Acknowledgements
We thank S. Naeem, J. T. Wootton, J. Bergelson, C. Pfister and M. Huston for comments on
the manuscript, and J. Chase and J. Shurin for help in the field. The paper is a Kellogg
Biological Station contribution, and was supported by grants from the National Science
Foundation, US Department of Education, and the University of Chicago.
Competing interests statement
The authors declare that they have no competing financial interests.
Correspondence and requests for materials and should be addressed to A.L.D.
(e-mail: aldownin@owu.edu).
..............................................................
Low host specificity of herbivorous
insects in a tropical forest
Vojtech Novotny*, Yves Basset, Scott E. Miller, George D. Weiblen§,
Birgitta Bremerk, Lukas Cizek* & Pavel Drozd{
* Institute of Entomology, Czech Academy of Sciences and Biological Faculty,
University of South Bohemia, Branisovska 31, 370 05 Ceske Budejovice, Czech
Republic
 Smithsonian Tropical Research Institute, Apartado 2072, Balboa, Ancon,
Panama
 Department of Systematic Biology, National Museum of Natural History,
Smithsonian Institution, Washington DC 20560-0105, USA
§ Department of Plant Biology, University of Minnesota, 220 Biological Sciences
Center, 1445 Gortner Avenue, St Paul, Minnesota 55108-1095, USA
k Bergius Foundation at the Royal Swedish Academy of Sciences, PO Box 50017,
SE-104 05 Stockholm, Sweden
{ University of Ostrava, Department of Biology, 30. dubna 22, 701 03 Ostrava,
Czech Republic
.............................................................................................................................................................................
Two decades of research1­4
have not established whether tropical
insect herbivores are dominated by specialists or generalists. This
impedes our understanding of species coexistence in diverse
rainforest communities. Host specificity and species richness of
tropical insects are also key parameters in mapping global
patterns of biodiversity1,4,5
. Here we analyse data for over 900
herbivorous species feeding on 51 plant species in New Guinea
and show that most herbivorous species feed on several closely
related plant species. Because species-rich genera are dominant
in tropical floras, monophagous herbivores are probably rare in
tropical forests. Furthermore, even between phylogenetically
distant hosts, herbivore communities typically shared a third of
their species. These results do not support the classical view that
the coexistence of herbivorous species in the tropics is a consequence
of finely divided plant resources; non-equilibrium
models of tropical diversity6
should instead be considered. Low
host specificity of tropical herbivores reduces global estimates of
arthropod diversity from 31 million (ref. 1) to 4­6 million
species. This finding agrees with estimates based on taxonomic
collections, reconciling an order of magnitude discrepancy
between extrapolations of global diversity based on ecological
samples of tropical communities with those based on sampling
regional faunas7,8
.
Host specificity is difficult to measure, and the limitations of
existing studies include sampling only certain taxonomic groups
rather than entire guilds, or sampling limited numbers of host plant
species and lineages. Studies are often of insufficient duration,
producing samples too small for quantitative analysis, or insects
are sampled destructively, which precludes feeding experiments and
the study of immature stages. Further, previous studies2
failed to
consider the phylogenetic relationships of host plants by using
measures of host specificity that relied on counts of higher plant
taxa (for example, genera or families). This approach can be
misleading when taxonomic ranks are not commensurate with
plant lineages. We examined the impacts of sampling bias and
phylogenetic effects on estimates of host specificity by analysing the
largest available data set of its kind. The leaf-chewing insect
community on 51 plant species was characterized by using a sample
Table 1 Overlap between leaf-chewing communities from closely and distantly
related host plants
Host plant Herbivores r So (mean ^ s.e.m.)
.............................................................................................................................................................................
Ficus spp. Coleoptera 20.182 0.51 (^0.008)
Lepidoptera 2 0.274 0.52 (^0.010)
Orthopteroids 2 0.267 0.48 (^0.019)
Total 2 0.370 0.51 (^0.007)
Psychotria spp. Coleoptera 0.58 (^0.020)
Lepidoptera 0.57 (^0.059)
Orthopteroids 0.54 (^0.076)
Total 0.57 (^0.029)
Plant genera Coleoptera 2 0.237 0.45 (^0.006)
Lepidoptera 2 0.328 0.09 (^0.005)
Orthopteroids 0.018 0.53 (^0.007)
Total 2 0.165 0.37 (^0.004)
.............................................................................................................................................................................
r, Spearman correlation between the phylogenetic distance of plants and the overlap of their
herbivore communities measured by the Sorensen index. Significant values (P , 0.05, Mantel
test) are in bold; data for Psychotria were too limited for calculation. So, average value of the
Sorensen index for all pairwise comparisons between communities from different hosts.
Figure 1 Phylogenetic relationships of host plants included in the study.
letters to nature
NATURE | VOL 416 | 25 APRIL2002 | www.nature.com 841 2002 Macmillan Magazines Ltd
of 50,734 herbivores comprising 935 species in a tropical lowland
forest in New Guinea.
An analysis of herbivore specificity compared the degree of faunal
overlap between communities feeding on closely and distantly
related plants (Fig. 1). Any two host species from within the genera
Ficus and Psychotria shared ,50% of their herbivorous species
(Fig. 2, Table 1). This overlap between herbivore communities
decreased gradually with increasing phylogenetic distance between
hosts. The same was true for each of the three major herbivorous
groups, including moths and butterflies (Lepidoptera), beetles
(Coleoptera) and grasshoppers and stick insects (orthopteroids).
Of 210 Ficus and Psychotria herbivores with known host range, only
one species of Coenobius (Chrysomelidae) feeding on Ficus nodosa
was monophagous. Large overlap among neotropical herbivore
communities on congeneric hosts was found also in Passiflora9
(but see ref. 10).
We postulate that monophagous herbivores are rare in tropical
forests because large genera tend to dominate tropical floras on all
spatial scales, and our data demonstrate that monophagy is scarce in
such genera. For example, Rubiaceae, the world's largest family of
predominantly tropical woody plants, includes 15 genera with at
least 100 species, representing 42% of 10,200 species in the family11
.
These genera also represent 50% of Rubiaceae in Costa Rica12
, 39%
in New Guinea13
and 36­58% in four neotropical forest commu-
nities14
. Furthermore, large genera feature prominently in most
tropical plant families11
.
The overlap between herbivore communities from phylogenetically
distant hosts was also high, although lower than in congeneric
hosts (Fig. 2, Table 1). Communities of Coleoptera and orthopteroids
maintained high overlap (more than 40%) even between
distantly related hosts. Most species of Coleoptera (96%) fed on
the foliage as adults only. The key determinant of the distribution
for these species is probably the host range of their poorly known
wood-boring or root-feeding larvae15
. The overlap between orthopteroid
communities did not even depend on phylogenetic distance
between hosts, as many species had broad host ranges spanning
multiple plant families. In contrast, communities of Lepidoptera
feeding on plants from different genera and families were highly
distinct, with a low overlap (less than 20%) that decreased steeply
towards zero with increasing phylogenetic distance between hosts.
The narrow host specificity of larval Lepidoptera in New Guinea is
consistent with data from the neotropics3,16
.
Large (speciose) plant genera hosted more distinct leaf-chewing
communities than small genera, as indicated by a negative correlation
(Spearman r  20.38, P , 0.05, n  29, Independent Contrasts)
between the size of the plant genus in New Guinea13
and the
average overlap (So) of its herbivore community with communities
feeding on other genera. The overall relationship was due to
Coleoptera (r  20.45, P , 0.05) and Lepidoptera (r  20.39,
P , 0.05); no correlation existed in the orthopteroids (P . 0.5).
These findings contradict the hypothesis that taxonomically isolated
plants from small genera support particularly distinct herbivore
communities. We suggest that herbivores specialize more often on
large plant genera than small ones because the former represent a
large resource base, particularly because many herbivores have
broad congeneric host ranges.
Our study showed that most herbivores feed on several closely
related congeneric plant species. Many herbivores have even wider
host ranges, as communities from allogeneric hosts shared on
average 37% of all herbivorous species (Table 1). These results are
at variance with the notion of extremely specialized interactions
between herbivores and finely divided plant resources that provide
distinct ecological niches and permit the coexistence of numerous
herbivorous species in diverse tropical forests. Our findings suggest
that alternative, non-equilibrium models of tropical diversity6
should be seriously considered.
There were on average Sav  32.9 ^ 1.7 species of Coleoptera,
26.1 ^ 2.4 species of Lepidoptera and 20.6 ^ 0.9 species of orthopteroids
feeding on a single host tree species. Their effective special-
ization17
was respectively FT  0.24 for Coleoptera, 0.51 for
Lepidoptera and 0.14 for orthopteroids (Fig. 3) so that there were
SavFT  7:9 unique (effectively specialized) species of Coleoptera,
13.3 of Lepidoptera and 2.9 of orthopteroids per host tree species.
These results are similar to the two to five unique species of
Figure 2 Overlap in species composition of their leaf-chewing communities and
phylogenetic distance between Ficus spp. (a), Psychotria spp. (b) and plants from different
genera (c). Values of Sorensen index So (means ^ s.e.m.) are shown for all pairwise
comparisons between hosts separated by a particular phylogenetic distance. Coleoptera
(squares), Lepidoptera (circles) and orthopteroids (triangles) are shown separately.
Figure 3 Dependence of effective specialization (FT ) on the number of host tree species
studied. Samples from individual tree species, each from a different genus, were
amalgamated in randomized sequence; the mean from 100 random sequences is given.
Coleoptera (squares), Lepidoptera (circles), orthopteroids (triangles) and butterflies
(crosses) are shown separately.
letters to nature
NATURE | VOL 416 | 25 APRIL2002 | www.nature.com842  2002 Macmillan Magazines Ltd
Coleoptera per host tree reported from other tropical forests4,18
but
at variance with the 136 unique species postulated by Erwin1
. Our
estimates of effective specialization are robust, as they changed little
with increasing number of tree species included in the analysis (Fig.
3). In contrast, Sav values were underestimated, as the species
accumulation curves for individual tree species did not approach
an asymptote, indicating that the total species richness of their
herbivore assemblages had not been sampled.
These host specificity data, together with improved estimates of
some other parameters4
, permitted a revision of Erwin's1
estimate of
arthropod diversity from 31.0 to 4.9 million species (Table 2). Erwin
extrapolated the species richness of a herbivore community from a
single tree species to the total number of tropical tree species. We
refined Erwin's method in recognition of the fact that plant genera,
rather than species, are more appropriate units for estimating the
diversity of tropical herbivore communities. The average number of
herbivorous species per host species was multiplied by the 1,872
plant genera known from New Guinea13
to obtain an estimate of
arthropod species in New Guinea. The New Guinean diversity was
extrapolated to global, using an estimate19
that ,5% of the world
diversity in several well-known groups of organisms, including
flowering plants, various invertebrates and vertebrates, occur in
New Guinea. The total arthropod diversity was estimated at 3.7
million and 5.9 million species from data on Coleoptera and
Lepidoptera, respectively (Table 2). This method was verified by
using butterflies, a well-known taxon with 959 described species in
New Guinea20
and 15,000­20,000 worldwide5
. In our samples,
butterflies had Sav  0.63 species per host tree species and effective
specialization FT  1.0, leading to an estimate of 1,179 species in
New Guinea and 23,500 worldwide.
Our estimates of global diversity are similar to the 6.6 million
(ref. 18) and 4.8 million (ref. 4) species of arthropods extrapolated
from other communities of tropical beetles. Even more importantly,
they are also in agreement with estimates of ,5 million species
based on the analysis of regional faunas and the evaluation of
museum collections by taxonomists5,7,8,21
, reconciling the disparate
results of ecological and taxonomic approaches to the estimation of
global species diversity. A
Methods
Insects
Leaf-chewing insects feeding on 51 woody species (Fig. 1 and the following species of
Euphorbiaceae s.l.: Breynia cernua, Endospermum labios, Homalanthus novoguineensis,
Mallotus mollissimus, Melanolepis multiglandulosa and Pimelodendron amboinicum) from
all major lineages of flowering plants and all stages of rainforest secondary succession22
were studied at four study sites in a diverse lowland rainforest (.150 woody species per
hectare) near Madang in Papua New Guinea in 1994­2000 (ref. 23). Insects were collected
by hand from foliage of each plant species for at least 1 year. Collecting effort was 1,500 m2
of foliage sampled and .1,000 tree inspections (a particular tree sampled at a particular
time) per species. This sampling effort represented ,1,000 person-days of fieldwork. Each
insect was provided with leaves of the plant species from which it was collected and only
those that fed were retained in the analyses. Larvae were reared to adults whenever
possible. All insects were assigned to morphospecies and later verified and identified by
specialists as far as possible (http://www.nmnh.si.edu/new_guinea/). Vouchers are
deposited in the Bishop Museum, Honolulu, and the Smithsonian Institution,
Washington.
Feeding
All feeding records limited to a single individual on a particular host species were excluded
as poorly supported. Monophagous species were defined as those represented by at least 10
individuals, all feeding on a single plant species. Faunal overlap was examined in 45 host
taxa with known phylogenetic relationships. Overlap was measured by the Sorensen
coefficient So  2a=2a  b  c; where a is the number of species common to the two
compared samples, and b and c are the numbers of species present only in the first and
second samples respectively. So quantifies the proportion of species in a sample that also
occurs in the other sample, averaged over the two samples.
The average proportion of herbivorous species feeding on a particular host plant that
was unique (effectively specialized17
) to that plant was estimated as FT  ST/HT, that is, as
the ratio of the total number of herbivorous species found on all T hosts studied (ST),
divided by the number of host-plant records involving these hosts (HT). The effective
specialization and species richness of herbivores were estimated from communities
feeding on 36 tree species, each representing a different genus.
Phylogenetic relationships
Phylogenetic distance between plants was estimated as the number of nodes in the
phylogeny between these plant species. Separate correlation analyses of 13 Ficus spp., 4
Psychotria spp. and 30 plants from different genera were performed. Ficus wassa and
Psychotria micralabastra represented their respective genera in the latter analysis.
COMPARE 4.4 software was used to calculate independent contrasts.
Phylogenetic relationships of plant species were obtained from refs 24­28. All
Rubiaceae species were sequenced for large subunit of ribulose-1,5-bisphosphate
carboxylase/oxygenase gene (rbcL) and 30S ribosomal protein S16 gene (rps16). The data
were analysed under parsimony with PAUP, version 4.0d64* (D. Swofford, unpublished).
All relationships were highly supported and in agreement with earlier analyses of the
family29
. Accession numbers from the EMBL/GenBank data bases for the rps16 intron and
the rbcL gene, respectively, follow (see Fig. 1 for full plant names): G. hansemannii,
AJ320077, AJ318446; M. citrifolia chromoplast, AJ320078, AJ318448; M. scratchleyi,
AJ320079, AJ318447; N. orientalis, AJ320080, AJ318449; N. clemensiae, AJ320081,
AJ318450; P. platyclada, AJ320082, AJ318451; P. leptothyrsa, AJ320083, AJ318452; P.
micralabastra, AJ320084, AJ318453; P. micrococca, AJ320085, AJ318454; P. ramuensis,
AJ320086, AJ318455; R. schumanniana, AJ320087, AJ318456; T. buruensis, AJ320088,
AJ318457; T. timon, AJ320089, AJ318458; V. cauliflora, AJ320090, AJ318459.
Received 6 September 2001; accepted 12 January 2002.
1. Erwin, T. L. Tropical forests: their richness in Coleoptera and other species. Coleopterist's Bull. 36,
74­75 (1982).
2. Basset, Y. Host specificity of arboreal and free-living insect herbivores in rain forests. Biol. J. Linn. Soc.
47, 115­133 (1992).
3. Barone, J. A. Host-specificity of folivorous insects in a moist tropical forest. J. Anim. Ecol. 67, 400­409
(1998).
4. degaard, F. How many species of arthropods? Erwin's estimate revised. Biol. J. Linn. Soc. 71, 583­597
(2000).
5. Stork, N. E. How many species are there? Biodivers. Conserv. 2, 215­232 (1993).
6. Hubbell, S. P. The Unified Neutral Theory of Biodiversity and Biogeography (Princeton Univ. Press,
Princeton, 2001).
7. May, R. M. in Nature and Human Society: The Quest for a Sustainable World (eds Raven, P. H. &
Williams, T.) 30­45 (National Academy Press, Washington DC, 2000).
8. Nielsen, E. S. & Mound, L. A. in Nature and Human Society: The Quest for a Sustainable World (eds
Raven, P. H. & Williams, T.) 312­322 (National Academy Press, Washington DC, 2000).
9. Thomas, C. D. Herbivore diets, herbivore colonization, and the escape hypothesis. Ecology 71,
610­615 (1990).
10. Marquis, R. J. in Plant­Animal Interactions: Evolutionary Ecology in Tropical and Temperate Regions
(eds Price, P. W., Lewinsohn, T. M., Fernandes, G. W. & Benson, W. W.) 179­208 (Wiley, London,
(1991).
11. Mabberley, D. J. The Plant Book (Cambridge Univ. Press, Cambridge, 1987).
12. Burger, W. & Taylor, C. M. Flora Costaricensis: Family #202 Rubiaceae. Fieldiana Botany, N.S. 33,
1­333 (1993).
Table 2 Estimates of the global number of arthropod species
Variable Coleoptera Lepidoptera
1 2 3
...................................................................................................................................................................................................................................................................................................................................................................
A. No. of herbivore species per tree species 682* 32.9 32.9 30.8{
B. Effective specialization 0.20* 0.24 0.24 0.51
C. Correction for non-herbivorous species 1.20* 1.20* 1.20* 1.00
D. Proportion of species from arthropods 0.40* 0.23 0.23 0.10#
E. Proportion of canopy fauna from total 0.66* 0.42 0.42 1.00
F. No. of tropical tree species 50,000* 50,000*
G. No. of plant genera in New Guinea 1,872§ 1,872§
H. Proportion of New Guinea species from total 0.05k 0.05k
Global no. of arthropod species (A  B  C  F)/(D  E) 31,000,000 4,904,348
Global no. of arthropod spp. (A  B  C  G)/(D  E  H) 3,672,376 5,881,075
...................................................................................................................................................................................................................................................................................................................................................................
Sources: *, from ref. 1; , from this study; , from ref. 4; §, from ref. 13; k, from ref. 19; {, our estimate of 26.1 externally feeding leaf-chewer species multiplied by 1.18 to account for herbivorous species from
other guilds, based on ref. 30; #, from ref. 8. Coleoptera 1, estimate from ref. 1. Note that the Coleoptera 2 and Lepidoptera estimates are mutually independent because they are based on two entirely
different sets of parameters.
letters to nature
NATURE | VOL 416 | 25 APRIL2002 | www.nature.com 843 2002 Macmillan Magazines Ltd
13. Ho¨ft, R. Plants of New Guinea and the Solomon Islands. Dictionary of the Genera and Families of
Flowering Plants and Ferns (Wau Ecology Institute, Wau, 1992).
14. Gentry, A. H. (ed.) Four Neotropical Forests (Yale Univ. Press, New Haven, 1990).
15. Jolivet, P. & Hawkeswood, T. J. Host-Plants of Chrysomelidae of the World (Backhuys, Leiden,
1995).
16. Janzen, D. H. Ecological characterization of a Costa Rican dry forest caterpillar fauna. Biotropica 20,
120­135 (1988).
17. May, R. M. How many species? Phil. Trans. R. Soc. Lond. B 330, 293­304 (1990).
18. Basset, Y., Samuelson, G. A., Allison, A. & Miller, S. E. How many species of host-specific insects feed
on a species of tropical tree? Biol. J. Linn. Soc. 59, 201­216 (1996).
19. Sekhran, N. & Miller, S. E. (eds) Papua New Guinea Country Study on Biological Diversity (Papua New
Guinea Dept of Environment & Conservation, Waigani, Papua New Guinea, 1996).
20. Parsons, M. The Butterflies of Papua New Guinea (Academic, London, 1999).
21. Miller, S. E., Novotny, V. & Basset, Y. in Foundations of Tropical Biology: Key Papers and Commentaries
(eds Chazdon, R. L. & Whitmore, T. C.) (Univ. Chicago Press, Chicago, in the press).
22. Leps, J., Novotny, V. & Basset, Y. Habitat and successional optimum of plants and the composition of
their leaf-chewing herbivores in Papua New Guinea. J. Ecol. 89, 186­199 (2001).
23. Basset, Y., Novotny, V., Miller, S. E. & Pyle, R. Quantifying biodiversity: experience with
parataxonomists and digital photography in Papua New Guinea and Guyana. BioScience 50, 899­908
(2000).
24. Angiosperm Phylogeny Group An ordinal classification for the families of flowering plants. Ann.
Missouri Bot. Gard. 85, 531­553 (1998).
25. Weiblen, G. D. Phylogenetic relationships of functionally dioecious Ficus (Moraceae) based on
ribosomal DNA sequences and morphology. Am. J. Bot. 87, 1342­1357 (2000).
26. Backlund, M., Oxelman, B. & Bremer, B. Phylogenetic relationships within the Gentianales based on
ndhF and rbcL sequences, with particular reference to the Loganiaceae. Am. J. Bot. 87, 1029­1043
(2000).
27. Soltis, D. E. et al. Angiosperm phylogeny inferred from 18S rDNA, rbcL, and atpB sequences. Bot.
J. Linn. Soc. 133, 381­461 (2000).
28. Qiu, Y.-L. et al. The earliest angiosperms. Evidence from mitochondrial, plastid and nuclear genomes.
Nature 402, 404­407 (1999).
29. Bremer, B. et al. More characters or more taxa for a robust phylogeny--case study from the coffee
family (Rubiaceae). Syst. Biol. 48, 413­435 (1999).
30. Powell, J. A., Mitter, C. & Farrell, B. in Lepidoptera, Moths and Butterflies Vol. 1: Evolution, Systematics,
and Biogeography (ed. Kristensen, N. P.) 403­422 (Walter de Gruyter, Berlin, 1998).
Acknowledgements
We thank parataxonomists J. Auga, W. Boen, C. Dal, S. Hiuk, B. Isua, M. Kasbal, R. Kutil,
M. Manumbor and K. Molem and also K. Darrow and N. Heidari for assistance; many
insect collectors and taxonomists are acknowledged elsewhere. N. Stork, T. Lewinsohn,
J. Zrzavy, J. Leps and A. Stewart commented on the manuscript. This work was supported
by the National Science Foundation (USA), Christensen Fund (USA), Grant Agency of the
Czech Republic, Czech Academy of Sciences, the Swedish Natural Science Research
Council, Czech Ministry of Education, Otto Kinne Foundation, Darwin Initiative (UK),
International Centre of Insect Physiology and Ecology (ICIPE) and Bishop Museum.
Competing interests statement
The authors declare that they have no competing financial interests.
Correspondence and requests for materials should be addressed to V.N.
(e-mail: novotny@entu.cas.cz).
..............................................................
Developmental constraints versus
flexibility in morphological evolution
Patri´cia Beldade, Kees Koops & Paul M. Brakefield
Institute of Evolutionary and Ecological Sciences, Leiden University,
PO Box 9516, 2300 RA Leiden, The Netherlands
.............................................................................................................................................................................
Evolutionary developmental biology has encouraged a change of
research emphasis from the sorting of phenotypic variation by
natural selection to the production of that variation through
development1
. Some morphologies are more readily generated
than others, and developmental mechanisms can limit or channel
evolutionary change2
. Such biases determine how readily populations
are able to respond to selection3
, and have been postulated
to explain stasis in morphological evolution4
and unexplored
morphologies5
. There has been much discussion about evolutionary
constraints6­8
but empirical data testing them directly are
sparse9,10
. The spectacular diversity in butterfly wing patterns11
is
suggestive of how little constrained morphological evolution can
be. However, for wing patterns involving serial repeats of the
same element, developmental properties suggest that some directions
of evolutionary change might be restricted12,13
. Here we
show that despite the developmental coupling between different
eyespots in the butterfly Bicyclus anynana, there is great potential
for independent changes. This flexibility is consistent with the
diversity of wing patterns across species and argues for a
dominant role of natural selection, rather than internal constraints,
in shaping existing variation.
Figure 1 Response to artificial selection on the size of the dorsal forewing eyespots of
B. anynana. a, Eyespot diameter/wing size relative to unselected control values are given
for the different directions of selection. AP and ap are coupled directions; Ap and aP are
uncoupling directions. Each point represents the mean (^standard error) for the two
replicate lines for each generation. Solid lines join points covering the first 11 consecutive
generations, all starting from the same original population (centre of graphic, G0). Broken
lines join the points for G11 and G25 phenotypes. b, Enlargement of the central area of a
showing the behaviour of individual replicate lines (filled and open symbols). c, Typical
dorsal surface of forewing of unselected female showing the anterior (A) and posterior (P)
eyespots. d, Representative G25 phenotypes (most ap females have no eyespots and
many AP females have extra, satellite eyespots). Responses in males were comparable to
those in females.
letters to nature
NATURE | VOL 416 | 25 APRIL2002 | www.nature.com844  2002 Macmillan Magazines Ltd