Content uploaded by Pedro Jordano
Author content
All content in this area was uploaded by Pedro Jordano
Content may be subject to copyright.
Content uploaded by Pedro Jordano
Author content
All content in this area was uploaded by Pedro Jordano
Content may be subject to copyright.
LETTERS
Non-random coextinctions in phylogenetically
structured mutualistic networks
Enrico L. Rezende
1
, Jessica E. Lavabre
1
, Paulo R. Guimara
˜
es Jr
1,2
, Pedro Jordano
1
& Jordi Bascompte
1
The interactions between plants and their animal pollinators and
seed dispersers have moulded much of Earth’s biodiversity
1–3
.
Recently, it has been shown that these mutually beneficial inter-
actions form complex networks with a well-defined architecture
that may contribute to biodiversity persistence
4–8
. Little is known,
however, about which ecological and evolutionary processes gen-
erate these network patterns
3,9
. Here we use phylogenetic meth-
ods
10,11
to show that the phylogenetic relationships of species
predict the number of interactions they exhibit in more than one-
third of the networks, and the identity of the species with which
they interact in about half of the networks. As a consequence of the
phylogenetic effects on interaction patterns, simulated extinction
events tend to trigger coextinction cascades of related species. This
results in a non-random pruning of the evolutionary tree
12,13
and a
more pronounced loss of taxonomic diversity than expected in the
absence of a phylogenetic signal. Our results emphasize how the
simultaneous consideration of phylogenetic information and net-
work architecture can contribute to our understanding of the
structure and fate of species-rich communities.
Plant and animal species establish mutually beneficial interactions
such as pollination and seed dispersal that can form complex net-
works of dependency. Recent work has characterized the architecture
of mutualistic networks, with the ultimate goal of understanding
their formation and maintenance and the coevolution of species
within them. These networks are very heterogeneous (some species
have a much larger number of interactions than expected by
chance)
4
, and nested (specialists interact with proper subsets of the
species that generalists interact with)
5
, and are built on weak and
asymmetric links (for example, if a plant species depends strongly
on an animal species, the dependence of the animal on that plant is
much weaker)
6
. The next step on the road to understanding these
networks is to disentangle the contribution of different evolutionary
and ecological processes in generating their patterns
3,9,14
.
Here we ask to what extent network architecture is associated with
species phylogenetic relationships (Fig. 1), and whether coextinction
cascades following a species disappearance
7
involve phylogenetically
related (that is, non-randomly sampled) species. The presence of a
phylogenetic signal, where patterns of interactions between species can
be partly explained by phylogenetic relatedness, would suggest that net-
work patterns are partially dependent on past evolutionary history, and
so cannot be exclusively explained by current ecological processes
14–17
.
We compiled the largest data set of plant–animal mutualistic inter-
actions, comprising 36 plant–pollinator and 23 plant–frugivore
mutualistic networks spanning a broad geographic range (data sets
are available as Supplementary Information). For each network, we
reconstructed the phylogenies of the animals and plants (see Supple-
mentary Methods). We then characterized two components of net-
work architecture. First, we considered the number of interactions
per species, that is, species degree
4,18
, and its quantitative extension,
species strength. The strength of a plant species, for instance, is
defined as the sum of dependences or interaction weights of the
animal species on that plant
6
. These simple components of network
architecture reflect the generalization level of a given species and its
quantitative relevance in terms of how other species depend on it.
Second, we considered the identity of each species’ interactors
(Methods). Both the number of interactions per species, and the
identity of the species with which they interact have been identified
as major determinants of network architecture and robustness
7,19,20
.
With the phylogenies and these components of network structure at
hand, we applied phylogenetic statistical tools
10,11
to characterize the
extent to which closely related species tend to have similar patterns of
interactions (Fig. 1).
1
Integrative Ecology Group, Estacio
´
n Biolo
´
gica de Don
˜
ana, CSIC, Apdo. 1056, E-41080 Sevilla, Spain.
2
Instituto de Fisica ‘Gleb Wataghin’, UNICAMP, 13083-970, Campinas, Sa
˜
o Paulo,
Brazil.
b
c
a
Figure 1
|
A phylogenetic approach to mutualistic networks. We test to
what extent the architecture of coevolutionary networks is associated with
evolutionary history conveyed in the phylogenies of plants and animals. A
plant (green circles) and an animal (red squares) are linked if the latter is a
pollinator or a seed disperser of the former. Symbol size is proportional to its
number of links.
a–c, Examples where phylogeny accurately predicts the
number of interactions (
a), phylogeny does not predict the number of
interactions (
b), and the real correspondence in one network (c) (see NCOR
in the Supplementary Methods).
Vol 448
|
23 August 2007
|
doi:10.1038/nature05956
925
Nature
©2007
Publishing
Group
Randomization tests (see Methods) suggest a significant phylogen-
etic signal in species degree in 24.8% of the data sets (26 of 105
phylogenies, one-tailed P , 0.05; Fig. 2a). Power analyses indicate
that these estimates are highly conservative, suggesting that phylo-
genetic effects may be present in a larger proportion of the com-
munities (Supplementary Figs 1 and 2). Also, the probability of
detecting a signal in species degree seems to increase with phylogeny
size (for example, 54.5% of the 22 phylogenies with more than 70
species showed significant signal; Supplementary Fig. 1). The nega-
tive association between the amount of phylogenetic signal estimated
as K and phylogeny size (Fig. 2a) is probably an artefact. Even though
the expectation of K is 1 under the null hypothesis of the true phylo-
geny, its lower bound decreases with the number of species in the
phylogeny. Alternatively, Ornstein–Uhlenbeck branch-length trans-
formation methods
11
(Methods) supported a significant signal in
degree in 36.2% of the phylogenies where analyses converged (25
out of 69). P-values from randomization and branch-length trans-
formation tests were highly positively correlated (P , 0.001), indi-
cating that these two tests provide similar results. Thus these results
show that phylogenetically related species have a similar number of
interactions per species in at least one out of four phylogenies en-
compassing 39.0% of the networks.
Conversely, significant signal in species strength is present in only
2.6% of the phylogenies according to randomization tests (1 of 38;
Fig. 2b), and in 20.8% of the phylogenies where branch-length trans-
formation converged (5 of 24). The am ount of phylogenetic signal for
species strength was significantly lower than estimates for degree
(paired t-test between log-transformed K values, t
37
5 1.806, one-
tailed P 5 0.039; Fig. 2c). In spite of the significant positive correla-
tion between degree and strength
6
, estimates of species strength may
be subject to higher levels of uncertainty associated with proximate
factors such as species abundance variability, changes associated with
phenological sequences, and sampling errors that tend to decrease
phylogenetic signal
11
. This hypothesis could be tested by determining
how estimates of species degree and strength vary over time in the
same communities, the expectation being that species strength would
show larger fluctuations than would degree. Alternatively, one could
test whether the signal for strength increases after normalizing by
species abundance. Nonetheless, our results suggest that species
degree has stronger phylogenetic signal than strength.
Turning now to our second component of network architecture—
which species interacts with which—we tested whether phylogenetic
relatedness correlates with ecological similarity. The ecological sim-
ilarity of any two species is defined as the number of species with
which they both interact divided by the total number of species with
which they interact (Methods). Phylogenetic and ecological distance
matrices are positively and significantly correlated in 42.7% of the
phylogenies (44 of 103, one-tailed Mantel test, P , 0.05). This means
that phylogenetically related species tend to interact with a similar set
of species. To determine whether this result is a consequence of the
phylogenetic signal in degree reported above, we repeated these tests,
controlling for differences in the number of interactions per species
(partial Mantel test, see Methods). The results remained qualitatively
similar. Partial Mantel correlations are significant in 46.6% of the
phylogenies, supporting the idea that phylogeny is associated with
the identity of the species’ interactors after controlling for degree.
The association between phylogenetic resemblance and ecological
similarity tends to be more common among animals: 60.8% of
Mantel correlations between ecological and phylogenetic distance
matrices were significant for animals, whereas 25.0% were significant
for plants (Fig. 3). In addition, comparison of the Mantel coefficients
Z for plants and animals indic ated that animal phylogenies were
more strongly associated with species interaction patterns than were
plant phylogenies (paired t-test between log-transformed values,
t
43
5 3.218, one-tailed P 5 0.001; Fig. 3). Further, these results are
robust even when the number of animal and plant species is statist-
ically controlled (analysis of covariance, ANCOVA, F
1,100
5 10.16,
P 5 0.02). Results from partial Mantel tests controlling for degree
were qualitatively similar (paired t-test, t
43
5 2.576, one-tailed
P 5 0.014). It would be interesting to investigate multiple alternative
hypotheses, such as differences in mobility and evolvability
21–23
,to
determine the cause of this difference.
Although network structure seems significantly more associated
with animal phylogenies, structure may be driven by the evolutionary
history of both plants and animals (7 of 44 communities), only plants
or only animals (3 and 21 communities, respectively), or neither
plants nor animals (the remaining 13 communities). This highlights
the large variability across networks, indicated by the residual vari-
ation of K or Z after controlling for phylogeny size (Figs 2 and 3;
Supplementary Table 1). Part of this variability is related to the taxo-
nomic diversity of the plant lineages, though apparently not to the
diversity of frugivores (fruit-eaters) or pollinators according to
10 100 400
Animals
Plants
–0.2
0.0
0.2
0.4
0.6
Signal degree – signal strength
–0.4
0.8
Ph
y
lo
g
eniesNumber of species
1
0.1
1
0.5
0.5
Amount of phylogenetic signal
0.1
5
b
ca
Figure 2
|
Magnitude of phylogenetic signal on the number and strength of
mutualistic interactions.
Relationship between the magnitude of
phylogenetic signal K and phylogeny size, estimated for species degree
(
a), and strength (b). Each data point represents a phylogeny: green circles
for plants and red squares for animals. Solid symbols indicate statistically
significant phylogenetic signals.
c, Comparison of phylogenetic signal for
species degree and strength. Green and dashed red bars correspond to plants
and animals, respectively. Estimates obtained for degree were significantly
higher.
10
–0.2
0.0
0.2
0.4
0.6
–0.2
0.0
0.2
0.4
0.6
–0.3
–0.2
–0.1
0.0
0.1
0.2
0.3
ca
b
100
Number of species
Correlation between phylogeny and ecology
Correlation animals – correlation plants
Communities
Pollination
Frugivory
400
Figure 3
|
Correlation between ecological similarity and phylogenetic
relatedness. Results of regular Mantel tests correlating phylogenetic and
ecological distance matrices plotted against phylogeny size, obtained for
plants (
a) and animals (b). Each data point corresponds to a phylogeny, and
a solid symbol indicates a statistically significant correlation.
c, Comparison
between Mantel Z estimates obtained separately for plants and animals
composing each network. Communities where interaction patterns are more
associated with animal phylogenies are depicted in red; those more
associated with plants in green. The phylogenetic structure of animals
correlates significantly better with interaction matrices than that of plants.
LETTERS NATURE
|
Vol 448
|
23 August 2007
926
Nature
©2007
Publishing
Group
several regression models (Methods). One plausible explanation for
this result is that flower or fruit morphology is variable across the
taxonomic groups examined, whereas phenotypic variation for fru-
givores and insects is more relevant at other taxonomic levels.
Partitioning the phenotypic variance across taxonomic groups may
address this hypothesis, and also clarify the mechanisms by which
evolutionary history translates into patterns of interactions.
We have shown that phylogenetically related species tend to have
similar roles in the network of interactions in almost half of the
communities studied here. These effects are not a reflection of differ-
ences between major taxonomic groups only, because they also
appear at finer scales of the phylogenies (Supplementary Fig. 3). In
light of this, we tested whether simulated coextinctions
7
involve taxo-
nomically related species more often than expected by chance for
cascades of identical size (Methods). This would result in a non-
random pruning of the evolutionary tree
12,13
. Simulations show that
the rate of taxonomic diversity loss is higher than expected in the
absence of phylogenetic signal (Fig. 4). Although these effects may
seem quantitatively small (partly owing to the averaging nature of
the index), they can actually encompass the extinction of entire fam-
ilies or higher taxonomic groups from the community. The overall
reduction in taxonomic diversity holds across communities, so that
values falling below the null expectation are significantly more fre-
quent than those above it (x
2
5 50.7, degrees of freedom, d.f. 5 1,
P , 0.0001; Fig. 4). Moreover, the contribution of phylogeny to spe-
cies patterns of interaction correlates with the magnitude of taxo-
nomic diversity loss across communities (P , 0.05 for parametric
and non-parametric correlations; Fig. 4c). Therefore, communities
in which species interactions have a strong phylogenetic component
are more prone to have closely related species going coextinct fol-
lowing an extinction event. We conclude that the interaction between
network and phylogenetic structures can ultimately result in non-
random coextinction patterns.
Plant–animal mutualisms form heterogeneous, nested networks
built on weak and asymmetric links among species, which may facil-
itate long-term species persistence
6,7
. Our results provide evidence
for the role of phylogenetic relationship as one determinant factor
shaping these patterns: phylogeny partly accounts for species’ pro-
pensities to interact in more than one-third of the networks, and the
identity of the species with which they interact in about half of the
networks. From a theoretical point of view, our results warrant the
inclusion of evolutionary history into mechanistic models of network
formation and maintenance
24
. From a conservation perspective, our
results show that cascading effects of coextinction may spread across
taxonomically related species, further increasing the erosion of taxo-
nomic diversity.
METHODS SUMMARY
The amount of phylogenetic signal in degree and strength was quantified with
the K statistic, which is the fraction of the amount of signal of the real data set
over that expected, assuming brownian motion and the same tree topology.
Significance of phylogenetic signal was tested with randomization and branch-
length transformation methods
11
.
We used Mantel tests to compare phylogenetic distance matrices with mat-
rices of ecological distance. Phylogenetic distance between pairs of plants (or
animals) was estimated as the expected covariance of the trait between the two
species
11,25
. Ecological distance was calculated as 1 2 S, where S is the Jaccard
index of similarity
26
. We also performed partial Mantel tests controlling for the
absolute difference in degree between two species.
Species removal simulations started from the most specialized to the most
generalized species
7
. After an extinction cascade, we calculated the community
taxonomic diversity as the average taxonomic distance between species
27
. Path
length weights between species increased the more distantly related they were
taxonomically (that is, species of the same genus have a distance of 1 whereas
species from different genera within the same family have a distance of 2, and so
on). Thus, the higher the index, the more diverse the community. To calculate
the decrease of taxonomic diversity of the real community with respect to the
expected decrease in the absence of phylogenetic signal, we replicated the coex-
tinction cascade 1,000 times after randomizing the taxonomic affiliation of
species going coextinct. The taxonomic diversity relative to the null expectation
was the ratio between real and null values, and the significance was estimated by
counting how often the real value fell below the randomization results. The
average rate of taxonomic loss per community was calculated as the slope of a
linear regression with an intercept forced through 1.
Full Methods and any associated references are available in the online version of
the paper at www.nature.com/nature.
Received 19 March; accepted 25 May 2007.
1. Ehrlich, P. R. & Raven, P. H. Butterflies and plants: a study in coevolution. Evol. I nt. J.
Org. Evol. 18, 586
–
608 (1964).
2. Thompson, J. N. The Geographic Mosaic of Coevolution (Univ. Chicago Press,
Chicago, 2005).
3. Waser, N. M. & Ollerton, J. (eds) Plant
–
Pollinator Interactions: from Specialization to
Generalization (Univ. Chicago Press, Chicago, 2006).
4. Jordano, P., Bascompte, J. & Olesen, J. M. Invariant properties in
coevolutionary networks of plant
–
animal interactions. Ecol. Lett. 6, 69
–
81
(2003).
Correlation between
p
h
y
lo
g
en
y
and ecolo
gy
Percentage of animals removed
0.94
0.96
0.98
1.00
1.02
1.04
0.94
0.96
0.98
1.00
1.02
–1.0
–0.5
0.0
0.5
Rate of taxonomic loss Animal taxonomic diversity Plant taxonomic diversity
–1.5
1.0
b
a
c
0
10020 40 60 80
Percentage of plants removed
0
10020 40 60 80
–0.1 0.0 0.1 0.2 0.3
Figure 4
|
Phylogenetic resemblance induces a higher loss of taxonomic
diversity after species extinctions.
Taxonomic diversity of plants (a) and
animals (
b) of five communities as a function of the number of extinct
species, removed from most specialized to most generalized. Each symbol
represents the removal step leading to the next generalization level.
Taxonomic diversity is the ratio between indexes from the real communities
over randomization results removing phylogenetic effects. Full symbols
indicate taxonomic losses below 95% of the null model.
c, Average (6s.e.m.)
rate of taxonomic loss of ten communities regressed against the magnitude
of phylogenetic effects.
NATURE
|
Vol 448
|
23 August 2007 LETTERS
927
Nature
©2007
Publishing
Group
5. Bascompte, J., Jordano, P., Melia
´
n, C. J. & Olesen, J. M. The nested assembly of
plant
–
animal mutualistic networks. Proc. Natl Acad. Sci. USA 100, 9383
–
9387
(2003).
6. Bascompte, J., Jordano, P. & Olesen, J. M. Asymmetric coevolutionary networks
facilitate biodiversity maintenance. Science 312, 431
–
433 (2006).
7. Memmott, J., Waser, N. M. & Price, M. V. Tolerance of pollination networks to
species extinctions. Proc. R. Soc. Lond. B 271, 2605
–
2611 (2004).
8. Va
´
zquez, D. P. & Aizen, M. A. Asymmetric specialization: a pervasive feature of
plant
–
pollinator interactions. Ecology 85, 1251
–
1257 (2004).
9. Thompson, J. N. Mutualistic webs of species. Science 312, 372
–
373 (2006).
10. Freckleton, R. P., Harvey, P. H. & Pagel, M. Phylogenetic analysis and comparative
data: a test and review of evidence. Am. Nat. 160, 712
–
726 (2002).
11. Blomberg, S. P., Garland, T. Jr & Ives, A. R. Testing for phylogenetic signal in
comparative data: Behavioral traits are more labile. Evol. Int. J. Org. Evol. 57,
717
–
745 (2003).
12. Purvis, A., Agapow, P.-M., Gittleman, J. J. & Mace, G. M. Nonrandom extinction
and the loss of evolutionary history. Science 288, 328
–
330 (2000).
13. Heard, S. B. & Mooers, A. Ø. Phylogenetically patterned speciation rates and
extinction risks change the loss of evolutionary history during extinctions. Proc. R.
Soc. Lond. B 267, 613
–
620 (2000).
14. Ives, A. R. & Godfray, C. J. Phylogenetic analysis of trophic associations. Am. Nat.
168, E1
–
E14 (2006).
15. Armbruster, W. S. Phylogeny and the evolution of plant
–
animal interactions.
Bioscience 42, 12
–
20 (1992).
16. Herrera, C. M. Historical effects and sorting processes as explanations for
contemporary ecological patterns: character syndromes in Mediterranean woody
plants. Am. Nat. 140, 421
–
446 (1992).
17. Jordano, P. Angiosperm fleshy fruits and seed dispersers: a comparative analysis
of adaptation and constraints in plant
–
animal interactions. Am. Nat. 145, 163
–
191
(1995).
18. Waser, N. M., Chittka, L., Price, M. V., Williams, N. M. & Ollerton, J. Generalization
in pollination systems, and why it matters. Ecology 77, 1043
–
1060 (1996).
19. Albert, R., Baraba
´
si, A.-L. & Jeong, H. The Internet’s Achilles’ Heel: Error and
attack tolerance in complex networks. Nature 406, 378
–
382 (2000).
20. Montoya, J. M., Pimm, S. L. & Sole
´
, R. V. Ecological networks and their fragility.
Nature 442, 259
–
264 (2006).
21. Wheelwright, N. T. Fruit size in a tropical tree species—variation, preference by
birds, and heritability. Vegetatio 108, 163
–
174 (1993).
22. Fenster, C. B., Armbruster, W. S., Wilson, P., Dudash, M. R. & Thompson, J. D.
Pollination syndromes and floral specialization. Annu. Rev. Ecol. Evol. Syst. 35,
375
–
403 (2004).
23. Bronstein, J. L., Alarco
´
n, R. & Geber, M. The evolution of plant
–
insect mutualisms.
New Phytol. 172, 412
–
428 (2006).
24. Cattin, M.-F., Bersier, L.-F., Banasek-Richter, C., Baltensperger, R. & Gabriel, J.-P.
Phylogenetic constraints and adaptation explain food-web structure. Nature 427,
835
–
839 (2004).
25. Garland, T. A. Bennett, F. & Rezende, E. L. Phylogenetic approaches in
comparative physiology. J. Exp. Biol. 208, 3015
–
3035 (2005).
26. Legendre, P. & Legendre, L. Numerical Ecology 2nd edn, 254
–
256 (Elsevier,
Amsterdam, 1998).
27. Clarke, K. R. & Warwick, R. M. A taxonomic distinctness index and its statistical
properties. J. Appl. Ecol. 35, 523
–
531 (1998).
Supplementary Information is linked to the online version of the paper at
www.nature.com/nature.
Acknowledgements We thank S. Armbruster, P. Buston, M. A. Fortuna,
P. Guimara
˜
es, M. Helmus, A. Ives, J. Olesen, D. Posada, A. G. Sa
´
ez, J. N. Thompson
and N. Waser for comments on a previous draft. This work was funded by the
European Heads of Research Councils, the European Science Foundation, and the
EC Sixth Framework Programme through a EURYI (European Young Investigator)
Award (J.B.), by the Spanish Ministry of Education and Science (P.J. and J.B.), by
the Junta de Andalucı
´
a (P.J. and J.B.), and by the Fundac¸a
˜o
de Amparo a
`
Pesquisa
do Estado de Sa
˜o
Paulo (P.R.G).
Author Contributions E.L.R. performed all the analysis and compiled the
phylogenies jointly with J.E.L. P.R.G. performed the extinction simulations. P.J. and
J.B. designed the study and compiled the interaction matrices. E.L.R. and J.B. wrote
a first version of the manuscript, and all authors contributed to the final draft.
Author Information Reprints and permissions information is available at
www.nature.com/reprints. The authors declare no competing financial interests.
Correspondence and requests for materials should be addressed to J.B.
(bascompte@ebd.csic.es).
LETTERS NATURE
|
Vol 448
|
23 August 2007
928
Nature
©2007
Publishing
Group
METHODS
Database. We compiled 59 qualitative mutualistic networks (36 plant–pollinator
and 23 plant–frugivore webs) describing the presence or absence of interactions.
The data set encompasses Mediterranean, tropical, temperate, subtropical and
Arctic communities from all continents except mainland Asia and Antarctica. Of
these networks, 22 (9 for pollination and 13 for frugivory) are quantitative,
describing the strength of each interaction or pairwise dependence (see data sets
in Supplementary Information). From these networks, we calculated species
degree and species strength for plants and animals separately, and recorded the
taxonomic affiliation of species forming each community.
Phylogenies. We assembled one animal and one plant phylogeny for each com-
munity. Phylogenies are based primarily on molecular data, with a few species
included according to taxonomic information (Supplementary Methods).
Phylogenies with less than ten species or with too many unresolved nodes were
not included in analyses. This resulted in 105 phylogenies for the following
groups: 35 insect phylogenies (Class Insecta; all pollinators), 18 bird phylogenies
(Class Aves; all frugivores), and 52 angiosperm phylogenies (Infraphylum
Angiospermae; 33 belonging to plant–pollinator and 19 to plant–frugivore
networks).
Phylogenetic statistical methods. We tested for the presence of phylogenetic
signal on species degree and strength with randomization and branch-length-
transformation tests
11
. These methods test whether species attributes are signifi-
cantly associated with phylogeny, using randomization or maximum-likelihood
procedures. Branch-length transformation tests were performed assuming the
Ornstein–Uhlenbeck model of stabilizing selection and a model in which char-
acter evolution can accelerate or decelerate (ACDC)
11
. Because ACDC models
did not converge in most cases, here we discuss results from the Ornstein–
Uhlenbeck model (all analyses are included in Supplementary Material for
completeness). Although these techniques provide similar information about
the presence of a phylogenetic signal, applying both methods can be useful in
determining how robust our results are and in overcoming limitations inherent
to each statistical test (Supplementary Figs 1 and 2). The amount of phylogenetic
signal was quantified with the K statistic (Fig. 2), which is the fraction of the
amount of signal of the real data set over that expected, assuming brownian
motion and the same tree topology.
We used Mantel tests to compare phylogenetic distance matrices with mat-
rices of ecological distances between species. Phylogenetic distance between pairs
of plants (or animals) was estimated as the expected covariance of the trait
between the two species
11,25
. Ecological distance was calculated as 1 2 S, where
S is the Jaccard index of similarity obtained from qualitative interaction mat-
rices
26
. The similarity between two species i and j is defined as S(i, j) 5
a/(a 1 b 1 c), where a, b and c represent the number of shared interacting
species, the number of interactions specific to species i, and the number of
interactions exclusive to species j, respectively.
Because differences in degree affect Jaccard estimates, we also performed
partial Mantel tests controlling for degree (the pairwise distance in degree was
calculated as the absolute difference in degree between two species; Supplemen-
tary Methods). Hence, this partial test can discern whether phylogeny strictly
affects the species with which species interact, independently of the total number
of interactions of each species. When necessary, estimates were log-transformed
to improve normality (or log-value 1 1 for statistics varying between 21 and 1,
as Mantel’s Z).
According to regression models controlling for phylogeny size, community
size, and number of interactions, phylogenetic signal was similar for frugivory
and pollination networks (P . 0.28 for K and for Mantel’s regular and partial Z),
hence results were pooled.
Taxonomic diversity and coextinction simulations. As a surrogate for phylo-
genetic diversity, we estimated taxonomic diversity of plants and animals in the
largest available phylogenies (23 plant and 27 pollinator phylogenies with more
than 30 species, and 15 bird phylogenies with more than 15 species; see
Supplementary Methods). The mean taxonomic distance between all species
was employed as an index of taxonomic diversity
27
in subsequent regressions.
Extinction cascades were simulated for the ten largest communities (all having
more than 40 animal and plant species) with available taxonomic affiliation,
following ref. 7. After one species is removed, species left without any interaction
go coextinct. Species removal started from the most specialized (least-linked) to
the most generalized (most-linked) species, which was proposed as a more
plausible extinction sequence because specialist species tend to be less abundant
than generalists
4,7,8
. In spite of several assumptions implicit in the model (for
example, all plants require animals for reproduction, and species cannot adapt to
new resources), this approach provides the first reasonable attempt to study
coextinction patterns in phylogenetically structured networks
7
. Although
these assumptions may affect the total number of species going extinct
7
, our
comparisons involve coextinction cascades of the same size, with and without
phylogenetic signal.
After an extinction cascade, we calculated the decrease of taxonomic diversity
of the real community respect to the expected decrease in the absence of phylo-
genetic signal. This was done by replicating the coextinction cascade after ran-
domizing the taxonomic affiliation of species going coextinct (that is, nodes
remain unchanged but their ‘name tags’ are shuffled). This null model removes
effects of phylogenetic relatedness
11
controlling for network structure and spe-
cies number. The relative taxonomic diversity is the ratio between real and null
values, and the average rate of taxonomic loss per community is the slope of a
linear regression with an intercept forced through 1 (that is, real values and the
null expectation are equal when no species are removed).
doi:10.1038/nature05956
Nature
©2007
Publishing
Group