suppressing the weaker ones (Fig. 1c). Suchbehaviour can be regarded as a simplifiedform of sensory attention, whereby the net-work selects the stimulus location based onstimulus strength. From a neurophysiologi-cal perspective, the selection of a single targetand the suppression of distractors is impor-tant, for example, in programming arm oreye movements.

The network’s ability to switch betweenlinear and nonlinear behaviour, based on itsinput, is very different from that of standardelectronics. Engineers usually require sepa-rate analogue and digital circuits to carry out linear amplification and nonlinearselection, respectively. Hahnloser et al.explain their hybrid analogue–digital circuitin terms of the set of neurons that are activein steady state, and a gain that depends onlyon the identities of the active neurons andnot on their analogue responses. Behaviourthat derives from a common set of activeneurons is linear in the input, whereasbehaviour that derives from a comparisonamong different sets of active neurons isnonlinear in the input.

Hahnloser et al. extend previous studiesof this class of recurrent networks byanalysing the stability of the dynamicalequations that model the network, and showthat there are inviolate constraints on theallowed network states. In particular, theyshow that if no single input can activate a setof neurons (the set cannot form a memory),then no input can activate a supergroup ofthese same neurons (no supergroup canform a memory). A limitation of the study,however, is the requirement that synapticconnections be symmetric — that is, thestrength of the connection from neuron A toneuron B must be the same as that from B toA. This assumption allows them to provenetwork stability, but is difficult to justifyneurobiologically. A fruitful area for futureresearch, therefore, is investigation of recur-rent cortical networks with non-symmetricsynaptic connections (see, for example, refs8–10 and Supplementary Information1).

What do Hahnloser et al., and others likethem, hope to accomplish by building siliconcircuits modelled on biology? First, they canlearn how to map neuronal ‘primitives’(such as neurons and synapses) onto silicon,and then how to compute using these primi-tives (see, for example, refs 10–12). Second,they can investigate how physical and tech-nological limits, such as wire density, signaldelays and noise, constrain neuronal com-putation. And third, they can learn aboutalternative models of computation. Biologyprovides examples of non-digital computingmachines that are incredibly space- andenergy-efficient, and that excel at findinggood solutions to ill-posed problems. Scien-tists may eventually decipher all of nature’selectrochemical circuits, but the work ofHahnloser et al. demonstrates that we

already know enough to begin building inte-grated circuits that compute like biology. ■

Chris Diorio is in the Department of ComputerScience and Engineering, University of Washington,Box 352350, 114 Sieg Hall, Seattle, Washington98195-2350, USA. e-mail: diorio@cs.washington.eduRajesh P. N. Rao is at the Sloan Center forTheoretical Neurobiology and the ComputationalNeurobiology Laboratory, Salk Institute forBiological Studies, 10010 N. Torrey Pines Road,California 92037, USA.e-mail: rao@salk.edu1. Hahnloser, R. H. R., Sarpeshkar, R., Mahowald, M. A., Douglas,

R. J. & Seung, H. S. Nature 405, 947–951 (2000).2. Salinas, E. & Abbott, L. F. Proc. Natl Acad. Sci. USA 93,

11956–11961 (1996).

3. Ben-Yishai, R., Bar-Or, R. L. & Sompolinsky H. Proc. Natl

Acad. Sci. USA 92, 3844–3848 (1995).

4. Somers, D. C., Nelson, S. B. & Sur, M. J. Neurosci. 8, 5448–5465

(1995).

5. Douglas, R. J. et al. Science 269, 981–985 (1995).

6. Stemmler, M., Usher, M. & Niebur, E. Science 269, 1877–1880

(1995).

7. Andersen, R. A., Essick G. K. & Siegel R. M. Science 230,456–458 (1985).

8. Abbott, L. F. & Blum, K. I. Cereb. Cortex 6, 406–416

(1996).

9. Rao, R. P. N. & Sejnowski, T. J. in Advances in Neural

Information Processing Systems Vol. 12 (eds Solla, S. A., Leen,

T. K. & Müller, K.-R.) 164–170 (MIT Press, Cambridge, MA,

2000).

10.Westerman, W. C., Northmore, D. P. M. & Elias, J. G. Analog

Integrated Circuits Signal Process. 18, 141–152 (1999).

11.Diorio, C., Hasler, P., Minch, B. A. & Mead, C. IEEE Trans.

Electron Devices 44, 2281–2289 (1997).

12.Mead, C. Analog VLSI and Neural Systems (Addison-Wesley,

Reading, MA, 1989).

news and views

892 NATURE | VOL 405 | 22 JUNE 2000 | www.nature.com

In tissues that can repair themselves, suchas skin and liver, dead cells can be replacedeither by the proliferation of nearby cells

or by the activation of resident stem cells —undifferentiated cells with the potential togenerate many different cell types. The brainapparently lacks this regenerative capacity,making it particularly vulnerable to injury or disease. Cells with stem-cell-like proper-ties are likely to occur throughout the adultcentral nervous system, but they normallygive rise to neurons in only a few, restrictedareas. These cells might represent a dormantcapacity for neuronal repair, if they could be mobilized to generate new functionalneurons in response to injury. On page 951 of this issue, Magavi, Leavitt and Macklis1

report the first intriguing evidence that thismay be possible.

In the adult brain, the generation of newneurons (neurogenesis) occurs in just tworegions2. The first is the subventricular zone(SVZ) in the wall of the lateral ventricle.Here, new interneurons are generated for theolfactory bulb (Fig. 1a), which is involved insensing odours. The second is the subgranu-lar zone of the dentate gyrus, which gives riseto a different type of neuron, the granule cell.In these areas, there is seemingly a continu-ous turnover of interneurons and granulecells, implying that the newborn neuronsreplace dying cells. But does neurogenesisactually help to replenish damaged neuronalcircuits? There is some evidence that damageto granule cells can trigger the increased pro-liferation and recruitment of new granulecells from resident progenitors3. And insults— such as seizures and inadequate bloodsupply to the cerebrum — that cause cells inthe hippocampus to die are accompanied byincreased neurogenesis in the dentate sub-granular zone4–6. But as yet there is no direct

evidence that the degenerated neurons areactually replaced by the new ones.

In other parts of the central nervous system, neural progenitors contribute to theongoing formation of new non-neuronalcells called astrocytes and oligodendrocytes,and may participate in the reaction of theseso-called glial cells to injury, as well as in scar formation7,8. Other progenitors appearto remain undifferentiated, or die7,8. A lowlevel of neurogenesis has been described inparts of the intact neocortex of adult monk-eys9, but the newly generated neuron-likecells appear to survive only transiently, andmay not differentiate into fully functionalneurons10.

In line with previous findings11,12, Magaviet al.1 have observed the continuous forma-tion of new cells in the intact neocortex of the adult mouse. These proliferating cellswere distinguished by labelling them withbromodeoxyuridine, a thymidine-base ana-logue that is incorporated into the new DNAformed during DNA replication in dividingcells. None of these newly formed cells, however, expressed any marker proteinscharacteristic of neurons.

To investigate the effects of neuronaldeath in the neocortex, Magavi et al.destroyed a subset of pyramidal neurons thatproject from the neocortex to the thalamus,which is a major subcortical relay stationinvolved in the control of cortical function.Their technique resulted in the slow death byapoptosis of the targeted cells only, withoutaffecting the surrounding cortical tissue.The number of new cells formed in the twoweeks after the lesion was similar in controland experimental mice. However, about1–2% of the newly formed cells in the dam-aged neocortex (about 50–100 cells per cubicmillimetre) expressed neuronal markers.

Neurobiology

Self-repair in the brainAnders Björklund and Olle Lindvall

© 2000 Macmillan Magazines Ltd

Such cells were not observed in the controlmice. The new neuronal cells occurred onlyin the cortical layer that was undergoingdegeneration, and some of them had a mor-phology characteristic of cortical pyramidalneurons. The newly formed neurons extend-ed processes to the original target sites in thethalamus, suggesting that they joined up thedamaged circuitry.

The observations are provocative, buthow certain can we be that the interpreta-tions are correct? As pointed out recently10,bromodeoxyuridine may label cells that arerepairing damaged DNA rather than repli-cating DNA ready for cell division. So there isa risk that neurons undergoing cell death, aswell as newly divided cells, may have incor-porated the label. But bromodeoxyuridine-labelled cells in the damaged neocortexexpressed markers — not found in controls— characteristic of migrating young neu-rons, and survived for at least six months. Soit seems unlikely that they were dying. It isalso conceivable that a few surviving neuronsmight be labelled as a result of reversibleDNA damage. This was not controlled for,but in this case one would expect less

bromodeoxyuridine to be incorporated thanwas observed.

These intriguing data raise several ques-tions. First, we need to know where the new-born neurons originate. Possibilities are theprogenitors present locally in the cortex, orcells in the underlying SVZ region (Fig.1b).During postnatal development, SVZ cellsmigrate to neocortex to form glial cells, butin the adult they migrate only to the olfactorybulb. Yet adult SVZ cells can respond to dam-age by increased proliferation and migrationtowards the lesion13,14. Magavi et al. observedcells migrating into the lesioned corticalarea. These cells, not present in controls,were apparently recruited from a distance bysignals produced as a result of the apoptoticcell death.

Second, we need to identify the signalsinvolved in the neurogenic response. In simi-lar experiments, Macklis and colleagues15

showed that the expression of several neu-rotrophic factors (required for neuronal survival, migration and differentiation) isupregulated in interneurons near the de-generating neocortical neurons. Similarly,insults leading to increased neurogenesis in

news and views

NATURE | VOL 405 | 22 JUNE 2000 | www.nature.com 893

Differentiatedpyramidal neuron

Glial cell

Migratingneuroblast

Neuralprogenitor

Stem cell

Cortex

Intact Lesion

SVZ

Rostralmigratory stream

1

2

ThalamusThalamus

a b

Figure 1 Does the brain have a dormant capacity for self-repair? a, In the neocortex of the adultmouse, stem cells and other neural progenitors occur locally in the cortical grey and white matter,and in the underlying subventricular zone (SVZ). The cortical progenitors give rise to glial cells (non-neuronal cells) only. The neurons generated in the SVZ migrate along the rostral migratory stream tothe olfactory bulb (not shown). b, Magavi et al.1 used a refined photolytic lesion to induce apoptoticcell death in a subset of neurons that project from the neocortex to the thalamus. The cells were filledfrom their terminals in the thalamus with nanospheres carrying a chromophore, which becomestoxic when irradiated with red laser light. During the following weeks, Magavi et al. detected newneurons with features of cortical pyramidal neurons in the layers undergoing degeneration. The newneurons appeared to be recruited from the resident cortical progenitors (pathway 1) and/or from theunderlying SVZ (pathway 2). A few of these cells seemed to connect back to the thalamus, suggestinga hitherto unknown capacity for neuronal replacement and functional repair.

100 YEARS AGOThe determination of the strength ofcollateral heredity is a problem of greatscientific importance, and it can only beachieved by co-operative action. I havefound so many teachers in all classes ofschools willing to give disinterested aid inthe cause of science that I venture to make a further appeal through Nature for moreassistance. Besides observations of physicaland mental characters, which can berecorded without measurement, my datapapers ask for certain head-measurements,which can, following the printedinstructions, be taken quite easily. I shall bemost glad to send sample papers to any onewilling to assist, and if, after consideringthese, they find themselves able to assist,say by filling in data papers for ten or morepairs of brothers or sisters, I will at oncedespatch a head-spanner, of which I haveseveral at the present time, free. The head-spanner should not be retained (unlessunder special circumstances) for more thana few weeks. Where the school is a smallone, one master has, as a rule, filled in thepapers entirely; in larger schools, one of thescience masters, or even the medical officer,has done the head-measurements, and theother data have been provided by house,form or consulting masters. Karl PearsonFrom Nature 21 June 1900.

50 YEARS AGODictionary of GeneticsThis book contains, or attempts to contain,every word connected with genetics in thewidest sense. It therefore spreads over(though it does not cover) the whole range of biology; psychology and anatomy,embryology and biochemistry are allrepresented… For some terms (such ascytomicrosome), the author proceeds bydescribing ignotum per ignotius; others (suchas mitoschisis or merostathmokinesis) havenever been used except by their anonymousinventors. Let us hope that others again (suchas thermocleistogamy, tachyauxesis andspermiocalyptrotheca) never will be used.Others which have a known meaning lose it(such as “mass mutation” in Œnothera) or donot appear at all (such as “sterility”). Thenagain, others (such as heterofertilization or parthenogamy) describe rare or evenimaginary phenomena. One supreme examplewhich is non-existent as a technical term(perultimate chromomere) appears to be atonce misplaced, misspelt and misdefined.From Nature 24 June 1950.

© 2000 Macmillan Magazines Ltd

the dentate gyrus induce dramatic changesin the expression of many neurotrophic factors16. The lesion-induced induction ofkey signalling molecules might representthe re-emergence of a programme that controls cortical development. To use thisresponse to neuronal death in reconstruct-ing functional neuronal circuits, we need to know much more about the sequence ofsignals that guide the migration, differentia-tion, integration and connections of neuralprogenitors.

These results raise the enticing possibilitythat the brain has a latent capacity for self-repair. However, the neurogenic responseobserved by Magavi et al.1 was limited, and it is inconceivable that the small fraction of damaged neurons that appeared to bereplaced by new neurons would allow sig-nificant functional recovery. We do not evenknow whether the new cells — even thoughthey seemed to form appropriate connec-tions — can take on the function of the cells they replace. There is a long way to go, but learning how to boost and guide neurogenesis from the stem-cell pool mighteventually lead to a powerful tool for brainrepair in human disorders of the central nervous system. ■

Anders Björklund and Olle Lindvall are at theWallenberg Neuroscience Center, Lund University,Sölvegatan 17, S-223 62 Lund, Sweden.e-mails: anders.bjorklund@mphy.lu.seolle.lindvall@neurol.lu.se1. Magavi, S. S., Leavitt, B. R. & Macklis, J. D. Nature 405,

951–955 (2000).

2. Gage, F. H. Science 287, 1433–1438 (2000).

3. Gould, E. & Tanapat, P. Neuroscience 80, 427–436

(1997).

4. Bengzon, J. et al. Proc. Natl Acad. Sci. USA 94, 10432–10437

(1997).

5. Parent, J. M. et al. J. Neurosci. 17, 3727–3738 (1997).

6. Liu, J., Solway, K., Messing, R. O. & Sharp, F. R. J. Neurosci. 18,

7768–7778 (1998).

7. Frisén, J., Johansson, C. B., Torok, C., Risling, M. & Lendahl, U.

J. Cell Biol. 131, 453–464 (1995).

8. Gensert, J. M. & Goldman, J. E. Neuron 19, 197–203 (1997).

9. Gould, E., Reeves, A. J., Graziano, M. S. A. & Gross, C. G.

Science 286, 548–552 (1999).

10.http://www.sciencemag.org/cgi/content/full/288/5467/771a

11.Altman, J. Anat. Rec. 145, 573–591 (1963).

12.Kaplan, M. S. Ann. NY Acad. Sci. 457, 173–192 (1985).

13.Weinstein, D. E., Burrola, P. & Kilpatrick, T. J. Brain Res. 743,

11–16 (1996).

14.Nait-Oumesmar, B. et al. Eur. J. Neurosci. 11, 4357–4366

(1999).

15.Wang, Y., Sheen, V. L. & Macklis, J. D. Exp. Neurol. 154,

389–402 (1998).

16.Lindvall, O., Kokaia, Z., Bengzon, J., Elmér, E. & Kokaia, M.

Trends Neurosci. 17, 490–496 (1994).

news and views

NATURE | VOL 405 | 22 JUNE 2000 | www.nature.com 895

In 1683, the Ottoman Turks’advance up the Danube wasturned back at Vienna. Acontemporary crustaceaninvader, however, is havingmore success. Jaimie Dick andDirk Platvoet have discoveredthat, in the freshwaterecosystems of the Netherlands,the native shrimp Gammarusduebeni is being wiped out by a menace from the east,Dikerogammarus villosus(Proc. R. Soc. Lond. B 267,977–983; 2000).

This species is native toeastern Europe and the Ukraine.But D. villosus has spread toWestern Europe through theDanube–Main canal (whichopened in 1992) and appearedin the Netherlands about fiveyears ago. Females, shownhere on zebra mussels, areabout 15 mm long, males being twice that size.

The alien’s method oftakeover is nothing if not direct — it eats the natives. In particular, male D. villosusconsume female G. duebeni,which are smaller than males

of the same species and lessable to resist attack. Theinvader is especially destructivebecause it can feed on its preybetween moults, when theexoskeleton is tough, and notjust on soft-skinned, recentlymoulted shrimps.

In some places, G. duebenihas already been displaced bya fast-breeding North Americanspecies, G. tigrinus, whichprobably reached Europe inship ballast-water. To an extentit could withstand this assault,because the two species preferdifferent habitats and salinitieswhich helps to keep them

apart. But D. villosus cantolerate a wide range ofdifferent conditions, and it isalso thought to be responsiblefor the recent sharp declines inpopulations of G. tigrinus.

As it becomes ever easierfor human beings to traversethe globe, exotic animals andplants will be introduced intonew environments, both onpurpose and as unseenhitchhikers. We face theprospect that ecosystems will become increasingly draband homogeneous, dominatedby a few super-competitivespecies. John Whitfield

Ecology

Shrimp-eat-shrimp

Atortuous quest involving physicists,chemists and biologists that hasendured for over 150 years has finally

ended with a paper by Rikken and Raupach1

on page 932 of this issue. They report the firstunequivocal use of a static magnetic field tobias a chemical process in favour of one of

two mirror-image products (left- or right-handed enantiomers). The chemistry of life is homochiral, being based almost exclusive-ly on L-amino acids and D-sugars, and theability of biological molecules to discrimi-nate between enantiomers is vital for livingsystems. The importance of handedness in nature is such that scientists have longwondered about its origin, and the processdemonstrated by Rikken and Raupach mayprovide a new clue.

The quest began in 1846 when Faradaymade the plane of polarization of a linearlypolarized light beam rotate by applying amagnetic field parallel to the beam. This discovery was of fundamental importancebecause it demonstrated conclusively theintimate connection between electromag-netism and light. But it also became a sourceof confusion to many scientists who failed toappreciate that there is a distinction betweenFaraday’s magnetic optical rotation and thenatural optical rotation discovered threedecades earlier by Arago and Biot in certaincrystals and fluids. Such natural opticalactivity is due to the handedness within themicrostructure of the crystals and fluids, asFresnel later showed.

The first to be misled was Pasteur, who in 1848 separated crystals of sodium ammo-nium tartrate into right- and left-handedforms, which gave equal and opposite naturaloptical rotations in solution. Following onfrom this epochal discovery, he attempted toinduce handedness in crystals by growingthem in a magnetic field2, which he mistak-enly thought, following Faraday’s discovery,to be a source of handedness. But LordKelvin, who first introduced the word ‘chirality’ into science, was under no suchmisapprehension, and stated quite firmlythat “the magnetic rotation has neither left-handed nor right-handed quality, that is to say, no chirality. This was perfectlyunderstood by Faraday, and made clear in

Chemistry

Chirality, magnetism and lightLaurence D. Barron

D. P

LAT

VO

ET

/J. D

ICK

/I. E

WA

RT

© 2000 Macmillan Magazines Ltd