visible targets viewed against darkness,observers wore prisms that deviated thelight from the targets, causing them to havelarger angular declinations (signalling closertargets) than in the initial experimentsdescribed above. The walking and pointingprocedure revealed a systematic increase inthe perceived visual direction as well as a cor-responding reduction in perceived distance.

In the same experiment, observers wereallowed to adapt to the prisms beforehand by walking around a lighted room whileavoiding obstacles, after which they per-formed blindfolded walking and pointing to targets. Adaptation to the prisms pro-duced shifts in perceived direction towardsthe horizon, with corresponding increases inperceived distance. A separate experiment,in which the adaptation procedure involvedthrowing beanbags at a target instead ofavoiding obstacles, produced similar results.This confirms that the shifts were real shiftsin perceived angular declination, not someafter-effect specific to the coupling betweenvision and walking.

The question still remained as to the exactmeaning of eye-level used by the nervous sys-tem in evaluating angular declination. Visual-ly perceived eye level (VPEL) — the directionin space that appears level with the eyes — isdetermined both by the visual scene and bynon-visual signals, such as those from theinner ear, which indicate the gravitationallyspecified horizontal plane8. In an additionalexperiment, Ooi et al. asked observers to posi-tion a visible spot viewed in darkness so that itappeared at eye level. Observers carried outthis procedure under the same sets of condi-tions used in the other experiments: with andwithout prisms, and while experiencing theafter-effects of prism adaptation. The shifts in VPEL matched those observed with thewalking and pointing procedure, indicatingthat angular declination is indeed measuredwith respect to VPEL, at least under these conditions of darkness. Another experiment,in which observers viewed a well-illuminatedscene, showed that viewing through prismsboth before and after adaptation producedshifts in perceived distance in the expecteddirections. So it seems that VPEL might alsoserve as the reference direction for angulardeclination in well-illuminated scenes.

Apart from showing the effectiveness ofangular declination as a distance cue andrevealing one aspect of its processing by the visual system, this research also has great methodological significance. Until nowthere has been no sure and simple way ofmeasuring the perceived three-dimensionallocation of a viewed target, at least onebeyond arm’s reach. With Ooi and col-leagues’ unambiguous demonstration thattheir refinement of visually directed actiondoes indeed measure perceived locationusing just a single response, the approach canbe used much more widely to measure the

effectiveness of different distance cues andtheir integration within the visual system. ■

Jack M. Loomis is in the Department of Psychology,University of California, Santa Barbara, California93106, USA.e-mail: loomis@psych.ucsb.edu 1. Sedgwick, H. A. in Handbook of Perception and Human

Performance, Vol. 1 (eds Boff, K. R., Kaufman, L. & Thomas,

J. P.) 21.1–21.57 (Wiley, Chichester, 1986).

2. Howard, I. P. & Rogers, B. J. Binocular Vision and Stereopsis

(Oxford Univ. Press, 1995).

3. Ooi, T. L., Wu, B. & He, Z. J. Nature 414, 197–200 (2001).

4. Thomson, J. A J. Exp. Psychol. Hum. Percept. Perform. 9,427–443 (1983).

5. Fukusima, S. S. et al. J. Exp. Psychol. Hum. Percept. Perform. 23,86–100 (1997).

6. Philbeck, J. W. et al. Percept. Psychophys. 59, 601–612 (1997).

7. Philbeck, J. W. & Loomis, J. M. J. Exp. Psychol. Hum. Percept.

Perform. 23, 72–85 (1997).

8. Matin, L. & Fox, C. R. Vis. Res. 29, 315–324 (1989).

news and views

156 NATURE | VOL 414 | 8 NOVEMBER 2001 | www.nature.com

Water continues to surprise us. Intu-itively, one would not expect water toenter a narrow hydrophobic pore,

such as that formed by a carbon nanotube,because of both the tube’s narrowness and its‘oily’, water-repellent properties. Such chemi-cal common sense stems from our experienceof the macroscopic world, however, and maynot apply at the nanometre scale.

On page 188 of this issue1, Hummer and colleagues report molecular-dynamicssimulations that examine the behaviour ofwater within carbon nanotubes, the internaldiameter of which is sufficient to accommo-date a single-file column of water molecules.Using standard parameters for the strengthof the weak attractive force — the van derWaals’ interaction — between water molec-ules and the carbon atoms of the nanotube,the authors show that water molecules enterthe pore, forming a column five molecules inlength. But if the carbon–water interaction is made a little less favourable, by reducingthe depth of the interaction’s energy well (thestrength of the interaction) from 10.114 kcalmol11 to 10.065 kcal mol11,there is a dramat-ic change in water behaviour, and the poreremains empty of water molecules for mostof the time.

The simulations are long by current standards (50 nanoseconds or more), therebyallowing the flow of water moleculesthrough the pore to be monitored. This workextends our understanding of how liquidsbehave on the nanoscale. Several studieshave examined the behaviour of liquidssqueezed into thin films, but comparabledata on nanoscale pores are harder to comeby, hence the value of simulations.

Many fluids behave abnormally whenconfined in a space of nanometre dimen-sions2. For example, simple organic liquidsbecome solid-like when squeezed betweentwo smooth surfaces into a film that is lessthan about five molecular layers thick3. Incontrast, if water is squeezed between two

mica surfaces, only small changes in viscosityoccur4. The nature of the surfaces betweenwhich the water is confined may also have aneffect. There have also been simulations5 toexamine the ‘nano-ice’ formed by water innanotubes of different dimensions, reveal-ing phases of ice that are not found underbulk conditions. This line of research is relevant not only to the science of carbonnanotubes. Water-filled pores of similardimensions to nanotubes, for instance aqua-porins and ion channels, are present in manymembrane-spanning transport proteins(Fig. 1). Aquaporins form water-permeablepores in many cell membranes, and ionchannels form ion-permeable pores thatgovern the electrical properties of nerve cells.

Hummer et al.1 describe simulations of anarrow — (6,6), in trade notation — carbonnanotube in a ‘box’ of 1,000 water molecules.A crucial feature of the simulations is thatwater molecules are allowed to enter andleave the tube freely, rather than simply being confined within it. Perhaps the mostinteresting aspect of the simulations is howthe wet–dry transition (that is, the entry andexit of water to and from the pore) can bemanipulated, and the consequences of suchmanipulation. In the simulations, the transi-tion could be induced by a minute change inthe strength of the water–carbon interactionpotential, implying that small changes in thepolarity (electrostatic state) and/or geome-try of a pore might drive transitions betweenfull and empty states.

How is water able to enter a hydrophobicchannel? The key lies in the energetics ofhydrogen bonding. If water molecules enterthe nanotube, they each lose on average twoout of their four hydrogen bonds. This wouldbe expected to be too high an energetic costto allow the water to enter. However, fluc-tuations in the number of hydrogen bondsper water molecule in the bulk aqueousphase mean that a significant fraction ofwater molecules are incompletely hydrogen-

Biophysics

Water at the nanoscaleMark S. P. Sansom and Philip C. Biggin

You would not expect water to enter a hydrophobic carbon nanotube. But computer simulations show that it can, and studying the processshould provide clues about the behaviour of biological pores.

© 2001 Macmillan Magazines Ltd

bonded, and thus have a low binding energy.This class of water molecules is not presentwithin the nanotube, which providesenough difference in chemical potential todrive water into it, where the water is shielded from fluctuations. Water within thenanotube moves through the tube in bursts— an average of about 17 water moleculesmove right through the tube each nano-second, but peaks of about 30 molecules ns11

are observed. As expected, water movesthrough the tube in a highly correlated, single-file manner, maintaining the chain ofhydrogen bonds.

Other simulations of water in nanotubes6

have also shown that the number of hydro-gen bonds within such tubes is reduced relative to conditions in bulk water. Interest-ingly, in the narrowest (6,6) nanotube, waterdiffuses about two times faster along the tube than it does in the bulk. This contrastswith the situation in other, more polar pores,in which water diffusion is reduced7, andshows the importance of the exact nature ofthe pore–water interaction.

The internal radius of a (6,6) nanotube is4.1 Å; that of an aquaporin is, on average,about 2 Å. The surface lining the aquaporinpore is a mixture of hydrophobic andhydrophilic regions and so might be expectedto exhibit even more complex behaviourthan a ‘simple’ nanotube. The observationthat entry of water into nanotubes can befine-tuned by controlling the exact strengthof water–tube interactions is directly rele-vant to the way in which evolution has fine-tuned different members of the aquaporinfamily in their relative permeability to waterand glycerol8.

Hummer and colleagues1 controlled thewet–dry transition of the nanotube by creat-ing a small change in the interaction potentialbetween the water molecules and carbon.Our own simulations9 of a simple model of a

membrane pore have shown that the exit andentry of water can also be controlled by smallchanges in pore radius — water is excludedfrom a pore of 4-Å radius but is freely able toenter one of 6-Å radius. The entry and exit of water could also be controlled by the presence or absence of a small dipole in anotherwise hydrophobic pore lining, such ascould be provided by a polar side chain in a biological channel. Such mechanisms forcontrolling pore wetting may be relevant tothe mechanisms for gating — opening andclosing — of biological channels.

These studies also demonstrate the powerof modern computer simulations to probethe properties of water in nanoscale environ-ments, revealing behaviour that would notbe expected from our understanding of thebulk properties of water. But the water models used in such simulations have theirlimitations. In particular, their parametershave been set to reproduce the experimentalproperties of bulk water, and so we need tofind out whether small changes in watermodels affect the simulation results.

Technologically, one of the most excitingaspects of Hummer and colleagues’ study is that it suggests that, by inducing smallchanges in nanotube structure, it may bepossible to drive the wet–dry transition andthus to gate a synthetic nanopore. This ismuch like using simulations to understandthe properties of biological pores at atomicresolution, because simulations allow us tomove beyond the somewhat static picturesyielded by conventional structural biology.From a sociological perspective, this studyconfirms that some of the most exciting mol-ecular science is happening at the interfacebetween physics and biology. We can lookforward to further fruitful encounters at thismeeting point. ■

Mark S. P. Sansom and Philip C. Biggin are in theLaboratory of Molecular Biophysics, Department of

news and views

NATURE | VOL 414 | 8 NOVEMBER 2001 | www.nature.com 157

Figure 1 Pore comparisons. Left, a biological pore (grey), as found in the bacterial water and glycerolchannel, GlpF. The pore is formed by the side chains of the protein backbone (seen here as colouredspaghetti). Right, a similar pore in a narrow carbon nanotube, such as that for which Hummer et al.1

carried out simulations of water conduction. It is hoped that nanotubes can be used in simulations to mimic and help understand the behaviour of aquaporins and ion channels.

100 YEARS AGOIn these days the search for somecharacteristic of the human body which willgive unequivocal evidence of the mentalnature of the individual still goes on… Theresult of my enquiries in this direction hasbeen to show that only two out of the sevenfeatures of the external ear which I haveinvestigated are correlated with a mental biastowards crime or insanity, viz. a retrogradedevelopment of the helix and a persistence ofthe ear tip… All that can be deduced from thepresent investigation is that a slightly greaterproportion of the people who have ear tips andretrograde helices give themselves over tocrime than those in which these two featuresare absent. The evidence is just sufficient tojustify the suspicion that a small proportion of criminals are criminals because of theirphysical constitution, and it is certainly theduty of every anatomist to discover how suchindividuals may be recognised. As yet all thecriminal marks we know of can only be statedin relative terms of the class, and have,unfortunately, no application to the individual.From Nature 7 November 1901.

50 YEARS AGOThe two main difficulties in internationalcollaboration… are, first, the tendency ofnumerous governments to express differentviews in different places on relatedinternational problems; and secondly, thefailure of all too many governments todischarge promptly and effectively theirsubstantive obligations as participants in aco-operative international system. The firstnot only creates confusion but may alsoaffect the soundness of the plans which areformulated. The second, however, couldmake the formulation and execution of plansa complete waste of time. It is the growingdisregard for international obligations andthe unilateral repudiation of negotiatedagreements that is the most disturbingfeature of to-day. If it is unchecked, it canhalt all schemes for economic and technicalassistance to backward countries.Technicians and scientific advisers cannotwork confidently and effectively apart fromsome rule of law and guarantee of stability,nor indeed will it be possible to recruit themfor such work. If they cannot be recruited,then the programme stops. In this connexion,the challenge which present events in Persiaand Egypt thus presents to internationalinstitutions does not appear to have receivedthe attention it demands.From Nature 10 November 1951.

© 2001 Macmillan Magazines Ltd

Biochemistry, University of Oxford, South ParksRoad, Oxford OX1 3QU, UK.e-mail: mark@biop.ox.ac.uk1. Hummer, G., Rasaiah, J. C. & Noworyta, J. P. Nature 414,

188–190 (2001).

2. Granick, S. Science 253, 1374–1379 (1991).

3. Klein, J. & Kumacheva, E. J. Chem. Phys. 108, 6996–7009

(1998).

4. Raviv, U., Laurat, P. & Klein, J. Nature 413, 51–54 (2001).5. Koga, K., Gao, G. T., Tanaka, H. & Zeng, X. C. Nature 412,

802–805 (2001).6. Martí, J. & Gordillo, M. C. Phys. Rev. E 64, 021504 (2001).7. Tieleman, D. P., Biggin, P. C., Smith, G. R. & Sansom, M. S. P.

Q. Rev. Biophys. (in the press).8. Fu, D. et al. Science 290, 481–486 (2000).9. Beckstein, O., Biggin, P. C. & Sansom, M. S. P. J. Phys. Chem. B

(in the press).

quickly than Ab40. It seems plausible, then,that drugs that reduce Ab42 levels mightprove useful in treating Alzheimer’s disease.However, recently described inhibitors of g-secretase decrease the generation of all Ab-peptides and, more problematically, thecleavage of Notch as well6,8.

The new results3 hint that NSAIDs mightcontrol g-secretase specifically to block thegeneration of Ab42 only. The authors testedthe effects of three NSAIDs — ibuprofen,indomethacin and sulindac sulphide — onthe processing of APP in cultured cells, andfound that the ratio of Ab42 to Ab40 wassubstantially decreased. Moreover, ibupro-fen lowered the levels of Ab42 in the brains of mice that had been engineered to providea model of Alzheimer’s disease (described inref. 9). Importantly, these drugs do not seemto interfere with the processing of Notch3.

The results shed light on why differentanti-inflammatory drugs have differenteffects on Alzheimer’s disease10. As men-tioned above, several retrospective studiessuggested that some NSAIDs may be ben-eficial1,2. In a small ‘prospective’ trial, indo-methacin slowed the cognitive decline seenin Alzheimer’s patients1,2. And ibuprofenreduces both amyloid plaques and inflam-mation in mouse models of the disease9. Butclinical trials using other classes of anti-inflammatory compound, such as hydroxy-chloroquine11 or the steroid prednisone12,did not show clear benefits to Alzheimer’spatients. Similarly, the best known NSAID— aspirin — had an unconvincing effect inepidemiological studies10. This all seemedrather puzzling at the time: if inflammationis involved in the progression of Alzheimer’sdisease, why would one anti-inflammatorydrug benefit patients while another did not?Weggen et al.’s results offer an answer: we

news and views

NATURE | VOL 414 | 8 NOVEMBER 2001 | www.nature.com 159

Some of the drugs used to treat diseasessuch as rheumatoid arthritis have anunexpected benefit: they reduce the

risk of Alzheimer’s disease1,2. Until now itwas thought that the anti-inflammatoryproperties of the drugs could explain thisodd side effect. But on page 212 of this issueWeggen and colleagues3 offer a differentexplanation, based on their studies of miceand cultured cells. If the findings can beextended to people, these drugs could even-tually form the basis of treatments for thisdevastating neurodegenerative disorder.

Alzheimer’s disease is characterized byseveral changes in the brain, including thebuild-up of protein deposits known as amyloid plaques outside nerve cells, and thebreakdown of neurons. The plaques consistof protein fibres, some 7–10 nanometresthick, that are mixed with small peptidescalled amyloid-b (Ab) peptides. Matureplaques also contain degenerating nerveendings, and are surrounded by active astro-cytes and microglia — cells that help to clearup debris in the brain.

The presence of these cells and certaininflammatory proteins has suggested thatinflammation contributes to Alzheimer’s dis-ease. Moreover, ‘retrospective’ epidemiologi-cal studies1,2 have shown that patients who aretaking certain nonsteroidal anti-inflamma-tory drugs (NSAIDs) have a lower than normal risk of developing this disorder. But,as Weggen et al.3 show, it is not only the anti-inflammatory effects of the compounds thatare important here. Instead, the authors pro-pose that some of these drugs directly affectthe generation of the aggressive 42-amino-acid-long form of the Abpeptide, Ab42.

This Ab42 peptide, together with a seriesof slightly shorter and longer variants, is generated by the cleavage of the amyloid pre-cursor protein (APP; Fig. 1) — a membraneprotein with unknown function that isexpressed in a wide range of tissues. Cleavageis carried out by enzymes that were untilrecently known only by aliases: b-secretaseand g-secretase. But considerable progress

has now been made in identifying the pro-teins behind these names4,5. For instance, the g-secretase is apparently a multiprotein complex, containing presenilin proteins, thatcleaves APP in its transmembrane domain4,5.It also cleaves other transmembrane molec-ules, most importantly Notch6 — a proteinthat is crucial during development.

The g-secretase complex can cleave APPat different positions in its transmembranedomain, which explains why Ab peptides of different lengths are generated (Fig. 1). In patients with the inherited form ofAlzheimer’s disease, mutations in the APPgene or in the presenilin genes often occur,and these mutations cause small but signifi-cant changes in the preferred cleavage site in APP7. This increases the production of theAb42 peptide relative to the other majorspecies — the slightly shorter Ab40. Thissubtle change in ratio is thought to be at the root of Alzheimer’s disease4,5, because the Ab42 peptide precipitates much more

Alzheimer’s disease

An inflammatory drug prospectBart De Strooper and Gerhard König

There is no known cure for Alzheimer’s disease. But new hope (for miceat least) comes from an in-depth investigation of a class of drugs used to treat inflammatory diseases.

Figure 1 Alzheimer’s and amyloid. The lower part of the figure shows amino-acid sequences in single-letter code. Amyloid-b peptides (Ab ) come in a variety of sizes, of which the 42-amino-acid form(Ab42) is thought to contribute significantly to the development of Alzheimer’s disease. Thesepeptides are produced from the amyloid precursor protein (APP) by enzymatic activities known as a-, b - and g -secretases. The g -secretase (or g -secretases) can cleave APP at several positions,generating Ab peptides of different lengths (38, 40 or 42 amino acids). This relaxed enzymaticspecificity is poorly understood. Asterisks indicate the positions of mutations in the APP gene thatcorrelate with the development of Alzheimer’s disease; most of these mutations cause the balance ofAb peptides to shift in favour of Ab42. The numbers 671, 687 and 713 indicate the positions of theamino acids in the protein chain. The p3 peptide is generated by combined a- and g-secretase activity.

K M D A E F R H D S G Y E V H H Q K L V F F A E D V G S N K G A I I G L M V G G V V I A T V I V I T L V M L K K ...

L V F F A E D V G S N K G A I I G L M V G G V V

D A E F R H D S G Y E V H H Q K L V F F A E D V G S N K G A I I G L M V G G V V I A

D A E F R H D S G Y E V H H Q K L V F F A E D V G S N K G A I I G L M V G G V V

D A E F R H D S G Y E V H H Q K L V F F A E D V G S N K G A I I G L M V G G

713671 687

Amyloid precursor protein

Aβ38

Aβ40

Aβ42

p3

γ-secretase

α-secretaseβ-secretase

Transmembranedomain

© 2001 Macmillan Magazines Ltd