unexplained, time-series methods result inexcessively wide prediction intervals whenused for long-term forecasting.

Judgemental methods can be used to correct or constrain such broad predictionintervals. Expert judgement is also usedwhen expected values and correspondingprediction intervals are hard to obtain byformal methods. In demographic forecast-ing, the expert-judgement method has beenpioneered by Lutz and colleagues5,6,10. Agroup of experts is asked to indicate theprobability that a key parameter (such as theaverage number of children per woman,or life expectancy) in some future year fallswithin a certain prespecified range. A weak-ness of this approach is that experts, oftenbeing unduly confident, tend to give over-optimistically narrow prediction intervals11.When the forecasts are later compared withactual data, the intervals turn out to fit theobserved trends much less frequently thanthe probabilities suggested.

Finally,empirical errors observed for pastforecasts may be extrapolated to predict theexpected errors for the current forecast. Forthe NRC report9, this method was includedin the statistical model to calculate the uncertainty of United Nations populationforecasts for every country. A problem here is that forecasts prepared in the 1960s or earlier were poorly documented, so data onhistorical errors do not stretch back as far asone would like.

Elements of the three methods are often used in combination. For instance,time-series methods involve some degree of subjectivity, perhaps in choosing theextrapolation model or the length of the historical data series. These decisions maystrongly influence the prediction intervals.And the intervals, whether obtained by time-series methods or expert opinion, are frequently checked against historical error patterns4,7. Lutz et al.1 combined expertjudgement with time-series models.

Irrespective of the method used, proba-bilistic forecasts of the youngest and oldestage groups show the most uncertainty,because fertility and mortality are hard topredict. Short-term forecasts of the demog-raphy of developing countries are also sub-ject to great uncertainty, because of the poorquality of their demographic data. Finally,prediction intervals are always narrowerwhen variables are aggregated (for example,when the populations of all countries areadded together) rather than looked at indi-vidually, because the errors tend to canceleach other out. These kinds of uncertaintyassessments are crucial, and statistical agencies would do a great service to users offorecasts if they would adopt probabilisticmethods rather than more traditionalmethods that do not take uncertainty intoaccount.Yet those making use of populationforecasts should also keep in mind that

probabilistic statements themselves areuncertain to some extent. ■

Nico Keilman is in the Department of Economics,University of Oslo, Blindern, N-0317 Oslo, Norway.e-mail: nico.keilman@econ.uio.no1. Lutz, W., Sanderson, W. & Scherbov, S. Nature 412, 543–546

(2001).

2. Horiuchi, S. Nature 405, 744–745 (2000).

3. Keilman, N. J. Official Stat. 13, 245–277 (1997).

4. Lee, R. & Tuljapurkar, S. J. Am. Stat. Assoc. 89, 1175–1189

(1994).

5. Lutz, W. & Scherbov, S. Eur. J. Pop. 14, 1–17 (1998).6. Lutz, W. & Scherbov, S. Z. Bevölkerungswissenschaft 23, 83–109

(1998).7. Alho, J. A Stochastic Forecast of the Population of Finland

Reviews 1998/4 (Statistics Finland, Helsinki, 1998).8. De Beer, J. & Alders, M. Working Paper no. 45

http://www.unece.org/stats/documents/1999.05.projections.html9. National Research Council Beyond Six Billion: Forecasting the

World’s Population (Nat. Acad. Press, Washington DC, 2000).10.Hanika, A., Lutz, W. & Scherbov, S. Statist. Nachrichten 12,

984–988 (1997).11.Armstrong, J. From Crystal Ball to Computer 2nd edn (Wiley,

New York, 1985).

news and views

NATURE | VOL 412 | 2 AUGUST 2001 | www.nature.com 491

The science of manipulating matter onthe nanometre scale — nanofabrication— has provided researchers with a

remarkably diverse set of tools for probingthe behaviour of biological structures. At arecent meeting* on the “Biological Appli-cations of Nanotechnology”, scientists at the

forefront of this research showed how thesetools are already being used to sequencegenomes, diagnose disease and watch singlebiomolecules perform their natural func-tions in real time. But there were also a fewhints of a fresh theme emerging from all thiswork: the design of artificial nanostructuresthat can interact with and replace natural,biological materials. Studies that aim at

Nanotechnology

Boning up on biologyT. Andrew Taton

Attempts to tailor nanometre-scale objects to mimic and interact withnatural materials raise the question of how to predict the biologicalresponse to these tiny creations.

500 nm

100 µm

10 nm

Bone-marrowstem cell

Bone-formingosteoblastcell

Collagentriple helix

Hydroxyapatitecrystals

Extracellularbone matrix

Figure 1 The many scales of organization in natural bone. Collagen triple helices spontaneously formnanoscale bundles of protein, which act as a template for the crystallization of hydroxyapatitenanocrystals. The collagen matrix is also recognized by undifferentiated bone-marrow stem cells,which become bone-forming osteoblasts after signals from bone-specific proteins in the matrix.To duplicate the extraordinary durability and strength of natural bone, researchers must designsynthetic materials that match these structural levels of organization from the nanoscale upwards.(Note that this figure shows the dimensional, not sequential, principles of bone construction.)

*American Chemical Society ProSpectives Conference, Berkeley,

California, USA, 3–6 June 2001.

© 2001 Macmillan Magazines Ltd

replicating the complex nanoscale structureof cell membranes, tissue and bone are lead-ing to promising materials with potentialuses as implants and therapies. At the sametime, the work has the potential to illustratehow cells interact with nanometre-sizedobjects in their own world.

Bone is one example of a natural materialwhose properties depend intimately on itsnanoscale structure. Bone is an inorganic–bioorganic composite material consistingmainly of collagen proteins and hydroxy-apatite (a crystalline form of calcium phos-phate). Collagen spontaneously forms fibrilsof aligned protein helices1, on which tinyhydroxyapatite crystals (10–50 nanometresin length) can grow2 (Fig.1).Both the size andthe orientation of the crystals are dictatedspecifically by the collagen template, and theprecise structural relationship between thecollagen and hydroxyapatite is critical tobone’s resilience and strength.

A group at Northwestern University hasbeen trying to design self-assembling, syn-thetic substitutes for collagen, which can also act as templates for hydroxyapatite crys-tallization (S. Stupp, Northwestern Univ.,Evanston, Illinois).Stupp demonstrated thatamphiphilic molecules — bearing a longhydrophobic alkyl group on one end and ahydrophilic peptide on the other — sponta-neously assemble into cylindrical structuresthat resemble collagen fibrils. More impor-tantly, these cylinders guide the formation ofhydroxyapatite crystallites with orientationsand sizes similar to those in natural bone. Itremains to be seen whether Stupp’s compos-ite materials replicate the mechanical prop-erties of bone, and whether the nanoscaleorder in these materials translates intomacroscale toughness in a prosthetic bonereplacement. Nevertheless, the work repre-sents a step towards synthetically matchingthe complex,hierarchical structure of bone.

Stupp’s work also raises a question in thecontext of biologically orientated nanotech-nology research.Can we synthesize materialsthat not only replicate the properties foundin nature but also prompt biological sys-tems to build on these materials? In naturalbone formation, for example, bone-specificproteins3,4 and possibly some types ofcollagen5 have peptide sequences that signalbone-marrow stem cells to differentiate into bone-forming osteoblast cells (Fig. 1).The osteoblasts then supply the calciumrequired to form the crystalline hydroxy-apatite. The initial results reported by Stupp were obtained with amphiphilic mol-ecules terminated by Arg-Gly-Asp (RGD),a peptide sequence known to be a general promoter of cell adhesion and growth6,7.However, these molecules cannot promotethe differentiation of marrow stem cells.

The next challenge in this research will beto synthesize amphiphilic assemblies bear-ing other peptide sequences that specifically

bind marrow cells and trigger osteoblast dif-ferentiation.Once again, the nanoscale orga-nization of these signalling sequences will becrucial to this effort; it has already beenshown that the presence of the sequencesalone is not enough to promote osteoblastdifferentiation or bone formation8. Thetechnological goal of this research would beto replace solid bone grafts with an organicscaffold that would foster the growth of new,natural bone. In theory, an appropriatelydesigned material could achieve this only by organizing the inorganic component ofbone and by communicating specific mes-sages to bone-forming cells.

Unfortunately, several other speakers atthe conference stressed just how difficult itcan be to design nanostructured materials to interact specifically with cell surfaces. Forexample, merely attaching peptides to nano-scale materials does not completely dictatetheir interactions with cellular membranereceptors; other properties, such as generalbiocompatibility, solubility and nonspecificinteractions with cell membranes, have to be correctly engineered as well. These issueshave made it particularly challenging tocoerce cells to specifically recognize andingest inorganic nanoparticles known asquantum dots (M. Bruchez, Quantum DotCorp., Hayward, California; W. Parak andA.Alivisatos,Univ.California,Berkeley).

In the world of organic dendrimers —highly branched polymers that form spheri-cal shapes in solution — even small varia-tions in the size of dendrimer particles, overthe range 1–10 nm, can have significanteffects on the ability of these materials tocarry molecules across cell membranes(D. Tomalia, Univ. Michigan). Because cellsinteract with many nanoscale objects in theirown world (such as vesicles and viruses),attempts to engineer unnatural nanostruc-tures such as quantum dots and dendrimerspromise to reveal more about why naturalnanostructures have the properties they do. Admittedly, there is still some way to gobefore nanotechnology and biology marchhand in hand. Nonetheless, the science ofsynthesizing nanostructured materials thatspecifically recognize biological materials is advancing quickly, and points towards a future when nature will welcome ournanoscale inventions. ■

T. Andrew Taton is in the Department of Chemistry,University of Minnesota, Minneapolis, Minnesota55455, USA.e-mail: taton@chem.umn.edu1. Holmgren, S. K., Taylor, K. M., Bretscher, L. E. & Raines, R. T.

Nature 392, 666–667 (1998).

2. Weiner, S. & Wagner, H. D. Annu. Rev. Mat. Sci. 28, 271–298 (1998).

3. Niklason, L. E. Nature Biotechnol. 18, 929–930 (2000).

4. Bruder, S. P., Fink, D. J. & Caplan, A. I. J. Cell. Biochem. 56,

283–294 (1994).

5. Mizuno, M., Fujisawa, R. & Kuboki, Y. J. Cell Physiol. 184,

207–213 (2000).

6. Pierschbacher, M. D. & Ruoslahti, E. Nature 309, 30–33 (1984).

7. Roberts, C. et al. J. Am. Chem. Soc. 120, 6548–6555 (1998).

8. Staatz, W. D. et al. J. Biol. Chem. 206, 7363–7367 (1991).

news and views

492 NATURE | VOL 412 | 2 AUGUST 2001 | www.nature.com

100 YEARS AGOBoomerangs may be studied for theiranthropological interest as examples ofprimitive art, or for the manner in which theyillustrate dynamical principles. But there isextraordinary fascination in making andthrowing them, and in watching theremarkable and always graceful curvesdescribed in their flight; accordingly mychief object in the following paper has beento diminish the practical difficulties of thesubject by giving some of the results of tenyears’ experimental acquaintance with it…It is rather difficult to give sufficient spin tokeep the motion stable through a long flight,and I have found it advantageous to windround the wood about 60 grammes weight of copper wire in three equal portions, ofwhich one is near the middle and one neareach end. This materially increases themoment of inertia about the centre of gravitywithout interfering seriously with otherdetails. I have thrown a loaded boomerang ofthis type 167 metres, and my range with a spherical ball of half the weight is only 63 metres.From Nature 1 August 1901.

50 YEARS AGOThe removal of the nucleus causes a slowdecrease of the ribonucleic acid content ofthe cytoplasm of Amoeba proteus. Thisindicates that the ribonucleo-proteinparticles (microsomes) subsist in thecytoplasm only in the presence of thenucleus. However, this nuclear control is notimmediate, for the ribonucleic acid contentof non-nucleated fragments dropssignificantly only after the fourth dayfollowing enucleation. It appeared of interestto find out whether the oxygen uptake isunder nuclear control; a rapid decrease ofrespiration after enucleation, as stated byClark, might mean that respiratory enzymes,known to be predominantly bound tomitochondria, are dependent on the nucleusfor the maintenance of their activity… One is therefore led to believe that enucleationleads to a break in the normal couplingbetween oxidations and phosphorylations:the non-nucleated amoeba would thusbehave like cells treated with dinitrophenol,which interrupts synthesis by blocking thecoupling between oxidation andphosphorylation. Work is now in progress tofind out if the nucleus, and in particular thenucleolus, takes any part in the synthesis ofthe coenzymes necessary for this coupling.From Nature 4 August 1951.

© 2001 Macmillan Magazines Ltd