Initiation of DNA replication in eukaryotic cells. Annu. Rev. Cell Biol. 13, 293–332 (1997). Mechanic, L. E., Hall, M. C. & Matson, S. W. Escherichia coli DNA helicase II is active as a monomer. J. Biol. Chem. 274, 12488–12498 (1999). Velankar, S. S., Soultanas, P., Dillingham, M. S., Subramanya, H. S. & Wigley, D. B. Crystal structures of complexes of PcrA DNA helicase with a DNA substrate indicate an inchworm mechanism. Cell 97, 75–84 (1999). Petit, M. A. et al. PcrA is an essential DNA helicase of Bacillus subtilis fulfilling functions both in repair and rollingcircle replication. Mol. Microbiol. 29, 261–273 (1998). Zuccola, H. J., Filman, D. J., Coen, D. M. & Hogle, J. M. The crystal structure of an unusual processivity factor, herpes simplex virus UL42, bound to the C terminus of its cognate polymerase. Mol. Cell 5, 267–278 (2000). Fan, L., Sanschagrin, P. C., Kaguni, L. S. & Kuhn, L. A. The accessory subunit of mtDNA polymerase shares structural homology with aminoacyl-tRNA synthetases: implications for a dual role as a primer recognition factor and processivity clamp. Proc. Natl Acad. Sci. USA 96, 9527–9532 (1999). Doublie, S. & Ellenberger, T. The mechanism of action of T7 DNA polymerase. Curr. Opin. Struct. Biol. 8, 704–712 (1998). Johnson, A. A., Tsai, Y., Graves, S. W. & Johnson, K. A. Human mitochondrial DNA polymerase holoenzyme: reconstitution and characterization. Biochemistry 39, 1702–1708 (2000). Sanders, G. M., Kassavetis, G. A. & Geiduschek, E. P. Dual targets of a transcriptional activator that tracks on DNA. EMBO J. 16, 3124–3132 (1997). Leipe, D. D., Aravind, L., Grishin, N. V. & Koonin, E. V. The bacterial replicative helicase DnaB evolved from a RecA duplication. Genome Res. 10, 5–16 (2000). A comprehensive phylogenetic and structural analysis of RecA-like proteins and the DnaB family of hexameric helicases indicates that these proteins share a common ancestor that had the ability to form oligomeric rings and undergo NTP-driven conformational changes. Egelman, E. A ubiquitous structural core. Trends Biochem. Sci. 25, 183–184 (2000).

Acknowledgements We thank all researchers who provided images, including J. Kuriyan and D. Jeruzalmi (E. coli β clamp, yPCNA, T4 gp45); Y. Shamoo and T. Steitz (RB69 clamp and polymerase); T. Ellenberger (T7 gp4); E. Gogol (T4 gp41); E. Egelman (E. coli DnaB, RuvB, rho, BPV E1, phage λ β protein, hDmc1, hRad52); C. San Martin (SV40 T antigen); I. Hickson (BLM helicase); J. Chong and B. Stillman (M. th MCM); Y. Ishimi (human MCM); P. Ahnert (T7 gp4 on DNA); R. Kovall and B. Matthews (phage λ exonuclease); M. Redinbo and W. Hol (human topoI); A. Mondragón (E. coli topoI and topoIII); J. Wang (yeast topoII); J. Y. Lee and S. W. Suh (DNA ligase); A. Anston (TRAP); J. Carrascosa (φ29 connector) and M. Kasai (human translin). We also thank M. Davey and K. Picha for discussions. Work supported by a grant from the NIH to M.O.

www.nature.com/reviews/molcellbio

| OCTOBER 2000 | VOLUME 1

REVIEWS

LIPID RAFTS AND SIGNAL TRANSDUCTION Kai Simons*‡ and Derek Toomre‡ Signal transduction is initiated by complex protein–protein interactions between ligands, receptors and kinases, to name only a few. It is now becoming clear that lipid microenvironments on the cell surface — known as lipid rafts — also take part in this process. Lipid rafts containing a given set of proteins can change their size and composition in response to intra- or extracellular stimuli. This favours specific protein–protein interactions, resulting in the activation of signalling cascades. EXOPLASMIC LEAFLET

Lipid layer facing the extracellular space.

*Max Planck Institute for Molecular Cell Biology and Genetics, Pfotenhauerstrasse 110, D01307 Dresden, Germany. ‡Cell Biology/Biophysics Programme, European Molecular Biology Laboratory (EMBL), Meyerhofstrasse 1, D-69117 Heidelberg, Germany. Correspondence to K.S. e-mail: Simons@EMBLHeidelberg.de

The Singer–Nicholson fluid mosiac concept is still the textbook model of how the cell membrane is organized. It proposes that the lipid bilayer functions as a neutral two-dimensional solvent, having little influence on membrane protein function. But biophysicists find that lipids exist in several phases in model lipid bilayers, including gel, liquid-ordered and liquid-disordered states, in order of increasing fluidity1. In the gel state lipids are semi-frozen, whereas at the other extreme, the liquid-disordered state, the whole lipid bilayer is fluid, as proposed by the Singer–Nicholson model. In the liquidordered phase, phospholipids with saturated hydrocarbon chains pack tightly with cholesterol (BOX 1) but nevertheless remain mobile in the plane of the membrane2. Despite a detailed biophysical characterization of model membranes, it has been difficult to show that lipids exist in these different phases in the complex environment of the cell. Lipid rafts

A turning point came when the lipid raft hypothesis was formulated more than ten years ago1,3,4. It originated from studies on epithelial cell polarity, and its central postulate was the existence of lipid rafts, consisting of dynamic assemblies of cholesterol and sphingolipids (BOX 1), in the EXOPLASMIC LEAFLET of the bilayer. The preponderance of saturated hydrocarbon chains in cell sphingolipids allows for cholesterol to be tightly intercalated, similar to the organization of the liquid-ordered state in model membranes. The inner leaflet is probably rich in phospholipids with saturated fatty acids and cho-

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

lesterol5, but its characterization is still incomplete. It is also not clear how the inner leaflet is coupled to the outer leaflet. One possibility is that long fatty acids of sphingolipids in the outer leaflet couple the exoplasmic and cytoplasmic leaflets by interdigitation. Transmembrane proteins could also stabilize this coupling. The membrane surrounding lipid rafts is more fluid, as it consists mostly of phospholipids with unsaturated, and therefore kinked, fatty acyl chains and cholesterol. In other words, lipid rafts form distinct liquid-ordered phases in the lipid bilayer, dispersed in a liquid-disordered matrix of unsaturated glycerolipids1,6. The raft concept has long been controversial, largely because it has been difficult to prove definitively that rafts exist in living cells. But recent studies with improved methodology have dispelled most of these doubts (BOX 2). One of the most important properties of lipid rafts is that they can include or exclude proteins to variable extents. Proteins with raft affinity include glycosylphosphatidylinositol (GPI)-anchored proteins1,7, doubly acylated proteins, such as Src-family kinases or the α-subunits of heterotrimeric G proteins8, cholesterol-linked and palmitoylated proteins such as Hedgehog9, and transmembrane proteins, particularly palmitoylated ones1. GPI-anchored proteins or proteins that carry hydrophobic modifications probably partition into rafts owing to preferential packing of their saturated membrane anchors. It is not yet clear why some transmembrane proteins are included into rafts, but mutational analysis has shown that amino acids in the transmembrane domains near the exoplasmic leaflet are critical10.

VOLUME 1 | OCTOBER 2000 | 3 1

REVIEWS

Box 1 | Glossary of lipid structures Phospholipids

Cholesterol

Sphingolipids H3C

O R

CH2

(CH2)12

H C

C H

H C HO

H C

CH2

R''

NH R

O HO

H2C

R''

O– R, R', Hydrocarbon chains of fatty acids R'', Head-group

TRANSCYTOSIS

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

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

The surface of an epithelial cell that adjoins underlying tissue. SOMATODENDRITIC MEMBRANE

The surface of a neuron that surrounds the cell body and dendrites.

Palmitoylation can increase a protein’s affinity for rafts, but it is not sufficient for raft association11. It is likely that a given protein can associate with rafts with different kinetics or partition coefficients. For instance, a monomeric transmembrane protein may have a short residency time in rafts, spending most of its time outside rafts. But when the same protein is crosslinked or otherwise oligomerized, its affinity for rafts increases12. As we will discuss below, a common theme is that the clustering of separate rafts exposes proteins to a new membrane environment, enriched in specific enzymes, such as kinases, phosphatases and perhaps palmitoylases and depalmitoylases. Even a small change of partitioning into a lipid raft can, through amplification, initiate signalling cascades. These dynamic features of initial raft association have so far received little attention, but we predict that they are crucial for the activation of many signal transduction pathways.

BIOSYNTHETIC 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. ENDOCYTIC PATHWAY

Macromolecules are endocytosed at the plasma membrane. They first arrive in early endosomes, then late endosomes, and finally lysosomes where they are degraded by hydrolases. Recycling back to the plasma membrane from early endosomes also occurs.

Caveolae

One subset of lipid rafts is found in cell surface invaginations called caveolae (BOX 3). These flask-shaped plasma membrane invaginations were first identified on the basis of their morphology by Palade13 and Yamada14 in the 1950s. Caveolae are formed from lipid rafts by polymerization of caveolins — hairpin-like palmitoylated integral membrane proteins that tightly bind cholesterol15,16. The general function of caveolae is not clear. They have been implicated in endocytosis15 and TRANSCYTOSIS of albumin and other proteins across the endothelial monolayer17. In developing myocytes, ribbons of many caveolae organize into T-tubules18, which are required for the calcium regulation of muscle contraction. Caveolae also function during signal transduction19, but they are not absolutely required as several

Box 2 | Key papers on the existence of rafts in cell membranes • Fluorescence resonance energy transfer measurements using fluorescent folate to show interactions of folate receptors when they are in proximity in rafts in living cells31. • Biochemical crosslinking of GPI-anchored proteins when they are in proximity in rafts32. • Antibody crosslinking of raft proteins into patches segregating from non-raft proteins12. • Photonic force microscopy measurements of the size of rafts in living cells30. • Visualization of rafts and clustered rafts in IgE signalling by electron microscopy81.

| OCTOBER 2000 | VOLUME 1

cell types that lack caveolin, such as lymphocytes and neurons, can nevertheless signal through rafts. Raft distribution and trafficking

The distribution of lipid rafts over the cell surface depends on the cell type. In polarized epithelial cells and neurons, lipid rafts accumulate in the APICAL and axonal plasma membrane, respectively. BASOLATERAL and SOMATODENDRITIC MEMBRANES also contain rafts, but in smaller amounts4. Interestingly, caveolae are present mainly on the basolateral side of epithelial cells20, which faces the blood supply and is more active during signal transduction. In lymphocytes and fibroblasts, rafts are distributed over the cell surface without obvious polarity. We can roughly estimate the fraction of the cell surface covered by rafts by comparing the ratio of the main raft and non-raft exoplasmic leaflet lipids, sphingolipids and phosphatidylcholine, respectively. Typically, sphingolipids make up about 45% of the cell surface in fibroblasts21 and roughly 30% in lymphocytes22, but these values are upper limits and may also be cell-type dependent. Raft lipids are most abundant at the plasma membrane, but can also be found in the BIOSYNTHETIC and ENDOCYTIC PATHWAYS. Whereas cholesterol is synthesized in the endoplasmic reticulum (ER), sphingolipid synthesis and head-group modification are completed largely in the Golgi23. As these data predict, cholesterol–sphingolipid rafts first assemble in the Golgi1. Movement of lipid rafts out of the Golgi seems to be mainly towards the plasma membrane, as vesicles going back to the ER contain little sphingomyelin and cholesterol24. The inclusion of proteins into rafts is important for polarized delivery to the cell surface in many cell types4,25,26. Lipid raft trafficking does not end with surface delivery — rafts are continuously endocytosed27 from the plasma membrane. From early endosomes, rafts either recycle directly back to the cell surface or return indirectly through recycling endosomes, which could also deliver rafts to the Golgi28. Raft size

One reason why it has been so difficult to prove that rafts exist in cells is that they are too small to be resolved by standard light microscopy. But if raft components are crosslinked with antibodies or lectins in living cells, then raft protein and lipid components cluster together, and

www.nature.com/reviews/molcellbio

REVIEWS

Box 3 | Confusion about the relationship between caveolae and rafts The problem has arisen primarily because of the difficulty of isolating pure caveolae. Several methods have been used — the first was simply Triton X-100 extraction at 4 °C (REF. 82). When it became clear that this fraction contains not only plasma membrane caveolae but rafts (detergent-resistant membranes or DRMs) from all cellular membranes83, new methods were devised. The most frequently used of these is based on density gradient centrifugation84 but it is also not capable of isolating pure caveolae35. Immuno-isolation has been used with conflicting results85–87, so that the only safe method now used to identify caveolar proteins is double-label immunoelectron microscopy80, 88. Confusion persists because definitions of caveolae remain vague. In a recent review, Anderson defined caveolae as “meant to encompass a membrane system with specific functions essential for normal cell behavior”19. Lisanti et al. describe rafts, in the absence of caveolae, as “caveola-related domains”16. To resolve these issues, we propose to distinguish between rafts, DRMs and caveolae. We reserve the use of the term “caveolae” for morphologically defined cell surface invaginations (containing caveolin), as originally proposed half a century ago13, 14.

have, for example, reinforced the conclusion that the fatty acids that mediate protein binding to rafts are usually saturated. Indeed, feeding cells with polyunsaturated fatty acids leads to the replacement of saturated fatty acids with unsaturated ones in acylated proteins, causing these proteins to dissociate from rafts36. Similarly, the addition of exogenous GANGLIOSIDEs to cells can lead to their incorporation into rafts and, as a result, also cause proteins to dissociate from rafts37. The lack of standardized methodology has led to confusion in the current nomenclature between rafts, detergent-resistant membranes and caveolae (TABLE 2). In the hope of dissipating some of this confusion, we summarize the advantages and disadvantages of some methods used to study rafts in TABLE 1, and propose a more standardized nomenclature in TABLE 2. Rafts in signal transduction

raft and non-raft components separate into micronsized quilt-like patches12,29. In fibroblasts, raft proteins rapidly diffuse in assemblies of roughly 50 nm diameter30, corresponding to about 3,500 sphingolipid molecules. The number of proteins in each raft depends on the packing density, but is probably not more than 10–30 proteins. We do not yet know whether individual raft proteins are randomly distributed between different rafts, or whether they are grouped in specialized rafts. Clusters of up to 15 molecules of the same protein have been observed within the same raft31,32, supporting the view that some proteins are distributed non-randomly. However, other studies indicate that such clusters may only represent a small population33. Either way, given its small size, a raft can statistically contain only a subset of all available raft proteins. This conclusion may have profound consequences on how signalling through rafts can be dynamically activated by raft clustering, as will be discussed later. Methods to study rafts

SUCROSE GRADIENT CENTRIFUGATION

Allows separation of cellular membranes according to their size and/or density by centrifugation. GANGLIOSIDES

Anionic glycosphingolipids that carry, in addition to other sugar residues, one or more sialic acid residues. MAST CELL

A type of leukocyte, of the granulocyte subclass. BASOPHIL

Polymorphonuclear phagocytic leukocyte of the myeloid series.

The formulation of the raft hypothesis was influenced by the discovery that, on entering the Golgi, some proteins form large complexes with lipids, which resist solubilization by non-ionic detergents34. Detergent-resistant membrane (DRM) complexes float to low density during SUCROSE GRADIENT CENTRIFUGATION and are enriched in raft proteins and lipids, providing a simple means of identifying possible raft components. Despite the ease and usefulness of non-ionic detergent extraction, this method is not without pitfalls7. A raft protein can be connected to the cytoskeleton, so it will not float after detergent extraction. Or its association with rafts can be so weak that it is solubilized by the detergent. Moreover, changes in detergents and extraction conditions can produce strikingly different results7,29,35. One useful approach in raft research has been the manipulation of raft lipid constituents (BOX 4). This treatment leads to the dissociation of proteins from rafts, which can be readily detected by common methods used to analyse raft association (TABLE 1). This type of methodology has greatly contributed to our understanding of raft function in vivo. Such experiments

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

The most important role of rafts at the cell surface may be their function in signal transduction (TABLE 3). It is well established that, in the case of tyrosine kinase signalling, adaptors, scaffolds and enzymes are recruited to the cytoplasmic side of the plasma membrane as a result of ligand activation38. One way to consider rafts is that they form concentrating platforms for individual receptors, activated by ligand binding. If receptor activation takes place in a lipid raft, the signalling complex is protected from non-raft enzymes such as membrane phosphatases that otherwise could affect the signalling process. In general, raft binding recruits proteins to a new micro-environment, where the phosphorylation state can be modified by local kinases and phosphatases, resulting in downstream signalling. To highlight these principles, examples of signalling pathways that involve lipid rafts are described below. Immunoglobulin E signalling. The first signalling process convincingly shown to involve lipid rafts was immunoglobulin E (IgE) signalling during the allergic immune response39–41 (FIG. 1a). This signalling pathway is activated when IgE binds through its Fc segment to receptors (FcεRI) residing in the plasma membrane of MAST CELLS and BASOPHILS. FcεRI is monomeric and binds one IgE molecule. The receptor is activated by the bindBox 4 | Common tools to disrupt rafts Cholesterol sequestration • Antibiotics: Filipin | Nystatin | Amphotericin • Pore-forming agents: Saponin | Digitonin | Streptolysin O Cholesterol depletion • Methyl-β-cyclodextrin Inhibition of cholesterol biosynthesis • Lovastatin Perturbation of raft stability • Exogenous cholesterol • Exogenous gangliosides • Exogenous polyunsaturated fatty acids

VOLUME 1 | OCTOBER 2000 | 3 3

REVIEWS

Table 1 | Techniques to identify rafts Approach*

Information available

Live cells

Comments

References

Flotation of detergent– resistant membranes (DRMs)

Identifies putative raft association Identifies possible raft proteins

• Easy to do • Most common approach for identifying putative proteins involved in signalling • Artefacts possible • Weak associations with rafts are difficult to detect

1, 4, 7

Antibody patching and immunofluorescence microscopy

Identifies putative raft association

• Easy to do • Common approach • Better than flotation for detecting weak raft associations • Cell–cell variability makes quantification difficult

12, 29

Immunoelectron microscopy

Determines location of raft components

• Promising results • Requires technical expertise

80, 81, 88

Chemical crosslinking

Identifies native raft protein complexes

Yes

• Straightforward • Choice of appropriate conditions and reagents is semi-empirical

Single fluorophore tracking microscopy

Monitors the diffusion and dynamics of individual raft proteins or lipids

Yes

• Requires highly specialized equipment and expertise

Photonic force microscopy

Determines the diffusion constant, size and dynamics of individual rafts

Yes

• Very informative technique • Requires highly specialized equipment and technical expertise • Time-consuming acquisition and analysis

Fluorescence resonance energy transfer (FRET)

Detects whether two raft components are spatially close (for example, <10 nm)

Yes

• Powerful approach • Choice of appropriate donor and acceptor probes is important

31, 33

*The disruption of rafts by cholesterol depletion or sequestration is especially useful as a control for each of these approaches.

METHYL-β-CYCLODEXTRIN

Carbohydrate molecule with a pocket for binding cholesterol. MAJOR HISTOCOMPATIBILITY COMPLEX

A complex of genetic loci, occurring in higher vertebrates, encoding a family of cellular antigens that help the immune system to recognize self from non-self. ANTIGEN-PRESENTING CELL

A cell, most often a macrophage or dendritic cell, that presents an antigen to activate a T cell.

ing of oligomeric antigens to receptor-bound IgE. Crosslinking of FcεRI by oligomeric antigens activates the transmembrane signalling process, ultimately leading to release of the chemical mediators of allergic reactions. The Fc receptor is a tetramer composed of one α-, one β- and two γ-chains41. The α-chain binds IgE and the β- and the γ-chains contain immune receptor tyrosinebased activation motifs (ITAMs), common to all multisubunit immune recognition receptors. Crosslinking of two or more of these receptors by antigens recuits the doubly acylated non-receptor Src-like tyrosine kinase Lyn, which is thought to initiate the signalling cascade by phosphorylating ITAMs so that they can bind to Syk/ZAP-70 family tyrosine kinases through their phosphotyrosine residues39,40. Syk is activated by phosphorylation and this, in turn, leads to activation of phospholipase Cγ (PLCγ). Finally, downstream signalling results in increased calcium levels in the proximity of the membrane, and this triggers the release of histamine from nearby granules. IgE signalling was initially thought to be based on protein–protein interactions alone42. But several observations indicate that rafts are involved in this process41. The first hint came from the finding that FcεRI is soluble in Triton X-100 at steady state, but becomes insoluble in low concentrations of this detergent after crosslinking39. Moreover, FcεRI crosslinking causes the redistribution of raft components, including gangliosides and GPI-anchored proteins, to patches that are large enough to be seen by fluorescence microscopy43,44. This observation also indicates that raft clustering takes place following receptor activation. A last indication came from the finding that IgE signalling is abolished if

| OCTOBER 2000 | VOLUME 1

surface cholesterol is depleted with METHYL-β-CYCLODEX45 . One working model for IgE signalling is that crosslinking increases the raft affinity of FcεRI. In this model, a change in receptor partioning would lead to increased phosphorylation of its ITAMs by raft-associated Lyn kinase, possibly due to the exclusion of inhibitory phosphatases. FcεRI crosslinking could, in addition, bring small individual rafts together. Linker proteins, such as members of the BASH (B cell adaptor containing SH repeats) family and LAT (linker for activation of T cells), are good candidates for this job46. As a result of amplification, even small changes in receptor partitioning could produce strong signals. One key issue to be explained is how FcεRI aggregates as small as dimers42 can initiate the raft clustering process to activate the allergic response. TRIN

T-cell antigen receptor signalling. The T-cell antigen receptor (TCR) is another multisubunit immune recognition receptor that engages lipid rafts during signalling47,48 (FIG. 1b). The TCR is composed of αβ-heterodimers which associate with the CD3 (γδε) complex and the ζ-homodimer. Whereas the α- and β-subunits contain the extracellular binding site for peptides that are presented by the MAJOR HISTOCOMPATIBILITY COMPLEX (MHC) class I and II proteins on the surface of ANTIGENPRESENTING CELLS (APCs), the CD3 and ζ-subunits contain cytoplasmic ITAM motifs. The earliest signalling event after TCR engagement is the phosphorylation of ITAM tyrosine residues by the doubly acylated non-receptor Src-like tyrosine kinases, Lyn and Fyn47,48. When ZAP-70 binds to phosphorylated ITAMs it is activated and, in

www.nature.com/reviews/molcellbio

REVIEWS

III. DRMs

Components • Glycosphingolipids • Cholesterol • Lipid-modified proteins containing saturated acyl chains: – GPI-anchored proteins – Doubly acylated Src-type kinases • Transmembrane proteins

I. Rafts

• Rafts clustered by: – Antibody – Lectin – Adjacent cell proteins – Physiological crosslinking proteins

• Rafts remaining insoluble after • Raft proteins and lipids treatment on ice with • Caveolins detergent‡§: Triton X-100 (most popular), Brij-58, CHAPS, NP-40

IV. Caveolae

Properties

• 50 nanometres in diameter • Mobile (~10–8 cm–2 sec–1) • Liquid-ordered phase

• Large, often hundreds of nanometres to micrometres in size • Often bound to cytoskeleton

• Float to low density in sucrose or OptiprepTM density gradients

Comments

• Native rafts are only detected in living cells

• Clustering is used both artificially and physiologically to trigger signalling cascades

• Non-native (aggregated) raft • Raft subcategory • Variable effects depending on: • Highly specialized – Detergent type – Detergent:lipid ratio – Cell type

• Morphological ‘cave-like’ invaginations on the cell surface

* DRM, detergent-resistant membrane; DIG, detergent-insoluble glycolipid-rich domain; DIC, detergent-insoluble complex; LDM, low-density membrane; DIM, detergent-insoluble material; GEM, glycolipid-enriched membrane; TIFF, Triton X-100 insoluble floating fraction. ‡Care should be taken when choosing solubilization conditions for co-immunoprecipitation experiments, as these popular detergents do not solubilize rafts on ice. Co-localization of proteins in rafts or DRMs could be mistaken for direct protein–protein interactions if rafts are not completely solubilized. §Rafts can be solubilized in octyl glucoside or in the detergents listed above at raised temperatures.

turn, phosphorylates LAT, a transmembrane protein that couples TCR activation to several signalling pathways49–51. Several GPI-linked proteins52,53 and accessory molecules54–56 help to amplify the T-cell activation events. Phosphatases are also required to switch these pathways on and off 57. One remarkable feature of T-cell activation is that only 10–100 cognate peptide–MHC complexes, among a total pool of 104–105 MHC molecules expressed on an APC, have to be recognized by receptors on the T cell to generate an immune response58. This is possible only because the same TCR can be activated over and over again. This process can take hours to complete and is facilitated by the assembly of an immunological synapse, several micrometres in diameter, at the contact zone with the APC. A complex series of events involving the actin cytoskeleton59 leads to the formation of the immunological synapse60 — a contact zone between APC and T cells, where T-cell activation takes place47,58,61. During the formation of the immunological synapse, the T cell polarizes its actin and microtubule networks towards this contact site, and also directs membrane traffic in this direction59. Evidence from several laboratories indicates that clustering of rafts is an essential feature in the formation of an immunological synapse47,48. As with IgE receptors, monomeric TCR complexes have only weak steady-state raft affinity29,54. After receptor crosslinking, their raft residency increases and they become partly insoluble in detergent. Although we still need to understand the precise mechanism of initial TCR engagement in vivo, artificial crosslinking of the TCR-associNATURE REVIEWS | MOLECUL AR CELL BIOLOGY

ated CD3 subunits55, or of CD3 and CD28 with antibody-coated beads, can be used experimentally to activate TCR signalling56. Under these conditions many proteins, including the hyperphosphorylated TCR multisubunit complex, and cytoplasmic proteins such as ZAP-70, Vav, PLCγ, Grb2 and phosphatidylinositol 3-OH kinase become detergent resistant, indicating possible raft association55. Consistent with this interpretation, cholesterol depletion by methyl-β-cyclodextrin dissociates these proteins from rafts and inactivates the signalling cascade52,55. The activation of Lck by the TCR could furthermore lead to raft clustering, perhaps

Table 3 | Signal transduction processes involving rafts Protein

Selected references

FcεRI receptor

T-cell receptor

47, 48

B-cell receptor

EGF receptor

35, 91

Insulin receptor

EphrinB1 receptor

Neurotrophin

GDNF

63, 65

Hedgehog

H-Ras

Integrins

95, 96

eNOS

97, 98

VOLUME 1 | OCTOBER 2000 | 3 5

REVIEWS through protein adaptors that are constitutively associated with rafts, such as LAT49,50. This cascade of interactions, including scaffolding proteins and adaptors, builds up the immunological synapse on the surface of the T cell. Moreover, rafts also function to concentrate MHC class II molecules, loaded with specific peptides, on the APC side of the synapse62. GDNF signalling. The glial-cell-derived neurotrophic factor (GDNF) family of ligands is important for the development and maintenance of the nervous system. In addition, they function during differentiation of the kidney and spermatogonia. GDNF binds to a multicomponent receptor complex that is composed of the GPI-linked GDNF receptor-α (GFRα) and the transmembrane tyrosine kinase, RET. The receptor subunits GFRα and RET are not associated with each other in the absence of ligand63. But after extracellular GDNF a FcεRI signalling Multivalent antigen IgE 1

α Extracellular

γ γ β

Fc receptor

LAT

Raft Lyn ITAM motifs

4 P

P 3 Syk

Downstream signals

b TCR signalling

MHC

Antigen presenting cell

1 β α

Raft

LAT

TCR/CD3

Lck 4

– 3

T cell

Fyn

CBP

ZAP

5 ? Csk

Downstream signals

Figure 1 | Initial signalling events in rafts for a | IgE receptor (FcεRI)- and b | T-cell antigen receptor (TCR)-mediated signalling. For clarity, only a small subset of involved proteins are shown. A likely sequence of the key initial events is indicated numerically. 1 | Ligand-induced receptor dimerization of the Fc receptor or TCR/CD3 probably increases its raft association, which leads to 2 | phosphorylation of the receptors’ immune receptor tyrosinebased activation motifs (ITAMs) by Src-family protein tyrosine kinases (for example, Lyn, Lck and Fyn). 3 | Phosphorylated ITAMs act as a membrane docking site for cytoplasmic Syk/ZAP70; these are also tyrosine kinases and are activated in the raft by tyrosine phosphorylation. 4 | Syk/ZAP-70 can, in turn, activate other proteins such as LAT, a raft-associated adaptor. Through crosslinking, LAT can recruit other proteins into the raft and further amplify the signal. The complex cascade of later downstream signalling events is not shown. 5 | One possible way of downregulating the signal may occur by binding of the cytosolic kinase Csk to the raftassociated protein CBP. Csk may then inactivate the Src-family kinases through phosphorylation57.

| OCTOBER 2000 | VOLUME 1

stimulation, RET moves into rafts, where it associates with GFRα. Signal transduction depends on the colocalization of RET and GFRα in lipid rafts, as cholesterol depletion with methyl-β-cyclodextrin decreases GDNF signalling63. As with IgE receptor and TCR signalling, the issue of how signalling is activated on GDNF binding is unresolved. GDNF is a dimer and is sufficient to trigger the initial events. Being a dimer, it could crosslink its receptor, but whether this is really the signal-initiating event has not yet been shown. It is also not known whether signalling can occur within a single raft, or whether it requires raft clustering to reach a signalling threshold. If both GFRα and RET are necessary for GDNF signalling, you would predict that the receptor subunits should localize together in vivo. But GFRα is more widely expressed than RET in neural tissue. In fact, GDNF can also signal through GFRα in a RET-independent way64,65. Moreover, autophosphorylated RET can trigger different signalling pathways depending on whether it is inside or outside rafts (C. Ibáñez, personal communication). Ras signalling. The small GTPase Ras is central to many signalling processes. It acts as a switch that, when activated, recruits serine/threonine kinases of the Raf family to the plasma membrane. These, in turn, activate the ERK–MAP kinase pathway and other targets. The two Ras isoforms, K-Ras and H-Ras, are almost identical in sequence but have different signalling properties66. Both isoforms have a carboxy-terminal prenylated CAAX sequence, but whereas K-Ras has a polybasic region required for plasma membrane localization, H-Ras is palmitoylated67 and therefore more likely to partition into lipid rafts. Roy et al.66 showed that expression of a dominant-negative mutant of caveolin strongly inhibited H-Ras-mediated Raf activation, but had no effect on its activation by K-Ras. The expression of this mutant led to a decrease in the number of caveolae on the cell surface, and depleted cell surface cholesterol. The mutant phenotype could be mimicked by depleting cholesterol with methyl-β-cyclodextrin and it could be rescued by addition of exogenous cholesterol. One interpretation of these results is that expression of the caveolin mutant reduces the cholesterol content of the plasma membrane and therefore the number of functional lipid rafts. As H-Ras can signal only through rafts, it can no longer activate Raf. But K-Ras, which does not operate in rafts, is not affected. Hedgehog signalling. Drosophila melanogaster Hedgehog and its mammalian homologues act as short-range morphogens during tissue patterning. In the absence of Hedgehog signalling, the sterol-sensing membrane protein Patched represses the constitutive signalling activity of a second membrane protein, Smoothened, by forming an inactive Patched–Smoothened complex68. Hedgehog binding to Patched releases Smoothened, which activates a signalling cascade that culminates in the upregulation of a specific set of nuclear transcripts. Hedgehog is an interesting signalling molecule, as it

www.nature.com/reviews/molcellbio

REVIEWS is post-translationally modified to introduce a cholesterol moiety at the carboxyl terminus69 and a palmitate moiety at the amino terminus70. Cholesterol-modified Hedgehog is membrane bound, and has been shown to associate with lipid rafts in Drosophila embryos9. The cholesterol modification restricts the signalling range of Hedgehog, making it a short-range morphogen. If Hedgehog is mutated to lose its hydrophobic anchor, it is secreted and can activate cells much further away than normal71. So how is membrane-bound Hedgehog released from the cell where it is synthesized, to act as a signal for a target cell several cell layers away? It seems that the association of Hedgehog with rafts is important for its function, but it is not sufficient68. If cholesterol is replaced with a GPI-anchor — which should still localize the protein to rafts — Hedgehog is no longer released from the surface of the expressing cells71. Another sterolsensing protein, Dispatched, is also required for the release of Hedgehog71. The mechanism of release could A Model 1 a Activation in a raft

b Altered partitioning

Extracellular

Dimerization

Dimerization Antibodies, ligands

Antibodies, ligands

Signalling

B Model 2 Clustering of rafts triggers signalling GPI

GPI

Dimerization Clustering agent

Extracellular (antibodies, ligands) GPI GPI Membrane protein (for example, LAT) b c

Cytosolic (cytoskeletal elements, adaptors, scaffolds)

Figure 2 | Models of how signalling could be initiated through raft(s). A | In these models, signalling occurs in either single rafts (Model 1) or clustered rafts (Model 2). Following dimerization (or oligomerization) the protein becomes phosphorylated (blue circle) in rafts. In single rafts this can occur by activation a | within the raft, or b | by altering the partitioning dynamics of the protein. B | In the second model we assume that there are several rafts in the membrane, which differ in protein composition (shown in orange, purple or blue). Clustering would coalesce rafts (red), so that they would now contain a new mixture of molecules, such as crosslinkers and enzymes. Clustering could occur either extracellularly, within the membrane, or in the cytosol (a–c, respectively). Raft clustering could also occur through GPIanchored proteins (yellow), either as a primary or co-stimulatory response. Notably, models 1 and 2 are not mutually exclusive. For instance, extracellular signals could increase a protein’s raft affinity (for example, similar to the effect of single versus dual acylation) therefore drawing more of the protein into the raft where it can be activated and recruit other proteins, such as LAT, which would crosslink several rafts.

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

involve either displacement of the cholesterol tether or shedding of membrane vesicles from Hedgehog-producing cells. In conclusion, the requirement for lipid rafts during Hedgehog signalling is completely different from that described for other signalling processes. The cell biology of this fascinating signalling process is poorly understood, and awaits a detailed exploration. Models for signal initiation in rafts

Although we are at a stage at which we can build working hypotheses, we still do not know exactly how receptors signal through lipid rafts. As illustrated by the preceding examples (except for Hedgehog signalling) a common theme is that individual rafts cluster together to connect raft proteins and interacting proteins into a signalling complex. For instance, doubly acylated nonreceptor tyrosine kinases and G proteins from separate rafts could be brought into contact with a signalling receptor in this way. Receptors could behave in at least three different ways in rafts (FIG. 2). First, receptors associated at steady state with lipid rafts could be activated through ligand binding (FIG. 2A, a). Second, individual receptors with weak raft affinity could oligomerize on ligand binding, and this would lead to an increased residency time in rafts (FIG. 2A, b). Last, activated receptors could recruit crosslinking proteins that bind to proteins in other rafts, and this would result in raft coalescence (FIG. 2B). These models are not mutually exclusive. Through formation of a raft cluster, a network of interactions between adaptors, scaffolds and anchoring proteins would be built up to organize the signal complex in space and time. This signalling complex would be insulated within the raft clusters from the surrounding liquid-disordered lipid matrix. The formation of clustered rafts would lead to amplification through the concentration of signalling molecules, as well as to exclusion of unwanted modulators. The interactions that drive raft assembly are dynamic and reversible. Raft clusters can be disassembled by negative modulators and/or by removal of raft components from the cell surface by endocytosis. The coalescence of individual rafts to form raft clusters has been observed repeatedly, for example, when crosslinking raft components with antibodies12,29. The movement and behaviour of the raft clusters can also be influenced by interaction with cytoskeletal elements44,72,73 and second messengers such as the phosphoinositide PtdIns(4,5)P2, which help organize actin assemblies on the cytoplasmic surface of the rafts74,75. Many open issues

Several aspects of raft structure and function still need to be explained. One important area is raft composition and the question of whether more than one kind of raft exists on the cell surface of different cell types76,77. Not only do we need to identify raft-associated proteins, but we also have to determine the lipid composition in both the exoplasmic and cytoplasmic leaflets of rafts. As detergent extraction undoubtedly leads to raft aggregation, it is not easy to isolate individual rafts or ligand-activated

VOLUME 1 | OCTOBER 2000 | 3 7

REVIEWS raft clusters in such a way that their native state is preserved. Understanding the nature of individual rafts will probably require the development of new methods. The question of how single rafts are crosslinked to form clustered rafts during signal transduction also requires detailed exploration. This will require real-time imaging of the assembly of signalling complexes under normal conditions and during cholesterol depletion. A pressing issue is to clarify the function of caveolae during signal transduction. We know little about how proteins move into caveolae. Why does crosslinking of GPI-anchored proteins or gangliosides lead to their enrichment in caveolae78–80? Could the clustering of rafts induce the formation of caveolae in caveolin-containing cells? What are the protein signals required for caveolar trapping? Caveolae can be internalized, but how is this process regulated? What is the role of actin? These are

3. 4. 5.

10.

11.

12.

13. 14.

15. 16. 17.

Brown, D. A. & London, E. Functions of lipid rafts in biological membranes. Annu. Rev. Cell. Dev. Biol. 14, 111–136 (1998). Sankaram, M. B. & Thompson, T. E. Interaction of cholesterol with various glycerophospholipids and sphingomyelin. Biochemistry 29, 10670–10675 (1990). Simons, K. & van Meer, G. Lipid sorting in epithelial cells. Biochemistry 27, 6197–6202 (1988). Simons, K. & Ikonen, E. Functional rafts in cell membranes. Nature 387, 569–572 (1997). Fridriksson, E. K. et al. Quantitative analysis of phospholipids in functionally important membrane domains from RBL-2H3 mast cells using tandem highresolution mass spectrometry. Biochemistry 38, 8056–8063 (1999). Schroeder, R., London, E. & Brown, D. Interactions between saturated acyl chains confer detergent resistance on lipids and glycosylphosphatidylinositol (GPI)-anchored proteins: GPI-anchored proteins in liposomes and cells show similar behavior. Proc. Natl Acad. Sci. USA 91, 12130–12134 (1994). Hooper, N. M. Detergent-insoluble glycosphingolipid/cholesterol-rich membrane domains, lipid rafts and caveolae. Mol. Membr. Biol. 16, 145–156 (1999). Resh, M. D. Fatty acylation of proteins: new insights into membrane targeting of myristoylated and palmitoylated proteins. Biochim. Biophys. Acta 1451, 1–16 (1999). Rietveld, A., Neutz, S., Simons, K. & Eaton, S. Association of sterol- and glycosylphosphatidylinositol-linked proteins with Drosophila raft lipid microdomains. J. Biol. Chem. 274, 12049–12054 (1999). Scheiffele, P., Roth, M. G. & Simons, K. Interaction of influenza virus haemagglutinin with sphingolipidcholesterol membrane domains via its transmembrane domain. EMBO J. 16, 5501–5508 (1997). Melkonian, K. A., Ostermeyer, A. G., Chen, J. Z., Roth, M. G. & Brown, D. A. Role of lipid modifications in targeting proteins to detergent-resistant membrane rafts. Many raft proteins are acylated, while few are prenylated. J. Biol. Chem. 274, 3910–3917 (1999). Harder, T., Scheiffele, P., Verkade, P. & Simons, K. Lipid domain structure of the plasma membrane revealed by patching of membrane components. J. Cell Biol. 141, 929–942 (1998). The first demonstration that clusters of rafts segregate away from non-raft proteins. Palade, G. E. The fine structure of blood capillaries. J. Appl. Phys. 24, 1424 (1953). Yamada, E. The fine structure of the gall bladder epithelium of the mouse. J. Biophys. Biochem. Cytol. 1, 445–458 (1955). Parton, R. G. Caveolae and caveolins. Curr. Opin. Cell Biol. 8, 542–548 (1996). Smart, E. J. et al. Caveolins, liquid-ordered domains, and signal transduction. Mol. Cell. Biol. 19, 7289–7304 (1999). Schnitzer, J. E., Oh, P., Pinney, E. & Allard, J. Filipinsensitive caveolae-mediated transport in endothelium: reduced transcytosis, scavenger endocytosis, and capillary permeability of select macromolecules. J. Cell Biol. 127, 1217–1232 (1994).

| OCTOBER 2000 | VOLUME 1

important questions if we want to understand how raftassociated receptors are downregulated. This is an exciting time for researchers studying cellular membranes, but the issues at hand can be clarified only by a multidisciplinary approach. After long neglect, the dynamic organization of lipid bilayers is finally back at centre stage. Links DATABASE LINKS

Src kinase | caveolin | IgE | FcεRI | Lyn | Syk | ZAP-70 | PLCγ | LAT| TCR | CD3 | Fyn | Vav | Grb2 | Lck | GDNF | RET | Ras | Raf | ERK | Hedgehog | Patched | Smoothened | Dispatched FURTHER INFORMATION Simons lab homepage ENCYCLOPEDIA OF LIFE SCIENCES

Lipids| Membrane proteins

18. Parton, R. G., Way, M., Zorzi, N. & Stang, E. Caveolin-3 associates with developing T-tubules during muscle differentiation. J. Cell Biol. 136, 137–154 (1997). 19. Anderson, R. G. The caveolae membrane system. Annu. Rev. Biochem. 67, 199–225 (1998). 20. Vogel, U., Sandvig, K. & van Deurs, B. Expression of caveolin-1 and polarized formation of invaginated caveolae in Caco-2 and MDCK II cells. J. Cell Sci. 111, 825–832 (1998). 21. Renkonen, O., Kaarainen, L., Simons, K. & Gahmberg, C. G. The lipid class composition of Semliki forest virus and plasma membranes of the host cells. Virology 46, 318–326 (1971). 22. Levis, G. M. & Evangelatos, G. P. Lipid composition of lymphocyte plasma membrane from pig mesenteric lymph node. Biochem. J. 156, 103–110 (1976). 23. van Meer, G. Lipid traffic in animal cells. Annu. Rev. Cell Biol. 5, 247–275 (1989). 24. Brugger, B. et al. Segregation from COPI–coated vesicles of sphingomyelin and cholesterol. J. Cell Biol. (in the press). 25. Keller, P. & Simons, K. Post-Golgi biosynthetic trafficking. J. Cell Sci. 110, 3001–3009 (1997). 26. Ledesma, M. D., Simons, K. & Dotti, C. G. Neuronal polarity: essential role of protein–lipid complexes in axonal sorting. Proc. Natl Acad. Sci. USA 95, 3966–3971 (1998). 27. Mukherjee, S. & Maxfield, F. Role of membrane organization and membrane domains in endocytic lipid trafficking. Traffic 1, 203–211 (2000). 28. Puri, V. et al. Cholesterol modulates membrane traffic along the endocytic pathway in sphingolipid-storage diseases. Nature Cell Biol. 1, 386–388 (1999). 29. Janes, P. W., Ley, S. C. & Magee, A. I. Aggregation of lipid rafts accompanies signaling via the T cell antigen receptor. J. Cell Biol. 147, 447–461 (1999). 30. Pralle, A., Keller, P., Florin, E. L., Simons, K. & Horber, J. K. Sphingolipid–cholesterol rafts diffuse as small entities in the plasma membrane of mammalian cells. J. Cell Biol. 148, 997–1008 (2000). Individual raft size is measured by photonic force microscopy. 31. Varma, R. & Mayor, S. GPI-anchored proteins are organized in submicron domains at the cell surface. Nature 394, 798–801 (1998). 32. Friedrichson, T. & Kurzchalia, T. V. Microdomains of GPIanchored proteins in living cells revealed by crosslinking. Nature 394, 802–805 (1998). 33. Kenworthy, A. K., Petranova, N. & Edidin, M. Highresolution FRET microscopy of cholera toxin B-subunit and GPI-anchored proteins in cell plasma membranes. Mol. Biol. Cell 11, 1645–1655 (2000). 34. Brown, D. A. & Rose, J. K. Sorting of GPI-anchored proteins to glycolipid-enriched membrane subdomains during transport to the apical cell surface. Cell 68, 533–544 (1992). A pioneering demonstration that GPI-anchored proteins and influenza haemagglutinin remain associated with sphingolipids and cholesterol after Triton X-100 extraction. 35. Waugh, M. G., Lawson, D. & Hsuan, J. J. Epidermal growth factor receptor activation is localized within low-

buoyant density, non-caveolar membrane domains. Biochem. J. 337, 591–597 (1999). 36. Webb, Y., Hermida-Matsumoto, L. & Resh, M. D. Inhibition of protein palmitoylation, raft localization, and T cell signaling by 2-bromopalmitate and polyunsaturated fatty acids. J. Biol. Chem. 275, 261–270 (2000). Feeding cells with polyunsaturated fatty acids leads to dissociation of doubly acylated proteins from rafts. 37. Simons, M. et al. Exogenous administration of gangliosides displaces GPI-anchored proteins from lipid microdomains in living cells. Mol. Biol. Cell 10, 3187–3196 (1999). 38. Hunter, T. Signaling — 2000 and beyond. Cell 100, 113–127 (2000). 39. Field, K. A., Holowka, D. & Baird, B. Fc epsilon RImediated recruitment of p53/56lyn to detergent-resistant membrane domains accompanies cellular signaling. Proc. Natl Acad. Sci. USA 92, 9201–9205 (1995). 40. Sheets, E. D., Holowka, D. & Baird, B. Membrane organization in immunoglobulin E receptor signaling. Curr. Opin. Chem. Biol. 3, 95–99 (1999). 41. Baird, B., Sheets, E. D. & Holowka, D. How does the plasma membrane participate in cellular signaling by receptors for immunoglobulin E? Biophys. Chem. 82, 109–119 (1999). 42. Metzger, H. It’s spring, and thoughts turn to. . . allergies. Cell 97, 287–290 (1999). 43. Stauffer, T. P. & Meyer, T. Compartmentalized IgE receptormediated signal transduction in living cells. J. Cell Biol. 139, 1447–1454 (1997). 44. Holowka, D., Sheets, E. D. & Baird, B. Interactions between FcεRI and lipid raft components are regulated by the actin cytoskeleton. J. Cell Sci. 113, 1009–1019 (2000). 45. Sheets, E. D., Holowka, D. & Baird, B. Critical role for cholesterol in Lyn-mediated tyrosine phosphorylation of FcεRI and their association with detergent-resistant membranes. J. Cell Biol. 145, 877–887 (1999). This paper is the culmination of a series of studies showing the role of rafts in IgE receptor signalling. 46. Goitsuka, R. et al. A BASH/SLP-76-related adaptor protein MIST/Clnk involved in IgE receptor-mediated mast cell degranulation. Int. Immunol. 12, 573–580 (2000). 47. Janes, P. W., Ley, S. C., Magee, A. I. & Kabouridis, P. S. The role of lipid rafts in T cell antigen receptor (TCR) signalling. Semin. Immunol. 12, 23–34 (2000). 48. Langlet, C., Bernard, A. M., Drevot, P. & He, H. T. Membrane rafts and signaling by the multichain immune recognition receptors. Curr. Opin. Immunol. 12, 250–255 (2000). 49. Zhang, W., Trible, R. P. & Samelson, L. E. LAT palmitoylation: its essential role in membrane microdomain targeting and tyrosine phosphorylation during T cell activation. Immunity 9, 239–246 (1998). 50. Brdicka, T., Cerny, J. & Horejsi, V. T cell receptor signalling results in rapid tyrosine phosphorylation of the linker protein LAT present in detergent-resistant membrane microdomains. Biochem. Biophys. Res. Commun. 248, 356–360 (1998). 51. Lin, J., Weiss, A. & Finco, T. S. Localization of LAT in glycolipid-enriched microdomains is required for T cell

www.nature.com/reviews/molcellbio

REVIEWS

activation. J. Biol. Chem. 274, 28861–28864 (1999). 52. Moran, M. & Miceli, M. C. Engagement of GPI-linked CD48 contributes to TCR signals and cytoskeletal reorganization: a role for lipid rafts in T cell activation. Immunity 9, 787–796 (1998). 53. Stefanova, I., Horejsi, V., Ansotegui, I. J., Knapp, W. & Stockinger, H. GPI-anchored cell-surface molecules complexed to protein tyrosine kinases. Science 254, 1016–1019 (1991). 54. Montixi, C. et al. Engagement of T cell receptor triggers its recruitment to low-density detergent-insoluble membrane domains. EMBO J. 17, 5334–5348 (1998). 55. Xavier, R., Brennan, T., Li, Q., McCormack, C. & Seed, B. Membrane compartmentation is required for efficient T cell activation. Immunity 8, 723–732 (1998). Detailed characterization of several proteins participating in T-cell activation and their raft association. 56. Viola, A., Schroeder, S., Sakakibara, Y. & Lanzavecchia, A. T lymphocyte costimulation mediated by reorganization of membrane microdomains. Science 283, 680–682 (1999). Antibody-coated beads are used to activate clustering of raft components in T-cell signalling. 57. Cary, L. A. & Cooper, J. A. Molecular switches in lipid rafts. Nature 404, 945–947 (2000). 58. Lanzavecchia, A., Lezzi, G. & Viola, A. From TCR engagement to T cell activation: a kinetic view of T cell behavior. Cell 96, 1–4 (1999). 59. van der Merwe, A. P., Davis, S. J., Shaw, A. S. & Dustin, M. L. Cytoskeletal polarization and redistribution of cellsurface molecules during T cell antigen recognition. Semin. Immunol. 12, 5–21 (2000). 60. Grakoui, A. et al. The immunological synapse: a molecular machine controlling T cell activation. Science 285, 221–227 (1999). 61. Zhang, W. & Samelson, L. E. The role of membraneassociated adaptors in T cell receptor signalling. Semin. Immunol. 12, 35–41 (2000). 62. Anderson, H. A., Hiltbold, E. M. & Roche, P. A. Concentration of MHC class II molecules in lipid rafts facilitates antigen presentation. Nature Immunol. 1, 156–162 (2000). 63. Tansey, M. G., Baloh, R. H., Milbrandt, J. & Johnson, E. M. Jr GFRα-mediated localization of RET to lipid rafts is required for effective downstream signaling, differentiation, and neuronal survival. Neuron 25, 611–623 (2000). The demonstration that GDNF signalling is a raftdependent process. 64. Poteryaev, D. et al. GDNF triggers a novel retindependent src kinase family-coupled signaling via a GPI-linked GDNF receptor α1. FEBS Lett. 463, 63–66 (1999). 65. Trupp, M., Scott, R., Whittemore, S. R. & Ibanez, C. F. Ret-dependent and -independent mechanisms of glial cell line-derived neurotrophic factor signaling in neuronal cells. J. Biol. Chem. 274, 20885–20894 (1999). 66. Roy, S. et al. Dominant-negative caveolin inhibits H-Ras function by disrupting cholesterol-rich plasma membrane domains. Nature Cell Biol. 1, 98–105 (1999). This paper shows that H-Ras signals in rafts and K-

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

Ras signals outside rafts. 67. Hancock, J. F., Paterson, H. & Marshall, C. J. A polybasic domain or palmitoylation is required in addition to the CAAX motif to localize p21ras to the plasma membrane. Cell 63, 133–139 (1990). 68. Incardona, J. P. & Eaton, S. Cholesterol in signal transduction. Curr. Opin. Cell Biol. 12, 193–203 (2000). 69. Porter, J. A., Young, K. E. & Beachy, P. A. Cholesterol modification of hedgehog signaling proteins in animal development. Science 274, 255–259 (1996). 70. Pepinsky, R. B. et al. Identification of a palmitic acidmodified form of human Sonic hedgehog. J. Biol. Chem. 273, 14037–14045 (1998). 71. Burke, R. et al. Dispatched, a novel sterol-sensing domain protein dedicated to the release of cholesterol-modified hedgehog from signaling cells. Cell 99, 803–815 (1999). 72. Harder, T. & Simons, K. Clusters of glycolipid and glycosylphosphatidylinositol-anchored proteins in lymphoid cells: accumulation of actin regulated by local tyrosine phosphorylation. Eur. J. Immunol. 29, 556–562 (1999). 73. Laux, T. et al. GAP43, MARCKS, and CAP23 modulate PI(4,5)P2 at plasmalemmal rafts, and regulate cell cortex actin dynamics through a common mechanism. J. Cell Biol. 149, 1455–1472 (2000). 74. Pike, L. J. & Miller, J. M. Cholesterol depletion delocalizes phosphatidylinositol bisphosphate and inhibits hormonestimulated phosphatidylinositol turnover. J. Biol. Chem. 273, 22298–22304 (1998). 75. Rozelle, A. L. et al. Phosphatidylinositol 4,5-bisphosphate induces actin-based movement of raft-enriched vesicles through WASP-Arp2/3. Curr. Biol. 10, 311–320 (2000). 76. Iwabuchi, K., Yamamura, S., Prinetti, A., Handa, K. & Hakomori, S. GM3-enriched microdomain involved in cell adhesion and signal transduction through carbohydrate–carbohydrate interaction in mouse melanoma B16 cells. J. Biol. Chem. 273, 9130–9138 (1998). 77. Roper, K., Corbeil, D. & Huttner, W. B. Retention of prominin in microvilli reveals distinct cholesterol–based lipid microdomains within the apical plasma membrane of epithelial cells. Nature Cell Biol. 2, 582–592 (2000). 78. Mayor, S., Rothberg, K. G. & Maxfield, F. R. Sequestration of GPI-anchored proteins in caveolae triggered by crosslinking. Science 264, 1948–1951 (1994). 79. Parton, R. G. Ultrastructural localization of gangliosides; GM1 is concentrated in caveolae. J. Histochem. Cytochem. 42, 155–166 (1994). 80. Fujimoto, T. GPI-anchored proteins, glycosphingolipids, and sphingomyelin are sequestered to caveolae only after crosslinking. J. Histochem. Cytochem. 44, 929–941 (1996). 81. Wilson, B. S., Pfeiffer, J. R. & Oliver, J. M. Observing FceRI signaling from the inside of the mast cell membrane. J. Cell Biol. 149, 1131–1142 (2000). Clear visualization of raft clustering during IgE signalling by immuno-electron microscopy. 82. Sargiacomo, M., Sudol, M., Tang, Z. & Lisanti, M. P. Signal transducing molecules and glycosylphosphatidylinositol-linked proteins form a caveolin-rich insoluble complex in MDCK cells. J. Cell Biol. 122, 789–807 (1993). 83. Kurzchalia, T., Hartmann, E. & Dupree, P. Guilt by

84.

85.

86.

87.

88. 89.

90.

91.

92.

93.

94.

95.

96.

97.

98.

insolubility: Does a protein’s detergent insolubility reflect caveolar location. Trends Cell Biol. 5, 187–189 (1995). Smart, E. J., Ying, Y. S., Mineo, C. & Anderson, R. G. A detergent-free method for purifying caveolae membrane from tissue culture cells. Proc. Natl Acad. Sci. USA 92, 10104–10108 (1995). Schnitzer, J. E., McIntosh, D. P., Dvorak, A. M., Liu, J. & Oh, P. Separation of caveolae from associated microdomains of GPI-anchored proteins. Science 269, 1435–1439 (1995). Stan, R. V. et al. Immunoisolation and partial characterization of endothelial plasmalemmal vesicles (caveolae). Mol. Biol. Cell 8, 595–605 (1997). Oh, P. & Schnitzer, J. E. Immunoisolation of caveolae with high affinity antibody binding to the oligomeric caveolin cage. Toward understanding the basis of purification. J. Biol. Chem. 274, 23144–23154 (1999). Kurzchalia, T. V. & Parton, R. G. Membrane microdomains and caveolae. Curr. Opin. Cell Biol. 11, 424–431 (1999). Schutz, G. J., Kada, G., Pastushenko, V. P. & Schindler, H. Properties of lipid microdomains in a muscle cell membrane visualized by single molecule microscopy. EMBO J. 19, 892–901 (2000). Cheng, P. C., Dykstra, M. L., Mitchell, R. N. & Pierce, S. K. A role for lipid rafts in B cell antigen receptor signaling and antigen targeting. J. Exp. Med. 190, 1549–1560 (1999). Couet, J., Sargiacomo, M. & Lisanti, M. P. Interaction of a receptor tyrosine kinase, EGF-R, with caveolins. Caveolin binding negatively regulates tyrosine and serine/threonine kinase activities. J. Biol. Chem. 272, 30429–30438 (1997). Mastick, C. C., Brady, M. J. & Saltiel, A. R. Insulin stimulates the tyrosine phosphorylation of caveolin. J. Cell Biol. 129, 1523–1531 (1995). Bruckner, K. et al. EphrinB ligands recruit GRIP family PDZ adaptor proteins into raft membrane microdomains. Neuron 22, 511–524 (1999). Bilderback, T. R., Gazula, V. R., Lisanti, M. P. & Dobrowsky, R. T. Caveolin interacts with Trk A and p75(NTR) and regulates neurotrophin signaling pathways. J. Biol. Chem. 274, 257–263 (1999). Wary, K. K., Mariotti, A., Zurzolo, C. & Giancotti, F. G. A requirement for caveolin-1 and associated kinase Fyn in integrin signaling and anchorage-dependent cell growth. Cell 94, 625–634 (1998). Krauss, K. & Altevogt, P. Integrin leukocyte functionassociated antigen-1 mediated cell binding can be activated by clustering of membrane rafts. J. Biol. Chem. 274, 36921–36927 (1999). Shaul, P. W. et al. Acylation targets endothelial nitric-oxide synthase to plasmalemmal caveolae. J. Biol. Chem. 271, 6518–6522 (1996). Garcia-Cardena, G., Fan, R., Stern, D. F., Liu, J. & Sessa, W. C. Endothelial nitric oxide synthase is regulated by tyrosine phosphorylation and interacts with caveolin-1. J. Biol. Chem. 271, 27237–27240 (1996).

Acknowledgements We thank D. Brown, R. Parton, T. Harder and T. Kurzchalia for critical reading of this manuscript. C. Ibáñez provided helpful discussions on GDNF signalling.

VOLUME 1 | OCTOBER 2000 | 3 9

REVIEWS

• Lipid rafts consist of dynamic assemblies of cholesterol and sphingolipids in the exoplasmic leaflet of the lipid bilayer. • Lipid rafts can include or exclude proteins selectively, and the raft affinity of a given protein can be modulated by intra- or extracellular stimuli. • They are too small to be seen by standard microscope techniques. It is also not possible to isolate lipid rafts in their native state. Detergentresistant membranes, containing clusters of many rafts, can be isolated by extraction with Triton X-100 or other detergents on ice. • Raft association of proteins can be assayed by manipulating the lipid composition of rafts. If cholesterol or sphingolipids are depleted from membranes, lipid rafts are dissociated, and previously associated proteins are no longer in rafts. • There is great confusion in the nomenclature for lipid rafts, and Table 2 proposes a new nomenclature. • Rafts are involved in signal transduction. Crosslinking of signalling receptors increases their affinity for rafts. Partitioning of receptors into rafts results in a new micro-environment, where their phosphorylation state can be modified by local kinases and phosphatases, modulating downstream signalling. • Raft clustering could also be involved in signal transduction. Several rafts coalesce, resulting in amplification of the signal. • Some examples for such raft-dependent signalling processes are IgE signalling during the allergic response, T-cell activation and GDNF signalling. • Rafts are also necessary for Hedgehog signalling during development but the mechanism is very different. Hedgehog is a membranebound ligand and needs to be released from its cell of origin so it can signal to cells several layers away. It can be released from the cell when it is anchored in rafts through its cholesterol moiety.

Kai Simons received his M.D. Ph.D. degree in 1964 from the University of Helsinki, Finland. Simons then did postdoctoral research with A.G. Bearn at Rockfeller University in New York. In 1975, he became a Group Leader at the European Molecular Biology Laboratory (EMBL) in Heidelberg, Germany, and he started the Cell Biology Program, which became the focal point for molecular cell biology in Europe. He is the first President of the newly formed European Life Scientist Organization. In addition, he will head the new Max Planck Institute for Molecular Cell Biology and Genetics, which will start its operations in Dresden, Germany, in January 2001. His research interests have been concerned with the life cycle of Semliki Forest virus. This virus had the simplest biological membrane known. From this early work he moved to studies of epithelial polarity and intracellular protein and lipid transport. These studies laid the ground work for the lipid raft concept that now is dominating his research interests. Derek Toomre studied Glycobiology at the Univeristy of California, San Diego (UCSD), under the supervision of Ajit Varki. He obtained his Ph. D. in Biochemistry in 1996 and was subsequently attracted to the field of cell biology. He is a postdoc in Kai Simons’s lab on a Marie Curie Fellowship. His primary focus in the Simons lab has been the application of advanced videomicroscopy techniques including multicolour GFP imaging and total internal reflection fluorescence microscopy (TIRFM) to study the dynamic processes of membrane sorting, trafficking and fusion in living cells.

| OCTOBER 2000 | VOLUME 1

www.nature.com/reviews/molcellbio

REVIEWS Src kinase http://www.ncbi.nlm.nih.gov/LocusLink/LocRpt.cgi?l=1445

Ras http://www.ncbi.nlm.nih.gov/LocusLink/list.cgi?Q=ras*

Hedgehog http://www.ncbi.nlm.nih.gov/LocusLink/list.cgi?Q=hedge hog*&ORG=Hs

Raf http://www.ncbi.nlm.nih.gov/LocusLink/list.cgi?Q=raf* ERK http://www.ncbi.nlm.nih.gov/LocusLink/list.cgi?Q=erk

caveolin http://www.ncbi.nlm.nih.gov/LocusLink/list.cgi?Q=caveolin*&ORG=Hs

Hedgehog http://flybase.bio.indiana.edu/.bin/fbidq.html?FBgn0004644

IgE http://www.ncbi.nlm.nih.gov/LocusLink/LocRpt.cgi?l=3497

Patched http://flybase.bio.indiana.edu/.bin/fbidq.html?FBgn0003892

FcεR1 http://www.ncbi.nlm.nih.gov/LocusLink/LocRpt.cgi?l=2205

Smoothened http://flybase.bio.indiana.edu/.bin/fbidq.html?FBgn0003444

Lyn http://www.ncbi.nlm.nih.gov/LocusLink/LocRpt.cgi?l=4067

Dispatched http://flybase.bio.indiana.edu/.bin/fbidq.html?FBgn0029088

Syk http://www.ncbi.nlm.nih.gov/LocusLink/LocRpt.cgi?l=6850

LAT http://www.ncbi.nlm.nih.gov/LocusLink/LocRpt.cgi?l=27040

Encyclopedia of Life Sciences: Lipids http://www.els.net/elsonline/fr_loadarticle.jsp?available=1&ref=A0000711&orig=searching&page_number=1&page=search&Sitemap=Lipid&searchtype=freetext&searchlevel=4 membrane proteins http://www.els.net/elsonline/fr_loadarticle.jsp?available=1&r ef=A0000624&orig=searching&page_number=1&page=sear ch&Sitemap=Lipid&searchtype=freetext&searchlevel=4

TCR http://www.ncbi.nlm.nih.gov/LocusLink/list.cgi?Q=TCR*&O RG=Hs&V=0

lab homepage http://www.emblheidelberg.de/ExternalInfo/simons/index.html

ZAP-70 http://www.ncbi.nlm.nih.gov/LocusLink/LocRpt.cgi?l=7535 PLCγ http://www.ncbi.nlm.nih.gov/LocusLink/LocRpt.cgi?l=5335

CD3 http://www.ncbi.nlm.nih.gov/LocusLink/list.cgi?Q=CD3* Fyn http://www.ncbi.nlm.nih.gov/LocusLink/LocRpt.cgi?l=2534 VAV http://www.ncbi.nlm.nih.gov/LocusLink/LocRpt.cgi?l=7409 Grb2 http://www.ncbi.nlm.nih.gov/LocusLink/LocRpt.cgi?l=2885 Lck http://www.ncbi.nlm.nih.gov/LocusLink/LocRpt.cgi?l=3932 GDNF http://www.ncbi.nlm.nih.gov/LocusLink/LocRpt.cgi?l=2668 RET http://www.ncbi.nlm.nih.gov/LocusLink/LocRpt.cgi?l=5979

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 1 | OCTOBER 2000 | 4 1

REVIEWS

NEW TARGETS FOR INHIBITORS OF HIV-1 REPLICATION John P. Moore* and Mario Stevenson‡ Despite the success of protease and reverse transcriptase inhibitors, new drugs to suppress HIV-1 replication are still needed. Several other early events in the viral life cycle (stages before the viral genome is inserted into host cell DNA) are susceptible to drugs, including virus attachment to target cells, membrane fusion and post-entry events such as integration, accessory-gene function and assembly of viral particles. Among these, inhibitors of virus–cell fusion and integration are the most promising candidates.

BIOAVAILABILITY

The rate/extent to which a drug is absorbed and becomes available at its site of action. VIRION

A complete viral particle, comprising the nucleic-acid core and protein capsid, enclosed by a glycoproteincontaining membrane envelope in some species.

* Department of Microbiology and Immunology, Weill Medical College of Cornell University, 1300 York Avenue, New York, New York 10021, USA. e-mail: jpm2003@mail. med.cornell.edu ‡ Program in Molecular Medicine and Department of Molecular Genetics and Microbiology, University of Massachusetts Medical School, 373 Plantation Street, Worcester, Massachusetts 01605, USA. e-mail: Mario.Stevenson@ ummed.edu

Death rates due to HIV-1 infection have fallen considerably since specific inhibitors were developed to antagonize the viral reverse transcriptase and protease enzymes — at least in those countries that can afford combination antiviral therapies (roughly $10,000 per year, per patient)1. Although over a dozen drugs are now on the market, they all target only these two viral enzymes. But up to 20% of patients cannot tolerate antiviral cocktails in the short term2; there is increasing concern about the long-term metabolic side-effects of protease inhibitors (notably, poorly understood problems with fat metabolism2); and drug-resistant HIV-1 variants are emerging and spreading at an increasing rate3. Because antiviral therapy cannot eradicate HIV-1 from infected people4, we need to identify new classes of drugs suitable for long-term use that can supplement, or partially replace, existing drug regimes. Several stages of the viral life cycle are potentially vulnerable to specific inhibitors (FIG. 1). These can be divided into the entry steps, which involve viral-envelope glycoproteins and their receptors, and the postentry steps involving viral accessory-gene products and the cellular proteins with which they interact. Although both viral and cellular factors can be targeted, it is usually less toxic to attack a viral factor than to disturb the function of a host protein. Many factors influence the path of a drug to the clinic. The easiest (but rarely trivial) stage of drug development is usually the identification of lead compounds that inhibit HIV-1 replication in vitro. But other consid-

erations are often limiting factors. For example, an inhibitor that cannot be manufactured in bulk to defined specifications, or which has poor BIOAVAILABILITY, a too limited half-life in vivo or obvious toxicity, cannot make a practical drug. All other things being equal, it is preferable for drugs to be taken by mouth rather than injection. However, a drug that has limited oral availability but is otherwise effective might still be useful as an injectable, especially for salvage therapy (for people who have no other treatment options). Alternatively, it could be applied topically as a vaginal or rectal microbicide to prevent HIV-1 transmission, as long as the compound is safe and can be manufactured cheaply5. Our aim here is not to list every compound shown to have an antiviral effect in vitro, but to highlight those inhibitors whose mechanism of action is instructive, or which we think have a reasonable chance of a successful transition into the clinic. Inhibitors of virus–cell attachment

A logical point of the viral life cycle at which to inhibit HIV-1 replication is the first — the process by which the infectious VIRION enters its target cells (FIG. 1). The ratelimiting step for retroviral infection is often the initial attachment of the virion to the cell surface, which occurs before glycoproteins on the viral envelope interact with specific receptors on the host cell to trigger fusion. Several mechanisms have been proposed for virus–cell attachment. The most credible of these involve the association of positively charged regions of www.nature.com/reviews/molcellbio

| OCTOBER 2000 | VOLUME 1

REVIEWS

Figure 1 | Entry steps in the HIV-1 replication cycle. The infectious particle, which contains two copies of RNA, first binds to target cells using semi- or non-specific interactions between the viral envelope and cell surface glycans or adhesion factors (not shown). The gp120 envelope glycoprotein then interacts with the CD4 antigen. This initiates a conformational change in gp120 that facilitates its binding to a co-receptor molecule. Further conformational changes in the gp120–gp41 complex then lead to exposure of the fusion-peptide region of gp41 and its insertion into the host cell membrane. The viral and cell membranes fuse, the viral capsid core disassembles (uncoating step) and viral nucleic acids enter the cytosol in association with virion proteins. Within this complex of viral RNA and virion proteins (reverse transcription complex), reverse transcriptase catalyses complementary DNA synthesis. The resulting complex containing viral cDNA (pre-integration complex) is transported to the host cell nucleus where the viral integrase enzyme catalyses integration of viral cDNA within host cell DNA to form the provirus. In some instances, viral cDNA that is translocated to the nucleus circularizes to form episomes containing one or two long terminal repeats (LTR). These circular forms of viral cDNA are dead-end products of viral replication. Animated online

the viral-envelope glycoproteins with negatively charged heparan sulphate proteoglycans6,7; the binding of glycan moieties on the envelope glycoproteins to cell-surface lectins such as DC-specific ICAM-3 grabbing nonintegrin (DC-SIGN)8; and the interaction of virion- and cell-associated adhesion factors such as intercellular adhesion molecule-1 (ICAM-1) and leukocyte cell-surface antigen-1 (LFA-1) respectively9. The precise mechanism used by HIV-1 to attach to cells may be strain dependent, and may differ between viruses with different tropism properties7. In principle, virus–cell attachment can be antagonized. Carrageenans derived from seaweed, lectins and the cyanovirin protein all interact semi- or non-specifically with the viral gp120 envelope glycoprotein to inhibit HIV-1 replication at this stage (some of these compounds can also interfere with later events in receptor-mediated fusion by virtue of their attachment to gp120)6,7,10. Strongly cationic peptides also inhibit viral entry, although they interact with the target cell and not

with the virus11. However, there is a question mark over the clinical value of most of the attachment/entry inhibitors listed above, mainly because of issues relating to production, delivery, pharmacology and toxicity. The better characterized, more easily produced compounds might be useful as topical microbicides5. Virus-entry inhibitors

gp120–CD4 binding. The first stage of the fusion process involves binding of the envelope glycoprotein complex to the cell-surface CD4 antigen through the gp120 glycoprotein (FIG. 2). This step is vulnerable to agents that bind to either gp120 or CD4. The crystal structure of gp120 has revealed details of the CD4binding site, including a recessed pocket that could accommodate a small-molecule inhibitor of CD4 attachment12 (although no such inhibitor has yet been reported). Binding of the polyanionic, bis-azo compound FP-21399 to the V3 loop region of gp120 inhibits gp120–CD4 binding and virus–cell fusion.

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 1 | OCTOBER 2000 | 4 1

REVIEWS

CHEMOKINES

Cytokines involved in specific inflammatory responses. They are differentiated into CC or CXC chemokines on the basis of their primary sequence.

However, it is not clear how FP-21399 works as the V3 loop is not part of the CD4-binding site; effects on the co-receptor-binding function of gp120 are probably also involved13. Likewise, other polyanions — for example, sulphated polysaccharides such as dextran sulphate — may work by generally destabilizing the functions of the envelope glycoprotein complex14. Negatively charged albumins also bind gp120 to prevent a stable association between HIV-1 and CD4+ cells, and perhaps block gp120–CD4 attachment15. A larger, and more specific, inhibitor that binds to gp120 and acts as a soluble receptor decoy to prevent its association with CD4 is the soluble CD4 molecule. In its original incarnation as a monomeric protein, soluble CD4 showed no notable antiviral action in vivo. This was because, despite its efficient neutralization of cell-culture-adapted viruses in vitro, it had no effect against primary HIV-1 isolates of the type found in infected people. However, a soluble CD4 derivative known as CD4-IgG2, a tetrameric immunoglobulin-G fusion protein, is considerably more potent in vitro than the parental monomer16. Designated PRO 542, CD4-IgG2 has shown some ability to reduce plasma levels of virus in phase I clinical trials16. Peptides that mimic the gp120-binding region of CD4 domain 1 also have an antiviral effect in vitro17,18, although their potency will probably need to be increased from the present micromolar range if they are to have antiviral activity in vivo. The binding of gp120 to CD4 can also be blocked by targeting CD4. Monoclonal antibodies against CD4, such as Leu3A or OKT4A, do this efficiently in vitro. However, concerns about immunosuppression through depletion of CD4+ cells will probably prevent the clinical use of such antibodies. These concerns may be lessened for one anti-CD4 monoclonal, 5A8, which blocks virus–cell fusion after CD4 binding and seems not to cause depletion of CD4+ cells in monkeys19. Whether 5A8 will have other immunosuppressive effects through its binding to CD4 is not clear. None of the above inhibitors of gp120–CD4 binding will make orally available drugs because they are protein-based compounds, although delivery by injection or subcutaneously is an alternative option. gp120–co-receptor interactions. After the virus has attached to CD4, conformational changes in the envelope glycoproteins expose a previously hidden region of gp120 that can now bind to a co-receptor12,20. The HIV1 co-receptors belong to the seven-transmembranespanning, guanine-nucleotide-binding (G)-proteincoupled receptor superfamily20. The normal function of these receptors is to bind chemotactic cytokines — or CHEMOKINES — which signal the responsive cells to migrate towards or away from sites at which chemokines are released as part of a continuing immune response20. About a dozen co-receptors mediate HIV-1 entry in vitro — that is, they allow one or more viruses to fuse with a cell line expressing an experimentally transfected co-receptor20. Among these coreceptors, however, probably only CCR5 and CXCR4 are important as front-line pharmacological targets,

because they are the main co-receptors used by HIV-1 to enter primary CD4+ T cells and macrophages21–23. The CCR5 co-receptor facilitates the entry of ‘macrophage-tropic’, non-syncytium-inducing HIV-1 isolates, whereas CXCR4 serves the same purpose for ‘T-cell-line-tropic’, syncytium-inducing isolates20. Such viruses have now been designated as R5 strains if they

Figure 2 | HIV-1 entry and inhibition by gp41 peptides. The fusion process for the trimeric HIV-1 envelope glycoprotein complex is depicted in schematic form. The sequential binding of the gp120 moieties (yellow ovals) to CD4 and a co-receptor on the cell membrane (not shown) drives conformational changes in the gp41 moieties (dark blue ovals). These changes cause the N-peptide region of gp41 (light blue cylinders) to translocate, and the fusion peptide (red lines) to insert into the cell membrane, forming the ‘pre-hairpin intermediate’. Subsequent conformational changes in gp41 may be necessary to create the ‘hairpin form’ in which the viral and cellular membranes are brought into close enough proximity for membrane coalescence and fusion to occur. These changes can be inhibited by the endogenous C peptides, T20 and T1249 (depicted as red or black squares). These bind to the N-peptide region of the pre-hairpin intermediate, preventing the natural C-peptide region of gp41 from doing so. T1249 is a derivative of T20 that has increased potency (D. P. Bolognesi, personal communication)53,56. (Figure based on information described in REFS 101, 102. An alternative model of how gp41 mediates membrane fusion is shown in REF. 103.)

www.nature.com/reviews/molcellbio

| OCTOBER 2000 | VOLUME 1

REVIEWS

Figure 3 | HIV-1 co-receptors and inhibitors. a | Models of the CXCR4 and CCR5 co-receptors viewed from the perspective of the incoming virus (that is, from above and outside the cell). The surface charges are highlighted (red, negative charge; white, neutral; blue, positive charge). CXCR4 is unusual among known HIV-1 co-receptors in that it has a strongly negative surface. The surface of CCR5 may, however, be slightly more negative than is depicted because of post-translational addition of tyrosine sulphate moieties to its amino-terminal region104. It is instructive to compare the charges on gp120s from X4 and R5 HIV-1 isolates that use CXCR4 and CCR5 with the charges on these receptors; X4 virus gp120s are more positively charged105. (Models of CXCR4 and CCR5 prepared by S. R. Durrel and M. S. Dimitrov.) b | The chemical structures of CXCR4 inhibitors (AMD3100, ALX40-4C and T22) and a CCR5 inhibitor (TAK-779), with positively charged atoms or functional groups highlighted in blue. Note that the CXCR4 inhibitors are much more positive than the CCR5 inhibitor. (Figure prepared with the help of R. W. Doms and T. P. Sakmar.)

use CCR5, X4 strains if they use CXCR4, or R5X4 strains if they use both co-receptors24. Notably, whereas R5 viruses target activated, CCR5+ CD4+ memory T cells25, X4 viruses cause the depletion of naive, CXCR4+ CD4+ T cells26. Blocking the function of CCR5 may have no negative side-effects in patients because roughly 1% of caucasians naturally lack this protein27. In fact, in vitro experiments28 and epidemiology studies27 indicate that a reduced level of CCR5 can be beneficial by decreasing the ability of HIV-1 to infect cells. Inhibiting CXCR4 may be more problematic, as a CXCR4 deletion prevents fetal development in mice29. However, a CXCR4specific inhibitor was not acutely toxic in adult mice30. The first inhibitors known to prevent co-receptor interactions of the viral-envelope glycoproteins were macrophage inflammatory protein (MIP)-1α, MIP-1β and RANTES, the natural CC-chemokine ligands of CCR5 (REF. 31). Likewise, the CXC-chemokine stromalderived factor-1α (SDF-1α) inhibits entry through CXCR4 (REFS 32,33). Variants of these chemokines, usually amino-terminal modifications to RANTES or SDF1α with increased potency and/or altered agonist properties in vitro were soon developed, as were short peptides with some antiviral activity, which were based on the amino-terminal chemokine domain34,35. The chemokine-based inhibitors can interfere with HIV-1 replication in many ways. First, the chemokine and the gp120 glycoprotein may compete for binding to the coreceptor; second, the co-receptor may be downregulated after chemokine binding and signal transduction;

and third, signalling may alter the differentiation state of the target cell and affect HIV-1 replication later in the viral life cycle34–38. The signalling capacity of chemokines will probably affect their clinical use. Although it is not desirable to use agonists that affect the target cells for HIV-1 replication, non-agonists cannot cause co-receptor downregulation and therefore have reduced potency34–37. In vitro, both CC- and CXC-chemokines can, under some conditions, enhance HIV-1 replication through their effects on target cells39–41, and a modified RANTES derivative promoted the evolution of X4 viruses from one R5 strain when tested in a mouse model system42. Add in that chemokines will not be orally available, and that they will probably have a very short half-life in vivo, and the practical obstacles to their clinical development as inhibitors become important. Monoclonal antibodies targeted against CXCR4 and CCR5 can have considerable potency and breadth of action as inhibitors of HIV-1 replication in vitro, without being agonists28,43,44. As with chemokines, monoclonals will not be orally available, but they could be used as injectable agents. In principle, monoclonals might cause cellular depletion through immune targeting, which will need to be evaluated as suitable candidates enter clinical trials. Several small-molecule inhibitors of HIV-1 entry through CXCR4 and CCR5 are now known, all of which are receptor antagonists that have no signalling capacity themselves. The CXCR4 inhibitors — T22, ALX40-4C and AMD3100 (and their derivatives) — are highly

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 1 | OCTOBER 2000 | 4 3

REVIEWS

D-PEPTIDES

Peptides comprising D-amino acids, rather than the naturally occurring L isomer.

cationic compounds45–49, and cationic peptides derived from the V3 loop region of HIV-1 gp120 also inhibit HIV-1 entry through CXCR4 (REF. 50). The only smallmolecule CCR5 inhibitor whose structure has been described is TAK-779, a non-agonist that blocks both virus entry and chemokine binding51 (FIG. 3). Another non-agonist, small-molecule inhibitor that targets CCR5, Sch-C, entered phase I clinical trials this summer. Sch-C is an orally available compound (B. Baroudy, personal communication). With its single positive charge, TAK-779 differs chemically from all the known CXCR4 inhibitors (BOX 1). This difference probably reflects the surface charges of the two co-receptors — whereas the surface of CXCR4 is strongly anionic, CCR5 has an almost neutral surface, at least before post-translational modifications (FIG. 3). The difference in the physicochemical properties of the two co-receptors also, no doubt, affects the types of compounds that are identified as inhibitors in screens of compound libraries. Highly polar compounds such as T22, ALX40-4C and AMD3100 often have limited bioavailability, which restricts their potential as drugs. This problem will be exacerbated for ALX40-4C and T22 (and its derivatives T134 and T140) because they are peptides46,47,49.

Box 1 | Inhibitor binding sites on CXCR4 and CCR5 The chemical differences between the CXCR4- and CCR5-specific inhibitors are reflected in their binding sites. T22, ALX40-4C and AMD3100 bind mainly to the extracellular domain of CXCR4, especially to the second extracellular loop, and anionic residues on the receptor surface are particularly important97,98. These compounds are not agonists; they antagonize signalling using SDF-1α (REFS 45–48). Each is specific for CXCR4, perhaps owing to the unique nature of CXCR4’s extracellular surface. The binding site for TAK-779 on CCR5 is different — it includes a pocket formed between four of the seven transmembrane helices99. The figure depicts the seven transmembrane helices, viewed from the perspective of an incoming virus. Amino acids where alanine substitutions impair the anti-HIV-1 activity of TAK-779 to either a large or a small extent are highlighted in red and orange, respectively99. Although TAK779 may also contact unidentified residues in the extracellular region, its bulk intercalates within the transmembrane region of CCR5. Once in place, TAK-779 prevents gp120 and chemokines from binding to CCR5, thereby inhibiting transmembrane signalling without itself activating signals or causing CCR5 downregulation51,99. The potential of the transmembrane region in chemokine receptors as an antiviral target is also shown by the inhibition of HIV-1 entry through CXCR4 that is caused by peptide mimics of the transmembrane helices100. The CCR5 binding pocket for TAK-779 is probably not unique — there could be similar pockets in other chemokine receptors, which would be attractive targets for drug development. TAK-779 also binds to CCR2, and can block entry of simian immunodeficiency virus through this co-receptor51,99. Perhaps CXCR4 also has a similar pocket for inhibitors, but these have not been found because compound library screens are dominated by charged compounds that bind electrostatically to the extracellular loops of CXCR4. (Figure prepared by T. P. Samar.)

Although AMD3100 has negligible oral bioavailability, when delivered intravenously or intraperitoneally by infusion pumps it showed antiviral activity in a mouse model, without acute toxicity30. AMD3100 is now being evaluated in human clinical trials (again using non-oral delivery mechanisms). A distamycin analogue, NSC 651016, apparently blocks the interactions of chemokines with many receptors — including both CXCR4 and CCR5 — and has anti-HIV-1 activity in vitro52. How it achieves this is not known. gp41-mediated fusion. The final stage of fusion is mediated by conformational changes in the trimeric gp41 glycoprotein, which force the amino-terminal fusion peptides into the target-cell membrane. The potential of this process as a drug target was first revealed when peptides derived from gp41 were found to inhibit HIV-1 infection in vitro53,54. These peptides substitute for one or more components of the gp41 trimer, thereby inhibiting the conformational changes and preventing the intermolecular interactions necessary for fusion55. One such peptide, T20, reduces viral load significantly in clinical trials — the first demonstration that a fusion inhibitor works in humans56. Because it is a peptide, T20 was initially developed as an injectable drug and it is now being delivered by the subcutaneous route. A related peptide with increased potency in vitro, T1249, is now in clinical trials. A possible target in gp41 is a cavity of a suitable size and in an appropriate location to act as the binding site for a small-molecule fusion inhibitor57,58 (BOX 2). Protease-resistant, and therefore relatively stable, Damino-acid-based peptides (D-PEPTIDES) that fit into this pocket have been identified, and inhibit membrane fusion in the 10–100-µM range57, 58. Improvements to these peptides seem possible, and they can be used in assays to screen conventional chemical libraries for small molecules able to recognize the gp41 pocket57, 58. A triterpene, betulinic-acid derivative designated RPR103611 inhibits fusion of X4 HIV-1 strains efficiently, although it has limited activity against R5 isolates59. Although there is no direct evidence as to the nature of the RPR103611 binding site, it is thought to recognize gp41 because escape mutants have one or more amino-acid changes in this glycoprotein59. However, this is not certain because the region of gp41 associated with RPR103611 resistance has been implicated in binding to gp120, and an amino-acid change in this region of gp41 creates a mutant virus that is resistant to antibodies against gp120 (REF. 60). Finally, a synthetic D-amino-acid analogue of the fusion peptide sequence at the amino terminus of gp41 can interact with the wild-type fusion peptides and block envelope-glycoprotein-mediated cell fusion at a late stage61. Escape pathways

When entry inhibitors are used in vivo, HIV-1 will escape from them. But which pathways will it use? For the gp41 peptides, such as T20, the most efficient escape pathway has been defined — sequence changes in two www.nature.com/reviews/molcellbio

| OCTOBER 2000 | VOLUME 1

REVIEWS

CAPSID PROTEIN

The main constituent of the viral core.

amino-acid residues, within the T20-binding region of gp41, are enough to confer resistance in vitro by reducing the binding affinity of gp41 for the peptide62. The escape pathway for RPR103611 in vitro involves a few amino-acid changes in gp41 (REF. 59). Negatively charged albumins that bind to gp120 to prevent its association with CD4 can be evaded by mutant viruses with sequence changes in and around the V3 loop region, which lower the strength of the albumin–gp120 interaction15. How HIV-1 might evade a CD4-based compound such as CD4-IgG2 is not clear, although one possibility is to mutate and become CD4-independent but retain the use of a co-receptor for entry. Wild-type immunodeficiency viruses with this property are known43. Many co-receptor-targeted inhibitors have been studied in vitro. Here, several mechanisms for escape are possible, depending on how the experiment is designed. For instance, the escape mutant may continue to use the same co-receptor in an inhibitor-insensitive manner; co-receptor switching may occur (so an R5 virus becomes able to use CXCR4, or vice versa); or an entirely different co-receptor may be used by the escape mutant (FIG. 4). So far, the first mechanism is the most common, whereas the third has not been found in vitro. Any structural alterations in the viral envelope glycoproteins that create a co-receptor inhibitor escape mutant must, of course, allow these proteins to retain their natural resistance to the action of neutralizing antibodies. Otherwise, the atypical sensitivity of the escape virus to the host’s humoral immune response would compromise its ability to persist. Escape-mutant studies have been done on several CXCR4-specific inhibitors. Initial studies with AMD3100 showed that, after 63 passages with increasing doses, the mutant virus had 15 amino-acid changes scattered throughout gp120, indicating that escape from AMD3100 is not easy63. Follow-up studies confirmed

Box 2 | Inhibitor binding sites on gp41 The core structure of the gp41 ectodomain is a target for drugs that inhibit HIV-1 entry. The complex between the N36 and C34 peptides is shown from a side view102. In this model, the N36 helices form an interior, trimeric coiled-coil (blue) that contains three prominent, symmetryrelated hydrophobic pockets, one of which is outlined by a box. Three C34 helices (red) are packed into three highly conserved hydrophobic grooves in an anti-parallel orientation on the surface of the N36 coiledcoil trimer. The amino termini of the N36 helices point towards the top of the figure; those of the C34 helices point towards the bottom. The T20 and T1249 peptides shown in FIG. 2 are free-peptide analogues of the C34 helix. The box contains a close-up of the N36 pocket region, represented as a molecular surface. The labelled C34 residues, shown in stick form, bind within this pocket, an event that can be blocked by small molecule fusion inhibitors 57,58. (Figure prepared by M. Lu.)

that the escape mutant still used CXCR4, but in an AMD3100-insensitive manner64,65. So there may be more than one way for gp120 to contact CXCR4, while still permitting fusion. A similar experiment was done with SDF-1α, with a similar outcome64. There was partial cross-resistance between the AMD3100- and SDF1α-resistant viruses, with about half the amino-acid substitutions in gp120 common to the two escape mutants64. Co-receptor switching was not possible in the above experiments because a CCR5-negative cell line was used. When AMD3100 resistance was selected for in isolates that could use either CCR5 or CXCR4 to enter peripheral blood mononuclear (PBM) cells (which express both co-receptors), R5 viruses rapidly dominated the cultures65. But when viruses able to use only CXCR4 were used, they mutated under drug-selection pressure to acquire CCR5 usage65. Escape-mutant studies with CCR5-specific inhibitors have been more limited. An initial report using RANTES derivatives in an animal model concluded that an R5 virus switched to use CXCR4 (REF. 42). But this may not happen with other CCR5-targeted compounds. For example, escape from a CCR5-specific monoclonal antibody in the same animal model did not involve co-receptor-switching (P. Parren and D. R. Burton, personal communication). An in vitro doseescalation study using MIP-1α and an R5 virus in a CCR5+ CXCR4+ T-cell line found that a four- to sixfold reduction in sensitivity to CC-chemokines occurred after three months without any switch to CXCR4 usage66. Preliminary studies in PBM cells with a smallmolecule inhibitor directed at CCR5 confirm this finding; the escape mutants still use CCR5, but in a druginsensitive manner(J. P. M. et al., unpublished results). Inhibitors of uncoating

Once the virion has fused with the target cell, the poorly understood process of ‘uncoating’ occurs, resulting in the release of viral nucleic acids from the capsid core into the host cell cytoplasm (FIG. 1). Cyclophilin A, the cellular ligand of the immunosuppressive drug cyclosporine A, may participate in uncoating. Cyclophilin A is necessary for HIV-1 replication; it binds to the HIV-1 CAPSID PROTEIN and so is packaged within viral particles67,68. Treatment of infected cells with cyclosporine A results in the production of virions which, despite their normal morphology, are not infectious. Similarly, mutations in capsid protein that inhibit the cyclophilin A interaction impair viral infectivity69. Although the immunosuppressive effects of cyclosporine A preclude its clinical use to treat HIV-1 infection, cyclophilin A may be a target for other inhibitors that block HIV-1 replication at the uncoating stage. Targeting a cellular protein avoids the problem of viral resistance, which limits the usefulness of drugs that bind viral products. However, propagation of HIV1 in the presence of cyclophilin A allows viral variants to evolve that no longer require cyclophilin A for infectivity69. Therefore, small-molecule inhibitors of the cyclophilin A–capsid interaction may not be useful.

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 1 | OCTOBER 2000 | 4 5

REVIEWS

Figure 4 | Escape pathways for a CCR5-targeted inhibitor of HIV-1 entry. Escape pathways that an R5 virus could theoretically use under the selection pressure of a CCR5 inhibitor are shown (the same pathways could also be used by an X4 virus to evade a CXCR4 inhibitor). We assume that several co-receptors are available on the same cell, although in vivo the virus could switch to use a co-receptor on a different cell. In pathway 1, the escape virus still uses CCR5, but binds to a new site on the co-receptor that is not affected by the inhibitor. In pathway 2, the escape virus has switched to use CXCR4, and in pathway 3, to use an unspecified co-receptor ‘X’.

Targets in reverse transcription

PLUS- AND MINUS-STRAND RNA

The polarity of each RNA molecule contained in the virion. LONG TERMINAL REPEAT

A repeated sequence, several hundred base pairs long, found at the two ends of the retroviral genome.

After the core is uncoated, the viral nucleic acids (PLUSAND MINUS-STRAND RNA in the form of a dimer) enter the cytosol, where the viral enzyme reverse transcriptase catalyses the conversion of viral RNA into doublestranded complementary DNA (FIG. 5). Because this reaction is essential in the life cycle of all retroviruses, reverse transcription is one of the most active targets for drug discovery70. One ‘non-traditional’ target in reverse transcriptase is its intrinsic ribonuclease H (RNase H) activity, which degrades the RNA component of the RNA/DNA hybrid intermediates of reverse transcription. Although several compounds that are active against recombinant RNase H in vitro have been described, none has been taken to pre-clinical development71. Viruses propagated in vitro in the presence of a non-nucleoside reverse transcriptase inhibitor, delavirdine, have defective RNase H activity, suggesting that such resistance mutations can indirectly affect the efficiency of RNase H cleavage and compromise viral replication72. Although reverse transcriptase is necessary and sufficient to drive cDNA synthesis from RNA and dNTPs in vitro, viral and cellular cofactors are probably also required in vivo. For example, the virion’s nucleocapsid protein, a product of the gag open reading frame, promotes the packaging of genomic viral RNA during virion assembly73. All retroviral nucleocapsid proteins contain either one or two zinc-fingers (C-X2–C-X4-H-X4-C, where cysteine and histidine residues are coordinated by a zinc cation), which are important for the RNA-packaging activity. Small molecules, such as disulphide benzamides74,75 and pyridinioalkanoyl thiolesters76, displace the zinc ion and impair viral RNA packaging and infectivity. Nucleocapsid proteins may also function during viral reverse transcription77,78, indicating that their

inhibitors may target further steps. Indeed, disulphide benzamides impair both RNA packaging and reverse transcription in vitro78. Pharmacological agents based on cyclosporine A may also target reverse transcription (FIG. 5). This could be useful, because the same drug acting on different stages of viral replication should have an additive overall effect. Lentiviruses such as HIV-1 cannot infect lymphocytes in the ‘resting’ G0 stage of the cell cycle because reverse transcription and nuclear import of viral cDNA are too inefficient79,80. Therefore, cellular factors induced on cell-cycle progression are required for completion of post-entry events. The transcription factor NF-ATc, which is important for expression of the interleukin-2 gene, may also promote reverse transcription of HIV-1 (REF. 81). To be activated, NF-ATc must be dephosphorylated by calcineurin and then translocated to the nucleus82. Inhibition of calcineurin with cyclosporine A prevents HIV-1 reverse transcription, probably by affecting activation of NF-ATc (REF. 81). However, it is unclear whether HIV-1 could evolve an NF-ATc-independent mode of reverse transcription. Targets in nuclear import

On completion of reverse transcription, viral cDNA must be translocated to the host cell nucleus, where it integrates into the host’s DNA. No small molecules have yet been shown to block nuclear import. However, a triple-stranded cDNA intermediate of reverse transcription, also referred to as a ‘DNA flap’, is necessary for this process83, so reverse transcriptase inhibitors might have an adverse effect. In an infected cell, viral cDNA synthesis occurs within the context of a reverse transcription complex comprising viral nucleic acids and virion proteins (FIG. 5). This complex, roughly the size of a ribosome, has to be shuttled into the nucleus by the action of nucleophilic virion proteins. In addition to these proteins, the 99-base-pair DNA flap, generated during the synthesis of plus-strand viral cDNA, promotes the efficiency with which viral cDNA is transported to the nucleus83. Mutations that block formation of the flap during reverse transcription impair nuclear uptake of viral DNA and strongly inhibit viral replication83,84. Knowing that the DNA flap is required for nuclear import may now help the targeting of this step by current reverse transcriptase inhibitors. Targets in integration

The final event of viral entry — integration of cDNA into a host chromosome — is catalysed by the viral integrase enzyme. Integrase remains associated with viral nucleic acids, within the context of a pre-integration complex (FIG. 5), as it translocates to the cell nucleus85. Here, integrase interacts with sequences at the end of the LONG TERMINAL REPEAT86. Integrase first catalyses removal of the terminal dinucleotide from each 3′-end of the viral cDNA, then mediates a strand-transfer reaction to link the 3′-end of the cDNA to cellular DNA86. Mutations that disrupt integrase activity produce virions with an infectivity defect87. But the identification of clinically useful integrase www.nature.com/reviews/molcellbio

| OCTOBER 2000 | VOLUME 1

REVIEWS

Figure 5 | Post-fusion targets. Uncoating involves disassembly of the capsid core and liberation of viral RNA with associated virion proteins into the cytoplasm. The cellular protein cyclophilin A (CypA), which interacts with the capsid, may promote disassembly of the core. Cyclosporine A and non-immunosuppressive analogues such as FK506 prevent interaction of CypA with capsid and impair viral infectivity, perhaps by interfering with uncoating. However, mutations in the capsid alleviate the requirement for CypA and may provide an escape mechanism that would permit viral replication in the presence of CypA inhibitors. Reverse transcription of viral nucleic acids by reverse transcriptase (RT) occurs in the context of a highmolecular-weight complex comprising viral RNA and proteins derived from the virion. The nucleocapsid (NC) protein, which promotes packaging of viral RNA, may also be involved in reverse transcription. Dithiobenzamides, which disrupt zinc-fingers within the nucleocapsid, may have further effects on viral cDNA synthesis. The cellular transcription factor NF-ATc may be a cellular co-factor for viral reverse transcription. NF-ATc activity is inhibited in the presence of cyclosporine A. This indicates that small-molecule inhibitors of CypA–capsid interaction may target reverse transcription as well as uncoating. Nucleophilic virion proteins within the preintegration complex (PIC) direct nuclear import of viral cDNA. Although compounds that block this step have not been identified, the integrase (IN) enzyme may facilitate nuclear translocation of viral cDNA as well as integration. A triple-stranded cDNA intermediate of reverse transcription (DNA flap) also seems to be important for nuclear uptake of viral cDNA. This suggests that inhibitors of reverse transcription may have secondary effects on nuclear uptake of the PIC. Integrase catalyses the integration of viral cDNA with cellular DNA. Compounds containing a diketo-acid moiety inhibit the strand-transfer step of integration. As with the other viral enzymatic functions, amino-acid substitutions within the enzyme cause inhibitor resistance. (MA, matrix protein; CA, capsid protein; NRTI, nucleoside RT inhibitor; NNRTI, nonnucleoside RT inhibitor.)

inhibitors has been slow. Early drug-screening attempts used in vitro integration assays that did not faithfully reconstitute the in vivo action of integrase. Compounds identified in these assays interfered with formation of the integrase or cDNA complex, rather than integrase’s catalytic activity88,89. As a result they were inactive against integrase activity in pre-integration complexes and had no antiviral effect88,89. To reconstitute an integration reaction more faithfully, Hazuda and colleagues90 screened compound libraries using a more sophisticated assay; this involved recombinant integrase pre-assembled on immobilized oligonucleotides. Potent and specific inhibitors identified by this screen block strand-transfer activity and virus infection in vitro90. Although they all contain a diketo-acid moiety, which confers undesirable pharmacological properties on the inhibitors, this moiety is not necessary for inhibition90. So the task ahead is to identify strand-transfer inhibitors that do not contain a diketo-acid group. There is evidence that these new compounds are bona fide integrase inhibitors. First, in the presence of diketo acids, circular forms of HIV-1 DNA accumulated in the nucleus90. This is similar to observations in cells infected with integrase mutants of HIV-1 — circular forms of viral DNA are selectively produced87. Presumably, linear viral cDNA must rapidly integrate before the free cDNA ends recombine or become ligated by cellular ligases to form replication-defective circular forms of cDNA (FIG. 5). Second, viruses cultured in the presence of diketo acids acquired drug resistance through mutations in integrase90. Integrase has also been implicated in promoting the nuclear import of viral cDNA91. Indeed, integrase contains a nuclear-localization signal92. One model is that the nucleophilic activity of integrase helps to chaperone viral cDNA to the nucleus93. The diketo-acid integrase inhibitors do not seem to block nuclear import, as nuclear forms of viral cDNA accumulate in cells in the presence of inhibitor90. Nevertheless, in future screens, compounds that interact with integrase will need to be investigated for further effects on the nuclear translocation of viral cDNA. Unexploited targets

Many steps in the HIV-1 life cycle remain to be dissected. For instance, the RNase H activity of reverse transcriptase has yet to be attacked70. Inhibitor screens have generally used assays that do not distinguish between reverse transcriptase and RNase H activities, although assays reconstituting specific RNase H activity are being developed94. These should be adaptable for large-scale screening of compound libraries for inhibitors that can complement the current reverse transcriptase drugs. Overcoming the lack of convenient screening assays is crucial to finding inhibitors of other key stages in HIV1 replication. For example, the HIV virion infectivity factor (Vif) is essential for replication in PRIMARY CELLS. Vif is thought to block the activity of a negative cellular factor that normally prevents the formation of infectious virions95,96. As this cellular ligand for Vif is unknown, assays that reconstitute Vif activity in vitro are not yet available.

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 1 | OCTOBER 2000 | 4 7

REVIEWS

PRIMARY CELLS

Cells taken directly from an organism, rather than from a cell culture.

An alternative approach is to identify compounds that interact with Vif, then do secondary screening assays for antiviral activity among the hits. Nuclear import of viral nucleic acids is also a potential target for intervention. Nucleophilic virion proteins and reverse-transcription intermediates seem to contribute to nuclear uptake93. The nucleophilic virion proteins may operate through cellular nuclear-import receptors such as importin B. Targeting cellular proteins is usually problematic from the perspective of toxicity, although compounds that inhibit a protein–protein interaction by binding exclusively to the viral partner may have antiviral activity without toxic effects. Regardless of the target or inhibitor screening strategy, experience with the development of integrase inhibitors has taught us that reconstitution of biologically relevant activities should drive the design of inhibitor screens. Otherwise, irrelevant compounds will be selected, and compounds with antiviral activity will be missed. Conclusions and perspective

“Needs must where the devil drives.” The need for new inhibitors of HIV-1 replication should not be underestimated, and neither should the ability of HIV-1 to evade antiviral therapies. But the more stages of the viral life cycles that we can antagonize, the less likely that enough viral replication can occur to generate escape mutants. In other words, more components are needed for effecLinks DATABASE LINKS DC-SIGN | ICAM-1 | CD4 | NF-ATc | Cyclophilin A | CCR5 | CXCR4 |

Palella, F. J. Jr et al. Declining morbidity and mortality, among patients with advanced human immunodeficiency virus infection. N. Engl. J. Med. 338, 853–860 (1998). 2. Lucas, G. M., Chaisson, R. E. & Moore, R. D. Highly active antiretroviral therapy in a large urban clinic: Risk factors for virologic failure and adverse drug reactions. Ann. Intern. Med. 131, 81–87 (1999). 3. Yerly, S. et al. Transmission of antiretroviral-drug-resistant HIV-1 variants. Lancet 354, 729–733 (1999). 4. Furtado, M. R. et al. Persistence of HIV-1 transcription in peripheral blood mononuclear cells in patients receiving potent antiretroviral therapy. N. Engl. J. Med. 340, 1614–1622 (1999). 5. Van Damme, L. & Rosenberg, Z. F. Microbicides and barrier methods in HIV prevention. AIDS 13, S85–S92 (1999). 6. Mondor, I., Ugolini, S. & Sattentau, Q. J. Human immunodeficiency virus type 1 attachment to HeLa CD4 cells is CD4-independent and gp120 dependent and requires cell surface heparans. J. Virol. 72, 3623–3634 (1998). 7. Moulard, M. et al. Selective interactions of polyanions with basic surfaces on human immunodeficiency virus type 1 gp120. J. Virol. 74, 1948–1960 (2000). 8. Geijtenbeek, T. B. H. et al. DC SIGN, a dendritic cell specific HIV-1-binding protein that enhances transinfection of T cells. Cell 100, 575–585 (2000). 9. Liao, Z., Roos, J. W. & Hildreth, J. E. K. Increased infectivity of HIV type 1 particles bound to cell surface and solid-phase ICAM-1 and VCAM-1 through acquired adhesion molecules LFA-1 and VLA-4. AIDS Res. Hum. Retroviruses 16, 355–366 (2000). 10. Esser, M. T. et al. Cyanovirin-N binds to gp120 to interfere with CD4 dependent human immunodeficiency virus type 1 virion binding, fusion, and infectivity but does not affect

11.

12.

13.

14.

15.

16.

17.

18.

19.

tive antiviral cocktails, because the probability that any single inhibitor will be the ‘magic bullet’ is, in practice, zero. But prospects for the development of new inhibitors are real. Antagonists of viral entry, directed at the gp120 (CD4-IgG2) or gp41 (T20, T1249) glycoproteins or the CCR5 and CXCR4 co-receptors, are already in (or rapidly approaching) human clinical trials. Injectable CCR5 inhibitors, such as the monoclonal antibody PRO 140, may also soon enter the clinic, and an injectable CXCR4 blocker, AMD3100, is already there, as is the orally available CCR5 inhibitor, Sch-C. The various entry inhibitors may be especially useful in combination — they could act synergistically. But the development of post-entry inhibitors is less certain. For example, the diketo-acid moiety of integrase inhibitors will have to be eliminated from the next generation of compounds. RNase H also warrants closer attention, although convenient in vitro RNase H assays need to be developed first. Inhibition of accessory proteins is hindered by our poor understanding of their functions. The Vif protein is probably the most attractive target, given its obligatory function in viral replication. Identification of cellular ligands for Vif as well as the other accessory proteins will aid the design of inhibitor screening assays. A discussion of other targets — including the regulatory proteins Tat and Rev, which mediate later stages in the viral replication cycle — is outside the scope of this review. Tat and Rev have not received much attention as drug targets, probably because it is hard to recreate their activities in vitro. Nevertheless, cellular ligands for Tat and Rev have been identified, and this should help in designing screening assays. The coming years should, therefore, see many more of HIV-1’s vulnerabilities exploited successfully.

the CD4 binding site on gp120 or soluble CD4-induced conformational changes in gp120. J. Virol. 73, 4360–4371 (1999). Delezay, O., Hammache, D., Fantini, J. & Yahi, N. SPC3, a V3 loop-derived synthetic peptide inhibitor of HIV-1 infection, binds to cell surface glycosphingolipids. Biochemistry 35, 15663–15671 (1996). Kwong, P. D. et al. Structure of an HIV gp120 envelope glycoprotein complex with the CD4 receptor and a neutralizing human antibody. Nature 393, 648–659 (1998). Zhang, J. L. et al. The bis-azo compound FP 21399 inhibits HIV-1 replication by preventing viral entry. Virology 244, 530–541 (1998). Bugelski, P. J., Ellens, H., Hart, T. K. & Kirsh, R. L. Soluble CD4 and dextran sulphate mediate release of gp120 from HIV-1: Implications for clinical trials. J. Acq. Immun. Def. Synd. 4, 923–924 (1991). Cabrera, C. et al. Resistance of the human immunodeficiency virus to the inhibitory action of negatively charged albumins on virus binding to CD4. AIDS Res. Hum. Retroviruses 15, 1535–1543 (1999). Jacobson, J. M. et al. Single-dose safety, pharmacology and antiviral activity of the human immunodeficiency virus (HIV) type 1 entry inhibitor PRO 542 in HIV-infected adults. J. Infect. Dis. 182, 326–329 (2000). Zhang, W. et al. Conformational changes of gp120 in epitopes near the CCR5 binding site are induced by CD4 and a CD4 miniprotein mimetic. Biochemistry 38, 9405–9416 (1999). Vita, C. et al. Rational engineering of a miniprotein that reproduces the core of the CD4 site interacting with HIV-1 envelope glycoprotein. Proc. Natl Acad. Sci. USA 96, 13091–13096 (1999). Reimann, K. A. et al. A humanized form of a CD4-specific monoclonal antibody exhibits decreased antigenicity and

20. 21.

22.

23.

24. 25.

26.

27.

28.

29.

prolonged plasma half-life in rhesus monkeys while retaining its unique biological and antiviral properties. AIDS Res. Hum. Retroviruses 13, 933–943 (1997). Berger, E. A. HIV entry and tropism: the chemokine receptor connection. AIDS 11, S3–S16 (1997). Michael, N. L. et al. Exclusive and persistent use of the entry coreceptor CXCR4 by human immunodeficiency virus type 1 from a subject homozygous for CCR5∆ 32. J. Virol. 72, 6040–6047 (1998). Zhang, Y.-J. & Moore, J. P. Will multiple coreceptors need to be targeted by inhibitors of human immunodeficiency virus type 1? J. Virol. 73, 3443–3448 (1999). Glushakova, S. et al. Preferential coreceptor utilization and cytopathicity by dual-tropic HIV-1 in human lymphoid tissue ex vivo. J. Clin. Invest. 104, R7–R11 (1999). Berger, E. A. et al. A new classification for HIV–1. Nature 391, 240 (1998). Veazey, R. S. et al. Identifying the target cell in primary simian immunodeficiency virus (SIV) infection: highly activated memory CD4(+) T cells are rapidly eliminated in early SIV infection in vivo. J. Virol. 74, 57–64 (2000). Blaak, H. et al. In vivo HIV-1 infection of CD45RA+CD4+ T cells is established primarily by syncytium-inducing variants and correlates with the rate of CD4+ T cell decline. Proc. Natl Acad. Sci. USA 97, 1269–1274 (2000). Dean, M. et al. Genetic restriction of HIV-1 infection and progression to AIDS by a deletion of the CKR5 structural allele. Science 273, 1856–1862 (1996). Wu, L. et al. CCR5 levels and expression pattern correlate with infectability by macrophage tropic HIV-1, in vitro. J. Exp. Med. 185, 1681–1691 (1997). Zou, Y. R., Kottmann, A. H., Kuroda, M., Taniuchi, I. & Littman, D. R. Function of the chemokine receptor CXCR4 in haematopoiesis and in cerebellar development. Nature 393, 595–599 (1998).

www.nature.com/reviews/molcellbio

| OCTOBER 2000 | VOLUME 1

REVIEWS 30. Datema, R. et al. Antiviral efficacy in vivo of the anti-human immunodeficiency virus bicyclam SDZ SID 791 (JM 3100), an inhibitor of infectious cell entry. Antimicrob. Agents Chemother. 40, 750–754 (1996). 31. Cocchi, F. et al. Identification of RANTES, MIP-1 alpha and MIP-1 beta as the major HIV suppressive factors produced by CD8+ T cells. Science 270, 1811–1815 (1995). 32. Bleul, C. C. et al. The lymphocyte chemoattractant SDF-1 is a ligand for LESTR/fusin and blocks HIV-1 entry. Nature 382, 829–833 (1996). 33. Oberlin, E. et al. The CXC chemokine SDF-1 is the ligand for LESTR/fusin and prevents infection by T-cell-line adapted HIV-1. Nature 382, 833–835 (1996). 34. Simmons, G. et al. Potent inhibition of HIV-1 infectivity in macrophages and lymphocytes by a novel CCR5 antagonist. Science 276, 276–279 (1997). 35. Heveker, N. et al. Dissociation of the signalling and antiviral properties of SDF-1-derived small peptides. Curr. Biol. 8, 369–376 (1998). 36. Amara, A. et al. HIV coreceptor downregulation as antiviral principle: SDF-1α-dependent internalization of the chemokine receptor CXCR4 contributes to inhibition of HIV replication. J. Exp. Med. 186, 139–146 (1997). 37. Mack, M. et al. Aminooxypentane-RANTES induces CCR5 internalization but inhibits recycling: A novel inhibitory mechanism of HIV infectivity. J. Exp. Med. 187, 1215–1224 (1998). 38. Trkola, A. et al. Genetic subtype independent inhibition of human immunodeficiency virus type 1 replication by CC and CXC chemokines. J. Virol. 72, 396–404 (1998). 39. Kinter, A. et al. CC-chemokines enhance the replication of T tropic strains of HIV-1 in CD4+ T cells: Role of signal transduction. Proc. Natl Acad. Sci. USA 95, 11880–11885 (1998). 40. Marechal, V., Arenzana-Seisdedos, F., Heard, J.-M. & Schwartz, O. Opposite effects of SDF-1 on human immunodeficiency virus type 1 replication? J. Virol. 73, 3608–3615 (1999). 41. Trkola, A. et al. The CC chemokine RANTES increases the attachment of human immunodeficiency virus type 1 to target cells via glycosoaminoglycans, and also activates a signal transduction pathway that enhances viral infectivity. J. Virol. 73, 6370–6379 (1999). 42. Mosier, D. E. et al. Highly potent RANTES analogues either prevent CCR5-using human immunodeficiency virus type 1 infection in vivo or rapidly select for CXCR4-using variants. J. Virol. 73, 3544–3550 (1999). 43. Endres, M. J. et al. CD4-independent infection by HIV-2 is mediated by fusin/CXCR4. Cell 87, 745–756 (1996). 44. Olson, W. C. et al. Differential inhibition of human immunodeficiency virus type 1 fusion, gp120 binding, and CC chemokine activity by monoclonal antibodies to CCR5. J. Virol. 73, 4145–4155 (1999). 45. Schols, D. et al. Inhibition of T-tropic HIV strains by selective antagonization of the chemokine receptor CXCR4. J. Exp. Med. 186, 1383–1388 (1997). 46. Murakami, T. et al. A small molecule CXCR4 inhibitor that blocks T cell line-tropic HIV-1 infection. J. Exp. Med. 186, 1389–1393 (1997). 47. Doranz, B. J. et al. A small-molecule inhibitor directed against the chemokine receptor CXCR4 prevents its use as an HIV-1 coreceptor. J. Exp. Med. 186, 1395–1400 (1997). 48. Donzella, G. A. et al. AMD3100, a small molecule inhibitor of HIV-1 entry via the CXCR4 co-receptor. Nature Med. 4, 72–77 (1998). 49. Arakaki, R. et al. T134, a small-molecule CXCR4 inhibitor, has no cross drug resistance with AMD3100, a CXCR4 antagonist with a different structure. J. Virol. 73, 1719–1723 (1999). 50. Sakaida, H. et al. T-tropic human immunodeficiency virus type 1 (HIV-1) derived V3 loop peptides directly bind to CXCR-4 and inhibit T-tropic HIV-1 infection. J. Virol. 72, 9763–9770 (1998). 51. Baba, M. et al. A small molecule nonpeptide CCR5 antagonist with highly potent and selective anti-HIV-1 activity. Proc. Natl Acad. Sci. USA 96, 5698–5703 (1999). 52. Howard, O. M. Z. et al. Inhibition of in vitro and in vivo HIV replication by a distamycin analogue that interferes with chemokine receptor function: A candidate for chemotherapeutic and microbicidal application. J. Med. Chem. 41, 2184–2193 (1998). 53. Wild, C., Oas, T., McDanal, C., Bolognesi, D. & Matthews, T. A synthetic peptide inhibitor of human immunodeficiency virus replication: Correlation between solution structure and viral inhibition. Proc. Natl Acad. Sci. USA 89, 10537–10541 (1992). 54. Jiang, S., Lin, K., Strick, N. & Neurath, A. R. HIV–1 inhibition by a peptide. Nature 365, 113 (1993). 55. Jones, P. L., Korte, T. & Blumenthal, R. Conformational changes in cell surface HIV-1 envelope glycoproteins are triggered by cooperation between cell surface CD4 and co-receptors. J. Biol. Chem. 273, 404–409 (1998). 56. Kilby, J. M. et al. Potent suppression of HIV-1 replication in humans by T-20, a peptide inhibitor of gp41-mediated

virus entry. Nature Med. 4, 1302–1307 (1998). 57. Ferrer, M. et al. Selection of gp41-mediated HIV-1 cell entry inhibition from biased combinational libraries of nonnatural binding elements. Nature Struct. Biol. 6, 953–959 (1999). 58. Eckert, D. M., Malashkevich, V. N., Hong, L. H., Carr, P. A. & Kim, P. S. Inhibiting HIV-1 entry: Discovery of D-peptide inhibitors that target the gp41 coiled coil pocket. Cell 99, 103–115 (1999). 59. Labrosse, B., Treboute, C. & Alizon, M. Sensitivity to a nonpeptidic compound (RPR103611) blocking human immunodeficiency virus type 1 env-mediated fusion depends on the sequence and accessibility of the gp41 loop region. J. Virol. 74, 2142–2150 (2000). 60. Wilson, C. et al. The site of an immune-selected point mutation in the transmembrane protein of human immunodeficiency virus type 1 does not constitute the neutralization epitope. J. Virol. 64, 3240–3248 (1990). 61. Pritsker, M., Jones, P., Blumenthal, R. & Shai, Y. A synthetic all D-amino acid peptide corresponding to the Nterminal sequence of HIV-1 gp41 recognizes the wild-type fusion peptide in the membrane and inhibits HIV 1 envelope glycoprotein-mediated cell fusion. Proc. Natl Acad. Sci. USA 95, 7287–7292 (1998). 62. Rimsky, L. T., Shugars, D. C. & Matthews, T. J. Determinants of human immunodeficiency virus type 1 resistance to gp41-derived inhibitory peptides. J. Virol. 72, 986–993 (1998). 63. De Vreese, K. et al. The molecular target of bicyclams, potent inhibitors of human immunodeficiency virus replication. J. Virol. 70, 689–696 (1996). 64. Schols, D., Esté, J. A., Cabrera, C. & De Clercq, E. T-cellline tropic human immunodeficiency virus type 1 that is made resistant to stromal cell-derived factor 1α contains mutations in the envelope gp120 but does not show a switch in coreceptor use. J. Virol. 72, 4032–4037 (1998). 65. Esté, J. A. et al. Shift of clinical human immunodeficiency virus type 1 isolates from X4 to X5 and prevention of emergence of the syncytium-inducing phenotype by blockade of CXCR4. J. Virol. 73, 5577–5585 (1999). 66. Maeda, Y., Foda, M., Matsushita, S. & Harada, S. Involvement of both the V2 and V3 regions of the CCR5tropic human immunodeficiency virus type 1 envelope in reduced sensitivity to macrophage inflammatory protein 1alpha. J. Virol. 74, 1787–1793 (2000). 67. Franke, E. K., Yuan, H. E. H. & Luban, J. Specific incorporation of cyclophilin A into HIV-1 virions. Nature 372, 359–362 (1994). 68. Thali, M. et al. Functional association of cyclophilin A with HIV-1 virions. Nature 372, 363–365 (1994). 69. Braaten, D. et al. Cyclosporine A-resistant human immunodeficiency virus type 1 mutants demonstrate that gag encodes functional target of cyclophilin A. J. Virol. 70, 5170–5176 (1996). 70. Montaner, J. S., Montessori, V., Harrigan, R., O’Shaughnessy, M. & Hogg, R. Antiretroviral therapy: ‘the state of the art’. Biomed. Pharmacother. 53, 63–72 (1999). 71. Gerondelis, P. et al. The P236L delavirdine-resistant human immunodeficiency virus type 1 mutant is replication defective and demonstrates alterations in both RNA 5′end- and DNA 3′-end-directed RNase H activities. J. Virol. 73, 5803–5813 (1999). 72. Darlix, J. L., Lapadat-Tapolsky, M., de Rocquigny, H. & Roques, B. P. First glimpses at structure-function relationships of the nucleocapsid protein of retroviruses. J. Mol. Biol. 254, 523–537 (1995). 73. Tummino, P. J. et al. The in vitro ejection of zinc from human immunodeficiency virus (HIV) type 1 nucleocapsid protein by disulfide benzamides with cellular anti- HIV activity. Proc. Natl Acad. Sci. USA 93, 969–973 (1996). 74. Rice, W. G. et al. Inhibitors of HIV nucleocapsid protein zinc fingers as candidates for the treatment of AIDS. Science 270, 1194–1197 (1995). 75. Turpin, J. A. et al. Synthesis and biological properties of novel pyridinioalkanoyl thiolesters (PATE) as anti-HIV-1 agents that target the viral nucleocapsid protein zinc fingers. J. Med. Chem. 42, 67–86 (1999). 76. Tanchou, V. et al. Role of the N-terminal zinc finger of human immunodeficiency virus type 1 nucleocapsid protein in virus structure and replication. J. Virol. 72, 4442–4447 (1998). 77. Carteau, S., Gorelick, R. J. & Bushman, F. D. Coupled integration of human immunodeficiency virus type 1 cDNA ends by purified integrase in vitro: stimulation by the viral nucleocapsid protein. J. Virol. 73, 6670–6679 (1999). 78. Berthoux, L., Pechoux, C. & Darlix, J. L. Multiple effects of an anti-human immunodeficiency virus nucleocapsid inhibitor on virus morphology and replication. J. Virol. 73, 10000–10009 (1999). 79. Zack, J. A. et al. HIV-1 entry into quiescent primary lymphocytes: molecular analysis reveals a labile, latent viral structure. Cell 61, 213–222 (1990). 80. Stevenson, M., Stanwick, T. L., Dempsey, M. P. & Lamonica, C. A. HIV-1 replication is controlled at the level of T cell activation and proviral integration. EMBO J. 9,

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

1551–1560 (1990). 81. Kinoshita, S., Chen, B. K., Kaneshima, H. & Nolan, G. P. Host control of HIV-1 parasitism in T cells by the nuclear factor of activated T cells. Cell 95, 595–604 (1998). 82. Flanagan, W. M., Corthesy, B., Bram, R. J. & Crabtree, G. R. Nuclear association of a T-cell transcription factor blocked by FK-506 and cyclosporine A. Nature 352, 803–807 (1991). 83. Zennou, V. et al. HIV-1 genome nuclear import is mediated by a central DNA flap. Cell 101, 173–185 (2000). 84. Charneau, P. et al. HIV-1 reverse transcription. A termination step at the center of the genome. J. Mol. Biol. 241, 651–662 (1994). 85. Bukrinsky, M. I. et al. Association of integrase, matrix, and reverse transcriptase antigens of human immunodeficiency virus type 1 with viral nucleic acids following acute infection. Proc. Natl Acad. Sci. USA 90, 6125–6129 (1993). 86. Hansen, M. S., Carteau, S., Hoffmann, C., Li, L. & Bushman, F. Retroviral cDNA integration: mechanism, applications and inhibition. Genet. Eng. 20, 41–61 (1998). 87. Wiskerchen, M. & Muesing, M. A. Human immunodeficiency virus type 1 integrase: effects of mutations on viral ability to integrate, direct viral gene expression from unintegrated viral DNA templates, and sustain viral propagation in primary cells. J. Virol. 69, 376–386 (1995). 88. Hazuda, D. et al. Equivalent inhibition of half-site and fullsite retroviral strand transfer reactions by structurally diverse compounds. J. Virol. 71, 807–811 (1997). 89. Neamati, N. et al. Salicylhydrazine-containing inhibitors of HIV-1 integrase: implication for a selective chelation in the integrase active site. J. Med. Chem. 41, 3202–3209 (1998). 90. Hazuda, D. J. et al. Inhibitors of strand transfer that prevent integration and inhibit HIV-1 replication in cells. Science 287, 646–650 (2000). 91. Gallay, P., Hope, T., Chin, D. & Trono, D. HIV-1 infection of nondividing cells through the recognition of integrase by the importin/karyopherin pathway. Proc. Natl Acad. Sci. USA 94, 9825–9830 (1997). 92. Pluymers, W., Cherepanov, P., Schols, D., De Clercq, E. & Debyser, Z. Nuclear localization of human immunodeficiency virus type 1 integrase expressed as a fusion protein with green fluorescent protein. Virology 258, 327–332 (1999). 93. Stevenson, M. HIV nuclear import: What’s all the flap? Nature Med. 6, 9–10 (2000). 94. Rausch, J. W., Sathyanaryana, B. K., Bona, M. K. & Le Grice, S. F. Probing contacts between the ribonuclease H domain of HIV-1 reverse transcriptase and nucleic acid by site specific photocrosslinking. J. Biol. Chem. 3, 16015–16022 (2000). 95. Simon, J. H., Gaddis, N. C., Fouchier, R. A. & Malim, M. H. Evidence for a newly discovered cellular anti-HIV phenotype. Nature Med. 4, 1397–1400 (1998). 96. Madani, N. & Kabat, D. An endogenous inhibitor of human immunodeficiency virus in human lymphocytes is overcome by the viral Vif protein. J. Virol. 72, 10251–10255 (1998). 97. Labrosse, B. et al. Determinants for sensitivity of human immunodeficiency virus coreceptor CXCR4 to the bicyclam AMD3100. J. Virol. 72, 6381–6388 (1998). 98. Murakami, T. et al. Inhibitory mechanism of the CXCR4 antagonist T22 against human immunodeficiency virus type 1 infection. J. Virol. 73, 7489–7496 (1999). 99. Dragic, T. et al. A binding pocket for a small molecule inhibitor of HIV-1 entry within the transmembrane helices of CCR5. Proc. Natl Acad. Sci. USA 97, 5639–5644 (2000). 100. Tarasova, N. I., Rice, W. G. & Michejda, C. J. Inhibition of G–protein coupled receptor functions by disruption of transmembrane domain interactions. J. Biol. Chem. 274, 34911–34915 (1999). 101. Weissenhorn, W., Dessen, A., Harrison, S. C., Skehel, J. J. & Wiley, D. C. Atomic structure of the ectodomain from gp41. Nature 387, 426–430 (1997). 102. Chan, D. C., Fass, D., Berger, J. M. & Kim, P. S. Core structure of gp41 from the HIV envelope glycoprotein. Cell 89, 263–273 (1997). 103. Caffrey, M. et al. Three-dimensional solution structure of the 44 kDa ectodomain of SIV gp41. EMBO J. 17, 4572–4584 (1998). 104. Farzan, M. et al. Tyrosine sulfation of the amino terminus of CCR5 facilitates HIV-1 entry. Cell 96, 667–676 (1999). 105. Moulard, M. et al. Selective interactions of polyanions with basic surfaces on human immunodeficiency virus type 1 gp120. J. Virol. 74, 1948–1960 (2000).

Acknowledgements The authors thank B. Mellor for preparing the illustrations, and T. P. Sakmar, R. W. Doms, M. S. Dimitrov, S. Durell and M. Lu for making scientific contributions to the figures. The authors’ work in this area is funded by NIH grants and by the Pediatric AIDS Foundation (J.P.M.).

VOLUME 1 | OCTOBER 2000 | 4 9

REVIEWS

WALKING ON TWO HEADS: THE MANY TALENTS OF KINESIN Günther Woehlke and Manfred Schliwa The gallop of a race horse and the minute excursions of a cellular vesicle have one thing in common: they are based on the directional movement of proteins termed molecular motors — many trillions in the case of the horse, just a few in the case of the cell vesicle. These tiny machines take nanometre steps on a millisecond timescale to drive all biological movements. Over the past 15 years new biochemical and biophysical approaches have allowed us to take a giant step forward in understanding the molecular basis of motor mechanics.

Adolf-Butenandt-Institut, Zellbiologie, University of Munich, Schillerstrasse 42, 80336 Munich, Germany. e-mails: guenther.woehlke@lrz.unimuenchen.de; schliwa@bio.med.unimuenchen.de

“Where there’s life, there’s movement.” This catch phrase, which could just as well be reversed, epitomizes the fundamental realization that movement is one of the most characteristic features of life. Most biological movements are accomplished by ingenious protein machines termed molecular motors1. Some motors may occur in large ensembles as in our skeletal muscles, whereas others may operate as single molecules. Some undergo linear motion along a substrate, and others rotate about their axis (BOX 1). Some of the linear motors drive subcellular transport ranging from just a few micrometres up to several metres in certain neurons of large animals. Notwithstanding their diversity, all molecular motors have in common the fact that they undergo energy-dependent conformational changes that result in unidirectional movement. Among the best-studied molecular motors are those that use cytoskeletal fibres as a track. Three classes of cytoskeletal motors are known: myosin, which interacts with actin filaments, and two types of microtubule motors, dynein and kinesin (FIG. 1). All cytoskeletal motors possess a catalytic motor domain, also referred to as the ‘head’, characterized by the presence of two binding sites, one for ATP and one for the track. This domain is surprisingly small in the kinesins (about 350 amino acids), of intermediate size in the myosins (about 800 amino acids), and large in the dyneins (over 4,000 amino acids). Outside the motor domain the

Figure 1 | Overview of three molecular motor ‘prototypes’. The actin-based motor skeletal muscle myosin in the centre is flanked by the microtubule motors conventional kinesin on the left and cytoplasmic dynein on the right. All three motors consist of a dimer of two heavy chains whose catalytic domains are shown in yellow, whereas the stalks, which form extended coiled-coils in both myosin and kinesin, are shown in blue. Associated polypeptides (four light chains in skeletal muscle myosin, two light chains in conventional kinesin, and a complex set of intermediate, light-intermediate and light chains in dynein) are shown in purple. The ‘antennae’ extending from the dynein heads contain the microtubule binding site, which in myosin and kinesin is part of the compact head. (Drawn roughly to scale.)

www.nature.com/reviews/molcellbio

| OCTOBER 2000 | VOLUME 1

REVIEWS

Box 1 | A motor by any other name... Many protein machines undergo ordered conformational changes to execute vectorial processes. Ion pumps, translocation pores that move proteins across membranes, ribosomes, DNA helicases, nuclear pores — in a sense, they all are motors. Even though the design principles differ and certainly do not suggest a common origin of these machines, some properties may be shared by otherwise distinct devices. Therefore conventional kinesins and many DNA or RNA polymerases have in common the ability to move along their respective tracks (microtubules or DNA) for long distances without dissociating, taking hundreds or even thousands of ‘steps’ in the process38,80,83–85. This characteristic feature is termed ‘processivity’ (see main text). In the case of polymerases, contact with the track is facilitated by protrusions that clamp around the DNA strand 86,87 (see the review by Hingorani and O’Donnell on page 22 of this issue). A different type of movement results when the molecules that constitute the ‘track’ are arranged in a circular fashion and the motor rotates in the centre. Such is the case in ATP synthase, where a single γ-subunit rotates against a surrounding cylinder of three α- and three β-subunits, a process driven by a proton gradient88. This is the smallest known ‘rotary’ motor. A larger but even more remarkable rotary machine is the bacterial flagellar motor, where a corkscrew-like flagellum is attached to a rotating ring-shaped assembly inserted into the membrane. This complex is powered by proton or sodium gradients, and although it is only composed of about 20 proteins, it can rotate at rates exceeding 1,000 Hz, propelling a bacterium at speeds of several hundred micrometres per second89,90. A ‘motor’ of completely different design is used by certain intracellular pathogenic bacteria or viruses. They exploit the complex cellular machinery normally used for lamellipodial extension during cell migration and modify it to create an intracellular ‘rocket propulsion’ system91. By nucleating the assembly of a network of actin filaments and actin-associated proteins near their surface to generate a cushion of crosslinked fibres, the intruder is pushed through the cytoplasm92,93. This machinery differs from the other motors described here in that it is composed of a massive three-dimensional cytoskeletal network rather than a compact macromolecule.

P-LOOP-TYPE ATP-BINDING SITE

‘Phosphate-binding loop’; a nucleotide-binding consensus motif (GXXXXGKT/S) at the ATP-binding site.

three classes of motors differ considerably, suggesting that functional diversity is in part embodied in the nonmotor domains. The discovery of numerous motors over the past 10–20 years has led to the realization that all three motors constitute superfamilies with dozens of members. However, functional characterization has not kept up with the pace of discovery and, as a result, many are known only as a piece of sequence or a twig on a phylogenetic tree. To appreciate the diversity of motors, consider the fact that a mammalian organism harbours at

least 40 kinesin motors2,3, probably just as many myosins, and at least half a dozen dyneins. Consider, in addition, that some of these motors may have associated polypeptides that help specify function. For example, animal conventional kinesins are associated with light chains, of which several isoforms are known4,5, leading to further functional diversification. If a similar scenario applies to other types of motors, eukaryotic cells may easily harbour hundreds of motor complexes with highly specific functions. In this review we discuss the molecular basis of motor protein function, using kinesin as a model. We centre on two of the most fascinating properties of kinesins: the capacity to decide which way to move along the track, and the ability to move long distances without dissociating (termed processivity). Wherever possible, we include comparisons with other cytoskeletal motors to emphasize common principles of action. Different makes, same engine

On the basis of phylogenetic analysis of the motor domain, the kinesin superfamily comprises at least ten families, some of which can even be divided into distinct subfamilies (FIG. 2). Many motors cannot be assigned to any of the existing groups and therefore are referred to as ‘orphans’6,7. New members are still being found, but with the advances in genome sequencing projects, the pace of discovery of new motors is slowing down. The common denominator of kinesin motors is the catalytic motor domain, which shows at least 35% sequence identity among all the kinesins found so far. It possesses a P-LOOP-TYPE ATP-BINDING SITE and a number of signature sequences that are found only in kinesins8,9. Some of the latter are now known to be responsible for the interaction with microtubules, and others are of unassigned function. As also shown in FIG. 2, the size of kinesin motors may differ widely, even among members of the same family, whereas other families are more conserved in size. Likewise, the speed of motors may vary considerably, indicating a high degree of functional adaptation.

Figure 2 | Overview of the domain organization, heavy chain molecular weight, polarity of movement and velocity of the main kinesin families. The abbreviations of the diverse families, which are named after certain prototypes of motors characteristic of each family, correspond to the nomenclature of the kinesin home page. The conserved motor domain is shown in yellow, domains that include coiled-coil segments are shown in blue, and predominantly globular domains with familyspecific and presumably varied functions are shown in purple. The velocity of MCAK, which actually may be a microtubule depolymerizer rather than a motor, is controversial. Only one example of an ‘orphan’ kinesin (Smy1) is shown. (For further characteristics of the kinesin families see the kinesin home page.) (ND, not determined).

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 1 | OCTOBER 2000 | 5 1

REVIEWS The regions outside the motor domain are family specific and share little, if any, sequence homology. They often include one or more segments predicted to form a coiled-coil that may facilitate oligomerization. The diversity of these non-motor regions correlates with the range of biological functions for kinesin motors, which include not only the transport of membrane-bound organelles, protein assemblies and messenger RNA, but also cell division, chemosensory and signal transduction functions, microtubule dynamics, neuronal plasticity and embryonic development (for recent reviews, see REFS 3,10). Interaction with their respective cargoes is probably mediated through the non-motor domains but, with few exceptions, little is known about these interactors and how they link up with the motor. Getting started

Members of the conventional kinesin family have been extensively used for the study of motor mechanisms. Conventional kinesin derives its designation from the fact that it was the first kinesin to be identified and purified from cell extracts. It has developed into a ‘gold

standard’ with which the properties of other kinesins, discovered later, are compared. Conventional kinesin is a homodimer of two heavy chains, each of which possesses an amino-terminal motor domain, a long (60–80 nm) stalk with alternating flexible and coiled-coil segments, and a small globular tail domain (FIG. 3). The atomic structure of the kinesin motor domain11 reveals a completely unexpected structural homology with myosin and small G-proteins, which are also P-loop nucleotidases. This suggests that, in these three classes of proteins, the catalytic core has a similar function as an engine that exploits the energy of ATP hydrolysis to drive a conformational change. In kinesin motors this is controlled by the interaction with the microtubule. Free in solution, the ATPase is inactive and the molecule rests in its ADP-bound form. The hydrolytic cycle can only be initiated upon binding to the microtubule, a process resembling the activation of myosin by actin or the regulation of G-proteins by activators and exchange factors12,13. However, if it were for this feature alone, the catalytic domain would be no more than an allosteric enzyme.

Box 2 | Force generation in kinesin and myosin Kinesin and myosin have long been considered to be unrelated molecules because they use different tracks, possess motor domains of different size, and show no marked sequence homology. Therefore the least expected outcome of the crystallographic studies11 was a striking structural homology between the kinesin motor domain and the core of the myosin head. The core structural elements, a set of seven β-sheets sandwiched between three α-helices on either side, are essentially superimposable. In addition, several amino acids around the nucleotide-binding site are conserved positionally, indicating that both classes of motors may originate from a common ancestral nucleotidase94. The structural similarities in the catalytic cores of kinesin and myosin indicate that the conformational changes upon ATP hydrolysis are initiated in the same way in what seem to be homologous domains. However, the (much larger) conformational changes that lead to a step along the respective tracks are executed in domains carboxy-terminal to the catalytic core. In skeletal muscle myosin, an α-helix stabilized by two associated light chains and based in a domain termed the ‘converter’ undergoes a nucleotide-dependent angular rotation. In conventional kinesin, conformational changes in a flexible subdomain, the ‘neck linker’, which alternates between a mobile state and a docked position on the catalytic core, are associated with the forward motion of the kinesin head45. The communication between the nucleotide-binding site and the mechanical element at the carboxyl terminus of the head is mediated by analogous elements in kinesin and myosin: an αhelix that contacts the catalytic site at one end passes by the polymer binding face and ends near the mechanical amplifier, the converter in myosin or the neck linker in kinesin. So myosin and kinesin have developed different ways of converting small ATPdependent conformational changes (on a scale of ångströms) into large changes of conformation (on a scale of nanometres) of an associated mechanical element. Nevertheless, the sequence of events in the respective motor domains follows remarkably similar pathways. The figure shows the sequence of events during force generation in the myosin (left) and kinesin (right) heads. In both motor domains, a conformational change is initiated by the binding of ATP (green) in the catalytic site (top row). This information is transmitted through analogous elements, the relay helix in myosin and the switch II helix in kinesin (middle row), to a mechanical element. There, the initially small conformational change is translated by structurally unrelated elements, the converter in myosin and the neck linker in kinesin (bottom row), into a much larger conformational change.

www.nature.com/reviews/molcellbio

| OCTOBER 2000 | VOLUME 1

REVIEWS mechanical element, the lever arm (BOX 2). Knowing which way to go

Figure 3 | Domain organization of the conventional kinesin heavy-chain dimer, showing the crystal structure of the catalytic domains and the neck96. The structure of the stalk and tail are inferred from electron microscopic images and coiled-coil prediction analyses. Regions predicted to form coiled-coils (neck, coil 1, coil 2, coiled-coil tail) and flexible regions (hinge, kink, stalk–tail linker) are indicated.

Further domains are required to turn the enzyme into a motor. In conventional kinesins, the neck and neck linker, and probably also the hinge (FIG. 3), translate the small conformational change that is generated in the ATPbinding site into a much larger mechanical movement. Studies on truncated and mutated kinesins show that both the velocity and the ability to stay on track are affected when these domains are deleted or changed14–16. Because motility is not entirely abolished in these artificial molecules, neck and neck linker seem to amplify events in the core motor domain to produce physiological behaviour. In this respect kinesin resembles myosin, where small conformational changes in the catalytic site are translated into a large conformational change of a

Figure 4 | Crystal structures of dimeric Ncd and conventional kinesin. Ncd23 (left) and conventional kinesin96 (right) are shown in side view (top) and top view (bottom). These views clearly show the different positions of the catalytic motor domains relative to the neck regions in these two motors, which move in opposite directions along microtubules. ADP, bound in the active site, is shown in green.

Considering the similarity between the core motor domains of different kinesins, you might assume that, except for variations in speed and processivity, all members of the kinesin superfamily would behave similarly. This is not the case. One kinesin family shows a particularly surprising feature — its members move in the opposite direction to conventional kinesin. Microtubules are intrinsically polar assemblies of α/β-tubulin dimers (BOX 3). Conventional kinesins, and most other kinesins, move towards the plus end (FIG. 2). So the discovery of a minus-end-directed kinesin, nonclaret disjunctional (ncd), named after a long-known spindle mutant of the fruitfly Drosophila melanogaster, came as a surprise17,18. Because other members of the ncd family of motors also move towards the minus end19, and all of these motors have the motor domain at the carboxyl terminus, these two features are probably linked. The finding of ‘reverse’ motility immediately challenged our understanding of how kinesin motors generate movement. It raised the question of where the ‘gear and transmission’ are that revert the direction of stepping, and which protein domains are important for this movement. To answer these questions, artificial chimeric motors combining parts of plus-end-directed conventional kinesins with those of ncd have been generated20–22. Using the motor domain of ncd attached to a conventional kinesin stalk, it was possible to reverse ncd’s physiological minus-end motility, an indication that regions outside the catalytic core confer directionality. The reverse experiment, making kinesin move backwards, turned out to be more difficult but finally revealed regions that were responsible for minus-end directionality in the ncd neck. So the neck linker and neck regions emerge as being primarily responsible for directional determination. Although the common kinesin core seems to possess a subtle intrinsic bias that is sufficient for very slow plusend motility16, the conserved helix preceding the motor core of ‘reverse’ motors is able to override this bias, forcing the molecule to move towards microtubule minus ends. Conversely, the neck and neck linker of conventional kinesin amplify the intrinsic bias into robust and fast plus-end motility. How can neck and neck-linker domains cause opposite movement? A look at the three-dimensional structures of dimeric conventional kinesin and ncd reveals that the respective necks position the two motor domains differently (FIG. 4). In ncd, hydrogen bonds cause the heads to lie close to the neck coiled-coil, generating a 180° rotational symmetry around its axis in a shape that resembles two oppositely oriented ‘P’s23. With only a small angular variation, this crystallographic structure can be fitted into three-dimensional images of motordecorated microtubules obtained by electron microscopy. Conventional kinesin, in contrast, dimerizes through a neck that points away from the core motor domains, and the two heads include an angle of about

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 1 | OCTOBER 2000 | 5 3

REVIEWS

Box 3 | Microtubule structure Microtubules are built from α/β-tubulin dimers that are stacked in linear arrays termed protofilaments, 13 of which form the wall of a microtubule in most cell types. Owing to the stereotyped stacking of subunits, these protofilaments (and therefore the microtubules) possess an intrinsic molecular polarity, with one end exposing the α-subunit, and the other the β-subunit. Although the end exposing the α-subunit, called the minus end, is usually anchored near the centrosome, the cell’s microtubuleorganizing centre, a microtubule can grow or shrink rapidly at the end exposing the β-subunit, called the plus end. The motor domain of kinesin possesses a microtubulebinding face that interacts with tubulin dimers (mostly the β-subunit) in the microtubule wall in always the same orientation, thereby recognizing (and exploiting) the intrinsic molecular polarity of microtubules.

120°. This spatial arrangement is incompatible with electron microscopic images24,25. Although cryo-electron microscopic reconstructions visualize the bound head, they do not clearly resolve the second head26–28. This may be due to disorder (or movement) of the second head, but variations in the pattern of kinesin decoration cannot be excluded29. Whatever the reason, there is a clear difference in the binding patterns of conventional kinesin and ncd, resulting from differences in the neck region, which helps to establish a structural asymmetry and, as a consequence, generates a directional bias. The mechanism by which the bound head is moved in the right direction along microtubules — plus end for conventional kinesin, minus end for ncd — is largely unknown. One important point to consider is that, in the molecular world, brownian motion generates a large positional noise on a nanometre scale and causes the tethered head to fluctuate around an average position. A motor protein might just limit these fluctuations of the unbound head to a location near a new microtubule-binding site and allow it to ‘find’ this site by a diffusive mechanism. Alternatively, the neck may actively push the unbound head towards the next binding site. These models represent two possible principles for understanding molecular motility. In the first, translocation would essentially be driven by biased diffusion, with the motor acting as a molecular ratchet. The second model explains directional force generation by conformational changes of rigid mechanical elements. These models are not mutually exclusive and allow for a combination of both mechanisms during stepping (see online animation). Which mechanisms are used by the other two classes of molecular motors, myosin and dynein, to decide which way to go? As shown by the studies using the kinesin–ncd pair, it is extremely useful if representatives that move in opposite directions are available. Such a myosin that ‘goes the other way’ has recently been found. So how does it work? Recall that in skeletal muscle myosin, the structural change initiated in the catalytic core is translated into a swing of the lever arm whose motion is coordinated by the converter domain30 (BOX 2). Theoretically, by changing the connectivity at the base of the lever, you should be able to engineer a myosin whose mechanism of transduction from the active site is unchanged, but whose converter region causes the lever to swing in the opposite direction31.

Figure 5 | Processive catalysis of conventional kinesin. After initial binding, kinesin is able to ‘walk’ along the microtubule without dissociating. It is thought that a twoheaded motor is necessary for this processive behaviour because at each time point, the dimeric molecule needs to remain tethered to the microtubule through one head. This is achieved by the coordinated catalysis in the two heads: throughout the entire catalytic cycle, one of the heads is kept in the tight microtubule-bound form (containing either ATP or no nucleotide), while the other head is in transit in an ADPbound form. During a brief phase of the cycle both heads may be bound to the microtubule. D and T inidicate bound ADP and ATP, respectively.

Nature has apparently done just that. The ‘reverse’ myosin, a member of the myosin VI family, moves along actin filaments in the opposite direction because the lever swings the other way32. The only main difference from the motor domain of muscle myosin is in the sequence of the converter region, which contains a large insertion. Because all members of the class VI myosins possess this insertion, it is conceivable that they all move in the opposite direction. A dynein that goes the other way has not been isolated, but there are reasons to believe that ‘reverse’ dyneins may also exist. Dynein is the motor that powers ciliary and flagellar beating33, but cytoplasmic isoforms involved in mitosis and organelle transport also exist34. All cytoplasmic dyneins characterized so far move towards the microtubule minus end, but in the cell processes of an unusual giant amoeba35, the characteristics of ATP-driven organelle transport along uniform bundles of microtubules are dynein-like in both directions36. Another good place to look for a reverse dynein should be the www.nature.com/reviews/molcellbio

| OCTOBER 2000 | VOLUME 1

REVIEWS

Box 4 | Duty ratio and processivity The velocity of a molecular motor is restricted by the velocity of the catalytic events that supply the energy. Many kinesins, myosins and dyneins presumably move in a stepwise manner along their respective tracks. If one step is coupled to one ATP hydrolysis event, as generally believed81,82, the stepping frequency cannot be higher than the ATPase rate. To calculate the gliding velocity from ATPase rates, you must consider the working distance per step. For conventional kinesin it is about 8 nm, so given an ATPase rate of 20–40 ATP per second per head, we obtain a gliding velocity of 320–640 nm s–1 for a twoheaded kinesin42–44. In the case of muscle myosin, the ATPase rate of about 20 s–1 and a step-size of 5.5 nm gives 110 nm s–1 — a value 80-fold too small to explain the observed velocity of movement. The answer to this paradox lies in the concept of the duty ratio. It is defined as the fraction of time that a motor remains attached to its track during one full CROSSBRIDGE CYCLE95. All molecular motors are proposed to undergo a working stroke during the attached phase, and then to recover their initial conformation in a detached (or weakly bound) phase. Conventional kinesin has a high duty ratio, meaning that the attached phase is long (over half of the full cycle), allowing a single molecule to transport a microtubule for several micrometres without falling off. Skeletal muscle myosin, on the other hand, has a low duty ratio, as shown by its inability to work as a single molecule. Rather, several tens of molecules are required to generate continuous movement. This seems to be an adaptation to the arrangement in sarcomeres, where myosin filaments interdigitate with large ensembles of actin filaments, ensuring the close proximity of many potential binding sites for myosin heads. The duty ratio is intimately linked to the ability of a motor to operate processively, that is, to undertake many steps in succession without dissociating from the track. For kinesin, a high duty ratio and high processivity are thought to be an adaptation to its function as a single-molecule motor that transports cargo over long distances along a microtubule. Processivity is believed to require two heads that move in a ‘hand-overhand’ fashion where the chemo-mechanical coordination between the duty cycles of the two motor domains ensures that there is always one head bound.

axoneme of eukaryotic cilia and flagella, a complex microtubule-based motile machinery. Axonemes may contain as many as a dozen dyneins, few of which have been characterized. It is pure speculation, but it would not be surprising if axonemes were to require ‘reverse’ motors for coordination of their intricate behaviour. Staying on track

CROSSBRIDGE CYCLE

The sequence of structural changes of a myosin head coordinated with the hydrolysis of one molecule of ATP. DUTY RATIO

The fraction of time that a motor molecule remains attached to the track during one full ATP hydrolysis cycle. GLIDING ASSAY

Optical assay for the movement of cytoskeletal filaments over a ‘lawn’ of motor molecules attached to a coverslip. OPTICAL TRAP ASSAY

A focused laser beam that traps refractile particles (for example, polystyrene beads) with attached motor molecules, allowing determination of step size and force per step.

Kinesin moves along microtubules in steps that bridge the distance between adjacent tubulin subunits and can take many steps from one dimer to the next without falling off 37,38. This form of movement is dubbed processive and is linked to the DUTY RATIO (BOX 4). The ability to stay on track allows a single kinesin molecule to move a microtubule in a GLIDING ASSAY. In contrast, for skeletal muscle myosin to move an actin filament, several molecules (probably more than 20) need to cooperate. Processivity of conventional kinesin requires two motor domains that are linked by heavy-chain dimerization. A motor that possesses just one head, but otherwise has an unchanged neck and stalk region, is still active in multiple-motor gliding assays, but it fails to operate as a single molecule because it seems to fall off after a single step39–41. Therefore it is thought that conventional kinesin motility relies on a precise coordination between the two motor domains where one head proceeds to the next binding site while the motor remains tethered to the microtubule through the other, attached head. Conventional kinesin can move processively because, at each time point, at least one head is

microtubule bound. How is this amazing natural clockwork synchronized? How does a head ‘know’ when to hold on and when to let go? The answer is still open, though kinetic models have set a framework for processivity models. According to the ‘alternating site model’, kinesin uses a nucleotidedependent change in its affinity for the microtubule to regulate the behaviour of the two heads42–44 (FIG. 5). In solution, a kinesin dimer contains one ADP per head. Upon microtubule binding, only one head (say, head A) locks onto the microtubule and loses its ADP. Head A can detach again only if it binds and hydrolyses a new ATP molecule. During this hydrolysis process, head A allows head B to find the next microtubule binding site, where it loses its ADP and holds on tight. How head A acts on head B is crucial for an understanding of kinesin’s motility mechanism, but precisely when and how force is being produced still remains an open issue. After the attachment of head B, head A finishes hydrolysis in a weakly bound ADP state and detaches from the microtubule while head B holds on. At this point, the heads have exchanged their roles (see online animation). So processive kinesin movement is achieved by three steps: first, a modulation of microtubule affinity through ATP hydrolysis, second, a mechanism that keeps the two heads out of phase and, last, a ‘power stroke’ linked to the hydrolysis cycle. In conventional kinesin, the power stroke entails conformational changes in the neck linker region45, whereas in myosin it is the swing of the lever arm in conjunction with the converter region (BOX 2). Stepping of other motors

The concept of how nucleotide binding and hydrolysis are linked to molecular motion was first developed for skeletal muscle myosin46. However, a single myosin molecule is not processive and seems to hop along its track, making contact with the actin filament for only a short period of time (BOX 4), in contrast to the behaviour of conventional kinesin. Does that imply that kinesins are, in general, processive, and myosins are not? An answer to this question requires single-molecule studies of other motors. The few that have been looked at indicate that generalizations may be premature. For example, the dimeric kinesin-like motor Ncd is not processive47–49. It is not known whether other members of the family of ‘reverse’ kinesins are, or whether minus-end motility is incompatible with processivity; this is a question worth pursuing. On the other hand, at least one dimeric myosin of class V has now been shown to be processive50. In analogy to conventional kinesin, myosin V may drive vesicular transport as a single dimeric molecule51,52. On the basis of an OPTICAL 53 54 TRAP ASSAY and electron microscopic visualization , myosin V appears to move in large steps (thanks to a long neck region) of about 36 nm, even against an opposing force exerted by the trap50. So, physiologically, myosin V is more similar to conventional kinesin than to its close relative, muscle myosin. More surprising is the observation that a single-headed motor can be processive as well. The monomeric

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 1 | OCTOBER 2000 | 5 5

REVIEWS cessive at low ATP concentrations, therefore resembling kinesin58. Unlike kinesin, however, the motor shows frequent backward steps, and at higher ATP concentrations becomes nonprocessive altogether — a behaviour not found in any other motor so far. A single-headed dynein of Chlamydomonas reinhardtii flagella59 has also been reported to be processive, but this motor poses a riddle: it behaves as a motor with a low duty ratio, which is incapable of processive movement in gliding assays, but in a laser trap it behaves as a single motor that can take eight or nine continuous steps, even against a weak retaining force. The mechanism is unclear, but this behaviour may, in principle, result from the coordination of two independent microtubule-binding sites within the large dynein head. So processivity can be based, it seems, on mechanisms distinct from the strict ‘hand-over-hand’ coordination model of dimeric kinesin. Getting (in)activated

Figure 6 | Model for how cargo binding might be linked to motor activation. When not bound to cargo, the globular tail and the adjacent cargo-binding site (red segment) are located close to the motor domain (step 1). Docking onto cargo activates the cargo interaction site in the tail coiled-coil (step 2). Cargo docking is proposed to be transmitted to the globular tail domain (step 3) by as yet unknown mechanisms, initiating a conformational change (step 4) that relieves the inhibition of the motor domain. This sequence of events is indicated by red coloration of the domains involved. Whether this model leads to complete unfolding of the motor, as is generally assumed, or where it allows the motor to retain a modified folded conformation, remains to be shown.

RETAINING FORCE

Force exerted by a laser trap on a motor-carrying bead, moving along a microtubule. molecule of ATP.

mouse kinesin KIF1A, implicated in axonal transport55, may promote processive motility despite being a monomer56. Its motile behaviour, however, differs from that of conventional kinesins in that KIF1A shows phases of back-and-forth movement with a net directional bias. Contact with the microtubule surface is apparently maintained by a positively charged surface loop in the head that interacts with the negatively charged carboxyl terminus of tubulin57. This interaction may allow for one-dimensional diffusion along microtubules during the weak binding state, though the mechanism that sustains the directional bias towards the microtubule plus end remains unclear. Given this diffusional step, it should be interesting to see whether the motor is still processive when it moves against a RETAINING FORCE. Dyneins have also been analysed for processivity, and they show unusual behaviour. Single dynein molecules from Tetrahymena cilia move in 8-nm steps and are pro-

Kinesin’s enzymatic activities and cellular functions are probably regulated at several levels60, including associated light chains, phosphorylation, binding to cargo and, as a more recent addition, intramolecular folding. Light chains, which interact with the heavy chains near the globular carboxyl terminus, have long been suspected to mediate kinesin function. Mutations in light chains result in the same phenotype as heavy-chain mutations61,62, indicating that both molecules may cooperate in the same cellular pathway. This idea is supported by experiments with antibodies against light chains, which interfere with kinesin binding to vesicles63 and organelle movements in vitro64,65, albeit in both directions. Precisely how light chains affect kinesin function is not yet known, but one study indicates that they may inhibit binding of the heavy chains to microtubules66, possibly in a phosphorylation-dependent manner67,68. They may also be involved in cargo targeting and cargo binding, because in one study a splice variant of light chains has been found to be associated specifically with mitochondria69. Perhaps the most basic level of kinesin regulation occurs within the kinesin dimer itself and involves an intramolecular interaction of the head and tail that is mediated by folding70,71. In the compact conformation, which prevails at physiological ionic strength, the ATPase activity of the motor domain is inhibited72. Self-inhibition requires neither associated proteins nor post-translational modifications73 but critically depends on the presence of the flexible kink in the middle of the stalk74–76. Upon cargo binding — even an artificial cargo such as silica beads77 — tail inhibition is relieved. The ability of unloaded kinesin to bind to, and move along, microtubules is not abolished75, but movement is initiated less frequently and terminated earlier. The pronounced inhibition of the ATPase activity of a folded motor is due to the selective inhibition of the initial productive interaction with a microtubule78. Once bound, subsequent processive cycles are not strongly inhibited75. At the molecular level, folding requires an interaction between a domain near the carboxyl terminus and a region near the motor domain73,76. In the folded state, www.nature.com/reviews/molcellbio

| OCTOBER 2000 | VOLUME 1

REVIEWS the globular tail domain is placed in close proximity to the catalytic motor domain. A conserved motif in the globular tail79 may be directly involved in modulating the ATPase activity of the motor domain73,76, whereas a cargo-binding region has been located in a coiled-coil next to the globular tail domain76. Its binding partner on the cargo is not known, but it is important that this domain is adjacent to the regions thought to act during folding and regulation. A model of how cargo binding might be linked to motor activation is presented in FIG. 6. The tail inhibition model offers a reasonable explanation for the behaviour of conventional kinesins in vitro and in vivo, and may represent the most basic level of regulation. The reduction of the spontaneous activity of folded kinesin that the model proposes helps to explain how excessive movements of unloaded motor are prevented. However, because folding does not suppress movement completely, other factors are also likely to contribute. Conclusions and outlook

This year kinesin celebrates its fifteenth birthday, but the field has advanced far past the ‘puberty’ stage and has reached a degree of maturity previously unforeseen. From atomic structures, the realization has come that the actin- and microtubule-based motors, myosin and kinesin, are closely related9,11,92. This finding has led to the now widely held view that largely homologous conformational changes in the catalytic site are translated into motion by a diverse set of structural elements in different motors30,32,45 and result in steps of different size, duration or direction23,30,56. Because only a handful of motors have been looked at in detail so far, analyses of other motor classes may reveal more variations on the theme of mechanical amplification. Tremendous advances have been made in the development of techniques for single-molecule analysis, contributing information on step size, duty cycle, force generation and energy consumption per step37,80,81,82. Many valuable contributions to the study of motor mechanics have also been made possible through ingenuity (and intuition) in the design of mutant motors and their expression in suitable host cells.

1. 2.

Spudich, J. A. How molecular motors work. Nature 372, 515–518 (1994). Hirokawa, N., Noda, Y. & Okada, Y. Kinesin and dynein superfamily proteins in organelle transport and cell division. Curr. Opin. Cell Biol. 10, 60–73 (1998). Goldstein, L. B. & Philp, A. V. The road less traveled: emerging principles of kinesin motor utilization. Annu. Rev. Cell Dev. Biol. 15, 141–183 (1999). Cyr, J. L., Pfister, K. K., Bloom, G. S., Slaughter, C. A. & Brady, S. T. Molecular genetics of kinesin light chains: generation of isoforms by alternative splicing. Proc. Natl Acad. Sci. USA 88, 10114–10118 (1991). Wedaman, K. P., Knight, A. E., Kendrick-Jones, J. & Scholey, J. M. Sequences of sea urchin kinesin light chain isoforms. J. Mol. Biol. 231, 155–158 (1993). Moore, J. D. & Endow, S. A. Kinesin proteins: a phylum of motors for microtubule-based motility. Bioessays 18, 207–219 (1996). Goodson, H. V., Kang, S. J. & Endow, S. A. Molecular phylogeny of the kinesin family of microtubule motor proteins. J. Cell Sci. 107, 1875–1884 (1994). Goldstein, L. S. With apologies to Scheherazade: tails of

10.

11.

12. 13.

14.

Nevertheless, uncharted territory still lies ahead. The diversity of kinesin motors, reflected in the existence of at least ten kinesin families, has not yet been fully exploited to learn more about different ways, or unexpected subtleties, of force generation. Mutant motors generated by rational design and/or random mutagenesis will continue to help in this endeavour. Although the likely binding site for tubulin on the kinesin surface has been mapped and models of kinesin docking onto microtubule exist, details of this mechanism, in particular possible repercussions of the tubulin docking site on conformational changes in the motor head, have not yet been explored. Possibly the largest gap in our knowledge of kinesin function is in the way that motors are attached to their respective cargoes. The diversity of kinesin tails suggests a similar range in cargo attachment mechanisms. In vitro assays of cargo binding will probably yield answers, but reliable assays do not yet exist. The question of cargo binding is intimately linked to the mechanism of its regulation, and the regulation of kinesin activity in general. Folding and tail inhibition are of basic importance, but as means of motor regulation they are, so far, restricted to conventional kinesins. Other regulatory factors must exist, and some of them are known, but they are likely to constitute only the tip of the iceberg. The existence of a complex regulatory machinery would be in line with the fact that all vital cellular processes are controlled by complex regulatory networks. Consequently, the elucidation of motor regulation may constitute the biggest challenge for years to come. Links DATABASE LINKS myosin | dynein | kinesin | ncd | myosin VI | myosin V | KIF1A FURTHER INFORMATION Kinesin home page | Structure and function of microtubules | Online animation: Kinesin stepping ENCYCLOPEDIA OF LIFE SCIENCES Dynein and kinesin | Cytoskeleton | Intracellular transport | ATP-binding motifs

1001 kinesin motors. Annu. Rev. Genet. 27, 319–351 (1993). Vale, R. D. Switches, latches, and amplifiers: common themes of G proteins and molecular motors. J. Cell Biol. 135, 291–302 (1996). Succinct discussion of the surprising similarities between these protein families. Lane, J. D. & Allan, V. Microtubule-based membrane movement. Biochim. Biophys. Acta. 1376, 27–55 (1998). Excellent overview of the diverse cellular functions of motor proteins. Kull, F. J., Sablin, E. P., Lau, R., Fletterick, R. J. & Vale, R. D. Crystal structure of the kinesin motor domain reveals a structural similarity to myosin. Nature 380, 550–555 (1996). Sadhu, A. & Taylor, E. W. A kinetic study of the kinesin ATPase. J. Biol. Chem. 267, 11352–11359 (1992). Gilbert, S. P. & Johnson, K. A. Pre-steady-state kinetics of the microtubule-kinesin ATPase. Biochemistry 33, 1951–1960 (1994). Romberg, L., Pierce, D. W. & Vale, R. D. Role of the

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

15.

16.

17.

18.

19.

20.

kinesin neck region in processive microtubule-based motility. J. Cell. Biol. 140, 1407–1416 (1998). Grummt, M. et al. Importance of a flexible hinge near the motor domain in kinesin-driven motility. EMBO J. 17, 5536–5542 (1998). Case, R. B., Rice, S., Hart, C. L., Ly, B. & Vale, R. D. Role of the kinesin neck linker and catalytic core in microtubule-based motility. Curr. Biol. 10, 157–160 (2000). McDonald, H. B., Stewart, R. J. & Goldstein, L. S. The kinesin-like ncd protein of Drosophila is a minus enddirected microtubule motor. Cell 63, 1159–1165 (1990). Walker, R. A., Salmon, E. D. & Endow, S. A. The Drosophila claret segregation protein is a minus-end directed motor molecule. Nature 347, 780–782 (1990). Endow, S. A. et al. Yeast Kar3 is a minus-end microtubule motor protein that destabilizes microtubules preferentially at the minus ends. EMBO J. 13, 2708–2713 (1994). Henningsen, U. & Schliwa, M. Reversal in the direction of movement of a molecular motor. Nature 389, 93–96 (1997).

VOLUME 1 | OCTOBER 2000 | 5 7

REVIEWS

21. Case, R. B., Pierce, D. W., Hom Booher, N., Hart, C. L. & Vale, R. D. The directional preference of kinesin motors is specified by an element outside of the motor catalytic domain. Cell 90, 959–966 (1997). 22. Endow, S. A. & Waligora, K. W. Determinants of kinesin motor polarity. Science 281, 1200–1202 (1998). 23. Sablin, E. P. et al. Direction determination in the minusend-directed kinesin motor ncd. Nature 395, 813–816 (1998). Analysis of the crystallographic structure of dimeric ncd. 24. Sack, S. et al. X-ray structure of motor and neck domains from rat brain kinesin. Biochemistry 36, 16155–16165 (1997). 25. Mandelkow, E. & Hoenger, A. Structures of kinesin and kinesin-microtubule interactions. Curr. Opin. Cell Biol. 11, 34–44 (1999). 26. Hoenger, A. et al. Image reconstructions of microtubules decorated with monomeric and dimeric kinesins: comparison with x-ray structure and implications for motility. J. Cell Biol. 141, 419–430 (1998). 27. Arnal, I. & Wade, R. H. Nucleotide-dependent conformations of the kinesin dimer interacting with microtubules. Structure 6, 33–38 (1998). 28. Hirose, K., Cross, R. A. & Amos, L. A. Nucleotidedependent structural changes in dimeric ncd molecules complexed to microtubules. J. Mol. Biol. 278, 389–400 (1998). 29. Hoenger, A. et al. A new look at the microtubule binding patterns of dimeric kinesins. J. Mol. Biol. 297, 1087–1103 (2000). 30. Dominguez, R., Freyzon, Y., Trybus, K. M. & Cohen, C. Crystal structure of a vertebrate smooth muscle myosin motor domain and its complex with the essential light chain: visualization of the pre-power stroke state. Cell 94, 559–571 (1998). 31. Becker, E. W. Kinetic equilibrium of forces and molecular events in muscle contraction. Proc. Natl Acad. Sci. USA 97, 157–161 (2000). 32. Wells, A. L. et al. Myosin VI is an actin-based motor that moves backwards. Nature 401, 505–508 (1999). First demonstration of a myosin that moves in the opposite direction. 33. Gibbons, I. R. Studies on the adenosine triphosphatase activity of 14 S and 30 S dynein from cilia of Tetrahymena. J. Biol. Chem. 241, 5590–5596 (1966). 34. Vallee, R. B., Wall, J. S., Paschal, B. M. & Shpetner, H. S. Microtubule-associated protein 1C from brain is a twoheaded cytosolic dynein. Nature 332, 561–563 (1988). 35. Euteneuer, U., Koonce, M. P., Pfister, K. K. & Schliwa, M. An ATPase with properties expected for the organelle motor of the giant amoeba, Reticulomyxa. Nature 332, 176–178 (1988). 36. Schliwa, M., Shimizu, T., Vale, R. D. & Euteneuer, U. Nucleotide specificities of anterograde and retrograde organelle transport in Reticulomyxa are indistinguishable. J. Cell Biol. 112, 1199–1203 (1991). 37. Svoboda, K., Schmidt, C. F., Schnapp, B. J. & Block, S. M. Direct observation of kinesin stepping by optical trapping interferometry. Nature 365, 721–727 (1993). First demonstration that conventional kinesin moves in 8-nm steps. 38. Howard, J., Hudspeth, A. J. & Vale, R. Movement of microtubules by single kinesin molecules. Nature 342, 154–158 (1989). 39. Young, E. C., Mahtani, H. K. & Gelles, J. One-headed kinesin derivatives move by a nonprocessive, low-duty ratio mechanism unlike that of two-headed kinesin. Biochemistry 37, 3467–3479 (1998). 40. Hancock, W. O. & Howard, J. Processivity of the motor protein kinesin requires two heads. J. Cell Biol. 140, 1395–1405 (1998). 41. Berliner, E., Young, E. C., Anderson, K., Mahtani, H. K. & Gelles, J. Failure of a single-headed kinesin to track parallel to microtubule protofilaments. Nature 373, 718–721 (1995). 42. Hackney, D. Evidence for alternating head catalysis by kinesin during microtubule-stimulated ATP hydrolysis. Proc. Natl Acad. Sci. USA 91, 6865–6869 (1994). Introduces the kinetic model of head–head interaction. 43. Ma, Y. Z. & Taylor, E. W. Interacting head mechanism of microtubule-kinesin ATPase. J. Biol. Chem. 272, 724–730 (1997). 44. Gilbert, S. P., Moyer, M. L. & Johnson, K. A. Alternating site mechanism of the kinesin ATPase. Biochemistry 37, 792–799 (1998).

45. Rice, S. et al. A structural change in the kinesin motor protein that drives motility. Nature 402, 778–784 (1999). Detailed analysis using an impressive array of techniques of the movements of the neck linker domain. 46. Lymn, R. W. & Taylor, E. W. Mechanism of adenosine triphosphate hydrolysis by actomyosin. Biochemistry 10, 4617–4624 (1971). 47. Mackey, A. T. & Gilbert, S. P. Moving a microtubule may require two heads: a kinetic investigation of monomeric Ncd. Biochemistry 39, 1346–1355 (2000). 48. Pechatnikova, E. & Taylor, E. W. Kinetics processivity and the direction of motion of Ncd. Biophys. J. 77, 1003–1016 (1999). 49. Foster, K. A. & Gilbert, S. P. Kinetic studies of dimeric Ncd: evidence that Ncd is not processive. Biochemistry 39, 1784–1791 (2000). 50. Metah, A. D. et al. Myosin-V is a processive actin-based motor. Nature 400, 590–593 (1999). 51. Tabb, J. S., Molyneaux, B. J., Cohen, D. L., Kuznetsov, S. A. & Langford, G. M. Transport of ER vesicles on actin filaments in neurons by myosin V. J. Cell Sci. 111, 3221–3234 (1998). 52. Rogers, S. L. et al. Regulation of melanosome movement in the cell cycle by reversible association with myosin V. J. Cell Biol. 146, 1265–1276 (1999). 53. Walker, M. L. et al. Two-headed binding of a processive myosin to F-actin. Nature 405, 804–807 (2000). 54. Finer, J. T., Simmons, R. M. & Spudich, J. A. Single myosin mechanics: piconewton forces and nanometre steps. Nature 368, 113–119 (1994). 55. Aizawa, H. et al. Kinesin family in murine central nervous system. J. Cell Biol. 119, 1287–1296 (1992). 56. Okada, Y. & Hirokawa, N. A processive single-headed motor: kinesin superfamily protein KIF1A. Science 283, 1152–1157 (1999). Experimental evidence for the directional movement of single monomeric kinesin molecules. 57. Okada, Y. & Hirokawa, N. Mechanism of the singleheaded processivity: diffusional anchoring between the Kloop of kinesin and the C terminus of tubulin. Proc. Natl Acad. Sci. USA 97, 640–645 (2000). 58. Hirakawa, E., Higuchi, H. & Toyoshima, Y. Y. Processive movement of single 22S dynein molecules occurs only at low ATP concentrations. Proc. Natl Acad. Sci. USA 97, 2533–2537 (2000). 59. Sakikabara, H., Kojima, H., Sakai, Y., Katayama, E. & Oiwa, K. Inner-arm dynein c of Chlamydomonas flagella is a single-headed processive motor. Nature 400, 586–590 (1999). 60. Thaler, C. D. & Haimo, L. T. Microtubules and microtubule motors: mechanisms of regulation. Int. Rev. Cytol. 164, 269–327 (1996). 61. Hurd, D. D., Stern, M. & Saxton, W. M. Mutation of the axonal transport motor kinesin enhances paralytic and suppresses Shaker in Drosophila. Genetics 142, 195–204 (1996). 62. Gindhart, J. G. Jr, Desai, C. J., Beushausen, S., Zinn, K. & Goldstein, L. S. Kinesin light chains are essential for axonal transport in Drosophila. J. Cell Biol. 141, 443–454 (1998). 63. Yu, H., Toyoshima, I., Steuer, E. R. & Sheetz, M. P. Kinesin and cytoplasmic dynein binding to brain microsomes. J. Biol. Chem. 267, 20457–20464 (1992). 64. Stenoien, D. L. & Brady, S. T. Immunochemical analysis of kinesin light chain function. Mol. Biol. Cell 8, 675–689 (1997). 65. Brady, S. T. & Pfister, K. K. Kinesin interactions with membrane bounded organelles in vivo and in vitro. J. Cell Sci. 14, S103–S108 (1991). 66. Verhey, K. J. et al. Light chain-dependent regulation of kinesin’s interaction with microtubules. J. Cell Biol. 143, 1053–1066 (1998). 67. Matthies, H. J., Miller, R. J. & Palfrey, H. C. Calmodulin binding to and cAMP-dependent phosphorylation of kinesin light chains modulate kinesin ATPase activity. J. Biol. Chem. 268, 11176–11187 (1993). 68. Hollenbeck, P. J. Phosphorylation of neuronal kinesin heavy and light chains in vivo. J. Neurochem. 60, 2265–2275 (1993). 69. Khodjakov, A., Lizunova, E. M., Minin, A. A., Koonce, M. P. & Gyoeva, F. K. A specific light chain of kinesin associates with mitochondria in cultured cells. Mol. Biol. Cell 9, 333–343 (1998). 70. Hisanaga, S. et al. The molecular structure of adrenal medulla kinesin. Cell Motil. Cytoskel. 12, 264–272 (1989).

71. Amos, L. A. Kinesin from pig brain studied by electron microscopy. J. Cell Sci. 87, 105–111 (1987). 72. Hackney, D., Levitt, J. & Suhan, J. Kinesin undergoes a 9 S to 6 S conformational transition. J. Biol. Chem. 267, 8696–8701 (1992). 73. Stock, M. F. et al. Formation of the compact conformer of kinesin requires a COOH-terminal heavy chain domain and inhibits microtubule-stimulated ATPase activity. J. Biol. Chem. 274, 14617–14623 (1999). 74. Coy, D. L., Hancock, W. O., Wagenbach, M. & Howard, J. Kinesin’s tail domain is an inhibitory regulator of the motor domain. Nature Cell Biol. 1, 288–292 (1999). 75. Friedman, D. S. & Vale, R. D. Single-molecule analysis of kinesin motility reveals regulation by the cargo-binding tail domain. Nature Cell Biol. 1, 293–297 (1999). 76. Seiler, S. et al. Cargo binding and regulatory sites in the tail of fungal conventional kinesin. Nature Cell Biol. 2, 333–338 (2000). 77. Coy, D. L., Wagenbach, M. & Howard, J. Kinesin takes one 8-nm step for each ATP that it hydrolyzes. J. Biol. Chem. 274, 3667–3671 (1999). 78. Hackney, D. D. & Stock, F. M. Kinesin’s IAK tail domain inhibits initial microtubule-stimulated ADP release. Nature Cell Biol. 2, 257–260 (2000). 79. Kirchner, J., Seiler, S., Fuchs, S. & Schliwa, M. Functional anatomy of the kinesin molecule in vivo. EMBO J. 18, 4404–4413 (1999). 80. Block, S. M., Goldstein, L. S. & Schnapp, B. J. Bead movement by single kinesin molecules studied with optical tweezers. Nature 348, 348–352 (1990). 81. Hua, W., Young, E. C., Fleming, M. L. & Gelles, J. Coupling of kinesin steps to ATP hydrolysis. Nature 388, 390–393 (1997). 82. Schnitzer, M. J. & Block, S. M. Kinesin hydrolyses one ATP per 8-nm step. Nature 388, 386–390 (1997). 83. Gilbert, S. P., Webb, M. R., Brune, M. & Johnson, K. A. Pathway of processive ATP hydrolysis by kinesin. Nature 373, 671–676 (1995). 84. Wang, M, D. et al. Force and velocity measured for single molecules of RNA polymerase. Science 282, 902–907 (1998). 85. Uptain, S. M., Kane, C. M. & Chamberlin, M. J. Basic mechanisms of transcript elongation and its regulation. Annu. Rev. Biochem. 66, 117–172 (1997). 86. Gelles, J. & Landick, R. RNA polymerase as a molecular motor. Cell 93, 13–16 (1998). 87. Jager, J. & Pata, J. D. Getting a grip: polymerases and their substrate complexes. Curr. Opin. Struct. Biol. 9, 21–28 (1999). 88. Kinosita, K. Jr, Yasuda, R., Noji, H., Ishiwata, S. & Yoshida, M. F1-ATPase: a rotary motor made of a single molecule. Cell 93, 21–24 (1998). 89. DeRosier, D. J. The turn of the screw: the bacterial flagellar motor. Cell 93, 17–20 (1998). 90. Ryu, W. S., Berry, R. M. & Berg, H. C. Torque-generating units of the flagellar motor of Escherichia coli have a high duty ratio. Nature 403, 444–447 (2000). 91. Tilney, L. G. & Portnoy, D. A. Actin filaments and the growth, movement, and spread of the intracellular bacterial parasite, Listeria monocytogenes. J. Cell Biol. 109, 1597–1608 (1989). 92. May, R. C. et al. The Arp2/3 complex is essential for the actin-based motility of Listeria monocytogenes. Curr. Biol. 9, 759–762 (1999). 93. Frischknecht, F. et al. Actin-based motility of vaccinia virus mimics receptor tyrosine kinase signalling. Nature 401, 926–929 (1999). 94. Kull, F. J., Vale, R. D. & Fletterick, R. J. The case for a common ancestor: kinesin and myosin motor proteins and G proteins. J. Muscle Res. Cell Motil. 19, 877–886 (1998). 95. Howard, J. Molecular motors: structural adaptations to cellular functions. Nature 389, 561–567 (1997). 96. Kozielski, F. et al. The crystal structure of dimeric kinesin and implications for microtubule–dependent motility. Cell 91, 985–994 (1997). First crystallographic structure of dimeric conventional kinesin that provides the basis for all considerations of kinesin mechanics.

Acknowledgements We thank U. Euteneuer and K. Hahlen for valuable comments. Work in the authors’ laboratory is supported by the Deutsche Forschungsgemeinschaft, the Volkswagen Stiftung, and the Fonds der Chemischen Industrie.

www.nature.com/reviews/molcellbio

| OCTOBER 2000 | VOLUME 1

REVIEWS

MULTILEVEL REGULATION OF THE CIRCADIAN CLOCK Nicolas Cermakian and Paolo Sassone-Corsi Living organisms adapt to light–dark rhythmicity using a complex programme based on internal clocks. These circadian clocks, which are regulated by the environment, direct various physiological functions. As the molecular mechanisms that govern clock function are unravelled, we are starting to appreciate simple patterns as well as exquisite layers of regulation.

Institut de Génétique et de Biologie Moléculaire et Cellulaire CNRS–INSERM–Université Louis Pasteur, 1, Rue Laurent Fries, 67404 Illkirch–Strasbourg, France. e-mail: paolosc@igbmc.ustrasbg.fr Correspondence to: P.S.-C.

Most, if not all, living organisms have in-built temporal programmes that allow them to adapt to cyclic environmental conditions — for example, day/night or annual variations in light and temperature1,2. Although a variety of physiological processes are cyclic, circadian rhythms are those with a period close to 24 hours. A critical feature of circadian clocks is that they can function autonomously, without any external input or time cues, although environmental signals (for example, day/night cycles) can reset or ‘entrain’ the pacemaker. In mammals, the central clock structure is located within the suprachiasmatic nucleus (SCN) of the anterior hypothalamus (BOX 1 and FIG. 1)3,4. Our aim here is to describe and illustrate the general concepts that underlie animal circadian clocks. In all the models studied so far (vertebrates, insects, fungi and bacteria), the rhythms are generated by feedback loops that act, in most cases, at the transcriptional level5. Although several characteristics governing clock function have been conserved during evolution, some detailed features of these loops and the way their function is regulated may differ from organism to organism. These differences may be related to the specific roles of proteins, post-translational modifications and subcellular localization. Clock genes

The clock can be considered as consisting of three components: the input pathways, the oscillator or pacemaker (which generates rhythmicity autonomously) and the output pathways (FIG. 2). Pacemaker genes are essential for creating and sustaining rhythms under constant

conditions — their inactivation generally leads to arrhythmicity or to a shorter or longer period. Many of these genes encode transcription factors or proteins that function in some way in gene regulation, emphasizing the idea that the generation and modulation of rhythms relies mainly on transcriptional feedback loops and on activation and repression of gene expression. The first clock mutant to be discovered was period (per), identified 30 years ago in the fruitfly, Drosophila melanogaster 6. In the 1990s, further Drosophila genes essential for pacemaker function were isolated through genetic screening (TABLE 1). These include timeless (tim), clock, cycle and doubletime (dbt)5. In contrast, mammalian clock genes have been identified mostly by homology to the fly genes; there are three homologues of per and one of tim5. One notable exception is Clock, which was first isolated in the mouse by a genetic screen, followed by positional cloning7. The homologue of dbt has also been identified in rodents through positional cloning of the mutated gene in tau mutant hamsters8. Bmal1 (the mammalian homologue of cycle) was also first described in mammals in a database search9 and then in a two-hybrid screen using mouse CLOCK as a bait10. A recurring structural feature in many clock proteins is the PAS domain — a protein–protein interaction module first identified in the PER, arylhydrocarbon receptor (AHR) and single-minded (SIM) proteins9. Input gene products are thought to sense external stimuli and relay the message to the oscillator to reset or entrain it. They include photoreceptors specific for the clock (for example, the cryptochrome (CRY) protein in Drosophila)11,12. Finally, output genes (or clock-controlled

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 1 | OCTOBER 2000 | 5 9

REVIEWS

Box 1 | The suprachiasmatic nucleus as a central autonomous clock The concept of circadian clock is half a century old2. For decades, just one or a few central clocks were thought to control all physiological circadian rhythms. Central clocks are located in the central nervous system — for example in the ventro-lateral neurons in Drosophila33 and in the suprachiasmatic nucleus (SCN) in mammals (FIG. 4)3,4. The central role of the SCN in the circadian system is supported by various observations3,4. First, lesions in the hamster SCN abolish many physiological rhythms. Second, isolated SCN (in vivo or in vitro) or cultured SCN cells maintain rhythms in firing rate and in glucose use. Third, transplantation of SCN tissue to SCN-lesioned animals restores their rhythmicity, but with characteristics of the donor, not of the host. The SCN is more complex than originally thought4. Different parts of this hypothalamic nucleus express different peptides, or have very different functions4. SCN neurons are independent oscillators with different phases and periods, as shown with dissociated neurons52 and using transgenic mice97. The output rhythm is an average of all these oscillators98, giving a coherent physiological clock.

genes) convey information from the oscillator to downstream physiological systems that should work in a rhythmic fashion. However, these definitions are an oversimplification, and some input or output genes also seem to be important for pacemaker function (TABLE 1). In other words, essential components of the input, the oscillator and the output may overlap13 (FIG. 2). Feedback loops

E-BOX ELEMENT

Promoter element recognized by transcription factors of the basic helix–loop–helix class.

The general view of how a clock protein elicits rhythmic effects is that — somehow — its activity needs to oscillate. This can be achieved by specific checkpoints at various levels (for instance, oscillation of messenger RNAs can be regulated at both the transcriptional and posttranscriptional levels). A clever molecular device to create rhythmicity is the feedback loop5, and we now know much about how transcriptional regulatory loops underlie the Drosophila (FIG. 3) and mammalian (FIG. 4) clockwork. There is both structural and functional homology among many of

VIP

c-fos

Figure 1 | Anatomical location of the suprachiasmatic nuclei. a | Histology of a mouse brain coronal section. b | In situ hybridization with a VIP probe. VIP encodes the vasoactive intestinal polypeptide and is specifically expressed in a subset of suprachiasmatic nucleus (SCN) neurons4. c | Light-induced c-fos expression in the SCN: in situ hybridization with a c-fos antisense probe on a brain section (only the SCN region is shown) from mice kept in the dark (left) or exposed to a 15-minute light pulse (right) before dissection.

the ‘clock proteins’ (TABLE 1). For example, a heterodimer of two basic helix–loop–helix (bHLH)–PAS factors, CLOCK and BMAL1, binds to E-BOX ELEMENTS in the promoters of other clock genes and activates their expression10,14. The products of these genes then enter the nucleus and repress their own transcription by inhibiting CLOCK–BMAL15,15,16. So the action of CLOCK–BMAL1 on other clock-controlled genes leads to their oscillation and, hence, to a rhythmic output17. A similar picture — but involving different proteins — was uncovered in the fungus Neurospora crassa and in the cyanobacterium Synechococcus5. Although there are similarities between the fly and mammalian systems, there are differences too — particularly in the negative limb of the loop. In Drosophila, for example, PER and TIM dimerize, enter the nucleus5 and inhibit the binding of CLOCK–BMAL1 to DNA16, thereby repressing gene expression (FIG. 3). Moreover, in the fly, CRY acts mainly as a photoreceptor11,12: it binds to the PER–TIM dimer in a light-dependent fashion18, mediating the light-dependent degradation of TIM5. TIM may be dispensable for transcriptional inhibition, as indicated by the use of mutant flies with an extremely long period and a mutated tim gene19. In mammals, nuclear translocation is preceded by dimerization among PER-family members, or between PERs and the two homologues of CRY, CRY1 or CRY215,20–22. These complexes then negatively feed back on activation of their own genes by CLOCK–BMAL115,20,21 (FIG. 4). In this system, the CRYs seem to have a minor role in phototransduction, but they are crucial components of the core clock mechanism, as shown by targeted inactivation of their genes23. Indeed, they seem to be directly and centrally involved in negative action on CLOCK–BMAL1: first by mediating nuclear translocation of the PER–CRY complexes15, and second, by directly binding to the activators, perhaps as an interface between PERs and CLOCK–BMAL120. The situation may even be more complex in other vertebrates — close homologues of BMAL1 have been identified in the zebrafish (BMAL2)24 and in humans (MOP9)25. These BMALs interact with CLOCK and co-localize with it in circadian-clock structures. It is still unclear how mammalian TIM participates in this loop, and whether it is the true functional homologue of Drosophila TIM26. The most likely picture is that the circadian clock uses several loops — either inter- or intragenic — organized in networks. In both flies and mammals, a positive intergenic loop superimposes on the intragenic negative loops described above21,27 (FIGS 3, 4). In Drosophila, PER and TIM positively regulate expression of the Clock gene27,28, whereas in mammals, PER2 positively feeds back on Bmal1 expression21. So Drosophila Clock and mammalian Bmal1 oscillate in antiphase with the Per genes28,29. If levels of the oscillating activator are limiting (as is the case for Drosophila CLOCK30), functional CLOCK–BMAL1 is at peak levels when the cycle restarts. In this model, the rhythms result from both the negative effect on CLOCK–BMAL1 activity and the positive effect on expression of Clock or Bmal1. There is an analogous situation in Neurospora, where the frequency (frq) gene is www.nature.com/reviews/molcellbio

| OCTOBER 2000 | VOLUME 1

REVIEWS a

b Input

Pacemaker

Input

Output

Pacemaker

anisms34. The positive regulation of FRQ on WC-1 in Neurospora also occurs at the post-translational level31.

Output

Figure 2 | The circadian clock. a | The circadian clock can be divided into three conceptual components: the pacemaker, dedicated to generating and sustaining rhythms, receives and integrates signals from input pathways, and affects physiology through output pathways. b | Analysis of the roles played by clock genes has revealed that the conceptual components of the clock can overlap. The scheme can be applied to central master clocks, but also to peripheral clocks (BOX 2). (Adapted from REF. 13.)

regulated by a dimer composed of the two WHITE COLLAR factors, WC-1 and WC-2. The FRQ protein negatively regulates expression of its own gene and positively regulates expression of the WC-1 protein31. An example of two interconnected loops — and so the beginning of a network — is what can be defined as the ‘CRY loop’ and the ‘PER loop’ in mammals (FIG. 4)21,32. The CRY loop seems to be central in the negative regulation of CLOCK–BMAL1 (possibly in both a PERdependent and PER-independent manner). The PER loop, which in mammals takes various shapes depending on which PER protein is involved, would have other roles, including the positive action of PER2 on Bmal1. Finally, there are cases where the rhythms are not generated at the transcriptional level — for example, the Drosophila output gene pigment-dispersing factor (pdf). Whereas protein levels oscillate, the mRNA does not — in other words, the PDF oscillation is generated at the post-transcriptional level33. Another example is regulation of the per gene in the silkmoth Antheraea pernyi. Here, PER and TIM are not found in the nucleus in pacemaker cells of the central brain, indicating that per rhythmicity is obtained by post-translational mech-

PHASE SHIFT

A shift in the endogenous circadian rhythms that occurs when an organism is placed in different lighting regimes (or other external stimuli). SUBJECTIVE NIGHT AND DAY

Correspond to the endogenous night–day circadian cycle of an organism, independent of the astronomical day and night.

Nucleus

Cytosol Degradation

Cycle (Bmal ) +

Clock

CRY

TIM

TIM CRY

TIM

PER

PER CLOCK CYCLE TIM tim

PER

DBT

CCG

PER

Output function

Degradation

P P

per

DBT

ccg

Figure 3 | Model of the molecular mechanisms involved in the Drosophila circadian clock. At least two interlocked feedback loops are involved in generating rhythms: a negative autoregulatory loop of PER and TIM proteins on their own genes (through inhibition of the transcriptional activator CLOCK–CYCLE), and a positive effect of these proteins on the expression of CLOCK. CRY and DBT proteins regulate the stability of the TIM and PER proteins, respectively. CLOCK–CYCLE also regulates clock-controlled genes (ccg), and hence rhythmic output.

Regulation of the clock

Although the relatively straightforward mechanism of negative-feedback loops is needed to establish and maintain circadian rhythms, the clock has further levels of complexity: post-transcriptional regulation, posttranslational modifications, chromatin remodelling, availability of clock proteins, involvement of as-yetunidentified co-regulators and regulation of intracellular localization. Amazingly, these processes collectively ensure the precise periodicity of the clock: the oscillation is close to 24 hours, and the average deviation from one free-running cycle to another is remarkably low — of the order of one minute2. These regulatory pathways also provide an interface that can be used as an entry point for stimuli that can reset or control the clock. So different signalling and structural machines act in conjunction to coordinate the intricate processes involved in clock function. Signalling to the clock. Environmental light signals are relayed to the SCN through the retino-hypothalamic tract (RHT), using the neurotransmitter glutamate35. Activation of ionotropic glutamate receptors in SCN cells initiates a cascade of events leading to PHASE SHIFT of the clock. Some transcription factors are final effectors of this pathway, and induce immediate early genes such as c-fos (FIG. 1), fos-B or NGFI-A (REFS 36,37), and clock genes such as mPer1 and mPer2 (REFS 38,39). Transcriptional induction of these genes seems to be important for phase shifting of the clock40,41. Several studies highlight the involvement of cyclicAMP-responsive factors in the circadian clock. The cAMP-response element-binding protein (CREB) is activated by phosphorylation in the SCN on light stimulation in mice, at times where this stimulus produces a phase shift (that is, the SUBJECTIVE NIGHT)42. Phosphorylation occurs at a cAMP-dependent protein kinase (PKA) consensus serine (Ser 133) phosphoacceptor site. This allows the co-activator CREB-binding protein (CBP) to be recruited. CREB-mediated transcription in the SCN is responsive to light at the same times of the day43. In Drosophila, a mutation in the dCREB2 gene affects the period and the amplitude of the oscillation of a CREB–luciferase reporter, as well as expression of the per gene, although neither the levels of dCREB2 protein nor its phosphorylation oscillate44. The CREB protein is a convergence end point of several signal transduction pathways (FIG. 5)45. In particular, activation of the mitogen-activated protein kinase (MAPK)/extracellular signal-regulated kinase (ERK) pathway results in powerful activation of CREB by phosphorylation at Ser 133. MAPKs are activated in the mouse SCN in response to light pulses during the subjective night, and phosphorylation of ERKs and CREB overlap, at least partly, in timing and anatomical location46. Moreover, a MAPK-specific inhibitor attenuates agonist-induced CREB phosphorylation in cultured

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 1 | OCTOBER 2000 | 6 1

REVIEWS SCN neurons43. Cell cultures also show a role for MAPKs in rhythmic clock gene expression in response to an extracellular signal47. Furthermore, basal ERK phosphorylation shows circadian rhythmicity in the SCN46. The link between the glutamate receptor and activation of MAPKs and CREB (FIG. 5) was made by monitoring glutamate-induced phase shifts in the firing rate of SCN neurons in rat-brain slices. An early step is activation of nitric-oxide synthase and the subsequent calcium influx in SCN cells35. This indicates that nitric oxide diffuses between SCN cells, perhaps contributing to synchronization among clock cells and amplification of the signal35. Why does light have opposite effects during the early and late night; that is, delay and advance, respec-

tively? In SCN slices, a cyclic-GMP-dependent protein kinase (PKG) inhibitor blocks glutamate-induced phase advance but not phase delay48. In contrast, release of intracellular calcium by the endoplasmic reticulum mediates phase delays (and not advances), as depletion of calcium stores and inhibitors of ryanodine receptor calcium channels selectively block glutamate- and lightinduced phase delays48. The molecular events downstream of nitric oxide and upstream of MAPKs within this pathway are still undefined. Their elucidation may tell us why light phase-shifts the clock during the subjective night but not during the subjective day. Other neurotransmitters are also involved in nonphotic signalling to the SCN49,50, in the communication

Table 1 | Clock genes in animals and fungi Gene

Organism

Characteristics of protein product

Function (in the circadian system)

cry

Drosophila

Similar to DNA photolyases

Conveys photic information to PER–TIM

Immediate early genes (c-fos)?

Mammals

Transcription factors

Induced by light in the suprachiasmatic nucleus; important for photic entrainment

per1, 2

Mammals

PAS domain

Induced by light in the SCN; Per1 important for photic entrainment

wc-1

Neurospora

PAS domain; GATA transcription factor Required for light-induced frq expression

Input genes

Pacemaker genes clock

Mammals, birds, fish, bHLH–PAS factor amphibians

Dimerizes with BMAL1 and activates clock and clock-controlled gene expression

per

Mammals (3 genes), fish, insects

Negatively regulates CLOCK–BMAL1-driven transcription; positively regulates bmal (mammals) or clock (Drosophila) expression

tim

Mammals (?), fish Drosophila

bmal1 (cycle)

Mammals (2 genes?) Fish (2 genes) Drosophila

bHLH–PAS factor

Dimerizes with CLOCK and activates CLOCK and CLOCK-controlled gene expression (function of MOP9, a close human homologue, is not defined)

cry

Mammals (2 genes) Drosophila

Similar to DNA photolyases

In Drosophila, binds TIM and inhibits PER–TIM dimer In mammals, negatively regulates CLOCK–BMAL1-driven transcription; translocates PER to the nucleus

CKIε (doubletime)

Mammals Drosophila

Serine/threonine protein kinase

Phosphorylates and destabilizes PER (at least PER1 in mammals)

vrille?

Drosophila

Similar to PAR leucine-zipper factors

Affects per and tim expression

dbp?

Mammals

PAR leucine-zipper transcription factor May be involved in per1 expression

frq

Neurospora

wc-1

Neurospora

PAS domain; GATA transcription factor Dimerizes with WC-2 and activates frq expression

wc-2

Neurospora

PAS domain; GATA transcription factor Dimerizes with WC-1 and activates frq expression

Drosophila

Neuropeptide

PAS domain

In Drosophila, dimerizes with PER to repress CLOCK– BMAL1-driven transcription and activate clock expression

Negatively regulates WC-1/WC-2-driven transcription of its own gene; positively regulates WC-1 expression

Output genes pdf

Links the molecular clock to behaviour

avp

Mammals

Vasopressin peptide

Controls the activity of various output pathways

vrille

Drosophila

Similar to PAR leucine-zipper factors

Indirectly controls PDF peptide oscillation

dbp

Mammals

PAR leucine-zipper transcription factor Controls rhythmic expression of various genes; influences circadian behaviour

CREM

Mammals

bZip transcription factor

The repressor isoform ICER is involved in the rhythmic expression of NAT, the key enzyme in melatonin synthesis

takeout

Drosophila

Similarity to ligand-binding proteins

Involved in output pathways that link the clock to feeding

www.nature.com/reviews/molcellbio

| OCTOBER 2000 | VOLUME 1

REVIEWS between SCN cells51,52 and in modulating the response to light53. These include GABA (γ-aminobutyric acid), dopamine, serotonin, neuropeptide Y and the hormone melatonin49,53–55. So a variety of signalling pathways must be activated in clock cells in response to these stimuli, and the signalling molecules activated vary according to the cell type and/or the incoming signal (FIG. 5).

HISTONE ACETYLTRANSFERASES AND DEACETYLASES

Enzymes that modify histones by adding and removing acetyl groups, a chemical modification thought to remodel chromatin structure.

Chromatin remodelling and the clock. Regulation of gene expression has been linked to changes in the chromatin architecture, such as histone modifications including acetylation, phosphorylation, methylation, ubiquitination and ADP-ribosylation56. These modifications are thought to provide plasticity to chromatin structure, inducing dynamic condensation and decondensation57. As clock function and integration of inputs rely on transcriptional regulation, is chromatin remodelled during circadian cycles or in response to signals that regulate the clock? Transcription factors are only some of the possible nuclear targets of signalling pathways. Histones are also end-points of signal-transduction cascades55. Modifications, mostly to the histone termini (or tails), are coupled to cellular phenomena such as transcription, silencing, mitosis, meiosis and chromatin assembly. Many transcriptional regulators — in particular coactivators and co-repressors — are HISTONE ACETYLTRANS-

Cytosol

Nucleus ? + ?

Bmal1

Cytoplasmic retention and degradation

PER2

Clock

PER ? CRY

PER

? P P

CRY

PER

CLOCK BMAL1

CKIε TIM mCry1

(HATs) or HISTONE DEACETYLASES57. Although the co-activators contacted by the CLOCK–BMAL1 heterodimer have not been characterized, there are clues from transcription factors of the same family. AHR and AHR nuclear translocator (ARNT) are two bHLH–PAS factors which, together, activate transcription of genes in response to toxic hydrocarbons such as dioxin. The AHR–ARNT heterodimer induces changes in chromatin structure58, and ARNT interacts with CBP and p300 (REF. 59), two large co-activators known to have HAT activity. It is tempting to speculate that the CLOCK–BMAL1 dimer would also modulate histone acetylation by the involvement of co-activators (for example, p300/CBP). In fact, ARNT proteins are not the only bHLH–PAS proteins to interact with p300/CBP. The nuclear hormone receptor co-activator SRC-1 (which shows HAT activity on its own too) also interacts with p300 (REF. 60). In some cases, signalling cascades may end in modification of both transcription factors and chromatin, offering a powerful and synergistic way to regulate gene expression. Rsk-2, a member of the p90rsk family of kinases activated by MAPKs, phosphorylates nuclear targets including CREB, leading to the activation of early-response genes such as c-fos (REF. 61). Rsk-2 is also required for EGF-stimulated phosphorylation of histone H3 in fibroblasts62. Could H3 be an end point of the light-response pathway in the SCN? In answering this we may find that light affects chromatin remodelling in cells of the circadian clock (FIG. 5). FERASES

CRY1

? CRY2

CKIε

mCry2 PER1 mPer1 PER2 mPer2 PER3 mPer3 CCG ccg Output function

Figure 4 | Model of the molecular mechanisms involved in the mammalian circadian clock. At least two interlocked feedback loops are involved in generating rhythms: a negative autoregulatory loop of CRY and possibly PER proteins on their own genes (through inibition of the transcriptional activator CLOCK–BMAL1), and a positive effect of PER2 on the expression of BMAL1. CKIε regulates the stability and the nuclear translocation of PER proteins. CLOCK–BMAL1 also regulates clock-controlled genes (ccg), and hence rhythmic output.

Transcript and protein stability. An essential prerequisite for obtaining oscillations from negative-feedback loops is a short half-life of mRNAs and/or clock proteins. Work in flies, on post-transcriptional and posttranslational regulation of clock elements63,64, fits theoretical models predicting that degradation of clock mRNAs and proteins is crucial to the control of oscillator periodicity65. For example, the extent to which gene expression is repressed by the PER–TIM heterodimer in Drosophila can be modulated through the stability of PER and TIM. A kinase encoded by the dbt gene reduces the stability of monomeric PER by phosphorylating it66. Flies with an inactivated dbt gene accumulate hypophosphorylated PER. Moreover, some point mutations of this gene lead to longer or shorter cycle periods. PER1 protein is phosphorylated and destabilized by casein kinase Iε (CKIε), the mammalian homologue of DOUBLETIME, and mutation of the gene in the hamster produces short-period animals8. There is an analogous situation in Neurospora, where phosphorylation of FRQ causes its degradation, and so influences period length67. Protein degradation can also integrate input signals. In Drosophila, TIM is degraded in response to light5. Degradation requires the light-dependent association of TIM with CRY, which acts as a photoreceptor18. In Neurospora, light triggers phosphorylation and, ultimately, degradation of WC-1 (REF. 68). A similar concept might apply to plants, as members of a new family of

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 1 | OCTOBER 2000 | 6 3

REVIEWS

UBIQUITIN LIGASE

Photic Non-photic cues cues

RHT

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

Median raphe

IGL

5-HT1B

Synapse Serotonin (5-HT)

Glu

NPY

PHOTOLYASE

A DNA repair enzyme that splits pyrimidine dimers (lesions caused by UV irradiation) into monomers.

NMDA (+ others)

cAMP

Ca2+ influx NOS ? (cGMP) GC

PKG (phase advance)

PKA

5-HT1A 5-HT7

Melatonin SCN neuron Mel1a(+ other?)

? PKC?

K+ channels

NO? Effect on CREB?

Ryanodine receptor Ca2+ release from ER (phase delay) ?

Phase shift

Phase shift + Acute inhibition of SCN neuronal firing

ERK Kinase (?) P H

P CREB

Chromatin remodelling ?

ieg Cytosol

CRE P H

P CREB

Per1, Per2 Nucleus

CRE Downregulation

Figure 5 | Signalling within a suprachiasmatic nucleus synapse and neuron. The suprachiasmatic nucleus (SCN) circadian clock is affected by light (photic cues) through the retino-hypothalamic tract (RHT) leading to use of glutamate (Glu) as a neurotransmitter. Glutamate receptor triggering induces various intracellular responses, leading ultimately to gene expression and phase shifts. Non-photic cues involve a variety of other neurotransmitter and signalling pathways (three of which are shown here). The photic and non-photic pathways can cross-regulate each other pre- and postsynaptically, this scheme is therefore an oversimplification. AC, adenylate cyclase; CRE, cAMP response element; ER, endoplasmic reticulum; ERK, extracellular signalregulated kinase; GC, guanylyl cyclase; H, histone; ieg, immediate early genes (for example, c-fos); IGL, nerve terminus from the intergeniculate leaflet (thalamus); NMDA, N-methyl-D-aspartate; NOS, nitric oxide synthase; NPY, neuropeptide Y; PKA, cAMPdependent protein kinase; PKC, protein kinase C; PKG, cGMP-dependent protein kinase.

proteins mediating light modulation of the circadian clock bear an F-box, a domain that recruits substrates to the UBIQUITIN-LIGASE complex69,70. Nucleo-cytoplasmic shuttling. Another regulatory checkpoint for circadian feedback loops is the entry of clock proteins to the nucleus. Whereas in Drosophila PER and TIM heterodimerize and enter the nucleus5,19, in mammals the translocation is preceded by associations among PERs and CRYs15,20–22. So Drosophila TIM19 and mammalian CRYs15 could be viewed as shuttling carriers of PER proteins. As well as decreasing the stability of PER, CKIε seems to control the nuclear entry of this protein. It is proposed that CKIε binds and phosphorylates PER1, causing a nuclear-localization signal to be masked so the protein is retained in the cytoplasm71. So CKIε and the CRYs could be seen as components of the clock with antagonistic effects on PER function, promoting retention of PERs in the cytoplasm and retardation of the cycle, or entry of PERs into the nucleus and closing of the negative-feedback loop, respectively. A balance of CKIε and CRY abundance and activity would

finely tune the period of the circadian cycle. Photoreception

How is the light signal transduced to the clock within the SCN? What are the clock photoreceptive cells? And what is the identity of the photoreceptor? The Drosophila CRY protein seems to act as a photoreceptor for the circadian clock11,12, being directly involved in regulating light-responsiveness of PER and TIM18. This pterin/flavin-coupled protein, similar to 32 PHOTOLYASES, absorbs energy from blue/ultraviolet light . A role for mouse CRY homologues in phototransduction is less evident. In the mouse, CRY proteins are central to the oscillator mechanism23 as they are involved in light-independent inhibition of the CLOCK/BMAL1 activators15,20,21. Moreover, mice mutated for the mCry1 and mCry2 genes still show activation of mPers upon light stimulation72,73. Although these observations do not rule out a role for these proteins in light reception in mammals, they have prompted a search for other candidates74. Light induces phase shifts of locomotor activity75 and www.nature.com/reviews/molcellbio

| OCTOBER 2000 | VOLUME 1

REVIEWS

OPSINS

Hydrophobic glycoproteins found in the visual pigments of vertebrates.

normal suppression of pineal melatonin synthesis76 in transgenic mice that lack both rods and cones. Removal of the eyes abolishes these responses74, indicating that phototransduction mediated by cones and rods is not necessary for light-regulated clock function and modulation of temporal physiology, and that there must be other photoreceptors in the retina. Action spectra for phase-shifting of circadian rhythms77,78 indicate that the photoreceptor involved may be an OPSIN coupled with a retinal-based photopigment74. Opsin-like molecules have been identified in the retina (and other photoreceptive tissues) of fish and amphibians79,80. One of these, melanopsin, has also been found in the human inner retina81. Moreover, an opsin has been found in the blind mole rat, a small rodent with a degenerated visual system, but whose circadian clock can be entrained by light82. Could one of these molecules act as a photoreceptor for the vertebrate clock? Independent peripheral clocks have also been found in the tissues of various animals (BOX 2). Is the same light-reception mechanism used in central and peripheral clocks? Considering its wide tissue distribution, CRY could be the photoreceptor of all — both central and peripheral — circadian clocks in Drosophila, although there is no experimental evidence for this. In

Box 2 | A multitude of independent clocks Independent peripheral Central clock clocks 12 The general view, of a single circadian pacemaker that 9 3 controls all circadian Humoral or 6 neuronal physiology (BOX 1), had to Direct signals be modified after clocks light were found in peripheral, signals 12 12 12 non-neuronal tissues. Peripheral 9 9 9 3 3 3 Clocks independent of the clocks 6 6 6 central pacemaker were found first in the hamster retina99, and subsequently in Drosophila wings, antennae100 or Malpighian Rhythms tubules101, in zebrafish driven by Rhythms driven by central hearts or kidneys102 and in peripheral clocks clocks rat fibroblasts87. Different organs from rats bearing a Per1 promoter–luciferase reporter transgene show rhythms in constant conditions in vitro, finally showing that the concept also applies to mammalian organs103. The circadian system Circadian rythms are generated by the central clock, which is under the influence of light (see figure). Peripheral clocks are probably under the direct control of the central clock, although they may also be regulated by other stimuli (for example, light in the zebrafish). Peripheral clocks are devoted to some tissue-specific outputs (such as the olfactory response in Drosophila104). Although they can function independently, even in vivo101, they can also receive inputs in alternative ways: either by direct light signals (as shown in Drosophila100,101 and in zebrafish83) or by signals from the central clock85,105, which can be neural85, diffusible or hormonal105. So the SCN directly induces a variety of physiological rhythms, while synchronizing or directing peripheral clocks, which may have distinctive characteristics. These may depend on the nature of the SCN signals they receive, on how they interpret these signals24 and on the time they take to react to them103.

the zebrafish, organs and cells in culture are also light responsive83, but further data are needed to define the contribution of cryptochromes. Mammals are not as translucent as the zebrafish or Drosophila, and one could speculate about the physiological significance of peripheral photoreception during evolution. However, light can reach inner parts of the mammalian body, including the brain84. Remaining questions include the possibility of circadian phototransduction in non-retinal mammalian tissues, and the presence of interplay between the photoreceptors involved in the different clocks. The clock outputs

The SCN affects physiology in different ways85. Output signals use efferent nerves, to various areas of the brain85 and to tissues, which connect through the autonomic nervous system86. In some cases, a diffusible signal has been invoked to justify the transmission of rhythmic information8. Peripheral clocks (BOX 2) may be controlled or synchronized by the central clock (for example, the SCN) through humoral signals — that is bloodborne signalling factors oscillating in a circadian fashion87. How is the output regulated at the molecular level? Examples of clock-controlled genes whose expression is activated by binding of the CLOCK–BMAL1 dimer to an E-box in their promoters include vrille in Drosophila88, and avp and dbp in the mouse17,89. Some output molecules are peptides. In Drosophila, the product of the pdf gene is involved in behavioural rhythms90. The promoter region of the mouse gene encoding the arginine vasopressin peptide (AVP) contains an E-box, and shows circadian activity that is severely blunted in Clock mutant mice17. AVP defines a subpopulation of SCN cells,and is thought to modulate the amplitude of firing-rate rhythm from the SCN. This drives rhythmic activity of the hypothalamo-pituitaryadrenal axis and controls levels of serotonin N-acetyltransferase and melatonin synthesis by the pineal gland17. Clock-controlled genes have been identified by screening for transcripts that undergo daily fluctuations88,91. Screens looking for Drosophila transcripts with phases similar to per and tim led to the isolation of a number of genes, some encoding transcription factors. One of them, vrille, seems to be more than just an output gene — it may be part of the oscillator itself88. Indeed, continuous expression of vrille alters the expression of per, tim and pdf. The VRILLE protein is similar to mammalian transcription factors of the proline and acidic-amino-acid-rich domain (PAR) leucine-zipper family. In fact, a well-characterized output protein of the mouse clock, D-site albumin promoter-binding protein (DBP), and other oscillating factors belong to this family of transcription factors92. DBP is rhythmically expressed in various tissues and drives the expression of several genes, particularly in the liver. Because expression of DBP is controlled by CLOCK–BMAL189, this transcription factor could be involved in output of the clock, driving the rhythmic expression of other clock-controlled genes. In support of this idea, mice mutated

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 1 | OCTOBER 2000 | 6 5

REVIEWS for the dbp gene, although rhythmic, show altered locomotor activity93. And DBP has been proposed to regulate mPer1 expression directly through a DBPbinding site in the promoter of the gene94. Another transcription factor involved in output clock function is CRE modulator (CREM), a member of the CREB family. One of the isoforms generated from the CREM gene, inducible cAMP early repressor (ICER), modulates melatonin synthesis in the mouse by directly regulating expression of the serotonin Nacetyltransferase gene, which contains a CRE in its promoter region95. Future directions

Discoveries and surprises about circadian clocks will undoubtedly continue at the same tremendous pace. What can we expect in the next few months and years? As far as molecular mechanisms in the core pacemaker are concerned, the roles of protein degradation and intracellular shuttling of clock molecules will show their regulatory functions. Similarly, there are probably many more co-regulators — as we know they exist in other transcriptional settings — whose function may be crucial for the clock. Regarding the organization of the circadian system in mammals, several questions stand out. First, how do SCN neurons communicate and synchronize, a problem linked to uncovering signalling pathways to the clock? Second, what are the individual functions of neuronal subpopulations within the SCN, particularly with

Menaker, M., Moreira, L. F. & Tosini, G. Evolution of circadian organization in vertebrates. Braz. J. Med. Biol. Res. 30, 305–313 (1997). 2. Pittendrigh, C. S. Temporal organization: reflections of a Darwinian clock-watcher. Annu. Rev. Physiol. 55, 16–54 (1993). An autobiographical essay by one of the fathers of modern chronobiology, covering the discovery of the basic principles of the discipline over fifty years. 3. Hastings, M. H. Central clocking. Trends Neurosci. 20, 459–464 (1997). 4. Moore, R. Y. & Silver, R. Suprachiasmatic nucleus organization. Chronobiol. Int. 15, 475–487 (1998). 5. Dunlap, J. C. Molecular bases for circadian clocks. Cell 96, 271–290 (1999). A comprehensive review on the genetics of the circadian clocks in cyanobacteria, fungi, Drosophila, mammals and plants. 6. Konopka, R. J. & Benzer, S. Clock mutants of Drosophila melanogaster. Proc. Natl Acad. Sci. USA 68, 2112–2116 (1971). 7. King, D. P. et al. Positional cloning of the mouse circadian Clock gene. Cell 89, 641–653 (1997). The conclusion of an impressive study that led to the identification of the first mammalian clock gene, named Clock. 8. Lowrey, P. L. et al. Positional syntenic cloning and functional characterization of the mammalian circadian mutation tau. Science 288, 483–492 (2000). 9. Hogenesch, J. B., Gu, Y. Z., Jain, S. & Bradfield, C. A. The basic-helix-loop-helix-PAS orphan MOP3 forms transcriptionally active complexes with circadian and hypoxia factors. Proc. Natl Acad. Sci. USA 95, 5474–5479 (1998). 10. Gekakis, N. et al. Role of the CLOCK protein in the mammalian circadian mechanism. Science 280, 1564–1569 (1998). A key paper about a basic device, the feedback loop: the positive elements of the loop were finally uncovered. 11. Emery, P., So, W. V., Kaneko, M., Hall, J. C. & Rosbash, M. CRY, a Drosophila clock and light-regulated cryptochrome,

12.

13. 14.

15.

16.

17.

18.

19.

20.

21.

22.

regard to afferent and efferent pathways? Third, what are the photoreceptors for entrainment of the clock? And finally, what mechanisms synchronize peripheral clocks with the central pacemaker? Another issue is whether clock proteins may elicit other functions not related to the clock. One candidate is mammalian Tim, which is probably not the true homologue of fly tim, and has a role in early development26. Similarly, BMAL1 can interact with other bHLH–PAS factors besides CLOCK9, increasing the combinatorial possibilities of its function. Finally, studies on cocaine sensitization in Drosophila clock mutants96 indicate new routes for exploring the connections between clock biology and behaviour. The circadian-clock field has entered an exciting phase. After several years in which many clock genes have been identified, it is now time to put the puzzle pieces together to understand how the signals are integrated into a precise clockwork. It is clear, though, that this puzzle is incomplete, and that new pieces will have to be identified through genetic screens and other molecular approaches. Links DATABASE LINKS Per | tim | clock | cycle | dbt |Bmal1 | PAS

domain | AHR | SIM | CRY | bHLH domain | CRY1 | CRY2 | MOP9 | pdf | c-fos | fos-B | NFGI-A | CREB | PKA | CBP | bHLH–PAS domain | PKG | ARNT | p300 | SRC-1 | Rsk-2 | H3 | CKIε | vrille | avp | dbp | CREM ENCYCLOPEDIA OF LIFE SCIENCES Circadian rhythms

is a major contributor to circadian rhythm resetting and photosensitivity. Cell 95, 669–679 (1998). Circadian photoreception in Drosophila is the duty of cryptochrome, a homologue of light-receiving proteins in plants. Stanewsky, R. et al. The cryb mutation identifies cryptochrome as a circadian photoreceptor in Drosophila. Cell 95, 681–692 (1998). Foster, R. G. & Lucas, R. J. Clocks, criteria and critical genes. Nature Genet. 22, 217–219 (1999). Darlington, T. K. et al. Closing the circadian loop: CLOCKinduced transcription of its own inhibitors per and tim. Science 280, 1599–1603 (1998). Kume, K. et al. mCRY1 and mCRY2 are essential components of the negative limb of the circadian clock feedback loop. Cell 98, 193–205 (1999). Lee, C., Bae, K. & Edery, I. PER and TIM inhibit the DNA binding activity of a Drosophila CLOCK-CYC/dBMAL1 heterodimer without disrupting formation of the heterodimer: a basis for circadian transcription. Mol. Cell. Biol. 19, 5316–5325 (1999). Jin, X. et al. A molecular mechanism regulating rhythmic output from the suprachiasmatic circadian clock. Cell 96, 57–68 (1999). This was the first paper establishing a molecular link between the central clockworks and the expression of an output gene. Ceriani, M. F. et al. Light-dependent sequestration of TIMELESS by CRYPTOCHROME. Science 285, 553–556 (1999). Rothenfluh, A., Young, M. W. & Saez, L. A TIMELESSindependent function for PERIOD proteins in the Drosophila clock. Neuron 26, 505–514 (2000). Griffin, E. A. Jr, Staknis, D. & Weitz, C. J. Light-independent role of CRY1 and CRY2 in the mammalian circadian clock. Science 286, 768–771 (1999). Shearman, L. P. et al. Interacting molecular loops in the mammalian circadian clock. Science 288, 1013–1019 (2000). Yagita, K. et al. Dimerization and nuclear entry of mPER proteins in mammalian cells. Genes Dev. 14, 1353–1363 (2000).

23. van der Horst, G. T. et al. Mammalian Cry1 and Cry2 are essential for maintenance of circadian rhythms. Nature 398, 627–630 (1999). 24. Cermakian, N., Whitmore, D., Foulkes, N. S. & SassoneCorsi, P. Asynchronous oscillations of two zebrafish CLOCK partners reveal differential clock control and function. Proc. Natl Acad. Sci. USA 97, 4339–4344 (2000). 25. Hogenesch, J. B. et al. The basic helix-loop-helix-PAS protein MOP9 is a brain-specific heterodimeric partner of circadian and hypoxia factors. J. Neurosci. 20, 1–5 (2000). 26. Gotter, A. L. et al. A time-less function for mouse timeless. Nature Neurosci. 3, 755–756 (2000). 27. Glossop, N. R., Lyons, L. C. & Hardin, P. E. Interlocked feedback loops within the Drosophila circadian oscillator. Science 286, 766–768 (1999). 28. Bae, K., Lee, C., Sidote, D., Chuang, K. Y. & Edery, I. Circadian regulation of a Drosophila homolog of the mammalian Clock gene: PER and TIM function as positive regulators. Mol. Cell. Biol. 18, 6142–6151 (1998). 29. Oishi, K., Sakamoto, K., Okada, T., Nagase, T. & Ishida, N. Antiphase circadian expression between BMAL1 and period homologue mRNA in the suprachiasmatic nucleus and peripheral tissues of rats. Biochem. Biophys. Res. Commun. 253, 199–203 (1998). 30. Bae, K., Lee, C., Hardin, P. E. & Edery, I. dCLOCK is present in limiting amounts and likely mediates daily interactions between the dCLOCK-CYC transcription factor and the PER-TIM complex. J. Neurosci. 20, 1746–1753 (2000). 31. Lee, K., Loros, J. J. & Dunlap, J. C. Interconnected feedback loops in the Neurospora circadian system. Science 289, 107–110 (2000). 32. Lucas, R. J. & Foster, R. G. Circadian clocks: A cry in the dark? Curr. Biol. 9, 825–828 (1999). 33. Park, J. H. et al. Differential regulation of circadian pacemaker output by separate clock genes in Drosophila. Proc. Natl Acad. Sci. USA 97, 3608–3613 (2000). 34. Gotter, A. L., Levine, J. D. & Reppert, S. M. Sex-linked period genes in the silkmoth, Antheraea pernyi: implications for circadian clock regulation and the evolution of sex chromosomes. Neuron 24, 953–965 (1999). 35. Ding, J. M. et al. Resetting the biological clock: mediation of

www.nature.com/reviews/molcellbio

| OCTOBER 2000 | VOLUME 1

REVIEWS

36.

37.

38.

39.

40.

41.

42.

43.

44.

45.

46.

47.

48.

49. 50.

51.

52.

53.

54.

55.

56. 57.

58.

59.

60.

nocturnal circadian shifts by glutamate and NO. Science 266, 1713–1717 (1994). Kornhauser, J. M., Nelson, D. E., Mayo, K. E. & Takahashi, J. S. Photic and circadian regulation of c-fos gene expression in the hamster suprachiasmatic nucleus. Neuron 5, 127–134 (1990). Morris, M. E., Viswanathan, N., Kuhlman, S., Davis, F. C. & Weitz, C. J. A screen for genes induced in the suprachiasmatic nucleus by light. Science 279, 1544–1547 (1998). Albrecht, U., Sun, Z. S., Eichele, G. & Lee, C. C. A differential response of two putative mammalian circadian regulators, mper1 and mper2, to light. Cell 91, 1055–1064 (1997). Shearman, L. P., Zylka, M. J., Weaver, D. R., Kolakowski, L. F. Jr & Reppert, S. M. Two period homologs: circadian expression and photic regulation in the suprachiasmatic nuclei. Neuron 19, 1261–1269 (1997). Akiyama, M. et al. Inhibition of light- or glutamate-induced mPer1 expression represses the phase shifts into the mouse circadian locomotor and suprachiasmatic firing rhythms. J. Neurosci. 19, 1115–1121 (1999). Wollnik, F. et al. Block of c-Fos and JunB expression by antisense oligonucleotides inhibits light-induced phase shifts of the mammalian circadian clock. Eur. J. Neurosci. 7, 388–393 (1995). Ginty, D. D. et al. Regulation of CREB phosphorylation in the suprachiasmatic nucleus by light and a circadian clock. Science 260, 238–241 (1993). Obrietan, K., Impey, S., Smith, D., Athos, J. & Storm, D. R. Circadian regulation of cAMP response element-mediated gene expression in the suprachiasmatic nuclei. J. Biol. Chem. 274, 17748–17756 (1999). Belvin, M. P., Zhou, H. & Yin, J. C. The Drosophila dCREB2 gene affects the circadian clock. Neuron 22, 777–787 (1999). de Cesare, D., Fimia, G. M. & Sassone-Corsi, P. Signaling routes to CREM and CREB: plasticity in transcriptional activation. Trends Biochem. Sci. 24, 281–285 (1999). Obrietan, K., Impey, S. & Storm, D. R. Light and circadian rhythmicity regulate MAP kinase activation in the suprachiasmatic nuclei. Nature Neurosci. 1, 693–700 (1998). Akashi, M. & Nishida, E. Involvement of the MAP kinase cascade in resetting of the mammalian circadian clock. Genes Dev. 14, 645–649 (2000). Ding, J. M. et al. A neuronal ryanodine receptor mediates light-induced phase delays of the circadian clock. Nature 394, 381–384 (1998). Hastings, M. H. et al. Non-photic signalling in the suprachiasmatic nucleus. Biol. Cell 89, 495–503 (1997). Maywood, E. S., Mrosovsky, N., Field, M. D. & Hastings, M. H. Rapid down-regulation of mammalian period genes during behavioral resetting of the circadian clock. Proc. Natl Acad. Sci. USA 96, 15211–15216 (1999). Shinohara, K., Hiruma, H., Funabashi, T. & Kimura, F. GABAergic modulation of gap junction communication in slice cultures of the rat suprachiasmatic nucleus. Neuroscience 96, 591–596 (2000). Welsh, D. K., Logothetis, D. E., Meister, M. & Reppert, S. M. Individual neurons dissociated from rat suprachiasmatic nucleus express independently phased circadian firing rhythms. Neuron 14, 697–706 (1995). Pickard, G. E. & Rea, M. A. Serotonergic innervation of the hypothalamic suprachiasmatic nucleus and photic regulation of circadian rhythms. Biol. Cell 89, 513–523 (1997). Liu, C. et al. Molecular dissection of two distinct actions of melatonin on the suprachiasmatic circadian clock. Neuron 19, 91–102 (1997). Liu, C. & Reppert, S. M. GABA synchronizes clock cells within the suprachiasmatic circadian clock. Neuron 25, 123–128 (2000). Cheung, P., Allis, C. D. & Sassone–Corsi, P. Signaling to chromatin through histone modifications. Cell (in the press). Workman, J. L. & Kingston, R. E. Alteration of nucleosome structure as a mechanism of transcriptional regulation. Annu. Rev. Biochem. 67, 545–579 (1998). Ko, H. P., Okino, S. T., Ma, Q. & Whitlock, J. P. Jr Dioxininduced CYP1A1 transcription in vivo: the aromatic hydrocarbon receptor mediates transactivation, enhancerpromoter communication, and changes in chromatin structure. Mol. Cell. Biol. 16, 430–436 (1996). Kobayashi, A., Numayama-Tsuruta, K., Sogawa, K. & FujiiKuriyama, Y. CBP/p300 functions as a possible transcriptional coactivator of Ah receptor nuclear translocator (Arnt). J. Biochem. 122, 703–710 (1997). Yao, T. P., Ku, G., Zhou, N., Scully, R. & Livingston, D. M. The nuclear hormone receptor coactivator SRC-1 is a specific target of p300. Proc. Natl Acad. Sci. USA 93,

10626–10631 (1996). 61. de Cesare, D., Jacquot, S., Hanauer, A. & Sassone-Corsi, P. Rsk-2 activity is necessary for epidermal growth factorinduced phosphorylation of CREB protein and transcription of c-fos gene. Proc. Natl Acad. Sci. USA 95, 12202–12207 (1998). 62. Sassone-Corsi, P. et al. Requirement of Rsk-2 for epidermal growth factor-activated phosphorylation of histone H3. Science 285, 886–891 (1999). 63. Dembinska, M. E., Stanewsky, R., Hall, J. C. & Rosbash, M. Circadian cycling of a PERIOD-beta-galactosidase fusion protein in Drosophila: evidence for cyclical degradation. J. Biol. Rhythms 12, 157–172 (1997). 64. So, W. V. & Rosbash, M. Post-transcriptional regulation contributes to Drosophila clock gene mRNA cycling. EMBO J. 16, 7146–7155 (1997). 65. Ruoff, P., Vinsjevik, M., Monnerjahn, C. & Rensing, L. The Goodwin oscillator: on the importance of degradation reactions in the circadian clock. J. Biol. Rhythms 14, 469–479 (1999). 66. Price, J. L. et al. double-time is a novel Drosophila clock gene that regulates PERIOD protein accumulation. Cell 94, 83–95 (1998). This was the first paper to identify a regulatory protein involved in the modification of other clock proteins. 67. Liu, Y., Loros, J. & Dunlap, J. C. Phosphorylation of the Neurospora clock protein FREQUENCY determines its degradation rate and strongly influences the period length of the circadian clock. Proc. Natl Acad. Sci. USA 97, 234–239 (2000). 68. Talora, C., Franchi, L., Linden, H., Ballario, P. & Macino, G. Role of a white collar-1-white collar-2 complex in blue-light signal transduction. EMBO J. 18, 4961–4968 (1999). 69. Nelson, D. C., Lasswell, J., Rogg, L. E., Cohen, M. A. & Bartel, B. FKF1, a clock-controlled gene that regulates the transition to flowering in Arabidopsis. Cell 101, 331–340 (2000). 70. Somers, D. E., Schultz, T. F., Milnamow, M. & Kay, S. A. ZEITLUPE encodes a novel clock-associated PAS protein from Arabidopsis. Cell 101, 319–329 (2000). 71. Vielhaber, E., Eide, E., Rivers, A., Gao, Z. H. & Virshup, D. M. Nuclear entry of the circadian regulator mPER1 is controlled by mammalian casein kinase I epsilon. Mol. Cell. Biol. 20, 4888–4899 (2000). 72. Okamura, H. et al. Photic induction of mPer1 and mPer2 in cry-deficient mice lacking a biological clock. Science 286, 2531–2534 (1999). 73. Vitaterna, M. H. et al. Differential regulation of mammalian period genes and circadian rhythmicity by cryptochromes 1 and 2. Proc. Natl Acad. Sci. USA 96, 12114–12119 (1999). 74. Foster, R. G. Shedding light on the biological clock. Neuron 20, 829–832 (1998). 75. Freedman, M. S. et al. Regulation of mammalian circadian behavior by non-rod, non-cone, ocular photoreceptors. Science 284, 502–504 (1999). Along with reference 76, this represented a big leap forward in the search for the mammalian circadian photoreceptor: it is neither the cones nor the rods, and so lies elsewhere in the retina. 76. Lucas, R. J., Freedman, M. S., Munoz, M., GarciaFernandez, J. M. & Foster, R. G. Regulation of the mammalian pineal by non-rod, non-cone, ocular photoreceptors. Science 284, 505–507 (1999). 77. Provencio, I. & Foster, R. G. Circadian rhythms in mice can be regulated by photoreceptors with cone-like characteristics. Brain Res. 694, 183–190 (1995). 78. Takahashi, J. S., DeCoursey, P. J., Bauman, L. & Menaker, M. Spectral sensitivity of a novel photoreceptive system mediating entrainment of mammalian circadian rhythms. Nature 308, 186–188 (1984). 79. Kojima, D., Mano, H. & Fukada, Y. Vertebrate ancient-long opsin: a green-sensitive photoreceptive molecule present in zebrafish deep brain and retinal horizontal cells. J. Neurosci. 20, 2845–2851 (2000). 80. Soni, B. G., Philp, A. R., Foster, R. G. & Knox, B. E. Novel retinal photoreceptors. Nature 394, 27–28 (1998). 81. Provencio, I. et al. A novel human opsin in the inner retina. J. Neurosci. 20, 600–605 (2000). 82. David-Gray, Z. K., Janssen, J. W., DeGrip, W. J., Nevo, E. & Foster, R. G. Light detection in a ‘blind’ mammal. Nature Neurosci. 1, 655–656 (1998). 83. Whitmore, D., Foulkes, N. S. & Sassone-Corsi, P. Light acts directly on organs and cells in culture to set the vertebrate circadian clock. Nature 404, 87–91 (2000). The striking message of this paper is that the peripheral organs and cells of zebrafish not only behave as independent clocks, but they have all the machinery to receive light and interpret these signals. 84. Wade, P. D., Taylor, J. & Siekevitz, P. Mammalian cerebral

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

85. 86. 87.

88. 89.

90.

91.

92.

93.

94. 95.

96.

97.

98.

99. 100.

101.

102.

103.

104.

105.

cortical tissue responds to low-intensity visible light. Proc. Natl Acad. Sci. USA 85, 9322–9366 (1988). LeSauter, J. & Silver, R. Output signals of the SCN. Chronobiol. Int. 15, 535–550 (1998). Ueyama, T. et al. Suprachiasmatic nucleus: a central autonomic clock. Nature Neurosci. 2, 1051–1053 (1999). Balsalobre, A., Damiola, F. & Schibler, U. A serum shock induces circadian gene expression in mammalian tissue culture cells. Cell 93, 929–937 (1998). This paper reports the first case of circadian oscillations in non-neuronal mammalian cells in culture. Blau, J. & Young, M. W. Cycling vrille expression is required for a functional Drosophila clock. Cell 99, 661–671 (1999). Ripperger, J. A., Shearman, L. P., Reppert, S. M. & Schibler, U. CLOCK, an essential pacemaker component, controls expression of the circadian transcription factor DBP. Genes Dev. 14, 679–689 (2000). Renn, S. C., Park, J. H., Rosbash, M., Hall, J. C. & Taghert, P. H. A pdf neuropeptide gene mutation and ablation of PDF neurons each cause severe abnormalities of behavioral circadian rhythms in Drosophila. Cell 99, 791–802 (1999). Sarov-Blat, L., So, W. V., Liu, L. & Rosbash, M. The Drosophila takeout gene is a novel molecular link between circadian rhythms and feeding behavior. Cell 101, 647–656 (2000). Mueller, C. R., Maire, P. & Schibler, U. DBP, a liver-enriched transcriptional activator, is expressed late in ontogeny and its tissue specificity is determined posttranscriptionally. Cell 61, 279–291 (1990). Lopez-Molina, L., Conquet, F., Dubois-Dauphin, M. & Schibler, U. The DBP gene is expressed according to a circadian rhythm in the suprachiasmatic nucleus and influences circadian behavior. EMBO J. 16, 6762–6771 (1997). Yamaguchi, S. et al. Role of DBP in the circadian oscillatory mechanism. Mol. Cell. Biol. 20, 4773–4781 (2000). Foulkes, N. S., Borjigin, J., Snyder, S. H. & Sassone-Corsi, P. Rhythmic transcription: the molecular basis of circadian melatonin synthesis. Trends Neurosci. 20, 487–492 (1997). Andretic, R., Chaney, S. & Hirsh, J. Requirement of circadian genes for cocaine sensitization in Drosophila. Science 285, 1066–1068 (1999). Kuhlman, S. J., Quintero, J. E. & McMahon, D. G. GFP fluorescence reports Period 1 circadian gene regulation in the mammalian biological clock. Neuroreport 11, 1479–1482 (2000). Liu, C., Weaver, D. R., Strogatz, S. H. & Reppert, S. M. Cellular construction of a circadian clock: period determination in the suprachiasmatic nuclei. Cell 91, 855–860 (1997). Tosini, G. & Menaker, M. Circadian rhythms in cultured mammalian retina. Science 272, 419–421 (1996). Plautz, J. D., Kaneko, M., Hall, J. C. & Kay, S. A. Independent photoreceptive circadian clocks throughout Drosophila. Science 278, 1632–1635 (1997). This paper presents the first description of independent non-neural peripheral clocks in animals. Giebultowicz, J. M., Stanewsky, R., Hall, J. C. & Hege, D. M. Transplanted Drosophila excretory tubules maintain circadian clock cycling out of phase with the host. Curr. Biol. 10, 107–110 (2000). Whitmore, D., Foulkes, N. S., Strahle, U. & Sassone-Corsi, P. Zebrafish Clock rhythmic expression reveals independent peripheral circadian oscillators. Nature Neurosci. 1, 701–707 (1998). Yamazaki, S. et al. Resetting central and peripheral circadian oscillators in transgenic rats. Science 288, 682–685 (2000). Krishnan, B., Dryer, S. E. & Hardin, P. E. Circadian rhythms in olfactory responses of Drosophila melanogaster. Nature 400, 375–378 (1999). Sakamoto, K. et al. Multitissue circadian expression of rat period homolog (rPer2) mRNA is governed by the mammalian circadian clock, the suprachiasmatic nucleus in the brain. J. Biol. Chem. 273, 27039–27042 (1998).

Acknowledgements We wish to apologize to all colleagues whose work, because of lack of space, could not be cited. We thank all the members of the Sassone-Corsi laboratory for helpful discussions. N.C. was supported by a Human Frontier Science Program Organization longterm fellowship and a Canadian Institutes of Health Research postdoctoral fellowship. Work in our laboratory is supported by grants from CNRS, INSERM, CHUR, Human Frontier Science Program, Organon Akzo/Nobel, Fondation pour la Recherche Médicale and Association pour la Recherche sur le Cancer.

VOLUME 1 | OCTOBER 2000 | 6 7

PERSPECTIVES TIMELINE

The past, present and future of molecular computing Adam J. Ruben and Laura F. Landweber Ever since scientists discovered that conventional silicon-based computers have an upper limit in terms of speed, they have been searching for alternative media with which to solve computational problems. That search has led them, among other places, to DNA.

2 1

vin 0 vout

When most people think of a ‘DNA computer’, the first image that springs to mind is a personal computer-like interface with microcentrifuge tubes lined up inside the central processor and a keyboard that plugs directly into the molecule’s 5′ end. Perhaps someday such a project will become reality. At the moment, however, ‘DNA computing’ is the slightly misleading title applied to experiments in which DNA molecules have computational roles. Often sequences of about 8–20 base pairs are used to represent bits, and numerous methods have been devised to manipulate and evaluate them. DNA is a convenient choice, as it is both self-complementary, allowing single-stranded DNA to select its own Watson – Crick complement, and can easily be copied. Also, molecular biologists have already built a toolbox for manipulating DNA, including restriction enzyme digestion, ligation, sequencing, amplification and fluorescent labelling, giving DNA a head start over alternative computational media. This unique combination of computer science and molecular biology has fascinated the world for nearly six years, perhaps because it finally links two popular disciplines that we

4 5

Figure 1 | An example of a seven-vertex, 13edge graph. The ‘salesman’ starts his journey at vertex 0 (vin) and ends at vertex 6 (vout).

have always hoped would be innately linked. Much of the human body is, in theory, a flood of binary operations. We could be staring into the abyss of a science as doomed as phrenology or mesmerism. But we may be at the forefront of a new and creative technology whose implications have not even been fully mapped out, let alone realized. Fifty years from now, molecular computing may conceivably be important in our lives — who, fifty years ago, could have predicted the personal computer? — and it all began with an experiment published by a computer scientist in Science in 1994.

ness of solutions is easy to check, and ‘NPcomplete’ problems are considered the hardest of the class because they require exponentially increasing amounts of time to solve.) The problem asks whether, given a set of n cities (‘vertices’) with m paths (‘edges’) connecting them, a ‘hamiltonian’ path exists that starts at a given vertex vin, passes through each vertex exactly once, and ends at vertex vout. For fairly small values of n, today’s computers easily solve this problem. However, when n becomes very large, the amount of time required to generate and check every possible solution increases exponentially, making very large calculations infeasible. The version Adleman solved contained only seven vertices (FIG. 1), which is no remarkable feat in the world of computer science. He used a simple, brute-force algorithm: generate random paths through the graph, discard any that do not begin at vin and end at vout , discard any that do not enter exactly n vertices, and discard any that do not pass through each vertex at least once. But the use of DNA to solve this small computational problem marked the genesis of a new scientific field. Adleman began by synthesizing a random 20-base-pair DNA oligonucleotide (20-mer) to represent each vertex, followed by another series of 20-mers to represent edges. The edge DNA had a certain built-in feature: in each 20-mer, the first ten nucleotides complemented the last ten of one vertex, and the last ten complemented the first ten of another vertex (FIG. 2). For example, a 20-base-pair edge connecting vertex one to vertex two would consist of the complements to vertex one’s last ten base pairs and vertex two’s first ten. That way, when the mixture of DNA is Path 1–2

The Travelling Salesman Problem

In a seminal paper, Leonard Adleman1 solved a simplified instance of a famous NP-complete computer problem called the Travelling Salesman Problem. (‘NP’ is the name given to the class of search problems for which correct-

Figure 2 | Edge DNA forming a splint bridging two vertices. The last five nucleotides of vertex 1, and the first five of vertex 2, complement the nucleotides on vertex 1–2.

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 1 | OCTOBER 2000 | 6 9

PERSPECTIVES denatured and then allowed to anneal and ligate, the edge will form a splint to ‘connect’ both vertices, and T4 DNA ligase can join together all possible such combinations. It is the bulk-annealing step that allows this molecular approach to test a massive number of possibilities in parallel. This initial denaturing/hybridization/ligation step generated over 1013 strands of DNA. Among these, it was probable that at least one encoded the hamiltonian path. Next, all sequences that began at vin and ended at vout were selectively amplified by the polymerase chain reaction (PCR) using primers recognizing those sequences. Any path that did not pass through exactly seven points was then eliminated by gel-purifying only the 140base-pair product (necessarily containing seven 20-mers). Finally, to eliminate solutions that did not pass through each vertex exactly once, the product from the previous step was affinity purified by denaturing the doublestranded paths and removing the biotinylated complementary strand on magnetic beads. The first affinity target was a single-stranded 20-mer complementary to the sequence of the second vertex, so only paths that contained the second ‘city’ were retained. This process was repeated for every vertex except the first and the last (as all paths were already bounded by vin and vout PCR primers). The presence of a DNA band at the end indicated that, for the given graph, a hamiltonian path exists. Confirmation that a path a

i 0

200

150

1 i h g f e d

100 c b

a 50

Figure 3 | Solving the Knight Problem. Multiplex linear PCR produces a ‘bar-code’ representing two configurations of knights on a 3 × 3 chessboard3.

“Here’s nature’s toolbox … a bunch of little tools that are dirt cheap; you can buy a DNA strand for 100 femtocents.” exists and ‘readout’ of the correct solution required graduated PCR on the final product. This involved a series of six different PCR reactions, each using the vin forward primer and a primer complementary to each of the other 20-mer vertices, which were then analysed in separate lanes on a gel. (So, for example, a readout showing bands of 40, x, 60, 80, 100 and 120 base pairs, where x represents the absence of a band in the second lane, would represent the path 0→1, 1→3, 3→4, 4→5, 5→6. This is not a hamiltonian path, as it misses the second vertex, and so would have been eliminated in the size-purification step.) Adleman’s path, however, did indeed pass through each vertex exactly once. Although several of these methods have been modified and improved, graduated PCR remains a simple and rapid method for readout that bypasses DNA sequencing. As this was the first experimental demonstration of DNA computing, Adleman reflected on some possible implications. This computation required about seven days of laboratory work, so for large n such a process could prove unwieldy, unless some sort of automation or alternative algorithm could be devised. However, a typical desktop computer can execute about 106 operations per second, and the fastest supercomputer can execute about 1012 — but Adleman’s molecular computation, if each ligation counts as an operation, did over 1014. Scaling up the ligation step could push the number over 1020 operations per second. Furthermore, it used an extremely small quantity of energy — 2 × 1019 operations (ligations) per joule, whereas the second law of thermodynamics dictates a theoretical maximum of 34 × 1019 operations per joule. Modern supercomputers only operate at 109 operations per joule1. Finally, one bit of information can be stored in a cubic nanometre of DNA, which is about 1012 times more efficient than existing storage media1. So molecular computers have the potential to be far faster and more efficient than any electronics we have developed. There is, of course, a possibility for error, although in this example the thoroughness of each step reduces the chance of survival of an incorrect path. However, in larger instances of the same algorithm, errors are more likely to propagate.

Six years later, DNA computing is still in its infancy. Researchers have developed several algorithms to solve classic problems, and a handful have been tested and work. Despite current progress, we are still a long way away from solving complex problems. Computing on surfaces

There is an interesting and potentially useful approach now available to the field of DNA computing: computing on surfaces2. This involves affixing the solution DNA strands, correct and incorrect, to a solid medium, and — in a subtractive algorithm, for example — selectively destroying the incorrect ones. Liu et al.2 used this approach to solve a small instance of an NP-complete satisfiability (SAT) problem involving Boolean logic. They began with a logical statement in four variables divided into four connected sections, or ‘clauses’. As each variable could be either 0 or 1, there existed 24 = 16 possible solutions to the problem, four of which were correct. A DNA strand corresponding to each possibility was synthesized. Each strand had the format 5′-FFFFvvvvvvvvFFFF-3′, where the eight bases labelled ‘F’ are fixed (for use in the PCR amplification step) and ‘vvvvvvvv’ is a unique octanucleotide corresponding to a different predetermined solution. The strands were bound to solid media (a maleimidefunctionalized gold surface), and a fixed 21mer spacer sequence was attached to the 5′ end in order to separate the solution sequence from the support. All DNA was single stranded. The algorithm used at this point involved a cycle of mark–destroy–unmark operations. First the sequences encoding correct solutions were marked (by hybridization to their complements), then all unmarked (singlestranded) sequences were destroyed using Escherichia coli exonuclease I, and finally the remaining solutions were unmarked by removing their complements. This cycle was then repeated for each clause of the problem. A distinct advantage of surface computing

Figure 4 | Five correct solutions to the Knight Problem and one incorrect solution. In the last solution, the white knight in square i (see FIG. 3 for key) threatens the knights in squares b and d.

| OCTOBER 2000 | VOLUME 1

www.nature.com/reviews/molcellbio

PERSPECTIVES is the readout step. This involves a second surface, which is a DNA chip — an addressed array of surface-bound oligonucleotides. The remaining solution strands were amplified by PCR (using FFFF-region primers) and fluorescein labelled. They were then hybridized to a DNA chip on which each of the original 16 solution molecules were bound in a predetermined spot. The four solutions hybridized to their four corresponding spots, and their presence was then detected by a fluorescence intensity of 10 to 777 times the background intensity of the other 12 spots. So a lot of technology was needed to recover the four sequences out of 16, but again it proved the principle. DNA chips allow more rapid throughput than direct sequencing, particularly in the crucial readout step. However, graduated PCR1 has the advantage that it can not only determine which strand is present, but it also measures its length, which is necessary for certain problems. The benefits of using the chip method are also sizeable. For example, each 16-base (FFFFvvvvvvvvFFFF) oligonucleotide ‘word’ can be extended to several words in sequence. This sort of flexibility allows for easy scaling up, meaning that a much more complex, chip-based DNA computer could be designed2. Another advantage is the ease of using solid-state media such as DNA chips. This allows the DNA to be easily separated from solutions, and so no column separation or nucleic acid precipitation steps are necessary. The only components bound to the chip are the DNA and anything attached to it. This greatly streamlines complex, repetitive chemical processes and, perhaps most importantly for the field of DNA computing, opens a door towards automation. RNA computing

Earlier this year our laboratory demonstrated the first use of RNA to solve a computational problem3. Known as the Knight Problem, this nine-bit SAT problem (the largest molecular computation solved so far) asks, given a 3 × 3 chessboard, what configurations of knights may be placed on the board such that none threatens any others. (The knight can attack any piece placed two spaces away from it in one direction and one space in the other — that is, moving in an ‘L’ shape.) There are 94 correct solutions (out of a possible 29 = 512), ranging from one solution with zero knights on the board to two solutions with five knights on the board. We generated a ten-bit DNA data pool, with a tenth bit included in case one of the previous nine should compute less reliably (similar to having ‘spare blocks’ on computer disks). Each RNA strand in the data pool con-

1 2 3 4

5 6 7

Macronucleus

Micronucleus 1

4 7

Figure 5 | DNA computing in vivo. Scrambled genes in some ciliates — microbial eukaryotes of the genus Stylonychia or Oxytricha — undergo massive rearrangement to form a functional gene in the macronucleus (top) that encodes a single protein product. Telomere repeats (yellow boxes) mark and protect the surviving ends. Dispersed coding segments 1–7 (blue boxes) become joined at their ends, analogous to the assembly of a seven-vertex path through a graph, such as the one in FIG. 1.

sisted of a series of ten 15-nucleotide bits (each of which can represent either 1 or 0, representing the presence or absence of a knight in this case) separated by nine 5-nucleotide spacers. The 5′ ‘prefix’ and 3′ ‘suffix’ sequences at the beginning and end permit PCR amplification and T7 RNA transcription of each library strand. As Liu et al.2 did, we separated the correct solutions from the data pool by ‘destroying’ the incorrect solutions. This time, however, the enzyme used was ribonuclease H (RNase H) because it digests RNA sequences hybridized to DNA, thus destroying specifically marked strands rather than unmarked ones. Furthermore, the use of RNase H in combination with any complementary DNA oligonucleotide is universally scalable, allowing specific digestion of many ‘words’ in parallel with a single enzyme. This flexibility in cleaving target sequences was an important incentive for choosing RNA. With DNA, computing can be done with the finite catalogue of restriction enzymes4, but these require specific, often palindromic, sequences and the use of several enzymes — that may not be compatible — in one pot. Our readout involved multiplex linear PCR, a process similar to Adleman’s graduated PCR1, except that we combined all the necessary primers in one reaction tube5, producing a bar-code pattern of all the amplified products in two lanes on a polyacrylamide gel (FIG. 3)3. Readouts were taken of 43 randomly sampled strands remaining at the end of the algorithm, of which 42 offered correct solutions to the Knight Problem (FIG. 4). Attention therefore focused on the forty-third solution: If it was incorrect, why had it not been cleaved by RNase H? As it turned out, the RNA strand contained an adjacent point mutation and a deletion in bit nine, preventing its hybridization to the complementary

DNA strand and hence digestion by RNase H. So accurate size purification of the data pool (with a resolution of 1–2 nucleotides) would effectively eliminate this most common source of error. However, the 2.3% error (incorrect placement of one knight out of 127 on 43 boards), although not bad in a normal molecular biology experiment, is still higher than that in a computer chip. Fortunately in this type of search problem one can easily check if the recovered solution is correct; hence it is robust to small amounts of error. Pros and cons

Molecular computing may or may not be a wave of the future, paving the way for technological advances in chemistry, computer science and biology. At its present stage of development it has several challenges. First, the materials used, whether DNA, RNA or proteins, are not reusable1. Whereas a silicon computer can operate indefinitely with electricity as its only input, a molecular computer would require periodic refuelling and cleaning. This illustrates a second important drawback to DNA computing — the molecular components used are generally specialized. A very different set of oligonucleotides is used to solve the Knight Problem3 than is used to find a hamiltonian path1. A third notable downside is the error rate — typically, in vivo, with all the cellular machinery functioning properly, a mismatch error occurs only once every billion base pairs or so6. However, when DNA is used to compute in vitro, the conditions are hardly as good as those in vivo, and although the error rate may seem acceptable for other experiments, computational problems generally require higher standards. In order to rival conventional computers, all of the procedures described here would need to lower their error rates to possibly unattainable levels.

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 1 | OCTOBER 2000 | 7 1

PERSPECTIVES Finally, the most apparent drawback is the time required for each computation. Whereas a simple desktop computer can solve the seven-city instance of the Travelling Salesman Problem in less than a second, Adleman took seven days1. The use of DNA chips2 or other approaches may eventually lead to automation, which would save considerable amounts of time, but fundamental DNA computing technology needs to advance far beyond its current bounds before it can be made practical. DNA computing has its advantages, though. One is its massive parallelism — that is, brute-force algorithms can search through quadrillions of molecules at the same time and find a correct solution, akin to in vitro selection3. Another is miniaturization. And once the procedures are under control, the raw materials cost less too. “Here’s nature’s toolbox,” commented Adleman7,“a bunch of little tools that are dirt cheap; you can buy a DNA strand for 100 femtocents.” The near future

Now is an exciting time in the field of DNA computing, as there is so much that has not been tried. In June, over 120 molecular biologists, computer scientists, mathematicians and chemists from around the world gathered in Leiden8 to discuss the latest in DNA computing technology. Clearly a next step is automation. McCaskill and colleagues in Germany have constructed a ‘microflow reactor’ on which they propose to solve a 20-bit satisfiability problem in an hour and a half 8. One could also construct a microfluidic device consisting of gated channels so small that only one molecule can pass through at a time 9, vastly improving readout8. And a team led by Adleman recently solved a 6-variable, 11clause satisfiability problem using a ‘dry’ computer consisting of thin, gel-filled glass tubes8. As for DNA chips, their future in DNA computing looks bright as well, because ‘universal’ DNA chips could contain every possible DNA sequence of a given length (probably about 8–12 base pairs). Hagiya and colleagues in Tokyo are finding creative uses for singlestranded DNA molecules that fold into intrastrand ‘hairpins’8,10. Winfree, Seeman and colleagues — responsible for construction of beautiful assemblies with DNA, such as a DNA nano-cube11 — have proposed the assembly of even more ordered structures that show patterned algorithmic supramolecular self-assembly8,11–13. Even a handful of mathematicians have lent a hand, proposing faster and more efficient algorithms tailored to the needs of DNA computing8.

Whether or not nucleic acid computers ultimately prove feasible, they have already contributed to multi-disciplinary science by causing us to question the nature of computing and to forge new links between the biological and computational sciences. For example, it has led us to focus on the nature of biological DNA computations, such as the assembly of modern genes from encrypted buildingblocks in the genomes of some single-celled ciliates (FIG. 5)14. After all, our bodies already contain millions of complicated, efficient, evolved molecular computers called cells. Adam J. Ruben and Laura F. Landweber are in the Department of Ecology and Evolutionary Biology, Princeton University, Princeton, New Jersey 08544, USA. e-mail: lfl@princeton.edu

Links

7. 8.

10. 11. 12.

FURTHER INFORMATION DNA computing: a

primer | Laura Landweber’s homepage 13. 1. 2.

Adleman, L. Molecular computation of solutions to combinatorial problems. Science 266, 1021–1023 (1994). Liu, Q. et al. DNA computing on surfaces. Nature 403,

14.

175–179 (2000). Faulhammer, D., Cukras, A., Lipton, R. J. & Landweber, L. F. Molecular computation: RNA solutions to chess problems. Proc. Natl Acad. Sci. USA 97, 1385–1389 (2000). Ouyang, Q., Kaplan, P. D., Liu, S. & Libchaber, A. DNA solution of the maximal clique problem. Science 278, 446–449 (1997). Henegariu, O., Heerema, N. A., Dlouhy, S. R., Vance, G. H. & Vogt, P. H. Multiplex PCR: Critical parameters and stepby-step protocol. Biotechniques 23, 504–511 (1997). Karp, G. Cell and Molecular Biology: Concepts and Experiments 2nd edn (John Wiley & Sons, New York, 1999). Seife, C. RNA works out knight moves. Science 287, 1182–1183 (2000). Condon, A. & Rozenberg, G. (eds) Prelim. Proc. 6th Int. Meet. DNA Based Computers (Leiden Univ., The Netherlands, 2000). Meller, A. et al. Rapid nanopore discrimination between single polynucleotide molecules. Proc. Natl Acad. Sci. USA 97, 1079–1084 (2000). Sakamoto, K. et al. Molecular computation by DNA hairpin formation. Science 288, 1223–1226 (2000). Seeman, N. C. DNA engineering and its application to biotechnology. Trends Biotechnol. 17, 437–443 (2000). Winfree, E. et al. in DNA Based Computers II. DIMACS Series in Discrete Mathematics and Theoretical Computer Science Vol. 44 (eds Landweber, L. F. & Baum, E. B.) (American Mathematical Soc., Providence, Rhode Island, 1999). Winfree, E., Liu, F., Wenzler, L. A. & Seeman, N. C. Design and self-assembly of two-dimensional DNA crystals. Nature 394, 539–544 (1998). Landweber, L. F., Kuo, T.-C. & Curtis, E. A. Evolution and assembly of an extremely scrambled gene. Proc. Natl Acad. Sci. USA 97, 3298–3303 (2000).

TIMELINE

Hayflick, his limit, and cellular ageing Jerry W. Shay and Woodring E. Wright Almost 40 years ago, Leonard Hayflick discovered that cultured normal human cells have limited capacity to divide, after which they become senescent — a phenomenon now known as the ‘Hayflick limit’. Hayflick’s findings were strongly challenged at the time, and continue to be questioned in a few circles, but his achievements have enabled others to make considerable progress towards understanding and manipulating the molecular mechanisms of ageing.

To set Hayflick’s discoveries in context, we need to go back to 1881 (TIMELINE, overleaf), when the German biologist August Weismann1 speculated that “death takes place because a worn-out tissue cannot forever renew itself, and because a capacity for increase by means of cell division is not everlasting but finite”. This concept, which was almost entirely forgotten by the time Hayflick began his work, was later challenged by the French Nobel-prize-winning surgeon Alexis Carrel, who suggested that all cells explanted

in culture are immortal, and that the lack of continuous cell replication was due to ignorance on how best to cultivate the cells. Carrel’s view was based on his and Albert Ebeling’s work, done at the Rockefeller Institute in New York City, in which they claimed that chick heart fibroblasts grew con-

Figure 1 | Leonard Hayflick in 1988. (Photograph: Peter Argentine.)

| OCTOBER 2000 | VOLUME 1

www.nature.com/reviews/molcellbio

Links

7. 8.

10. 11. 12.

FURTHER INFORMATION DNA computing: a

primer | Laura Landweber’s homepage 13. 1. 2.

Adleman, L. Molecular computation of solutions to combinatorial problems. Science 266, 1021–1023 (1994). Liu, Q. et al. DNA computing on surfaces. Nature 403,

14.

TIMELINE

Figure 1 | Leonard Hayflick in 1988. (Photograph: Peter Argentine.)

| OCTOBER 2000 | VOLUME 1

www.nature.com/reviews/molcellbio

PERSPECTIVES tinuously for 34 years2. This led to the general idea that all vertebrate cells could divide indefinitely in cell culture. However, Carrel’s original observations could not be reproduced by other scientists3,4, and may have been due to an experimental error4. The cells were fed with a daily extract of chick embryo tissue extracted under conditions that permitted the addition of fresh living cells to the culture at each feeding3. It has been suggested that Carrel knew about this error but never admitted it5,6, but even if this explanation is untrue, no one has ever confirmed Carrel’s work. The Carrel experiments were of great importance because, if valid, they meant that normal cells freed from in vivo control mechanisms could function normally and, apparently, forever. However, reports were beginning to emerge of difficulties in long-term cell culture when Leonard Hayflick (FIG. 1) and Paul Moorhead entered the field. They brilliantly got to the heart of the matter, demonstrating finite replicative capacity of normal human fibroblasts and interpreting the phenomenon as ageing at the cellular level3,4. These initial observations sparked Leonard Hayflick’s passion — which has lasted his entire career — to overturn the central dogma that all vertebrate cells grown in culture are immortal. But even today, there are sceptics. One is Harry Rubin7, who stated: “The concept of a genetically predetermined number of human fibroblast replications, and its implied extension to other cells, is based on an artefact resulting from the damage accumulated by the explanted cells during their replication in the radically foreign environment of cell culture.” Rubin is not alone in his opinion, and perhaps the truth lies somewhere in between. Nevertheless, the Hayflick limit is now generally accepted. How Hayflick found his limit

After obtaining his Ph.D. in 1956 from the University of Pennsylvania, Hayflick spent two years with one of the leading personalities in tissue culture at that time, Charles M. Pomerat, at the University of Texas in Galveston. In 1958, Hayflick was recruited to run the Wistar Institute’s cell-culture laboratory and also to initiate research on the possible viral aetiology of human cancer. He intended to expose normal human embryonic cells to cancer-cell extracts, in the hope of observing cancer-like changes in normal cells. When the normal cells no longer grew (FIG. 2), Hayflick thought he might have made a mistake in preparing the culture medium or washing glassware, or made some other technical oversight. He was assuming that Carrel was correct, and that cells could propagate indefinitely if provided with appropriate conditions. After

“The largest fact to have come from tissue culture in the last fifty years is that cells inherently capable of multiplying will do so indefinitely if supplied with the right milieu in vitro.” all, it had been 60 years since Ross Harrison had started the field of cell culture, and normal cultured cells were thought to be immortal. For Hayflick to propose that a cell-division counting mechanism could be involved in ageing was a completely new idea. But Hayflick was young and ambitious, and a series of carefully conducted experiments over about three years convinced him that the failure of his normal cells to replicate indefinitely was not due to technical errors. In 1961, working with the talented cytogeneticist Paul Moorhead, Hayflick did a series of experiments that challenged Carrel’s views. Hayflick and Moorhead showed that populations of cultured normal human fibroblasts doubled a finite number of times, after which the cells stopped dividing and entered what Hayflick termed the phase III phenomenon3. He called the primary culture phase I; the ten or so months of luxuriant growth, phase II; and the period when cell replication diminished and ultimately stopped, phase III (FIG. 3). These initial experiments showed that the previous interpretation — that all cells are immortal — was incorrect. The principle behind these experiments was simple: Hayflick and Moorhead mixed equal numbers of normal human male fibroblasts that had divided many times (cells at the fortieth population doubling) with female fibroblasts that had divided only a few times (cells at the tenth population doubling). Unmixed cell populations were kept as controls. When the male ‘control’ culture stopped dividing, the mixed culture was examined and only female cells were found. This showed that the old cells ‘remembered’ they were old, even when surrounded by young cells, and that technical errors or contaminating viruses were unlikely explanations as to why only the male cell component had died3. Hayflick was convinced that normal cells have a finite capacity to replicate, and appreciated that their behaviour differed profoundly from that of cultured cancer cells (for example, HeLa cells) and transplantable tumours, which are immortal. It was this insight that originated the concept of immortalization of

normal cells3,4. The experiment with mixed cells further assured Hayflick and Moorhead that culture artefacts could not explain their observations. They submitted a paper describing their findings to the Journal of Experimental Medicine but Peyton Rous, one of the journal’s editors, was not easily persuaded. After the paper had been peerreviewed, Rous included the following statement in his covering letter:“The largest fact to have come from tissue culture in the last fifty years is that cells inherently capable of multiplying will do so indefinitely if supplied with the right milieu in vitro.” The article was not accepted. Fortunately, the editors of Experimental Cell Research, where the paper was published3 in 1961, were less swayed by the dogma of the day. This work and subsequent studies (TIMELINE) changed the tenor of research, eventually leading Sir Macfarlane Burnett, Nobel laureate from Australia, to coin the phrase “the Hayflick limit” for the first time in his book Intrinsic Mutagenesis, published8 in 1974. Hayflick’s enduring impact

The durability and importance of Hayflick’s work are reflected in its citation history. Between 1961 and 1999 this paper was cited about 3,000 times. Of the roughly 70 million scientific papers published since 1945, only one in every 135,000 has been cited as many times or more than this paper. Eugene Garfield9, editor of Current Contents, stated

Figure 2 | Young and old human diploid cells (strain WI-38). a | Young cells in phase II at population doubling 20. b | Old cells in phase III at population doubling 55.

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 1 | OCTOBER 2000 | 7 3

PERSPECTIVES

Woodring Wright shows that the replicometer is located in the nucleus while studying for a Ph.D. in Leonard Hayflick’s laboratory12.

Timeline | Hayflick and his limit August Weismann proposes that worn-out tissues occur because cell division is finite and this leads to decline in organ performance1.

1881

Alexis Carrel refutes Weismann’s model2.

1907

1921

Ross Harrison describes the ability to maintain cells in culture.

Hayflick receives his Ph.D. in medical microbiology and chemistry from the University of Pennsylvania.

Hayflick and Moorhead discover the finite lifetime of cultured normal human cells and interpret this finding as a manifestation of human ageing at the cellular level3.

Hayflick recognizes that a direct relationship may exist between the population-doubling potential of cultured cells and the maximum lifespan of species from which they are taken39.

1956

1961

1973

1928

Leonard Hayflick is born on 20 May 1928 in Philadelphia.

in 1983: “By studying accelerated ageing under glass, as Hayflick calls it, we can learn a great deal about changes in ageing cells that could contribute to functional losses throughout our bodies. Therefore, it is not surprising that research on tissue culture in ageing research is one of the most active ageing research fronts.” A cellular counting mechanism

The existence of a counting mechanism is implied by two of Hayflick’s observations: first, that normal cultured human fetal cells only undergo a specific number of population

1958

After a post-doctoral fellowship at the University of Texas, Galveston, Hayflick returns to Philadelphia, where he spends ten years as an associate member of the Wistar Institute.

1965

Hayflick describes memory in cultured normal human cells: cells reconstituted from the frozen state remember at what population doubling level they were frozen and undergo further doublings only up to a predetermined maximum4,10.

doublings, and second, that cryogenically preserved cells can ‘remember’ how many times they have divided when they have frozen10. Although this mechanism has been referred to as a clock or timing mechanism, the replicative limit of normal cells is actually related to rounds of DNA replication, and not to the passage of time. Hayflick suggested the term “replicometer” be used to designate the putative molecular event counter11. So what is the molecular basis of the replicometer? In 1975, a doctoral student in Hayflick’s laboratory, Woodring Wright, showed that the replicometer was located in the nucleus12.

Box 1 | Hayflick’s achievements During his distinguished career, Hayflick has made several fundamental observations and is often credited with starting the field of cellular gerontology — the study of ageing at the cellular level. Hayflick, who is now a professor of anatomy on the faculty of the University of California, San Francisco, was Editor-in-Chief of Experimental Gerontology for 13 years, president of the Gerontological Society of America, chairman of the Scientific Review Board of the American Federation for Aging Research, and a founding member and chairman of the executive committee of the Council of the National Institute on Aging, NIH. He has received more than 25 major awards, is a fellow of the American Association for the Advancement of Science, an honorary member of the Tissue Culture Association, and author of over 225 scientific papers, reviews and the popular book How and Why We Age (Ballantine Books, New York, 1995). Hayflick’s achievements extend beyond cellular gerontology: he is also an accomplished microbiologist and was appointed Professor of Medical Microbiology at the Stanford University School of Medicine, Stanford, California, in 1968. He developed the first normal human diploid fibroblast cell strains. One of these, called WI-38, is still the most widely used and highly characterized normal human cell strain in the world10. He described the extraordinary sensitivity of cultured normal human fibroblasts to human viruses and suggested that these cells could be used for virus isolation, identification and vaccine production. He was the first to produce a vaccine (oral polio vaccine) from these cells44. WI-38 cells, or similar human-cell strains, are used today for the manufacture of most human virus vaccines throughout the world45, including rubella and the Salk polio vaccine. Over 750 million virus vaccine doses have been produced on WI-38 or similar diploid cell strains. Hayflick established international standards for the production of human biologicals in passaged cells, which are still used today by the biotechnology industry46. Hayflick is also known for discovering the cause of primary atypical pneumonia in humans. This type of pneumonia was thought to be of viral origin, but Hayflick showed that it is caused by Mycoplasma pneumoniae, a member of the smallest free-living class of microorganisms47. Mycoplasma pneumoniae was first grown by Hayflick on a medium that he developed48.

1974

1975

Macfarlane Burnett coins the phrase “the Hayflick limit” to describe Hayflick’s discovery that normal cells have finite capacity to replicate as opposed to cancer cells, which usually become immortal8.

1978

Elizabeth Blackburn discovers the sequence of the Tetrahymena telomere19.

Telomeres and telomerase

In the early 1970s it was realized that the properties of DNA replication prevent the cells from fully copying the ends of linear DNA, called telomeres. Because of the nature of lagging-strand synthesis, DNA polymerase cannot completely replicate the 3′ end of linear duplex DNA. This was referred to as the endreplication problem (FIG. 4) in 1972 by one of the discoverers of the double helix, James Watson13. At around the same time, Alexey Olovnikov, a Russian theoretical biologist, had heard a lecture in which Hayflick’s work was discussed. Olovnikov entered a Moscow subway station while wondering how normal cells might have a limited capacity to replicate, and, as the train stopped, he had a flash of insight. Olovnikov saw an analogy between the train representing the DNA polymerase and the track representing the DNA. If the train replicated the DNA track underneath the car, the first segment of DNA would not be replicated because it was underneath the engine at the start14. This was analogous to the end-replication problem described by Watson. Olovnikov realized that this repeated shortening of the DNA molecule at each round of DNA replication might explain Hayflick’s finding that normal cells can replicate only a specific number of times. Although published in both Russian and English15,16, Olovnikov’s ideas languished in the literature until the golden era of molecular biology emerged in the late 1970s. The presence of telomeres at the tips of chromosomes had been noted at least since a lecture given by Hermann Muller17 in 1938 and the work of Barbara McClintock18. However, the function of these structures in cell replication was unclear. There was evidence that telomeres prevented the ends of chromosomes from fusing to each other and

| OCTOBER 2000 | VOLUME 1

www.nature.com/reviews/molcellbio

PERSPECTIVES

Jerry Shay and Geron scientists show that telomerase is present in all cancer-derived cell lines and in 90% of primary human cancers28.

Carol Greider discovers telomerase26.

1981

1985

Hayflick and colleagues transform a normal human cell population to an immortal cell line with a chemical carcinogen and radiation40.

Roslin Institute scientists claim that the cloned sheep, Dolly, has shorter telomeres than an age-matched control41. Is Dolly a sheep in lamb’s clothing?

1990

1994

1998

Woodring Wright and Geron scientists show that ectopic expression of telomerase in normal fibroblasts and epithelial cells bypasses the Hayflick limit31, showing that the telomeres are the cellular replicometer.

Calvin Harley shows that telomeres shorten as cultured normal cells approach the Hayflick limit22.

Accumulated population doublings

that they allowed chromosome ends to attach to the nuclear envelope in some species. Fastforward to 1978, when Elizabeth Blackburn, working in Joseph Gall’s group, found that the telomeres of the ciliated protozoan, Tetrahymena thermophila, consisted of a simple sequence of hexameric repeats of the nucleotides TTGGGG19. The telomeres in human cells also consist of thousands of repeats, but in mammals the sequence is TTAGGG20. Once this sequence was known, the length of human telomeres could be measured. The first hints that human telomeres might shorten appeared in 1986, when it was shown that telomere lengths are not the same in all tissues21. These studies culminated in the demonstration that telomeres shorten as normal human fibroblasts divide in culture22. These initial observations and others23–25 supported the concept that telomere attrition limits normal cell proliferation in culture. If short telomeres limit the rate of cell growth, there had to be a solution to the telomere problem in immortal organisms, in the germline cells of higher organisms and in 60 50

Phase III (senescence)

40 Phase II

30 20 10

Phase I

0 10

130 170 210 Days in culture

1999

250

290

Figure 3 | Hayflick’s three phases of cell culture. Phase I is the primary culture; phase II represents subcultivated cells during the period of exponential replication. Phase III represents the period when cell replication ceases but metabolism continues. Cells may remain in this state for at least one year before death occurs.

The future — pharmaceutical companies are developing lead compounds for inhibiting telomerase in cancer cells. Methods for modulating telomere length in normal cells may have medical applications for treating age-related disease43.

2000

Advanced Cell Technology scientists report cows derived by nuclear transfer from populations of senescent donor somatic cells42, and that the telomeres are restored to at least normal lengths.

cancer cells. The solution originated again in studies with Tetrahymena by Carol Greider, a graduate student in Elizabeth Blackburn’s laboratory26. Greider and Blackburn discovered the enzyme — telomerase — that synthesizes and elongates telomeres. Telomerase was later found in extracts of immortal human cell lines27 and in most human tumours28. Telomerase contains an RNA template on which the new telomeres are made. This RNA component was cloned a few years later29 and subsequently the catalytic portion of the enzyme was cloned30. However, the idea that telomere shortening causes cell senescence has only recently been demonstrated31. Introduction of the telomerase catalytic protein component into normal human cells resulted in telomerase activity31. Normal human cells stably expressing transfected telomerase can maintain the length of their telomeres, and exceed their maximum lifespan by more than fivefold. So the normal longevity-determination mechanism of telomere shortening in human cells can be circumvented — evidence for the role of telomere shortening in cell senescence and that of telomerase expression in cell immortality. This discovery has profound theoretical and practical implications that include the immortalization of normal human cells for the production of commercially important proteins32. As there are sensitive methods for detecting telomerase in a single cell, the telomerase assay is a potential diagnostic tool for the detection of cancer cells in clinical specimens33. Telomerase inhibitors might be found that could, perhaps, be used for treating cancer34. From cells to the ageing organism

These observations on telomere biology in

cells propagated in culture have yet to be shown to be directly relevant to ageing of organisms. However, it is an attractive hypothesis that the replicative potential of human cells with an intrinsic capacity for replacement may be set to allow for normal growth, development, repair and maintenance, but not to allow the many divisions needed for cancer. Many cells (even in tissues noted for division) are not completely senescent — even in centenarians. But this does not contradict the role of senescent cells in ageing. Although cells can grow out of tissues obtained from elderly donors, this does not mean there are no senescent cells in that specimen. In fact, only a minority of cells in any tissue are likely to be senescent. However, the presence of some senescent cells may interfere with the function of otherwise normal somatic tissues35,36. Hayflick (BOX 1) proposes that telomere shortening may be the molecular equivalent of longevity determination37. Hundreds of physiological, molecular and behavioural changes in normal cultured human cells herald the approach of the Hayflick limit. These changes represent increasing molecular disorder, and all compromise the internal milieu, leading to loss of cell function. Hayflick suggests that the number of population doublings that a normal cell can undergo may be the in vitro expression of maximum potential longevity. This is never reached in vivo owing to the hundreds of molecular disorders that, in vitro, mark the approaching loss of replicative capacity, and, in vivo, increase vulnerability to disease and death11. Future challenges

Hayflick’s initial observations on cellular replicative senescence have focused attention 3' 5'

5' 3'

Leading strand

5' 3'

5' Lagging strand (Okazaki fragments)

No DNA for another priming event

Figure 4 | The end-replication problem. During DNA replication the leading strand is synthesized as a continuous molecule that can potentially replicate all the way to the end of a linear template. The lagging strand is made as a discontinuous set of short Okazaki fragments, each requiring a new primer to be laid down on the template, that are then ligated to make a continuous strand. The lagging strand cannot replicate all the way to the end of a linear chromosome, as there is no DNA beyond the end for a priming event to fill in the gap between the last Okazaki fragment and the terminus. This leaves a 3′ overhang. The leading strand is also probably processed to leave a 3′ overhang.

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 1 | OCTOBER 2000 | 7 5

PERSPECTIVES on telomere biology and its role in human ageing and cancer. Cancer cells need to maintain telomeres if they are to divide indefinitely, and reactivation of telomerase usually solves this problem. Hayflick’s idea that replicative senescence might be a barrier to tumorigenesis challenges us to determine whether this is true in all multicellular organisms and, if so, to what extent. For example, it is well established that short-lived organisms such as the inbred mouse have much longer telomeres and a higher incidence of cancer compared with humans. Are shortened telomeres serving an anti-cancer role in humans but not in mice38? As almost all cells and tissues, with the exception of post-mitotic cells such as neurons and cardiomyocytes, show progressive shortening of telomeres with increased age, organ failure may sometimes occur in chronic diseases of high cellular turnover. Although the ageing process is complex and cannot be explained solely on the basis of telomere biology, there is a growing consensus that we need to understand telomeres and telomerase with regard to ageing of organisms and cancer. The challenge is to determine whether or not telomere biology leads to an increase in vulnerability to ageing and to learn how to intervene in these processes. Hayflick believes that the most important question is rarely addressed:“Why are old cells more vulnerable to pathology than are young cells?”. Jerry W. Shay and Woodring E. Wright are at the University of Texas Southwestern Medical Center at Dallas, Department of Cell Biology, 5323 Harry Hines Boulevard, Dallas, Texas 75390-9039, USA. e-mail: shay@utsw.swmed.edu Correspondence to: J.W.S.

12. Wright, W. E. & Hayflick, L. Nuclear control of cellular ageing demonstrated by hybridization of anucleate and whole cultured normal human fibroblasts. Exp. Cell Res. 96, 113–121 (1975). 13. Watson, J. D. Origin of concatemeric T7 DNA. Nature New Biol. 239, 197–201 (1972). 14. Olovnikov, A. M. Telomeres, telomerase and aging: Origin of the theory. Exp. Gerontol. 31, 443–448 (1996). 15. Olovnikov, A. M. Principles of marginotomy in template synthesis of polynucleotides. Dokl. Akad. Nauk S.S.S.R. 201, 1496–1499 (1971). 16. Olovnikov, A. M. A theory of marginotomy: The incomplete copying of template margin in enzyme synthesis of polynucleotides and biological significance of the problem. J. Theor. Biol. 41, 181–190 (1973). 17. Muller, H. J. in Studies of Genetics: The Selected Papers of H. J. Muller 384–408 (Indiana Univ. Press, Bloomington, 1962). 18. McClintock, B. The stability of broken ends of chromosomes in Zea mays. Genetics 26, 234–282 (1941). 19. Blackburn, E. H. & Gall, J. G. A tandemly repeated sequence at the termini of the extrachromosomal ribosomal RNA genes in Tetrahymena. J. Mol. Biol. 120, 33–53 (1978). 20. Moyzis, R. K. et al. A highly conserved repetitive DNA sequence (TTAGGG)n, present at the telomeres of human chromosomes. Proc. Natl Acad. Sci. USA 85, 6622–6626 (1988). 21. Cooke, H. J. & Smith, B. A. Variability at the telomeres of human X/Y pseudoautosomal regions. Cold Spring Harb. Symp. Quant. Biol. 51, 213–219 (1986). 22. Harley, C. B., Futcher, A. B. & Greider, C. W. Telomeres shorten during ageing of human fibroblasts. Nature 345, 458–460 (1990). 23. Hastie, N. D. et al. Telomere reduction in human colorectal carcinoma and with ageing. Nature 346, 866–868 (1990). 24. DeLange, T. et al. Structure and variability of human chromosome ends. Mol. Cell Biol. 10, 518–527 (1990). 25. Lindsey, J., McGill, N. I., Lindsey, L. A., Green, D. K. & Cooke, H. J. In vivo loss of telomere repeats with age in humans. Mutat. Res. 256, 45–48 (1991). 26. Greider, C. W. & Blackburn, E. H. Identification of a specific telomere terminal transferase enzyme with two kinds of primer specificity. Cell 51, 405–413 (1985). 27. Morin, G. B. The human telomere terminal transferase enzyme is a ribonucleoprotein that synthesizes TTAGGG repeats. Cell 59, 521–529 (1989). 28. Kim, N.-W. et al. Specific association of human telomerase activity with immortal cells and cancer. Science 266, 2011–2015 (1994). 29. Feng, F. et al. The RNA component of human telomerase. Science 269, 1236–1241 (1995). 30. Nakamura, T. M. et al. Telomerase catalytic subunit

31.

32.

33. 34.

35.

36.

37.

38.

39. 40. 41. 42.

43.

44.

45.

46.

47.

48.

homologs from fission yeast and humans. Science 277, 955–959 (1997). Bodnar, A. G. et al. Extension of life span by introduction of telomerase into normal human cells. Science 279, 349–352 (1998). Shay, J. W. & Wright, W. E. The use of telomerized cells for tissue engineering. Nature Biotechnol. 18, 22–23 (2000). Shay, J. W. & Gazdar, A. F. Telomerase in the early detection of cancer. J. Clin. Path. 50, 106–109 (1997). Herbert, B.-S. et al. Inhibition of telomerase leads to eroded telomeres, reduced proliferation, and apoptosis. Proc. Natl Acad. Sci. USA 96, 14276–14281 (1999). Dimri, G. P. et al. A biomarker that identifies senescent human cells in culture and in ageing skin in vivo. Proc. Natl Acad. Sci. USA 92, 9363–9367 (1995). Shay, J. W. & Wright, W. E. in Ageing Vulnerability: Causes and Interventions (Novartis Foundation Proceedings, in the press). Hayflick, L. A brief overview of the discovery of cell mortality and immortality and of its influence on concepts about ageing and cancer. Pathol. Biol. 47, 1094–1104 (1999). Wright, W. E. & Shay, J. W. Telomere dynamics in cancer progression and prevention: Fundamental differences in human and mouse telomere biology. Nature Med. 6, 849–851 (2000). Hayflick, L. The biology of human aging. Am. J. Med. Sci. 265, 433–445 (1973). Namba, M. et al. in Monograph on Cancer Research Vol. 27, 221–230 (Tokyo Univ. Press, Tokyo, 1981). Shiels, P. G. et al. Analysis of telomere lengths in cloned sheep. Nature 399, 316–317 (1999). Lanza, R. P. et al. Extension of cell life-span and telomere length in animals cloned from senescent somatic cells. Science 88, 665–669 (2000). Shay, J. W. Telomerase in cancer — diagnostic, prognostic and therapeutic implications. Sci. Am. 4, S26–S34 (1998). Hayflick, L., Plotkin, S. A., Norton, T. W. & Koprowski, H. Preparation of poliovirus vaccines in a human fetal diploid cell strain. Am. J. Hyg. 75, 240–258 (1962). Hayflick, L., Moorhead, P., Pomerat, C. M. & Hsu, T. C. Choice of a cell system for vaccine production. Science 140, 760–763 (1963). Fletcher, M. A., Hessel, L. & Plotkin, S. A. in Developments in Biological Standardization Vol. 93, 97–107 (Basel, Kargert, 1998). Chanock, R. M., Hayflick, L. & Barille, M. F. Growth on artificial medium of an agent associated with atypical pneumonia and its identification as a PPLO. Proc. Natl Acad. Sci. USA 48, 41–49 (1962). Hayflick, L. Tissue cultures and mycoplasmas. Texas Rep. Biol. Med. 23, 285–303 (1965).

Links ENCYCLOPEDIA OF LIFE SCIENCES Ageing | Cell

OPINION

senesence in vitro FURTHER INFORMATION Jerry Shay’s homepage

Weismann, A. Collected Essays upon Heredity and Kindred Biological Problems (ed. Poulton, E. B.) (Clarendon, Oxford, 1889). 2. Carrel, A. & Ebeling, A. H. Age and multiplication of fibroblasts. J. Exp. Med. 34, 599–606 (1921). 3. Hayflick, L. & Moorhead, P. S. The serial cultivation of human diploid cell strains. Exp. Cell Res. 25, 585–621 (1961). 4. Hayflick, L. The limited in vitro lifetime of human diploid cell strains. Exp. Cell Res. 37, 614–636 (1965). 5. Witkowski, J. A. Dr. Carrel’s immortal cells. Med. Hist. 24, 129–142 (1980). 6. Witkowski, J. A. The myth of cell immortality. Trends Biochem. Sci. 10, 258–260 (1985). 7. Rubin, H. Telomerase and cellular lifespan: ending the debate? Nature Biotechnol. 16, 396–397 (1998). 8. Burnett, M. Intrinsic Mutagenesis (Medical and Technical Publishing Co., Lancaster, 1974). 9. Garfield, E. Current Comments. Curr. Contents 15, 5–8 (1983). 10. Hayflick, L. The coming of age of WI-38. Adv. Cell Cult. 3, 303–316 (1984). 11. Hayflick, L. How and why we age. Exp. Gerontol. 33, 639–653 (1998).

Cancer: looking outside the genome

Judah Folkman, Philip Hahnfeldt and Lynn Hlatky The ‘gene-centric’ approach has produced a wealth of information about the origins and progression of cancer, and investigators seek a full compilation of altered gene expressions for tumour characterization and treatment. However, the cancer genome appears to be far more unstable than previously thought. It may therefore be prudent to augment gene-level approaches with supra-genomic strategies that circumvent the genomic variability of cancer cells.

The idea that cancer may one day be fully unravelled at the molecular level largely dictates how investigations into the biology and treatment of cancer are conducted. Indeed, as Zhang et al.1 have stated, “much of cancer research over the past 50 years has been devoted to the analyses of genes that are expressed differently in tumour cells as compared with their normal counterparts”. A prevailing paradigm asserts that cancer arises from specific gene mutations, and that it may eventually be treatable by reversing these

| OCTOBER 2000 | VOLUME 1

www.nature.com/reviews/molcellbio

31.

32.

33. 34.

35.

36.

37.

38.

39. 40. 41. 42.

43.

44.

45.

46.

47.

48.

Links ENCYCLOPEDIA OF LIFE SCIENCES Ageing | Cell

OPINION

senesence in vitro FURTHER INFORMATION Jerry Shay’s homepage

Cancer: looking outside the genome

| OCTOBER 2000 | VOLUME 1

www.nature.com/reviews/molcellbio

PERSPECTIVES mutational alterations or by targeting them to eliminate the tumour cell. However, evidence mounts that epigenetic, cell–cell and extracellular influences are also pivotal in tumour progression2–4. In addition, recent reports suggest that alterations in cancer cell genomes may be too unstable to serve as ‘stand-alone’ therapeutic targets and too numerous to reverse (although the potential for diagnostic and prognostic applications is substantial). In a study of colorectal carcinoma, for example, Stoler et al.5 found 11,000 genomic alterations per cell. Although only a very small fraction of these alterations affect coding regions, it is still questionable whether a complete picture of cancer be painted from reductionist gene-centric approaches alone6–8. We propose that considerable insight and therapeutic benefit may be gained by additionally using a supra-genomic, constraintdriven approach to cancer that circumvents the problem of genomic instability. Here, the focus is not on the gene alterations within tumour cells, but on physiological constraints imposed on the overall tumour system. By viewing the tumour system as a whole, the extensive heterogeneity of individual gene expressions is transcended, and unifying principles not deducible from investigations at the genetic level emerge. An example of the insight that can be gained by a change in the level of exploration is the emergence of the organizing principle of surface tension, a property of water that cannot be deduced from a study of the individual properties of hydrogen and oxygen molecules. Rather, it was necessary to take a step back to reveal this bulk constraining principle. Similarly, identifying where cancer is constrained is fundamental, because it is at

100–200 µm

Viability threshold

Microvessel

Figure 1 | Diffusion constraints govern tumour growth. Depicted is the constraint that tumour cells must be within 100–200 µm, the consumption-adjusted oxygen diffusion distance, from a vessel to remain viable. Tumour cells are shown in various stages of deprivation as a function of distance from the vasculature. The green arrow represents oxygen and nutrients from the vessel lumen. The yellow arrow represents the estimated 20 or so known endothelial-derived mitogens and survival factors that influence tumour growth.

Box 1 | Targeting the tumour: anti-angiogenic therapy. The efficacy of anti-angiogenic therapy has been experimentally shown. For example, for just one of many angiogenesis inhibitors, there are more than 60 reports from different laboratories of anti-angiogenic inhibition of 33 different types of human, mouse, rat, hamster and rabbit tumours transplanted in animals, and of 23 different types of metastatic tumours, including human tumour metastases in mice, and rat and hamster metastases25. Tumour growth was inhibited at an average of 65% with a range of 43% to 100% (that is, complete regression, which was observed in seven tumour types in mice). Using anti-angiogenic therapy, spontaneous tumours in rats have been regressed26, and regression of transplanted murine and human tumours in mice has been accomplished without the development of acquired drug resistance27. Kerbel28 had earlier proposed that a lack of acquired drug resistance would be a major benefit of the anti-angiogenic approach. Furthermore, tumour growth in mice can be completely inhibited by transfection into the tumour cells of an anti-angiogenic protein (thrombospondin) that does not reduce the high proliferation rate of the tumour cells29.

the constraints that variability is reduced. With the consequent reduction in background noise, stable and predictable information can then be isolated. Predictability, in turn, offers a firm basis for therapeutic exploitation. A fundamental constraint on the tumour system is the need to distribute oxygen and nutrients to, and remove carbon dioxide and catabolites from, every cell. As we discuss here, investigation of how this constraint is accommodated brings with it new organizing perspectives on the cancer problem that may transcend genomic instability and restore predictive ability. Hundreds of diseases and counting

Cancer, already considered to be a large group of diseases classified by tissue of origin, is being further subdivided with the advent of new technology that can reveal a tumour’s genetic profile (such as DNA microarrays or chromosome dissection). The excitement of these technical achievements notwithstanding, they reinforce the prevailing reductionist perspective of cancer as hundreds of identifiable gene-driven ‘diseases’. Each of these diseases is thought to be potentially treatable through a full characterization and targeting of its individual genetic and signalling aberrations. To facilitate such massive characterizations, it is anticipated that there will be computational resources to “permit the oncologist to check the wiring of entire signal transduction or cell-cycle pathways”9. Furthermore, it is proposed that “performing relational analyses of all cancer-associated gene expression changes”10 will be feasible. But even computation experts agree this is a monumental undertaking. Although a description of the interaction of even a handful of gene expressions would already be a formidable challenge, recent reports, such as of thousands of genomic alterations in can-

cer cells5, raise concern about the feasibility of a quantitative endeavour of this sort. Based on the number of potentially interacting genes in the genome, bioinformatics pioneer Stuart Kauffman emphasizes that it is not going to be easy, and estimates that it will take 30–40 years before we understand “major chunks” of the problem11. But the issue goes beyond gene numbers. The time-dependent nature of the cancer genome complicates matters further. Aberrations continually accrue and alter the character of both the primary tumour and its metastases. Moreover, research pointing to the disruption of cell–cell4,12 (for example, parenchymal–stromal) interactions and other epigenetic events2 as being instrumental in cancer progression further obscures the logic of pursuing genetic analyses alone to fully characterize cancer. What does seem clear is that cancer is more than a disease of specific genes. Rather, instability of the genome as a whole is a hallmark of the cancer cell5,13,14. Genomic instability seems to provide the extensive genetic variability that is subsequently acted on by selection pressures to drive cancer progression5,14. Over time, this process generates extensive genomic heterogeneity among the cells of a tumour, as different lineages evolve in a spectrum of directions under the dictates of competitive selection. The occurrence of a specific sequence of gene mutations, as described in the multistep progression pathway of colon cancer elegantly delineated by Vogelstein and colleagues15, is now interpreted by these researchers and others as a consequence of natural selection on the genetic variants of unstable genomes5,14. Given the unstable nature of the genome, a reductionist’s assertion that cancer is hundreds of diseases may be an understatement. As with any degenerative process, a potentially unbounded number of aberrations is expected, limited only by the time they have

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 1 | OCTOBER 2000 | 7 7

PERSPECTIVES had to occur. This raises the concern that potential therapeutic attempts to target such expressions with specialized molecular attacks may be over-ascribing durable and exploitable mechanistic bases to what are, in fact, temporal or hypervariable events16. Even if these aberrations could be successfully targeted, they may not correct the fundamental degenerative nature of the disease. These emerging shortcomings of a strictly gene-centric approach to cancer have led some scientists to suggest a need for cancer paradigms that are anchored by fundamental underlying principles, which restore a unifying perception of the disease6,7. As it turns out, a study of endothelial-dependent tumour angiogenesis — the tumour’s ability to recruit its own private blood supply — brings with it a set of governing principles that arguably comprise one such paradigm. We here outline an ‘endothelio-centric’ paradigm and emphasize that molecular and non-molecular techniques alike remain invaluable exploratory tools within this or other constraint-based paradigms. The endothelio-centric paradigm allows the details of tumour-cell genetic presentation to be transcended. It focuses instead on general aspects of tumour behaviour, which are not subject to the vagaries of the high mutation rate of cancer cells. The question now is not how breast and colon carcinomas differ genetically, but in what fundamental ways these disparate tumours are similar. In this sense, the paradigm recasts cancer as a single disease. Cancer recast as a single disease

Like all cells of the body, tumour cells cannot remain viable beyond the effective 100–200-µm oxygen diffusion distance from the vasculature (FIG. 1). For a tumour to grow beyond a small size, angiogenesis is therefore required. The dependence of all tumours on angiogenesis is now well established. The tumour microvasculature, or more precisely, its endothelial cell lining, is responsible for oxygen and metabolite distribution. The optimized design and function of vascular endothelium is shown by the preservation of its basic form over evolutionary time. The diameter of capillary blood vessels, for example, is uniform across mammalian species. Because the endothelium functions as the control point for nutrient and metabolite exchange, it figures centrally in this constraint-based paradigm. Through a focus on endothelial control, a perception of cancer emerges that departs from the gene-centric view. What previously constituted a myriad of diseases is now seen

Tumour cell genome

Tumour-associated endothelial cell genome

Figure 2 | Endothelial and tumour cells as targets. The tumour cell and tumour-associated endothelial cell genomes are depicted as blue and red disks, respectively. Perturbations in the plane of the disk constitute normal adaptive changes in gene expression, and spiked projections out from the disk constitute genomic alterations. The tumour cell genome depicted is human colorectal carcinoma as reported by Stoler et al.5. Each spike includes 11 alterations, and there are 1,000 spikes, comprising 11,000 total genomic alterations per cell. The distribution of their heights corresponds to the numbers of alterations observed per PCR fragment. Data of St Croix et al.30 show that there are 79 significant (> 20fold) differences in gene expression between a tumour-associated endothelial cell and its ‘resting’ counterpart in normal tissue. These are indicated by perturbations in the plane of the disk (1 perturbation = 6 differences). Moreover, selected transcripts, which were highly expressed in tumour-associated endothelium, were also expressed in angiogenic endothelial cells associated with wound healing, yet expression of these transcripts was essentially absent in resting endothelium30, suggesting that these expression changes are probably normal, adaptive and reversible.

as a single (albeit multi-symptomatic) disorder of aberrant tissue mass, which continually stimulates the formation of new blood vessels to support its expansion. Even leukaemia, thought to be a ‘liquid tumour’, has been shown to induce angiogenesis in the bone marrow to support its growth17. This recasting of cancer as a single disease correspondingly suggests a general means of therapeutic address. Cancer may be controllable through the growth-limiting influence of its own endothelium. In contrast to the extremely heterogeneous and genomically unstable tumour cell population, the endothelium in the tumour bed presents a comparatively homogeneous and stable target, showing only normal adaptive plasticity and no propensity towards hypermutability (FIG. 2). Because all tumours rely on vasculature, endothelium should, in principle, be a reliable therapeutic target irrespective of either tumour type or genomic presentation. Endothelial control of tumours

Tumour angiogenesis is driven by positive and negative regulators of endothelial growth18. This process is analogous to the normal positive–negative control structure that fine-tunes vessel development during embryogenesis and wound healing. But unlike in embryogenesis and wound healing, most tumours persistently overexpress stimu-

lators, leading to continued recruitment of endothelium and continued expansion of the tumour mass. Ultimately, each newly recruited endothelial cell can support a large population of tumour cells. In cancer patients, this leverage may be exploited to therapeutic advantage. By administering angiogenesis inhibitors to shift the stimulatory climate in the tumour back to inhibition, the recruited endothelial cells may be removed, followed by the subsequent loss of the high relative numbers of supported tumour cells16,18,19. In this manner, the low-diversity population of proliferating endothelial cells, on which the tumour cells depend, serves as a targetable ‘weak link’ to which even a genetically unstable tumour-cell population has little or no evasive response. Results from laboratory work show ‘proof of principle’ of angiogenic control of tumour mass (BOX 1), and clinical investigations are underway. At the time of writing, at least 20 angiogenesis inhibitors are being tested in cancer patients in clinical trials in the United States and Europe. However, tumour endothelial targeting and tumour cell targeting should not be thought of as mutually exclusive. Antiangiogenic therapy can be added to chemotherapy, radiotherapy, immunotherapy, gene therapy or any other traditionally cancercell-directed modality. Clinical trials are in progress to explore the advantage of anti-

| OCTOBER 2000 | VOLUME 1

www.nature.com/reviews/molcellbio

PERSPECTIVES

“The question now is not how breast and colon carcinomas differ genetically, but in what fundamental ways these disparate tumours are similar.” angiogenic therapy combined with radiation or chemotherapy. More recently, it has been shown that the dose and schedule of chemotherapy can be maximized for antiangiogenic activity20,21. ‘Anti-angiogenic chemotherapy’ in tumour-bearing animals increases drug efficacy, reduces side-effects and bypasses even established drug resistance. Endothelial control of tissue mass

The control of tumour mass by vascular endothelial cells may represent a subset of a larger role for the endothelium in controlling the growth and final size of all tissue mass, normal or neoplastic22. Both tumours and organs may limit their own growth by increasing the production of angiogenic inhibitors with size, until a critical size is reached where further growth is actively selfinhibited. A dynamical theory of coordinated tumour/vascular growth16 describes how inhibition of tumour-derived angiogenesis can explain the observed deceleration in tumour growth. It also suggests that regulation of tumour growth may be a distorted, but essentially accurate, picture of how organ size might be governed in organogenesis. It seems that the principle of endothelial control of tissue mass may not only transcend the variability found within cancer, but the differences that exist among normal tissues as well.

open to simplifying alternatives, even to the extent that such alternatives may sidestep putatively important detail and so alter the very concept of solution. In this spirit we have asked whether, by additionally focusing attention on encompassing and stable aspects of tumour development, one might avoid the complexities and instabilities inherent in the probing of cancer solely through analysis of its genetic aberrations. Approaching cancer at the level where metabolic constraints are operative has, in fact, led to the simplifying idea that tumours depend on angiogenesis. This dependence reveals a supra-genomic cancer paradigm that provides insight into basic tumour biology. It also offers a therapeutic avenue for treatment that transcends difficulties presented by genetic instability or tumour-cell heterogeneity. Although the impetus for this Opinion article centres on the increasing awareness of the heterogeneity and instability of the cancer genome, it is possible that suppressing this degenerative process may itself comprise an alternative constraint-based paradigm. By reimposing a differentiated state, retinoic acid has had some success in this regard in leukaemia24. Furthermore, it has been proposed that a search for mutator genes14 may be prudent, with the prospect that their regulation may prevent destabilization of the genome. However, this approach would probably be more useful for disease prevention than after the instability has been unleashed. In any event, it seems clear that, by building paradigms upon constraints, one can create simplified — yet powerfully predictive — renderings of complex biological problems. Judah Folkman is in the Laboratory of Surgical Research, Department of Surgery, Children’s Hospital, and Departments of Cell Biology and Surgery, Harvard Medical School, Boston, Massachusetts 02115, USA. e-mail: foss@hub.tch.harvard.edu. Philip Hahnfeldt and Lynn Hlatky are in the Department of Adult Oncology, Dana-Farber Cancer Institute, and Department of Radiation Oncology, Harvard Medical School, Boston, Massachusetts 02115, USA. e-mails: Philip_Hahnfeldt@dfci.harvard.edu; lynn_hlatky@dfci.harvard.edu

So how do we get there from here?

Many scientists — including ourselves — believe that eventually there will be molecular solutions to the cancer problem. However, as Feinstein has pointed out23, in medical practice there is often a dissociation between understanding the cause of a disease and having an effective therapy for it. For example, “the exact cause of sickle-cell anaemia has been known, at the molecular level, for more than 40 years”, but this knowledge has been only partly exploited in current therapy. In contrast, in appendicitis, neither the epidemiology nor the bacteriology are well understood, yet the disease is commonly and easily curable. A lesson of this experience may be that, where treatment strategy is stymied by the complexity of the disease, one should be

6. 7. 8. 9. 10. 11. 12.

13. 14.

15. 16.

17.

18.

19.

20.

21.

22.

23.

24.

25.

26.

27.

Links ENCYCLOPEDIA OF LIFE SCIENCES Cancer.

28.

29. 1. 2.

Zhang, L. et al. Gene expression profiles in normal and cancer cells. Science 276, 1268–1272 (1997). Baylin, S. B. & Herman, J. G. DNA hypermethylation in tumorigenesis: epigenetics joins genetics. Trends Genet. 16, 168–174 (2000). Moinfar, F. et al. Concurrent and independent genetic alterations in the stromal and epithelial cells of mammary carcinoma: Implications for tumorigenesis. Cancer Res. 60, 2562–2566 (2000).

30.

Bissell, M. J. et al. Tissue structure, nuclear organization, and gene expression in normal and malignant breast. Cancer Res. 59, S1757–S1763 (1999). Stoler, D. L. et al. The onset and extent of genomic instability in sporadic colorectal tumor progression. Proc. Natl Acad. Sci. USA 96, 15121–15126 (1999). Strohman, R. C. The coming Kuhnian revolution in biology. Nature Biotechol. 15, 194–200 (1997). Hanahan, D. & Weinberg, R. A. The hallmarks of cancer. Cell 100, 57–70 (2000). Nurse, P. Reductionism: The ends of understanding. Nature 387, 657 (1997). Steeg, P. S. & Abrams, J. S. Cancer prognostics: past, present and p27. Nature Med. 3, 152–154 (1997). Szallasi, Z. Bioinformatics: Gene expression patterns and cancer. Nature Biotechol. 16, 1292–1293 (1998). Howard, K. The bioinformatics gold. Sci. Am. July, 58–63 (2000). Huang, S. & Ingber, D.E. The structural and mechanical complexity of cell-growth control. Nature Cell Biol. 1, E131–E138 (1999). Loeb, L. A. Many mutations in cancers. Cancer Surveys 28, 329–342 (1996). Cahill, D. P., Kinzler, K. W., Vogelstein, B. & Lengauer, C. Genetic instability and darwinian selection in tumours. Trends Cell Biol. 9, M57–M60 (1999). Kinzler, K. W. & Vogelstein, B. Lessons from hereditary colorectal cancer. Cell 87, 159–170 (1996). Hahnfeldt, P., Panigrahy, D., Folkman, J. & Hlatky, L. Tumor development under angiogenic signaling: A dynamical theory of tumor growth, treatment response and post-vascular dormancy. Cancer Res. 59, 4770–4775 (1999). Perez-Atayde, A. R. et al. Spectrum of tumor angiogenesis in the bone marrow of children with acute lymphoblastic leukemia. Am. J. Pathol. 150, 815–821 (1997). Hanahan, D. & Folkman, J. Patterns and emerging mechanisms of the angiogenic switch during tumorigenesis. Cell 86, 353–364 (1996). Good, D. J. et al. A tumor suppressor-dependent inhibitor of angiogenesis is immunologically and functionally indistinguishable from a fragment of thrombospondin. Proc. Natl Acad. Sci. USA 87, 6624–6628 (1990). Browder, T. et al. Antiangiogenic scheduling of chemotherapy improves efficacy against experimental drug-resistant cancer. Cancer Res. 60, 1878–1886 (2000). Klement, G. et al. Continuous low-dose therapy with vinblastine and VEGF receptor-2 antibody induces sustained tumor regression without overt toxicity. J. Clin. Invest. 105, R15–R24 (2000). Folkman, J. Is tissue mass regulated by vascular endothelial cells? Prostate as the first evidence. Endocrinology 139, 441–442 (1998). Feinstein, A. R. The santayana syndrome II: Problems in reasoning and learning about error. Perspect. Biol. Med. 41, 73-85 (1997). Sachs, L. The control of growth and differentiation in normal and leukemic blood cells. The 1989 Alfred P. Sloan Prize of the General Motors Cancer Research Foundation. Cancer 65, 2196–2206 (1990). Folkman, J. in Accomplishments in Cancer Research (eds Wells, S. A. Jr & Sharp, P. A.) 32–44 (J. B. Lippincott Williams and Wilkins, Pennsylvania 1998). Perletti, G. et al. Antitumor activity of endostatin against carcinogen-induced rat primary mammary tumors. Cancer Res. 60, 1793–1796 (2000). Boehm, T., Folkman, J., Browder, T. & O’Reilly, M. S. Antiangiogenic therapy of experimental cancer does not induce acquired drug resistance. Nature 390, 404–407 (1997). Kerbel, R. S. Inhibition of tumor angiogenesis as a strategy to circumvent acquired resistance to anticancer therapeutic agents. BioEssays 13, 31–36 (1991). Streit, M. et al. Thrombospondin-2: a potent endogenous inhibitor of tumor growth and angiogenesis. Proc. Natl Acad. Sci. USA 96, 14888–14893 (1999). St Croix, B. et al. Genes expressed in human tumor endothelium. Science 289, 1197–1202 (2000).

Acknowledgements We thank G. Anderson, R. Kerbel and B. Vogelstein for reading the manuscript and for their comments. We thank C. Lamont for figure graphics.

NATURE REVIEWS | MOLECUL AR CELL BIOLOGY

VOLUME 1 | OCTOBER 2000 | 7 9