Current Biology 17, 341–346, February 20, 2007 ª2007 Elsevier Ltd All rights reserved DOI 10.1016/j.cub.2006.12.039
Specialization, Constraints, and
in Mutualistic Networks
and Nils Blu
Department of Animal Ecology and Tropical Biology
University of Wu¨ rzburg
Wu¨ rzburg, 97074
Field Station Fabrikschleichach
University of Wu¨ rzburg
Institute of Molecular Neurobiology
Free University of Berlin
The topology of ecological interaction webs holds
important information for theories of coevolution, bio-
diversity, and ecosystem stability [1–6]. However,
most previous network analyses solely counted the
number of links and ignored variation in link strength.
Because of this crude resolution, results vary with
scale and sampling intensity, thus hampering a com-
parison of network patterns at different levels [7–9].
We applied a recently developed  quantitative and
scale-independent analysis based on information the-
ory to 51 mutualistic plant-animal networks, with inter-
action frequency as measure of link strength. Most net-
works were highly structured, deviating signiﬁcantly
from random associations. The degree of specializa-
tion was independent of network size. Pollination
webs were signiﬁcantly more specialized than seed-
dispersal webs, and obligate symbiotic ant-plant mutu-
alisms were more specialized than nectar-mediated
facultative ones. Across networks, the average spe-
cialization of animal and plants was correlated, but is
constrained by the ratio of plant to animal species in-
volved. In pollination webs, rarely visited plants were
on average more specialized than frequently attended
ones, whereas specialization of pollinators was posi-
tively correlated with their interaction frequency. We
conclude that quantitative specialization in ecological
communities mirrors evolutionary trade-offs and con-
straints of web architecture. This approach can be eas-
ily expanded to other types of biological interactions.
Results and Discussion
Ecological specialization in a food web or other interac-
tion networks is commonly deﬁned by the number of
realized ‘‘links.’’ For instance, predators are specialized
if they attack only a few prey species, and specialized
ﬂowers are those that are visited by few pollinator spe-
cies only. This concept has been extended to measure
the degree of specialization of entire networks (‘‘con-
nectance’’), where associations are classiﬁed as ‘‘pres-
ent’’ or ‘‘absent,’’ but all links are considered equally
important [1–3, 6, 8, 11–13]. However, such qualitative
measures ignore the importance of variation in interac-
tion strength for community dynamics [5, 14, 15]. More-
over, they are highly sensitive to sampling intensity and
network size [7–10, 15]. Therefore, weighted links have
been included in quantitative descriptors of different
types of webs [5, 14, 16]. In bipartite ecological net-
works, the frequency of an interaction between two spe-
cies is a meaningful measure of its strength (Figure 1)
and has been shown to represent a suitable surrogate
for mutualistic services such as pollination success
. In this article, we use two measures inspired by
information theory to quantify specialization within and
across networks. Technical properties of these indices
have been explored in a recent methodology article
 showing that—in contrast to other quantitative
measures—they are scale independent and largely in-
sensitive to sampling effort. Unlike previous measures,
we deﬁne the overall degree of specialization in each
web as the deviation from an expected probability distri-
bution of interactions (evaluated by the standardized
two-dimensional entropy H0
), and individual species’
specialization as the deviation from a conformity ex-
pected by the overall utilization of potential partners
(standardized Kullback-Leibler distance, d0
expected null distribution assumes that all species inter-
act with their partners in proportion to their total fre-
quencies, whereas the heterogeneity (evenness) of in-
teractions in previously proposed quantitative metrics
such as diversity indices [16, 18] varies with the partner
availabilities in an uncontrolled way and is thus less suit-
able in the context of network analyses (see also [10,
19]). On the basis of these standardized quantitative
measures, we explored 51 networks, covering four
types of mutualistic plant-animal associations, for pat-
terns of specialization on the level of the entire network
[2, 8], the community of each of the two parties (guild
level) , and the level of species .
Across all networks, the overall degree of specialization
) covered a broad range (Table S1 in the Supplemen-
tal Data available online). All networks showed a signiﬁ-
cantly higher degree of organization than simulated
networks, where partners were associated randomly
(all p %0.001), except for a single network of loosely as-
sociated ants and bromeliads  (p = 0.31). Pollination
mutualisms were signiﬁcantly more specialized than
seed-dispersal mutualisms (Figure 2), corroborating a
previous qualitative analysis  and expected on the
basis of evolutionary considerations . Plants may
Present address: Manchester Centre for Integrative Systems Biol-
ogy (MCISB), Manchester Interdisciplinary Biocentre (MIB), Man-
chester M1 7ND, United Kingdom.
beneﬁt disproportionately more from specialized polli-
nators, corresponding to the likelihood that each indi-
vidual pollinator successively visits conspeciﬁc plants
to maintain both male and female reproductive success
of a plant, thereby reducing maladaptive heterospeciﬁc
pollen transfer ([22, 23], but see ). In contrast to pol-
lination, the efﬁcacy of seed dispersal to suitable sites
does not depend on the specialization of the dispersal
agent . A broader spectrum of seed dispersers
may even be proﬁtable from the plant’s perspective to
avoid aggregation of seeds  and generate fat-tailed
dispersal kernels , which should be favored by natu-
ral selection under many conditions . Correspond-
ingly, obligate specialized mutualisms are known from
a number of pollination systems  but seem to be
largely absent in seed-disperser systems .
In ant-plant networks, there is an important distinction
between completely facultative associations, based on
extraﬂoral nectaries, and symbiotic associations, where
ant colonies, often obligatory, inhabit plants (myrmeco-
phytes) [27, 28]. For obligate and symbiotic mutualisms,
a higher degree of specialization is generally expected
[4, 29]. This differentiation is supported by our analysis:
Ant-plant mutualisms involving myrmecophytes were
signiﬁcantly more specialized than those involving ex-
traﬂoral nectaries (Figure 2). Obligate associations are
common among myrmecophytic associations, some-
times causing irreversible dependence on a single part-
ner species. Myrmecophytes represent a gradient from
plants that offer neither speciﬁc structures nor speciﬁc
food rewards to support their facultative ant inhabitants
[20, 28] to cases where only few ant species are adapted
to actively bite small entrance holes into preformed do-
matia and where colonies are fully supplied by nutritious
plant-produced food bodies and never forage outside
their host plants [27, 30]. Obligate-myrmecophytic sym-
bioses represent the most specialized networks across
all systems examined in this study. In contrast to other
networks, such associations often remain uninterrupted
for several generations, opening the opportunity for the
evolution of tight specialization. In contrast, extraﬂoral
nectaries usually attract a spectrum of largely opportu-
nistic ants, where the accessible nectaries seem to
offer little structural plasticity to facilitate specialization
except for some degree of biochemical differentiation
[31, 32]. This dissimilarity between the two types of
ant-plant mutualisms is particularly evident between
nectary-bearing and myrmecophytic species from the
same genus [27, 32]. The gradient from facultative to
obligate mutualisms is thus largely associated with an
The degree of specialization did not show a signiﬁcant
trend across networks of different dimensions (Figure 3)
(Spearman rank correlations for each of the four network
types, all 20.48 %r
%20.10, p R0.15). Given that H0
is mathematically independent of web size , the lack
of a correlation between web size and H0
species-rich and species-poor real biological systems
(or smaller fractions of a system) do not inherently differ
in their degree of specialization between partners. This
novel ﬁnding contrasts with the hyperbolic decline of
Figure 1. Visualization of Two Quantitative Networks
A pollinator web and an ant-plant association are displayed (webs 6
and 37 in Table S1). Widths of links are scaled in relation to interac-
tion frequencies, bar sizes to total interaction frequencies. Both
webs are regarded specialized, but the degree of specialization is
lower in the pollinator web (H0
= 0.46) compared to the myrmeco-
phyte web (H0
= 0.84). Note that the former web is asymmetric
(more pollinator than plant species), whereas the latter is symmetric.
Figure 2. Network-Level Specialization
Overall specialization (H0
) in 51 mutualistic
networks. Box plots show median, quartiles,
and range of the networks analyzed (number
of networks in parentheses). Asterisks show
signiﬁcant difference between types accord-
ing to a t test (*** p < 0.0001, both tR5.2,
Welsh corrected for unequal variances).
the qualitative connectance index over increasing net-
work size in different studies [2, 8, 33, 34], a decline
that was also found if applied to the dataset used here
(see Supplemental Data).
Within mutualistic networks, differences between the
average degree of specialization of both parties (i.e.,
plants versus animals) could be a consequence of con-
ﬂicting interests. Consumers would only beneﬁt from in-
creased specialization if this process went along with
greater resource-use efﬁcacy and/or reduced interspe-
ciﬁc competition, e.g., by improved resource detoxiﬁca-
tion, reduced handling effort, or speciﬁc search images,
and outweighed the costs of increased foraging time. If
resources were very similar, optimal foraging theory
would thus predict selection for generalization in both
frugivores and pollinators [11, 21, 23]—the latter con-
ﬂicting with the plant’s interest in specialized pollina-
tors. However, both parties did not vary independently
in their degree of specialization, and average specializa-
tion of plants hd0
iand animals hd0
iwas largely recipro-
cal (Pearson’s r
= 0.71, p < 0.0001, n = 51 webs). More-
over, differences between hd0
predicted by the asymmetry of the matrix (r
p < 0.0001) (Figure 4). In those webs where animal spe-
cies were more numerous than plants, animals showed
a lower degree of specialization (hd0
i) and vice
versa. This effect was even stronger (r
= 0.93) for simu-
lated networks with randomly assigned associations
(Figure S2). Pollinator and ant-nectar webs were
highly asymmetric, involving a much higher number of
Figure 3. Relationship between Network Size
Overall specialization (H0
) of 51 networks
plotted over network size (plant plus animal
species, log scale). Networks include pollina-
tion (yellow), seed-dispersal (black), ant-myr-
mecophyte (green), and ant-nectar plant (red)
Figure 4. Relationship between Network Asymmetry and Specialization
Asymmetry of the number of plant (I) and animal (J) species in each web (network asymmetry) is given as (J2I)/(I+J) and equals zero for bal-
anced webs (same number of animal and plant species). Specialization asymmetry between plants and animals is given as (hd0
i), based on weighted means across all species (plants ior animals j) in a web. Real networks include pollination (yellow), seed-dispersal
(black), ant-myrmecophyte (green), and ant-nectar plant (red) associations. The regression line is plotted for randomly generated networks (ﬁxed
total interactions per species, mean values from 100 randomizations per web, r=20.97).
Mutualistic Interaction Webs
pollinator species (usually insects) or ant species than
plant species (on average 3.6:1 and 3.8:1, respectively).
Consequently, pollinators were signiﬁcantly less spe-
cialized on plants than plants on pollinators, and ants
were signiﬁcantly less specialized on plants with nectar-
ies than vice versa (paired t test; pollinators: t= 3.8,
p = 0.001; ants: t= 3.2, p = 0.01). In contrast, networks
involving seed dispersers (mostly vertebrates) as well
as ant-myrmecophyte associations were usually more
symmetric (1.2:1 and 1.6:1, respectively) and did not
show signiﬁcant unequal specialization of both mutual-
ists (both p R0.38). Hence, the network architecture se-
verely constrains average specialization between two
parties irrespective of the type of association, a result
that is expected given the mathematical relationships
of the indices in their unstandardized form . Such
constraints on specialization have been largely over-
looked so far, but are important in other network metrics
as well, including the ‘‘number of links’’ or quantitative
‘‘dependences’’ used elsewhere  (see Supplemental
Data). However, with architectural constraints ac-
counted for, residual variation from the linear regression
(line shown in Figure 4) depicted differences between
networks depending on the type of association. Pollina-
tors were signiﬁcantly more specialized than expected
by the asymmetry (mean residuals > 0, t= 4.7,
p < 0.001), whereas ants visiting extraﬂoral nectaries
were more generalized than expected (residuals < 0,
t=22.4, p < 0.05). In seed-dispersal and ant-myr-
mecophyte networks, differences between animals
and plants in specialization did not deviate signiﬁcantly
from the expected on the basis of asymmetry (both
pR0.10). The increased residual specialization ob-
served in pollinators, but not seed dispersers, thus
corresponds to the plant’s differential interest in these
types of mutualists.
Although the average degree of specialization in a com-
munity may be constrained to a large degree, this does
not apply to single elements of the network, i.e., the local
population of each species. For example, disparities in
specialization of pollinators and plants were particularly
pronounced for the rarely interacting species in a net-
work. Across pollination networks, there was a signiﬁ-
cantly positive correlation between pollinator frequency
and specialization, but a signiﬁcant negative correlation
between plant frequency and specialization (Table 1).
We also found a signiﬁcantly positive correlation be-
tween ant frequency and specialization in ant-myrmeco-
phyte webs but not in any of the other networks investi-
gated. Previous qualitative network analyses showed an
invariable negative correlation between frequency and
specialization (estimated as the inverse of the number
of links), a correlation that can be explained by a null
model  and is strongly affected by sampling effort.
In contrast, our quantitative analysis demonstrates a
highly variable relationship between frequency and
quantitative specialization, one that differs between net-
work types. Our results suggest that plant populations
with low visitation frequencies, presumably those that
occur in low densities in a community, have a particularly
unconventional spectrum of visitors. Rare plants may be
particularly sensitive for two ﬁtness costs: subsequent
pollen deposition on (more common) plants and clog-
ging of the stigma by pollen from (common) plants
. Increased specialization and reduced overlap
with visitors of common ﬂowers may reduce such costs.
The positive correlation between animal abundance and
specialization indicates that resource partitioning is par-
ticularly pronounced among the most active species,
whereas rarely interacting species use their resources
Three general conclusions can be drawn from the re-
sults. (1) The network-level specialization is unaffected
by network size and form and depicts biologically mean-
ingful system-speciﬁc differences. Our results demon-
strate that the plant’s interest in specialized pollen trans-
fer but generalized fruit dispersal conformed to the
overall specialization of the respective networks. Net-
works involving facultative associations were less spe-
cialized than more obligate ones, particularly in ant-plant
webs. (2) The average degree of specialization of both
network parties is highly reciprocal, i.e., one party cannot
specialize or generalize on the other party without con-
comitant changes in the specialization within the other
party itself. Moreover, differences between network
parties are largely driven by constraints in the network
architecture (unequal species numbers). Such con-
straints cause unequal degrees of specialization as
well as asymmetric dependences between both parties.
Residual differences in specialization still contain mean-
ingful information, e.g., pollinators were more special-
ized than expected from architectural constraints only.
(3) Species-level specialization is less affected by these
constraints and may indicate differential roles of rare
and common species in a network. Such patterns may
potentially unveil density-dependent selection pres-
sures or feedback mechanisms between frequency and
The hypothesis that natural selection drives speciali-
zation between interacting mutualists or antagonists
has been debated for a long time [4, 11, 35]. Whereas
generalists are obviously much less limited by resource
or partner availability, specialists are usually better
adapted to effectively use their selected resources. For
antagonistic relationships (e.g., predator-prey, host-
parasite, and plant-herbivore interactions), defensive
Table 1. Relationship between Frequency and Specialization of
Plant and Animal Species
Pollination 20.20* (20.30 220.11)
(n = 20)
0.27* (0.15 20.37)
(n = 21)
Seed dispersal 0.06 (20.19 20.23)
(n = 7)
0.00 (20.27 20.24)
(n = 8)
Ant-myrmecophyte 0.14 (20.16 20.40)
(n = 14)
0.51* (0.24 20.71)
(n = 13)
Ant-nectar plant 20.10 (20.23 20.02)
(n = 7)
20.04 (20.27 20.16)
(n = 8)
Effect sizes derived from linear correlation coefﬁcients for each net-
work by using meta analysis based on Fisher’s z-transformation.
Mean back-transformed rvalues are shown with range of 95% boot-
strap conﬁdence intervals and number of webs (n) in parentheses.
Asterisks indicate signiﬁcant deviation from r=0.
mechanisms of hosts or prey substantially constrain the
choices of their enemies, enforcing specialization .
Trade-offs between specialization and generalization
may occur in food webs , but are also complex among
mutualists , where selective pressures on partner
choices may be variable and shaped by coevolutionary
complementarity or convergence . Reﬁned analyses
and more ﬁne-grained empirical data, particularly at
the level of individuals, may reveal additional insights
into the evolution of a broad spectrum of interaction
webs and their ecological fragility.
We analyzed the degree of specialization for 51 published and un-
published interaction webs that included frequency data, represent-
ing a broad range of mutualistic relationships between plant-based
resources and their consumers or inhabitants and covering six con-
tinents (Supplemental Data). Although all webs were obviously dom-
inated by mutualists, several datasets may contain nonmutualistic
species, e.g., nectar robbers and seed predators. Twenty-seven da-
tasets were obtained from the Interaction Web Database (http://
www.nceas.ucsb.edu/interactionweb). For each network containing
a total of Iplant and Janimal species, we obtained the two-dimen-
sional Shannon entropy for the observed association matrix  as
j=1pij ,ln pij ;with X
In this equation, irepresents one plant species and jone animal
species. The number of interactions between iand j(a
), e.g., the
number of recorded visits of pollinator jon plant i, is divided by
the total interaction frequencies recorded for the entire web, thus
Our specialization index H0
between the minimum
and maximum entropy for associations leading to the same matrix
row and column totals as
For quantiﬁcation of the degree of specialization of each species
(say plant i), the proportional distribution of the interactions with
each animal (j), p0
, was compared with the proportion of the total
number of interactions where jwas involved, q
, by using the Kull-
This measure was normalized as
Specialization of the plant community (guild level) was obtained as
the weighted mean hd0
i, for which each plant species iwas weighted
by its total number of interactions. Specialization of animals was cal-
culated in the same way (d0
i). Maximum and minimum
values for H
, and d
were computed algorithmically by using
the ﬁxed total number of interactions of each species as a constraint
. Resulting H0
, and d0
range between 0.0 for extreme gener-
alization and 1.0 for extreme specialization. For each network, H
was compared to a null model (randomly associating all species
with the total number of interactions being ﬁxed per species, 10
permutations) by using an established algorithm (, see ).
Fixed marginals have been advocated as suitable constraints for
null hypotheses in qualitative webs [6, 39]. In addition, we suggest
that total interaction frequencies may better reﬂect variation in ani-
mal activity or plant resource availability for the actual associations
than would external estimates of local population densities .
Such independent measures of the species’ local abundances for
both parties have not been provided by most empirical studies so
far. All calculations can be performed online at http://itb.biologie.
To analyze the relationship between specialization and interaction
frequency at the species level, we calculated linear correlation coef-
ﬁcients between log(total number of interactions) and arcsin(Od0
across all species of a guild per network and then quantiﬁed the
combined mean effect size from all networks of the same type by
using standard meta-analysis tools (MetaWin 2.0; Fisher’s z-trans-
formation, sample size as number of species, ﬁxed effects); 95%
conﬁdence intervals were based on bootstrapping with 999 itera-
tions, bias-corrected. To reduce a bias due to single, very large net-
works, we removed, prior to analysis, datasets where the number of
species was more than twice as large as in the second-largest
network (four cases).
Supplemental Data include additional results and data sources,
three ﬁgures, and one table and are available with this article online
We thank D. Va
´zquez, K. Fiedler, and K.E. Linsenmair for discussion
and helpful comments on earlier versions of the manuscript and the
Interaction Web Database for providing several of the networks
analyzed here. Field work of the N.B., T.H., and B.F. was supported
by the German Research Foundation (DFG).
Received: October 25, 2006
Revised: December 6, 2006
Accepted: December 6, 2006
Published online: February 1, 2007
1. May, R.M. (1972). Will a large complex system be stable? Nature
2. Jordano, P. (1987). Patterns of mutualistic interactions in polli-
nation and seed dispersal: Connectance, dependence asymme-
tries, and coevolution. Am. Nat. 129, 657–677.
3. Dunne, J.A., Williams, R.J., and Martinez, N.D. (2002). Food-web
structure and network theory: The role of connectance and size.
Proc. Natl. Acad. Sci. USA 99, 12917–12922.
4. Thompson, J.N. (2005). The Geographic Mosaic of Coevolution
(Chicago: University of Chicago Press).
5. Bascompte, J., Jordano, P., and Olesen, J.M. (2006). Asymmet-
ric coevolutionary networks facilitate biodiversity maintenance.
Science 312, 431–433.
6. Montoya, J.M., Pimm, S.L., and Sole
´, R.V. (2006). Ecological
networks and their fragility. Nature 442, 259–264.
7. Goldwasser, L., and Roughgar den, J. (1997). Sampling effects
and the estimation of food-web properties. Ecology 78, 41–54.
8. Olesen, J.M., and Jordano, P. (2002). Geographic patterns in
plant-pollinator mutualistic networks. Ecology 83, 2416–2424.
´zquez, D.P., and Aizen, M.A. (2003). Null model analyzes of
specialization in plant-pollinator interactions. Ecology 84,
10. Blu¨ thgen, N., Menzel, F., and Blu¨ thgen, N. (2006). Measuring
specialization in species interaction networks. BMC Ecol. 6,9.
11. Waser, N.M., Chittka, L., Price, M.V., Williams, N.M., and Oller-
ton, J. (1996). Generalization in pollination systems, and why it
matters. Ecology 77, 1043–1060.
Mutualistic Interaction Webs
12. Beckerman, A.P., Petchey, O.L., and Warren, P.H. (2006). Forag-
ing biology predicts food web complexity. Proc. Natl. Acad. Sci.
USA 103, 13745–13749.
˜es, P.R.J., Rico-Gray, V., dos Reis, S.F., and Thompson,
J.N. (2006). Asymmetries in specialization in ant-plant mutualis-
tic networks. Proc. R. Soc. Lond. B. Biol. Sci. 273, 2041–2047.
14. Berlow, E.L., Neutel, A.-M., Cohen, J.E., de Ruiter, P.C., Eben-
man, B., Emmerson, M., Fox, J.W., Jansen, V.A.A., Jones, J.I.,
Kokkoris, G.D., et al. (2004). Interaction strengths in food
webs: Issues and opportunities. J. Anim. Ecol. 73, 585–598.
15. Paine, R.T. (1988). Food webs: Road maps of interactions or
grist for theoretical development? Ecology 69, 1648–1654.
16. Bersier, L.F., Banasek-Richter, C., and Cattin, M.F. (2002). Quan-
titative descriptors of food-web matrices. Ecology 83, 2394–
´zquez, D.P., Morris, W.F., and Jordano, P. (2005). Interaction
frequency as a surrogate for the total effect of animal mutualists
on plants. Ecol. Lett. 8, 1088–1094.
18. Sahli, H.F., and Conner, J.K. (2006). Characterizing ecological
generalization in plant-pollination systems. Oecologia 148,
19. Hurlbert, S.H. (1978). Measurement of niche overlap and some
relatives. Ecology 59, 67–77.
20. Blu¨ thgen, N., Verhaagh, M., Goitı
´a, W., and Blu¨thgen, N. (2000).
Ant nests in tank bromeliads – an example of non-speciﬁc inter-
action. Insectes Soc. 47, 313–316.
21. Wheelwright, N.T., and Orians, G.H. (1982). Seed dispersal by
animals: Contrasts with pollen dispersal, problems of terminol-
ogy, and constraints on coevolution. Am. Nat. 119, 402–413.
22. Johnson, S.D., and Steiner, K.E. (2000). Generalization versus
specialization in plant pollination systems. Trends Ecol. Evol.
23. Pellmyr, O. (2002). Pollin ation by animals. In Plant-Animal Inter-
actions: An Evolutionary Approach, C.M. Herrera and O. Pellmyr,
eds. (Oxford, UK: Blackwell), pp. 157–184.
24. Russo, S.E., and Augspurger, C.K. (2004). Aggregated seed dis-
persal by spider monkeys limits recruitment to clumped pat-
terns in Virola calophylla. Ecol. Lett. 7, 1058–1067.
25. Higgins, S.I., and Richardson, D.M. (1999). Predicting plant
migration rates in a changing world: The role of long-distance
dispersal. Am. Nat. 153, 464–475.
26. Hovestadt, T., Messner, S., and Poethke, H.J. (2001). Evolution
of reduced dispersal mortality and ‘fat-tailed’ dispersal kernels
in autocorrelated landscapes. Proc. R. Soc. Lond. B. Biol. Sci.
27. Fiala, B., Jakob, K.E., Maschwitz, U., and Linsenmair, K.E.
(1999). Diversity, evolutionary specialization and geographic
distribution of a mutualistic ant-plant complex Macaranga and
Crematogaster in South East Asia. Biological Journal of the Lin-
nean Society 66, 305–331.
28. Heil, M., and McKey, D. (2003). Protective ant-plant interactions
as model systems in ecological and evolutionary research.
Annual Review of Ecology, Evolution, and Systematics 34,
29. Ollerton, J. (2006). ‘‘Biological barter’’: Patterns of specialization
compared across different mutualisms. In Plant-Pollinator Inter-
actions: From Specialization to Generalization, N.M. Waser and
J. Ollerton, eds. (Chicago: University of Chicago Press), pp.
30. Yu, D.W., and Davidson, D.W. (1997). Experimental studies of
species-speciﬁcity in Cecropia-ant relationships. Ecol. Monogr.
31. Blu¨ thgen, N., and Fiedler, K. (2004). Competition for composi-
tion: Lessons from nectar-feeding ant communities. Ecology
32. Heil, M., Rattke, J., and Boland, W. (2005). Postsecretory hydro-
lysis of nectar sucrose and specialization in ant/plant mutual-
ism. Science 308, 560–563.
´nek, M., and Stary
´, P. (1979). Connectance in real biotic
communities and critical values for stability of model ecosys-
tems. Nature 280, 311–313.
34. Montoya, J.M., and Sole
´, R.V. (2003). Topological properties
of food webs: From real data to community assembly models.
Oikos 102, 614–622.
35. Darwin, C. (1862). On the Various Con trivances by which British
and Foreign Orchids Are Fertilised by Insects (London: John
36. Jaenike, J. (1990). Host speci alization in phytophagous insects.
Annu. Rev. Ecol. Syst. 21, 243–273.
37. Bronstein, J.L., Alarco
´n, R., and Geber, M. (2006). The evolution
of plant-insect mutualisms. New Phytol. 172, 412–428.
38. Pateﬁeld, W.M. (1981). An efﬁcient method of generating ran-
dom RxC tables with given row and column totals. Appl. Stat.
39. Gotelli, N.J. (2000). Null model analysis of species co-occur-
rence patterns. Ecology 81, 2606–2621.
Supplemental Data S1
Specialization, Constraints, and
in Mutualistic Networks
¨thgen, Florian Menzel, Thomas Hovestadt,
Brigitte Fiala, and Nils Blu
Supplemental Results and Data Sources
All 51 mutualistic networks used in this analysis and their
sources are listed in Table S1. Previous analyses invari-
ably demonstrated a hyperbolic decline of the qualita-
tive ‘‘connectance’’ index with increasing network size,
a pattern that has considerably constrained the useful-
ness of this qualitative measure [S1]. Connectance has
been deﬁned as the number of realized links divided
by the number of cells in the association matrix (plant
species 3animal species) [S2, S3]. For the dataset
used here, connectance declined signiﬁcantly with in-
creasing matrix size within each of the network types
(all Spearman r
%20.64, p %0.04) except for the small
set of ant-nectar webs (r
=20.59, p = 0.13) (Figure S1).
Asymmetries in the average quantitative specializa-
tion of plants and animals are a function of the network
form (ratio of plant to animal species), an effect that is
particularly pronounced in random associations (Fig-
ure S2). Neither the reciprocity of average specialization
levels between guilds nor the effect of architectural con-
straints on specialization is conﬁned to the quantitative
metrics used here. Specialization as deﬁned in the tradi-
tional, qualitative sense (i.e., number of partner species)
is strictly dependent on the web architecture: In bal-
anced webs, when an equal number of animal species
(J) and plant species (I) is involved (I=J), the mean num-
ber of associated partners (links) for each animal (LJ)
equals that for plants (LI). Consequently, qualitative spe-
cialization is a direct function of the network asymmetry,
as LJ=LI=I=J. Differences in the number of partner spe-
cies also affect quantitative ‘‘dependences’’ between
plants and animals that were used in previous analyses
by Bascompte et al. [S4], an effect that has not been
shown previously. The dependence of plant ion animal
) and the reciprocal dependence of jon i(b
) is esti-
mated from of the interaction frequency (a
and bji =aij
It can be shown that the average dependence of a
guild across all putative interactions (I$J, including all
cases where a
= 0) is a simple function of the number
of species, because
Jand bji =1=ðIJÞ
Consequently, in square matrices (I=J), expected av-
erage differences between the two parties (b
zero, but become stronger with increasing asymmetry
in rectangular networks in both directions (I>Jor J>
I). This effect is also evident across empirical webs
S1. Blu¨ thgen, N., Menzel, F., and Blu¨thgen, N. (2006). Meas uring
specialization in species interaction networks. BMC Ecol. 6,9.
S2. Jordano, P. (1987). Patterns of mutualistic interactions in polli-
nation and seed dispersal: Connectance, dependence asym-
metries, and coevolution. Am. Nat. 129, 657–677.
S3. Olesen, J.M., and Jordano, P. (2002). Geographic patterns in
plant-pollinator mutualistic networks. Ecology 83, 2416–2424.
Figure S1. Relationship between Network Size and Connectance
Qualitative connectance (proportion of realized links of the total number of possible links) of 51 networks plotted over network size (plant plus
animal species, log scale). Networks include pollination (yellow), seed-dispersal (black), ant-myrmecophyte (green), and ant-nectar plant (red)
Figure S2. Relationship between Web Asymmetry and Specialization
Each point represents the mean value (6standard deviation [SD]) of 100 randomized networks simulated from the set of 51 natural networks
(maintaining the same total interaction frequencies per species). Asymmetry of the number of pla nt (I) and animal (J) species in each web is given
as (J2I)/(I+J), asymmetry in specialization betw eenplants iand animals jas (hd0
i). Regression line (r=20.97) was used for
Figure 4 in the main text and for calculating residuals of real networks.
Figure S3. Relationship between Web Asymmetry and Dependence
Asymmetry of the number of plant (I) and animal (J) species in each web is given as (J2I)/(I+J), and asymmetry in dependence between plants i
and animals jis given as (b
). Data were obtained from 26 networks [S4], including pollination (yellow) and seed-dispersal (black)
associations (means and 95% conﬁdence intervals for all realized interactions). The mean dependence asymmetry across all realized interac-
tions of a web (a
> 0) is strongly linearly predicted by network asymmetry (r= 0.97, p < 0.0001).
Table S1. Mutualistic Networks Analyzed
Number Reference Taxonomic Focus Location Plants Animals mhd0
1 Barrett and Helenurm [S5] (several families) Canada 12 102 550 0.61 0.40 0.55
2 Elberling and Olesen [S6] (several families) Sweden 23 118 383 0.48 0.29 0.33
3 Inouye and Pyke [S7] (several families) Australia 42 91 1459 0.54 0.59 0.60
4 Kato et al. [S8] (several families) Japan 91 679 2392 0.65 0.37 0.48
5 Memmott [S9] (several families) Britain 25 79 2183 0.27 0.16 0.24
6 Mosquin and Martin [S10] (several families) Canada 11 18 134 0.51 0.36 0.46
7 Motten [S11] (several families) USA 13 44 2225 0.43 0.34 0.43
8 Olesen et al. [S12] (several families) Azores 10 12 1139 0.50 0.46 0.53
9 Olesen et al. [S12] (several families) Mauritius 14 13 1512 0.19 0.25 0.38
10 Ollerton et al. [S13] Asclepiadaceae South Africa 9 56 594 0.43 0.27 0.43
11 Schemske et al. [S14] (several families) USA 7 32 299 0.33 0.18 0.34
12 Small [S15] (several families) Canada 13 34 992 0.54 0.39 0.55
13 Ssymank [S16] Syrphidae Germany 88 75 4837 0.45 0.47 0.47
´zquez and Simberloff [S17] (several families) Argentina 10 29 677 0.65 0.63 0.71
´zquez and Simberloff [S17] (several families) Argentina 9 33 613 0.64 0.55 0.78
´zquez and Simberloff [S17] (several families) Argentina 9 27 1130 0.90 0.65 0.85
´zquez and Simberloff [S17] (several families) Argentina 10 29 515 0.56 0.39 0.60
´zquez and Simberloff [S17] (several families) Argentina 8 35 719 0.53 0.26 0.57
´zquez and Simberloff [S17] (several families) Argentina 8 26 286 0.78 0.69 0.74
´zquez and Simberloff [S17] (several families) Argentina 7 24 761 0.65 0.84 0.79
´zquez and Simberloff [S17] (several families) Argentina 8 27 592 0.50 0.52 0.70
22 Beehler [S18] Birds Papua N.G. 31 9 1189 0.22 0.28 0.26
23 Engel [S19] Mammals Kenya 219 33 3730 0.20 0.43 0.39
24 Hovestadt [S20] Birds, Mammals Ivory Coast 34 48 17575 0.16 0.15 0.18
25 Kaufmann [S21] Ants West Malaysia 33 51 448 0.27 0.28 0.24
26 Poulin et al. [S22] Birds, Miconia,
Panama 17 20 492 0.16 0.17 0.21
27 Snow and Snow [S23] Birds Trinidad 65 14 2180 0.20 0.28 0.30
28 Snow and Snow [S24] Birds Britain 29 19 19946 0.21 0.29 0.30
29 Sorensen [S25] Birds Britain 11 14 7434 0.38 0.21 0.47
30 Blu¨thgen et al. [S26] Bromeliaceae Venezuela 4 13 39 0.23 0.14 0.23
31 N.B., unpublished data Melastomataceae Ecuador 5 5 71 0.16 0.23 0.27
32 Cabrera and Jaffe
´[S27] Melastomataceae Venezuela 7 14 111 0.44 0.23 0.40
33 Davidson et al. [S28] (several families) Peru 8 18 242 0.89 0.76 0.89
34 Dejean et al. [S29] Bromeliaceae,
Mexico 9 51 388 0.51 0.26 0.47
35 Fiala et al. [S30] Macaranga Borneo 9 7 349 0.72 0.77 0.80
36 Fiala et al. [S30] Macaranga Borneo 10 8 173 0.74 0.87 0.86
37 Fiala et al. [S30] Macaranga Borneo 6 6 98 0.74 0.76 0.84
38 Fiala et al. [S30] Macaranga Sumatra 7 4 78 0.57 0.79 0.80
39 Fiala et al. [S30] Macaranga West Malaysia 4 5 183 0.71 0.77 1.00
40 B.F., unpublished data Macaranga Borneo 6 7 88 0.96 0.97 0.99
41 B.F., unpublished data Macaranga Borneo 7 5 88 0.53 0.72 0.83
42 Fonseca and Ganade [S31] (several families) Brazil 16 25 417 0.72 0.78 0.80
43 Yu and Davidson [S32] Cecropia Peru 7 4 155 0.45 0.54 0.61
44 Blu¨thgen et al. [S33] Philodendron, Dioclea Venezuela 6 53 180 0.36 0.11 0.31
45 Blu¨thgen et al. [S34] (several fam.), incl.
Australia 51 41 644 0.18 0.20 0.13
46 B.F., unpublished data (several families) Borneo 15 14 267 0.14 0.17 0.19
48 B.F., unpublished data (several families) Borneo 22 28 324 0.34 0.21 0.23
47 B.F., unpublished data (several families) West Malaysia 11 16 121 0.37 0.26 0.33
49 B.F., unpublished data (several families) West Malaysia 24 35 315 0.31 0.19 0.21
50 Hossaert-McKey et al. [S35] Passiﬂora, Mimosa French Guiana 3 37 1661 0.24 0.08 0.23
51 Whalen and Mackay [S36] Euphorbiaceae Papua N.G. 5 17 246 0.22 0.13 0.24
References, taxonomic focus, number of plant and animal species, total interactions (m), weighted mean specialization of plants and animals
i), and overall specialization (H0
) are given.
S4. Bascompte, J., Jordano, P., and Olesen, J.M. (2006). Asym-
metric coevolutionary networks facilitate biodiversity mainte-
nance. Science 312, 431–433.
S5. Barrett, S.C.H., and Helenurm , K. (1987). The reproductive biol-
ogy of boreal forest herbs.1. Breeding systems and pollination.
Can. J. Bot. 65, 2036–2046.
S6. Elberling, H., and Olesen, J.M. (1999). The structure of a high
latitude plant-ﬂower visitor system: The dominance of ﬂies.
Ecography 22, 314–323.
S7. Inouye, D.W., and Pyke, G.H. (1988). Pollination biology in the
Snowy Mountains of Australia: Comparisons with montane
Colorado, USA. Aust. J. Ecol. 13, 191–210.
S8. Kato, M., Makutani, T., Inoue, T., and Itino, T. (1990). Insect-
ﬂower relationship in the primary beech forest of Ashu, Kyoto:
An overview of the ﬂowering phenology and seasonal pattern
of insect visits. Contributions from the Biological Laboratory,
Kyoto University 27, 309–375.
S9. Memmott, J. (1999). The structure of a plant-pollinator food
web. Ecol. Lett. 2, 276–280.
S10. Mosquin, T., and Martin, J.E.H. (1967). Observations on the
pollination biology of plants on Melville Island, N.W.T., Canada.
Can. Field Nat. 81, 201–205.
S11. Motten, A.F. (1986). Pollination ecology of the spring wildﬂower
community of a temperate deciduous forest. Ecol. Monogr. 56,
S12. Olesen, J.M., Eskildsen, L.I., and Venkatasamy, S. (2002). Inva-
sion of pollination networks on oceanic islands: Importance of
invader complexes and endemic super generalists. Divers.
Distrib. 8, 181–192.
S13. Ollerton, J., Johnson, S.D., Cranmer, L., and Kellie, S. (2003).
The pollination ecology of an assemblage of grassland ascle-
piads in South Africa. Ann. Bot. (Lond.) 92, 807–834.
S14. Schemske, D.W., Willson, M.F., Melampy, M.N., Miller, L.J.,
Verner, L., Schemske, K.M., and Best, L.B. (1978). Flowering
ecology of some spring woodland herbs. Ecology 59, 351–366.
S15. Small, E. (1976). Insec t pollinators of the Mer Bleue peat bog of
Ottawa. Can. Field Nat. 90, 22–28.
S16. Ssymank, A. (2001). Vegetation und Blu¨ tenbesuchende
Insekten in der Kulturlandschaft (Bonn: Bundesamt fu¨r
´zquez, D.P., and Simberloff, D. (2003). Changes in interac-
tion biodiversity induced by an introduced ungulate. Ecol.
Lett. 6, 1077–1083.
S18. Beehler, B. (1983). Frugivory and polygamy in birds of para-
dise. Auk 100, 1–12.
S19. Engel, T.R. (1999). Seed dispersal and forest regeneration in
a tropical lowland biocoenosis (Shimba Hills, Kenya). PhD
thesis, University of Bayreuth, Bayreuth, Germany.
S20. Hovestadt, T. (1997). Fruchtmerkmale, endozoochore Same-
nausbreitung und ihre Bedeutung fu¨r die Zusammensetzung
der Pﬂanzengemeinschaft. PhD thesis, University of Wu¨ rzburg,
Wu¨ rzburg, Germany.
S21. Kaufmann, E. (2002). Southeast Asian ant-gardens. PhD thesis,
University of Frankfurt (Main), Frankfurt (Main), Germany.
S22. Poulin, B., Wright, S.J., Lefebvre, G., and Calderon, O. (1999).
Interspeciﬁc synchrony and asynchrony in the fruiting phenol-
ogies of congeneric bird-dispersed plants in Panama. J. Trop.
Ecol. 15, 213–227.
S23. Snow, B.K., and Snow, D.W. (1971). The feeding ecology of
tanagers and honeycreeperes in Trinidad. Auk 88, 291–322.
S24. Snow, B.K., and Snow, D.W. (1988). Birds and Berries (Calton,
S25. Sorensen, A.E. (1981). Interactions between birds and fruit in
a temperate woodland. Oecologia 50, 242–249.
S26. Blu¨ thgen, N., Verhaagh, M., Goitı
´a, W., and Blu¨ thgen, N. (2000) .
Ant nests in tank bromeliads – an example of non-speciﬁc in-
teraction. Insectes Soc. 47, 313–316.
S27. Cabrera, M., and Jaf fe, K. (1994). A trophic mutualism between
the myrmecophytic Melastomataceae Tococa guianensis
Aublet and an Azteca ant species. Ecotropicos 7, 1–10.
S28. Davidson, D.W., Snelling , R.R., and Longino, J.T. (1989). Com-
petition among ants for myrmecophytes and the signiﬁcance
of plant trichomes. Biotropica 21, 64–73.
S29. Dejean, A., Olmsted, I., and Snelling, R.R. (1995). Tree-
epiphyte-ant relationships in the low inundated forest of Sian
Ka’an Biosphere Reserve, Quintana Roo, Mexico. Biotropica
S30. Fiala, B., Jakob, K.E., Maschwitz, U., and Linsenmair, K.E.
(1999). Diversity, evolutionary specialization and geographic
distribution of a mutualistic ant-plant complex Macaranga
and Crematogaster in South East Asia. Biological Journal of
the Linnean Society 66, 305–331.
S31. Fonseca, C.R., and Ganade, G. (1996). Asymmetries, compart-
ments and null interactions in an Amazonian ant-plant commu-
nity. J. Anim. Ecol. 65, 339–347.
S32. Yu, D.W., and Davidson, D.W. (1997). Experimental studies of
species-speciﬁcity in Cecropia-ant relationships. Ecol.
Monogr. 67, 273–294.
S33. Blu¨ thgen, N., Verhaagh, M., Goitı
´a, W., Jaffe
´, K., Morawetz, W.,
and Barthlott, W. (2000). How plants shape the ant community
in the Amazonian rainforest canopy: The key role of extraﬂoral
nectaries and homopteran honeydew. Oecologia 125, 229–240.
S34. Blu¨ thgen, N., Stork, N.E., and Fiedler, K. (2004). Bottom-up
control and co-occurrence in complex communities: Honey-
dew and nectar determine a rainforest ant mosaic. Oikos
S35. Hossaert-McKey, M., Orivel, J., Labeyrie, E., Pascal, L., Dela-
bie, J.H.C., and Dejean, A. (2001). Differential associations
with ants of three co-occurring extraﬂoral nectary-bearing
plants. Ecoscience 8, 325–335.
S36. Whalen, M.A., and Mackay, D.A. (1988). Patterns of ant and
herbivore activity on ﬁve understorey euphorbiaceous saplings
in submontane Papua New Guinea. Biotropica 20, 294–300.