Nature © Macmillan Publishers Ltd 1998

8

sons. First, climate models must be able tosimulate the full range of dynamical behav-iour of the climate system, and so the LGM ortransient and extreme climatic periods arethe most critical tests they can be exposed to.Second, models that estimate future changesmust be consistent with the sensitivity of theclimate system to altered forcing parameters.Third, these models, if correct, can provide amore detailed picture of past changes forregions or parameters for which no suitablepalaeoclimate archives are available. Finally,this will eventually contribute to a quantita-tive, model-based interpretation of palaeo-climatic proxy data.

Ganopolski et al. abandon the classicalapproach of using comprehensive three-dimensional general circulation models ofone of the two main components of the cli-mate system (the atmosphere or the ocean),while using a simplified representation of theother. Instead, they build on the progressmade in developing zonally averaged climatemodels3,4, which have become importanttools in palaeoclimatic research, and com-bine it with a new atmospheric model com-ponent. Such models permit the numbers ofruns required for optimal tuning. Thismeans that loosely constrained model para-meters or incompletely known boundaryconditions can be varied in such a way thatthe relevant atmosphere–ocean exchangefluxes of momentum, heat and fresh water,

simulated by the respective model compo-nents, are brought into adequate agreement.

Because of limitations on computer time,such systematic fine-tuning is not yet possi-ble for coupled three-dimensional models.So most of these models drift to unrealisticstates, and flux corrections must be applied.This remedy is undesirable but permissiblefor small and linear excursions from a well-defined climate state. But caution is calledfor when modelling climates as differentfrom today’s as the LGM.

The crucial component for a successfulsimulation is the hydrological cycle, which is notoriously difficult to simulate. Watervapour is the most important greenhousegas, and its reduction in level contributessignificantly to the global cooling of 6.2 7C inGanopolski and colleagues’ model. This iscolder than the usually accepted estimates ofabout 4 7C, but it is consistent with recentreconstructions of significantly lower temperatures for the LGM at high and low latitudes5–7.

The hydrological cycle also influences thecirculation of the deep ocean. Changes inpatterns of evaporation and precipitationdetermine the location of deep-water for-mation and the mix of waters from northernand southern origin in the North Atlantic8.Reduction in high-latitude precipitation,and an even bigger decrease in evaporationdue to a larger extent of the ice cover in thenorthern North Atlantic (the ice marginmoves from 757 N to 557 N in the model),mean that a surface freshwater anomaly candevelop; in consequence, the area of deep-water formation in the North Atlantic alsomoves about 207 south. The density of thesinking waters is reduced, with the conse-quence that they penetrate less deeply. In thisway, water masses of southern origin canextend far north in the Atlantic basin and fillmuch of the deep Atlantic. This meridionalshift in the distribution of water mass below1,000 m, seen in the model, is consistent withpalaeoceanographic evidence from stablecarbon isotope measurements on benthicforaminifera9.

As well as strengthening our confidencein reconstruction of proxy data, coupled climate models are also crucial in assessinghypotheses on which less complete modelsare based. Ganopolski et al. find that, duringthe LGM, the oceanic meridional heat flux inthe Atlantic was very different from today.Although this is not surprising (sea-surfacetemperatures were lower10, and sites of deep-water formation and sea-ice margins movedsouth), it contradicts recent assumptionsused for atmosphere-only models11. Thereduction of heat delivery into the NorthAtlantic by northward-flowing waters is alsoexpected, given the extremely cold tempera-tures on Greenland during the LGM (ref. 5).This leads to a pronounced asymmetry incooling for the LGM, with a fall of more than

news and views

338 NATURE | VOL 391 | 22 JANUARY 1998

Figure 1 CLIMAP10 reconstruction of sea-surface temperature in August at the Last Glacial Maximum(LGM), 21,000 years ago. Four areas that pose crucial questions for climate reconstruction bynumerical models are indicated as follows. 1, Are the Nordic Seas ice-free during the summer15? 2,What are the sea-surface salinities in the area of LGM deep-water formation16,17? 3, What is thereduction of sea-surface temperature in the tropical ocean7,18? 4, What are the surface-watercharacteristics in the Southern Ocean during the LGM? Other questions concern how theatmospheric temperature gradient changes with time, continental temperatures, the distribution oftracers and isotopes in the ocean, the water-mass mix, and the location of areas of deep-waterformation and their seasonality. (Reproduced from ref. 19.)

During most of the past 100,000 years,temperatures on Earth were muchcolder than they are now and climate

was very unstable. About 21,000 years agothat climatic period culminated in the LastGlacial Maximum (LGM), when about 50million km3 of water was locked in huge icesheets, lowering sea level by more than 120metres. Climate during the LGM was clearlyvery different from what it is today. But howdifferent? This is the subject of the paper byGanopolski et al. on page 351 of this issue1.

Using a properly tuned, simplified, coupled ocean–atmosphere climate model,they first verify the model’s ability to simu-late modern climate on the global scale. Afteradapting insolation, concentrations ofatmospheric greenhouse gases and ice-sheetdistribution to values typical for the LGM,they find that the same model yields a stableclimatic state with other atmospheric andoceanic characteristics that are reminiscentof the LGM. The simulated changes arebroadly consistent with what we know fromdecades of invaluable analyses of marine sed-iments, polar and tropical ice cores, treerings and groundwater — all of which arearchives that record past climatic changes.

Why should we be interested in simulat-ing the climate of the LGM, when that state isunlikely to return for another 50,000 years2

and Earth’s climate will instead most proba-bly warm considerably? There are four rea-

Palaeoclimatology

A glimpse of the glacialThomas F. Stocker

2. New, J. H., Sugiyama, T., Zaitseva, E. & Kowalczykowski, S. C.

Nature 391, 407–410 (1998).

3. Shinohara, A. & Ogawa, T. Nature 391, 404–407 (1998).

4. Benson, F. E., Baumann, P. & West, S. C. Nature 391, 401–404

(1998).

5. Kinzler, K. W. & Vogelstein, B. Nature 386, 761–763

(1997).

6. Wold, M. S. Annu. Rev. Biochem. 66, 61–92 (1997).

7. Mortensen, U. H., Bendixen, C., Sunjevaric, I. & Rothstein, R.

Proc. Natl Acad. Sci. USA 93, 10729–10734 (1996).

8. Yonesaki, T. & Minagawa, T. J. Biol. Chem. 264, 7814–7820

(1989).

9. Bai, Y. & Symington, L. S. Genes Dev. 10, 2025–2037 (1996).

10.Muris, D. F. R. et al. Curr. Genet. 31, 248–254 (1997).

11. Kanaar, R. & Hoeijmakers, J. H. J. Genes Funct. 1, 165–174 (1997).

12. Webb, B. L., Cox, M. M. & Inman, R. B. Cell 91, 347–356 (1997).

Nature © Macmillan Publishers Ltd 1998

8

Research into the molecular mecha-nisms of Alzheimer’s disease — themost common late-life dementia —

continues to illuminate intriguing issues inprotein biology, while seeking to identifytherapeutic targets. These twin outcomes areexemplified by the exciting results of DeStrooper and co-workers1, reported on page387 of this issue. The authors show that thepresenilin-1 (PS1) gene, mutations in whichcause an aggressive form of early-onsetfamilial Alzheimer’s disease2,3, regulates theunusual intramembranous proteolysis of the amyloid precursor protein (APP). PS1might, therefore, serve as an unexpected target for drugs against the disease.

In families with autosomal dominantAlzheimer’s disease, mutations in three differ-ent genes — APP, PS1 and PS2 — have beenidentified2,3. Missense mutations in the APPgene are located at, or close to, the recognitionsites of the APP-cleaving secretases (a, b andg; Fig. 1). These mutations lead to increasedproduction of amyloid-b peptide, particular-ly its amyloidogenic 42-residue form, theamyloid-b(1–42) peptide, which accumu-lates in the amyloid plaques of both ‘sporadic’and familial cases of Alzheimer’s disease3.Whereas mutations in APPare extremely rare,numerous mutations have been observed inthe PS1 and PS2 genes2,3 (which are homol-ogous to each other). All of the presenilinmutations analysed so far increase the levels of

secreted amyloid-b(1–42) peptide, evenpresymptomatically, supporting a central rolefor amyloid-b peptide in the pathogenesis ofAlzheimer’s disease2,3.

Presenilin proteins span the membranesix to eight times. They are proteolyticallyprocessed to produce stable amino- and carboxy-terminal fragments2, which form aheterodimeric complex that seems to be bio-logically relevant (Fig. 1)4. Work on thenematode Caenorhabditis elegans indicatesthat presenilins are involved physiologicallyin cell-fate decisions through the Notch signalling pathway2. In support of this, deletion of the PS1 gene in mice produces afatal developmental defect that closelyresembles the phenotype seen in Notchknockout mice5,6.

De Strooper et al.1 studied mice lackingboth PS1 alleles (PS1–/–). Previous work bythis group7 showed that, in mouse cells,endogenous APP is poorly processed intoamyloid-b peptide, whereas the human APPsequence is efficiently cleaved. To allow gen-eration of amyloid-b peptide in the PS1–/–

background to be analysed quantitatively,the authors infected hippocampal neuronscultured from the knockout embryos with arecombinant Semliki Forest virus encodinghuman APP. Strikingly, these cells producedsubstantially (~80%) less amyloid-b pep-tide, and the related peptide p3, than did thesame construct infected into cells expressing

news and views

NATURE | VOL 391 | 22 JANUARY 1998 339

Alzheimer’s disease

A technical KO of amyloid-b peptideChristian Haass and Dennis J. Selkoe

PS1. (The residual amyloid-b peptide/p3may have resulted from continued expres-sion of the homologous PS2 gene.)

In contrast to the presenilin mutationsthat are responsible for familial Alzheimer’sdisease — and specifically increase genera-tion of amyloid-b(1–42) peptide — the lackof PS1 reduced production of the 40- and 42-residue forms of amyloid-b peptide equally(Fig. 1). Carboxy-terminal fragments of APPbeginning at the b- and a-secretase sitesaccumulated to high levels in lysates of thePS1–/– neurons. But, in neurons from normalmice, these fragments were efficientlycleaved by g-secretase, giving rise to amy-loid-b peptide and p3.

These observations could indicate thatPS1 is the g-secretase, although this seemsunlikely because the protein has no sequencehomology to known proteases. Instead, De

Figure 1 The influence of presenilin-1 (PS1) onproteolytic processing of the amyloid precursorprotein (APP). b-secretase cleaves APP at theamino terminus of the amyloid-b (Ab ) peptidedomain to generate a carboxy (C)-terminalfragment containing the entire amyloid-bpeptide. Generation of amyloid-b peptide isdecreased by a-secretase, which cleaves in themiddle of the amyloid-b peptide domain tocreate a shorter carboxy-terminal fragment. De Strooper et al.1 have found that bothfragments accumulate in the PS1–/– mice. Thecarboxy-terminal fragments are the substrate forthe g-secretase(s), which liberate amyloid-bpeptide from the b-secretase-cleaved fragment,and p3 from the a-secretase-cleaved fragment.(Asterisks indicate the approximate locations ofthe APP mutations.)

2 3 4 5 6 81

7

2 3 4 5 61 7 8

Proteolytic processing

Presenilin-1

Heterodimer

Wild-type PS1

APP

FAD PS1

b

PS1

Membrane

Lumen

Cytoplasm

-/-

a

Ab

AbAb

40

42

p3

AccumulatingC-terminalfragments

CytoplasmLumen

g-secretase

Increased Ab(1—42)production

Decreased Ab(1—40),

Ab(1—42) and p3 production

Required for physiologicalAb and p3production

20 7C in northern polar regions but only 5 7Cin the south.

Much remains for palaeoclimatologistsand palaeomodellers to do. We need longerdata sets with higher time resolution; careful(re-)calibration of proxy indicators; betterdating of records, and improved synchroniza-tion of records that are located far apart; newanalytical techniques; and, not least, coupledlow-order and three-dimensional climatemodels with well-tested parametrizations12.

Figure 1 outlines four areas of the globewhere important issues have to be resolved.The time is clearly ripe for another look atCLIMAP10, the last large-scale integrated pro-ject to tackle the LGM, one which incorpo-rates the vast amount of palaeoclimatologicaldata accumulated over the past 20 years.Moreover, the ocean interior requires contin-ued attention, and initiatives such as IMAGES(International Marine Global ChangeStudy)13, or the more encompassing PAGES(Past Global Changes)14, are invaluable in thisrespect. After all, research into Earth’s climatichistory is a cornerstone of building the better computer models that are so urgently neededto assess what the future may hold.

Thomas F. Stocker is in the Division of Climate andEnvironmental Physics, Physics Institute, Universityof Bern, Sidlerstrasse 5, Bern, Switzerland.e-mail: stocker@climate.unibe.ch1. Ganopolski, A., Rahmstorf, S., Petoukhov, V. & Claussen, M.

Nature 391, 351–356 (1998).

2. Berger, A. & Loutre, M.-F. Ambio 26, 32–37 (1997).

3. Marotzke, J., Welander, P. & Willebrand, J. Tellus 40A, 162–172

(1988).

4. Wright, D. G. & Stocker, T. F. J. Geophys. Res. 97, 12707–12730

(1992).

5. Johnsen, S. J., Dahl-Jensen, D., Dansgaard, W. & Gundestrup,

N. Tellus 47B, 624–629 (1995).

6. Stute, M. et al. Science 269, 379–383 (1995).

7. Patrick, A. & Thunell, R. C. Paleoceanography 12, 649–657 (1997).

8. Stocker, T. F., Wright, D. G. & Broecker, W. S. Paleoceanography

7, 529–541 (1992).

9. Duplessy, J.-C. et al. Paleoceanography 3, 343–360 (1988).

10.CLIMAP Project Members Tech. Rep. Map & Chart Ser. MC-36

(Geol. Soc. Am., Boulder, CO, 1981).

11.Webb, R. S. et al. Nature 385, 695–699 (1997).

12. Joussaume, S. & Taylor, K. E. in Proc. 1st Int. AMIP Sci. Conf.,

Monterey: World Clim. Res. Prog. Rep. 92, 425–430 (World

Meteorol. Org., Geneva, 1995).

13.http://www.images.cnrs-gif.fr

14.http://www.pages.unibe.ch

15.Weinelt, M. et al. Palaeoclimates 1, 283–309 (1996).

16.Duplessy, J.-C. et al. Oceanologica Acta 14, 311–324 (1991).

17.de Vernal, A. et al. in AGU Fall Mtg Abstr. (Am. Geophys. Un.,

Washington, DC, 1997).

18.Bard, E. & Rostek, F. in AGU Fall Mtg Abstr. (Am. Geophys.

Un., Washington, DC, 1997).

19.http://iri.ldgo.columbia.edu