Report. Species coextinctions and the biodiversity crisis

Abstract

To assess the coextinction of species (the loss of a species upon the loss of another), we present a probabilistic model,
scaled with empirical data. The model examines the relationship between coextinction levels (proportion of species extinct)
of affiliates and their hosts across a wide range of coevolved interspecific systems: pollinating Ficus wasps and Ficus, parasites and their hosts, butterflies and their larval host plants, and ant butterflies and their host ants. Applying a
nomographic method based on mean host specificity (number of host species per affiliate species), we estimate that 6300 affiliate
species are “coendangered” with host species currently listed as endangered. Current extinction estimates need to be recalibrated
by taking species coextinctions into account.

Full text

DOI: 10.1126/science.1101101
, 1632 (2004); 305Science
et al.Lian Pin Koh,
Species Coextinctions and the Biodiversity Crisis
www.sciencemag.org (this information is current as of September 3, 2008 ):
The following resources related to this article are available online at
http://www.sciencemag.org/cgi/content/full/305/5690/1632
version of this article at:
including high-resolution figures, can be found in the onlineUpdated information and services,
http://www.sciencemag.org/cgi/content/full/305/5690/1632/DC1
can be found at: Supporting Online Material
http://www.sciencemag.org/cgi/content/full/305/5690/1632#otherarticles
, 3 of which can be accessed for free: cites 13 articlesThis article
35 article(s) on the ISI Web of Science. cited byThis article has been
http://www.sciencemag.org/cgi/content/full/305/5690/1632#otherarticles
4 articles hosted by HighWire Press; see: cited byThis article has been
http://www.sciencemag.org/cgi/collection/ecology
Ecology
: subject collectionsThis article appears in the following
http://www.sciencemag.org/about/permissions.dtl
in whole or in part can be found at: this article
permission to reproduce of this article or about obtaining reprintsInformation about obtaining
registered trademark of AAAS.
is aScience2004 by the American Association for the Advancement of Science; all rights reserved. The title
CopyrightAmerican Association for the Advancement of Science, 1200 New York Avenue NW, Washington, DC 20005.
(print ISSN 0036-8075; online ISSN 1095-9203) is published weekly, except the last week in December, by theScience
on September 3, 2008 www.sciencemag.orgDownloaded from

Species Coextinctions and the
Biodiversity Crisis
Lian Pin Koh,
1
*† Robert R. Dunn,
2
*‡ Navjot S. Sodhi,
1
§
Robert K. Colwell,
3
Heather C. Proctor,
4
Vincent S. Smith
5
㛳
To assess the coextinction of species (the loss of a species upon the loss of
another), we present a probabilistic model, scaled with empirical data. The
model examines the relationship between coextinction levels (proportion of
species extinct) of affiliates and their hosts across a wide range of coevolved
interspecific systems: pollinating Ficus wasps and Ficus, parasites and their
hosts, butterflies and their larval host plants, and ant butterflies and their host
ants. Applying a nomographic method based on mean host specificity (num-
ber of host species per affiliate species), we estimate that 6300 affiliate
species are “coendangered” with host species currently listed as endangered.
Current extinction estimates need to be recalibrated by taking species
coextinctions into account.
Rapid population declines and extinctions of
species following the widespread destruction
of natural habitats have been reported across
the natural world (1). Up to 50% of species
are predicted to be lost in the next 50 years (2,
3). This seemingly inevitable biodiversity cri-
sis has galvanized the study of population and
species extinctions (4). However, while in-
vestigations have focused on the pathology of
independent taxon-based extinctions, the pos-
sible cascading effects of species loss, while
acknowledged (5–7), have not been estimat-
ed quantitatively for extinct or endangered
taxa. Such a view underestimates the intricate
processes of species extinctions, especially in
complex ecosystems such as tropical rainfor-
ests, where many species obligately depend
on one another.
The term “coextinction” was first used to
describe the process of the loss of parasitic
insects with the loss of their hosts (5). The
concept has been expanded to describe the de-
mise of a broader array of interacting species,
including predators with their prey (6) and spe-
cialist herbivores with their host plants (7).
Here, we define coextinction as the loss of a
species (the affiliate) upon the loss of another
(the host). The most often cited example is that
of the extinct passenger pigeon (Ectopistes mi-
gratorius) and its parasitic louse (Columbicola
extinctus)(5), although the latter has been
shown to be alive and well on other hosts (8, 9).
More recently, the loss of tropical butterfly
species from Singapore was attributed to the
loss of their specific larval host plants (7).
Here, we apply a simple probabilistic
model to empirical “affiliation matrices”
(host by affiliate presence/absence matrices)
to examine the relationship between affiliate
and host extinctions across a range of co-
evolved interspecific systems: pollinating Fi-
cus wasps and Ficus, primate parasites
(Pneumocystis fungi, nematodes, and lice)
and their hosts, parasitic mites and lice and
their avian hosts, butterflies and their larval
host plants, and ant butterflies and their host
ants. The model estimates the number of
affiliate extinctions as a function of the num-
ber of host extinctions, assuming a random
order of host extinction (10). Figure 1 shows
the predicted coextinction curves for eight
relatively well studied affiliate-host systems.
The coextinction curve is linear for affiliate-
host systems in which each affiliate species
was associated with only one host, such as
Pneumocystis fungi and their primate hosts,
and curvilinear for systems in which at least
some affiliate species have multiple hosts,
such as butterflies and their larval host plants.
This probabilistic model relies on fully
specified affiliation matrices and thus is not
useful for estimating expected numbers of
affiliate extinctions in affiliate-host systems
for which host specificity distributions are
unavailable. To estimate coextinction levels
for these affiliate-host systems, we have de-
veloped a nomographic model of affiliate
extinctions that expresses affiliate extinction
1
Department of Biological Sciences, National Univer-
sity of Singapore, 14 Science Drive 4, Singapore
117543.
2
Department of Environmental Biology, Cur-
tin University of Technology, GPO Box U1987 Perth,
Western Australia 6845.
3
Department of Ecology and
Evolutionary Biology, University of Connecticut,
Storrs, CT 06269 –3043, USA.
4
Department of Biolog-
ical Sciences, University of Alberta, Edmonton, Al-
berta T6G 2E9, Canada.
5
Institute of Biomedical and
Life Sciences, University of Glasgow, Glasgow, G12
8QQ, United Kingdom.
*These authors contributed equally to this work.
†Present address: Department of Ecology and Evolu-
tionary Biology, Princeton University, Princeton, NJ
08544–1003, USA.
‡Present address: Department of Ecology and Evolu-
tionary Biology, University of Tennessee, 1416 Circle
Drive, 569 Dabney Hall, Knoxville, TN 37996 –1610,
USA.
§To whom correspondence should be addressed. E-
mail: dbsns@nus.edu.sg
㛳Present address: Illinois Natural History Survey, 607
East Peabody Drive, Champaign, IL 61820 – 6970,
USA.
Fig. 1. Proportion of affiliate species expected to go extinct through coextinction for a given
proportion of host extinction in eight affiliate-host systems: pollinating Agaonidae Ficus wasps–
Ficus, primate Pneumocystis fungi–primates, primate nematodes–primates, primate lice–primates,
seabird lice–seabirds, bird mites– birds, butterflies–host plants, and Lycaenidae ant butterflies–ants.
Coextinction curves were estimated with a rigorous probabilistic model. Briefly, we used an explicit
combinatorial model (20) as implemented in EstimateS 7 (21) to estimate, for each data set, the
number of affiliate species expected to survive, given a decreasing number of surviving host species.
The estimated number of affiliate extinctions for a given number of host extinctions was then
computed by subtracting the number of surviving species from the respective total number of
species. See (10) for details.
R EPORTS
10 SEPTEMBER 2004 VOL 305 SCIENCE www.sciencemag.org1632
on September 3, 2008 www.sciencemag.orgDownloaded from

probability as a function of host extinction
probability and mean host specificity (Fig. 2)
(10). This alternative approach is useful for
estimating coextinction levels because mean
host specificity is easier to approximate than
complete host specificity distributions for
many affiliate-host systems. The nomograph-
ic model reveals that affiliate extinction lev-
els can be expected to decrease approximate-
ly log-linearly as the mean number of hosts
increases, for any given level of host extinc-
tion (Fig. 2). The estimated affiliate extinc-
tion probability, A
, is described by the equa-
tion
A
⫽ (0.35E ⫺ 0.43)E1n(s) ⫹ E (1)
where E
is the host extinction probability
and s
is the mean host specificity of the
affiliate species. Affiliate extinction levels
estimated by this equation are highly con-
cordant (concordance correlation R
c
⬎
0.99) with those predicted by the probabi-
listic model for all 20 affiliate-host systems
we analyzed (10).
For selected affiliate-host groups, we
estimate the magnitude of historical affili-
ate extinctions due to the documented loss
of their hosts, as well as future affiliate
extinctions if all their currently endangered
hosts [International Union for Conservation
of Nature and Natural Resources (IUCN)
categories of “critically endangered,”“en-
dangered,” and “vulnerable” (11)] were to
go extinct. We estimate that at least 200
affiliate species have become extinct his-
torically from the extinction of their hosts
in these groups (Fig. 3A), and another 6300
affiliate species are currently “coendan-
gered”—likely to go extinct if their current-
ly endangered hosts in these groups become
extinct (Fig. 3B).
For all but the most host-specific affil-
iate groups (e.g., primate Pneumocystis
fungi and primates), affiliate extinction lev-
els may be modest at low levels of host
extinction but can be expected to rise
quickly as host extinctions increase to lev-
els predicted in the near future (2, 3). This
curvilinear relationship between host and
affiliate extinction levels may also explain,
in part, why so few coextinction events
have been documented to date (10). We
modeled extinction risk as a probability.
The actual numbers of affiliate extinctions
depend on the species richness of affiliate
groups at risk and can be expected to be
substantial for species-rich affiliate taxa
(e.g., beetles, Fig. 3B). Affiliate extinctions
may already have resulted from historical
extinctions of their hosts (Fig. 3A). How-
ever, only a small proportion of the number
of affiliate extinctions that we predict on
probabilistic grounds have been document-
ed (10). The study of the skins or other
remains of extinct potential host organisms
(e.g., birds and mammals) would likely
yield many more coextinct parasites or
mutualists.
Organisms with complex life histories
would be expected to have higher risks of
coextinction over evolutionary time than
those with simpler life histories. For exam-
ple, hummingbird flower mites face extinc-
tion if either the hummingbirds they use for
transport or the flowers on which the mites
depend for nectar and pollen go extinct
(12). Conversely, in interactions where
hosts are associated with many obligately
dependent affiliate species, the loss of the
host will result in the coextinctions of all its
affiliated organisms. For example, the army
ant, Eciton burchelli, hosts no fewer than
100 affiliate species, including springtails,
beetles, mites, and ant birds (13). Many of
these affiliate organisms would hence be
lost were E. burchelli to go extinct (14 ).
Because a disproportionate number of af-
filiate species obligately depend upon them
for their continued existence, species like
E. burchelli may be considered a “keystone
mutualist,” a keystone species in an evolu-
tionary sense (15 ). The ecological impor-
tance and conservation implications of
keystone mutualists deserve further inves-
tigation because their loss will likely result
in multiple extinction cascades.
It might be argued that there is no need
to focus on the endangerment of affiliate
species, because their protection follows
automatically from the protection of their
endangered hosts. Although this may be the
case for some categories of affiliates (e.g.,
obligate endoparasites with simple life cy-
cles), affiliates that depend on complex
ecological interactions between multiple
hosts, or affiliates that have demographic
thresholds more sensitive than those of
their hosts (7, 12) may be at greater risk of
extinction than their hosts. Further, some
affiliates may be lost when their hosts are
intentionally fumigated (16). On the other
hand, in some cases declines in host popu-
lations threatened by human activities may
be exacerbated by the negative effects of
affiliates. For example, forest habitat frag-
mentation in North America has favored
the parasitic brown-headed cowbird (Molo-
thrus ater) at the expense of some of its
declining hosts (17). In such cases, if other
threats to hosts cannot be remedied, control
of affiliates, even at the risk of their possi-
ble extinction, must be contemplated.
There is no point in attempting to save an
affiliate if its host(s) become extinct in the
process. Although this study is about spe-
cies coextinctions, we expect the loss of
host populations to result in the loss of
affiliate populations. For example, Koh et
al. (7 ) recently reported that local extinc-
Fig. 2. Nomographic model expressing affiliate extinction probability as a function of host
extinction probability and mean host specificity for 20 affiliate-host systems of varying mean host
specificities: pollinating Agaonidae Ficus wasps–Ficus, primate Pneumocystis fungi–primates, pri-
mate nematodes–primates, primate lice–primates, seabird lice–seabirds, bird mites (including
Avenzoariidae, Alloptidae, Analgidae, Proctophyllodidae, Pterolichidae, Pteronyssidae, Ptiloxenidae,
Syringobiidae, and Xolalgidae)– birds, butterflies (including Papilionidae, Nymphalidae, Pieridae,
Lycaenidae, and Hesperiidae)–host plants, and Lycaenidae ant butterflies–ants. See (10) for
method. The affiliate extinction levels predicted by the nomographic model were highly concordant
(concordant correlations R
c
⬎ 0.99) with those predicted from the probabilistic model (10).
Symbols and lines represent predicted affiliate extinction levels from the probabilistic and nomo-
graphic models, respectively.
R EPORTS
www.sciencemag.org SCIENCE VOL 305 10 SEPTEMBER 2004 1633
on September 3, 2008 www.sciencemag.orgDownloaded from

tions of butterfly species were significantly
correlated with local extinctions of specific
larval host plants (7). The issue of species
or population coextinction has immediate
implications for local conservation and
management decisions.
Species coextinction is a manifestation
of the interconnectedness of organisms in
complex ecosystems. The loss of species
through coextinction represents the loss of
irreplaceable evolutionary and coevolution-
ary history (18, 19). In view of the global
extinction crisis (3), it is imperative that
coextinction be the focus of future research
to understand the intricate processes of spe-
cies extinctions. While coextinction may
not be the most important cause of species
extinctions, it is certainly an insidious one.
References and Notes
1. B. W. Brook, N. S. Sodhi, P. K. L. Ng, Nature 424, 420
(2003).
2. S. L. Pimm, P. Raven, Nature 403, 843 (2000).
3. J. A. Thomas et al., Science 303, 1879 (2004).
4. M. Castelletta, N. S. Sodhi, R. Subaraj, Conserv. Biol.
14, 1870 (2000).
5. N. E. Stork, C. H. C. Lyal, Nature 366, 307 (1993).
6. L. A. N. Amaral, M. Meyer, Phys. Rev. Lett. 82, 652
(1999).
7. L. P. Koh, N. S. Sodhi, B. W. Brook, Biotropica 36, 272
(2004).
8. D. H. Clayton, R. D. Price, Ann. Entomol. Soc. Am. 92,
675 (1999).
9. R. D. Price, D. H. Clayton, R. J. Adams, J. Parasitol. 86,
948 (2000).
10. Materials and methods are available as supporting
material on Science Online.
11. C. Hilton-Taylor, 2000 International Union for Con-
servation of Nature and Natural Resources (IUCN)
Red List of Threatened Species (IUCN, Gland, Switzer-
land, 2000; www.redlist.org).
12. R. K. Colwell, S. Naeem, in Mites: Ecology and Evolu-
tionary Analyses of Life-History Patterns, M. A. Houck,
Ed. (Chapman, New York, 1994), pp. 23– 44.
13. W. H. Gotwald, Jr., The Cornell Series in Arthropod
Biology. Army Ants: The Biology of Social Predation
(Cornell Univ. Press, Ithaca, 1995).
14. G. P. Boswell, N. F. Britton, N. R. Franks, Proc. R. Soc.
Lond. B. Biol. Sci. 265, 1921 (1998).
15. L. E. Gilber, in Conservation Biology, M. E. Soule, B. A.
Wilcox, Eds. (Sinauer Associates, Sunderland, MA,
1980), pp. 11–33.
16. D. A. Windsor, Conserv. Biol. 9, 1 (1995).
17. T. M. Donovan et al., Ecology 78, 2064 (1997).
18. S. Nee, R. M. May, Science 278, 692 (1997).
19. A. Purvis, P. M. Agapow, J. L. Gittleman, G. M. Mace,
Science 288, 328 (2000).
20. R. K. Colwell, C. X. Mao, J. Chang, Ecology, in press.
21. R. K. Colwell, EstimateS: Statistical Estimation of Spe-
cies Richness and Shared Species from Samples. Ver-
sion 7 (http://viceroy.eeb.uconn.edu/estimates; per-
manent URL www.purl.oclc.org/estimates).
22. We thank K. Brennan, C. L. Nunn, S. V. Mironov, D. H.
Clayton, J.-P. Hugot, J. M. Morales, J. Dabert, V.
Novotny, and C. Mao for useful discussions. We also
thank three anonymous reviewers for their com-
ments. R.R.D. was funded by a Fulbright Fellowship,
N.S.S and L.P.K were supported by the National Uni-
versity of Singapore (R-154-000-210-112), R.K.C.
was supported by US-NSF grant DEB-0072702,
H.C.P.’s databasing work was funded by a Natural
Sciences and Engineering Research Council Canada
Discovery Grant, and V.S.S. was supported by a Well-
come Trust Biodiversity Fellowship.
Supporting Online Material
www.sciencemag.org/cgi/content/full/305/5690/1632/
DC1
Materials and Methods
Figs. S1 and S2
Tables S1 and S2
4 June 2004; accepted 29 July 2004
Fig. 3. Predictions of affiliate extinctions from the nomographic and combinatorial models. (A)
Estimated numbers of historically extinct affiliate species based on the number of host species
recorded as extinct. (B) Projected numbers of affiliate species extinctions, were all currently
endangered hosts to go extinct. The first value in parentheses represents the absolute number and
the second value the percentage of species extinct or endangered as predicted by the nomographic
model; the second set of values in parentheses represents predictions from the combinatorial
model for selected affiliate-host groups for which affiliation matrices are available. See (10) for
details.
R EPORTS
10 SEPTEMBER 2004 VOL 305 SCIENCE www.sciencemag.org1634
on September 3, 2008 www.sciencemag.orgDownloaded from