coupling strength that grows with systemsize. This synchronization can be achievedwithout forcing the dynamics to become, for example, periodic. Hence, the problem of spatial control of coupled dynamics,although it still involves stabilizing dynamicsthat are inherently unstable, is easier toachieve than control of chaotic into simpledynamics. Control of synchronization canusually be achieved by careful design of thecoupling, rather than resorting to feedbacktechniques. What then remains is to try tominimize the level of coupling required toachieve synchronization.

This is the problem that Wei, Zhan andLai1 have tackled. They have come up with anovel way of reducing the necessary couplingin an array by using wavelet decomposition ofthe matrix of coupling coefficients. Waveletsare mathematical functions that have beendeveloped over the past decade or so as a powerful tool for signal-processing andnumerical analysis. Wavelet analysis involvesreducing a signal into a series of coefficientsthat can be manipulated, analysed or used toreconstruct the signal. Wei et al. make a smallchange to the low-frequency components in the wavelet-transformed matrix, beforeapplying an inverse transform to obtain amodified coupling matrix. This turns out tobe an efficient strategy for achieving synchro-nization at much lower coupling strengths.

Wei et al. test their method by synchro-nizing a ring of coupled Lorenz systems. TheLorenz system is a set of three nonlinear dif-ferential equations showing chaotic behav-iour. In this proof-of-principle, a ring ofLorenz systems are coupled together linearly,their relations to each other represented by amatrix of coupling coefficients. A smallchange in this matrix (less than 2% for 64coupled systems), through the wavelet trans-form, produces a much lower threshold ofcoupling to achieve synchronization. Theauthors show that their technique is robusteven if the symmetry of nearest-neighbourcoupling is broken.

It will be interesting to see if this methodcan be extended to more general arrays ofcoupled systems, to better understand con-trol of spatial patterns. It may be that thework by Wei et al.1 will suggest new tech-niques and structures for the design of localand global coupling in such systems. ■

Peter Ashwin is in the School of MathematicalSciences, University of Exeter, Exeter EX4 4QE, UK.e-mail: P.Ashwin@ex.ac.uk

1. Wei, G. W., Zhan, M. & Lai, C.-H. Phys. Rev. Lett. 89, 284103

(2002).

2. Ott, E., Greboi, C. & Yorke, J. A. Phys. Rev. Lett. 64, 1196–1199

(1990).

3. Pyragas, K. Phys. Lett. A 170, 421–428 (1992).

4. Pikovsky, A., Rosenblum, M. & Kurths, J. Synchronization:

A Universal Concept in Nonlinear Sciences (Cambridge Univ.

Press, 2001).

Scientific discoveries often originate insurprising places. Some years ago, forinstance, researchers looking at how

the brain develops received help from anunexpected quarter: studies of patients withAlzheimer’s disease. This disease is character-ized in part by the abnormal accumulation, in the brain, of a protein called amyloid b-peptide (Ab), which is a fragment of a largerprotein, the amyloid precursor protein(APP), that sits across the outer membrane of nerve cells. Two enzymatic activities areinvolved in precisely snipping APP to pro-duce Ab, which is then shed into the brain.Curiously, one of these activities — dubbedg-secretase1 — was later discovered also tocleave Notch, a receptor protein that lies onthe cell surface, and thereby to affect the wayin which Notch regulates gene expressionduring normal development2. On page 438 ofthis issue, Takasugi and colleagues3 add to ourunderstanding of how APP and Notch areprocessed. Using genes and cells from flies

and humans, and the powerful new tech-nology of RNA interference, these authorsestablish specific roles for four different proteins underlying g-secretase activity.

For many years, much of the research intoAlzheimer’s disease has concentrated onidentifying and characterizing the protein(or proteins) that generate Ab. In the firststep of this process, APP is cleaved at a specificpoint by a so-called b-secretase activity; theprotein responsible for this activity was identified some four years ago. Cleavage bythe g-secretase activity then produces Aβ —but here the molecules at fault have beenharder to pin down. An early hint came fromthe finding that mutations in a gene encod-ing the presenilin-1 protein occur in severalfamilies with inherited Alzheimer’s disease;it was quickly shown that these mutationscause increased cleavage of APP to produceAb. So presenilin-1 was assumed to be the g-secretase.

A surprising link to brain development

news and views

NATURE | VOL 422 | 27 MARCH 2003 | www.nature.com/nature 385

Neurobiology

Ballads of a protein quartetMark P. Mattson

The fate of neurons in the developing brain and in Alzheimer’s disease maylie with a four-protein complex that regulates the cleavage of two moleculesspanning the cell membrane. The role of each protein is now being unveiled.

was then discovered when researchersknocked out the presenilin-1 gene in mice(reviewed in ref. 2). The animals died asembryos, and had severe defects in braindevelopment that were indistinguishablefrom the defects in mice lacking Notch. Thisis because presenilin-1 is required not only to cleave APP and generate Ab, but also tocleave Notch after Notch has detected andbound a partner protein. An intracellularfragment of Notch is then released, and regulates gene expression in the neuronalnucleus. It has been suggested4 that an intracellular fragment of APP, generated byg-secretase, likewise moves to the nucleusand regulates gene expression.

But it soon became clear that presenilin-1cannot work alone to cleave APP and Notch,and a search began for other proteins thatmight be involved. APP and Notch have beenhighly conserved during evolution, whichnot only attests to their physiological impor-tance, but also means that molecular-geneticanalyses of fruitflies and worms can be usedto investigate their cleavage. Such studieshave found that four proteins seem to contribute to g-secretase activity; these arepresenilin-1, nicastrin, APH-1 and PEN-2(Fig. 1, overleaf )5–7. It has just been shownthat g-secretase activity can be fully recon-stituted with only these four proteins8.

But what exactly do these proteins do? To begin to understand this, Takasugi andco-workers3 first generated fruitfly cells thatexpressed different combinations of fruitflynicastrin, APH-1 and PEN-2 and deter-mined the effects on cleavage of presenilin-1(this event having been previously associatedwith g-secretase activity). They found thatoverexpression of APH-1 — or APH-1 plusnicastrin — stabilized the four-protein complex and simultaneously reduced pre-senilin-1 cleavage, suggesting that APH-1inhibits the ability of g-secretase to cleaveany of its target proteins. They then showedthat, indeed, APH-1 reduces the g-secretasecleavage of APP as well.

To determine the role of PEN-2 in the g-secretase quartet, the authors used RNAinterference to target and degrade the messenger RNA encoding PEN-2, therebyreducing production of the protein, in fruit-fly cells, mouse and human brain neurons,and human tumour cells. This resulted indecreased g-secretase activity. Further experi-ments in which a fragment of APP was addedconfirmed that APH-1 inhibits, whereasPEN-2 promotes, the production of Ab.These findings advance our understandingof an enzyme activity that is important inboth brain development and Alzheimer’sdisease, and identify new protein targets for drugs to prevent or treat this disorder. But the results also raise new questions, and reveal further hurdles to treatingAlzheimer’s disease.

One general question is whether the

© 2003 Nature Publishing Group

functions of APH-1 and PEN-2 are the samein Notch cleavage as in APP cleavage. This isimportant not only for our understanding of Notch signalling, but also from a clinicalperspective: existing therapeutic g-secretaseinhibitors decrease Ab production, yet probably have serious side effects becausethey also inhibit Notch cleavage6 (which isimportant in adults as well as embryos). Inaddition, although various environmentalsignals — such as growth-factor, neuro-transmitter and cytokine molecules — affectthe expression and processing of APP andNotch9,10, we do not yet know if and how suchsignals affect g-secretase activity. Anotherquestion is whether the expression of thefour proteins varies during development,although this seems likely, given the obviousimportance of g-secretase during braindevelopment. And are any of the proteinsregulated at a level beyond the expression oftheir genes, for example by the covalentattachment of phosphate groups?

More specific questions concern how themembers of the g-secretase quartet interact,both physically and functionally, with oneanother and with their substrates. Physically,they seem to bind to one another; quite howis unknown, but they probably contain protein–protein interaction domains similarto those of other membrane proteins, such as subunits of ion channels. Functionally,

presenilin-1 is believed to be the enzyme that actually cleaves APP and Notch2. If so,then nicastrin, APH-1 and PEN-2 might regulate g-secretase activity by modifyingeither presenilin-1’s enzymatic activity or its association with substrates. Finally, the g-secretase complex presumably functionsin the plasma membrane, where Notch and

APP reside. But considerable evidence suggests that presenilin-1 occurs primarily in the endoplasmic reticulum, a network of internal membranes, so this presentsanother puzzle.

Returning to Alzheimer’s disease, oneburning question is how presenilin-1, nicastrin, APH-1 and PEN-2 contribute tothe aberrant APP processing and neuronaldegeneration seen in this disorder. Besidesincreasing Ab production, presenilin-1mutations lead to a large increase in the number of calcium ions in the endoplasmicreticulum; this may contribute to defects in neuronal communication and neuronaldeath11. It will be interesting to see if and how these two consequences of presenilin-1mutations are linked. It is not impossible that the alterations in g-secretase activity and APP processing caused by presenilinmutations and other genetic and environ-mental factors are secondary to a primary disturbance in calcium regulation or oxida-tive stress. It will be important to find out whether nicastrin, APH-1 or PEN-2modifies the effects of presenilin-1 mutationson neuronal calcium balance, and on the vulnerability of neurons. ■

Mark P. Mattson is in the Laboratory ofNeurosciences, National Institute on Aging,Gerontology Research Center, 5600 Nathan ShockDrive, Baltimore, Maryland 21224, USA.e-mail: mattsonm@grc.nia.nih.gov1. Haass, C. & De Strooper, B. Science 286, 916–919 (1999).2. Selkoe, D. J. Curr. Opin. Neurobiol. 10, 50–57 (2000).3. Takasugi, N. et al. Nature 422, 438–441 (2003).4. Leissring, M. A. et al. Proc. Natl Acad. Sci. USA 99, 4697–4702

(2002).5. Yu, G. et al. Nature 407, 48–54 (2000).6. Francis, R. et al. Dev. Cell 3, 85–97 (2002).7. Beher, D. & Shearman, M. S. Biochem. Soc. Trans. 30, 534–537

(2002).8. Edbauer, D. et al. Nature Cell Biol. (in the press).9. Baron, M. et al. Mol. Membr. Biol. 19, 27–38 (2002).10.Panegyres, P. K. Rev. Neurosci. 12, 1–39 (2001).11.Chan, S. L. et al. Neuromol. Med. 2, 167–196 (2002).

news and views

NATURE | VOL 422 | 27 MARCH 2003 | www.nature.com/nature 387

Ca2+

Ca2+

Nucleus

Endoplasmicrecticulum

Genes for neuronalplasticity andbrain development

Oxidativestress

Oxidativestress

Cell dysfunctionand death

Presenilin-1

Nicastrin

APH-1

PEN-2

?

APPAβ plaque

Aβ

Notch

BACE

APPcytoplasmicfragment

NICD

Insideneuron

γ-secretase

Figure 1 The g-secretase protein quartet, and its roles in brain development and Alzheimer’s disease.Presenilin-1, nicastrin, APH-1 and PEN-2 form a functional g-secretase complex, located in theplasma membrane and endoplasmic reticulum (ER) of neurons. The complex cleaves Notch (left) to generate a fragment (NICD) that moves to the nucleus and regulates the expression of genesinvolved in brain development and adult neuronal plasticity. The complex also helps in generatingthe amyloid b-peptide (Ab ; centre). This involves an initial cleavage of the amyloid precursor protein(APP) by an enzyme called BACE (or b-secretase). The g-secretase then liberates Ab , as well as anAPP cytoplasmic fragment, which may move to the nucleus and regulate gene expression. Mutationsin presenilin-1 that cause early-onset Alzheimer’s disease enhance g-secretase activity and Abproduction, and also perturb the ER calcium balance. Consequent neuronal degeneration may resultfrom membrane-associated oxidative stress, induced by aggregating forms of Ab (which create Abplaques), and by the perturbed calcium balance.

Quantum computing

Logic gateway Andrew Steane

Two groups have created logic gates using pairs of trapped ions. Ascomponents of a quantum computer, these gates have the potential toform part of a scaled-up, workable system.

Quantum computing often features inthese pages, owing to the lively pace ofresearch in this area. The main aim is

to build a large working quantum computer,but the path towards that goal promises toreveal some wonderful physics; subtle andsurprising effects are confidently expected,such as quantum error correction. Also, therich nature and behaviour of quantumentanglement — the special correlation thatcauses composite quantum systems to defydescription in terms of their constituent parts

— is not fully understood, even for systems of only a few particles, and experimentationis essential to understand larger systems. Thetechniques and language of quantum com-puting are very powerful. They will almostcertainly give birth to a new generation ofultra-precise experimental methods basedon the interference properties of entangledsystems, whether or not efficient quantumcomputers are eventually realized.

Two papers in this issue1,2 (on pages 408and 412) report notable progress in what is

© 2003 Nature Publishing Group