Content uploaded by Malena Sabatino
Author content
All content in this area was uploaded by Malena Sabatino on Mar 23, 2023
Content may be subject to copyright.
LETTER The phylogenetic structure of plant–pollinator networks
increases with habitat size and isolation
Marcelo A. Aizen,
1
* Gabriela
Gleiser,
1
Malena Sabatino,
2
Luis J.
Gilarranz,
3
Jordi Bascompte
4
and
Miguel Verd
u
5
Abstract
Similarity among species in traits related to ecological interactions is frequently associated with
common ancestry. Thus, closely related species usually interact with ecologically similar partners,
which can be reinforced by diverse co-evolutionary processes. The effect of habitat fragmentation
on the phylogenetic signal in interspecific interactions and correspondence between plant and ani-
mal phylogenies is, however, unknown. Here, we address to what extent phylogenetic signal and
co-phylogenetic congruence of plant–animal interactions depend on habitat size and isolation by
analysing the phylogenetic structure of 12 pollination webs from isolated Pampean hills. Phyloge-
netic signal in interspecific interactions differed among webs, being stronger for flower-visiting
insects than plants. Phylogenetic signal and overall co-phylogenetic congruence increased indepen-
dently with hill size and isolation. We propose that habitat fragmentation would erode the phylo-
genetic structure of interaction webs. A decrease in phylogenetic signal and co-phylogenetic
correspondence in plant–pollinator interactions could be associated with less reliable mutualism
and erratic co-evolutionary change.
Keywords
Area effect, co-phylogenetic correspondence, habitat islands, isolation, mutualistic networks,
Pampas, phylogenetic structure, pollination webs.
Ecology Letters (2016) 19: 29–36
INTRODUCTION
Habitat fragmentation can drive the direct extinction not only
of species, but also of the interspecific interactions that shape
the web of life (Bascompte & Jordano 2007; Aizen et al.
2012). In turn, interaction loss causes the disruption of diverse
ecological processes, which can further affect both short-term
species survivorship (e.g. Pauw 2007) and long-term evolution-
ary change (e.g. Galetti et al. 2013). Although consideration
of evolutionary and co-evolutionary processes adds an impor-
tant dimension to the values of conservation (Crandall et al.
2000; Moritz 2002), their study in the context of meta-com-
munities persisting in fragmented habitats is still in its infancy
(Peralta et al. 2015). We propose that the assessment of
the phylogenetic structure of interaction webs like plant–
pollinator networks, across habitat islands, can provide infor-
mation on what landscape factors and habitat characteristics
contribute to the preservation of long-term (co)evolutionary
processes.
Phenotypic traits shaping plant–animal interactions, particu-
larly plant–animal mutualisms, are often structured phyloge-
netically. As a consequence, plant or animal species sharing
common ancestry –and thus similar traits –tend to interact
with largely overlapping ecological assemblages of animal or
plant species, respectively, with matching traits (Rezende et al.
2007; G
omez et al. 2010). This type of phylogenetic structure
based on a total or partial phenotypic matching between
interacting partners can result from one-on-one and even mul-
ti-specific co-evolutionary processes (Marussich & Machado
2007; Segraves 2010; Guimar~
aes et al. 2011). Similarly, these
different modes of co-evolution, i.e. pair-wise and multi-
species, will strengthen the correspondence between plant and
animal phylogenies (e.g. Legendre et al. 2002). In addition,
trait matching between species that did not co-evolve, but co-
evolved elsewhere with close relatives of the species with
which they interact at present may also contribute to maintain
and reinforce the phylogenetic structure of plant and animal
interactions (e.g. Pearse et al. 2013). Therefore, we interpret
the presence of phylogenetic structure in mutualistic webs as
evidence of temporally persistent interactions between pairs or
(somewhat shifting) sets of partner species with co-adapted
traits, and its absence as evidence of less reliable mutualism
(i.e. mutualistic interactions not sustained over time) leading
to erratic selective pressures.
Both the phylogenetic signal in ecological interactions at
each trophic level, as well as the overall plant–animal phyloge-
netic congruence can be eroded under a scenario of habitat
fragmentation for at least two reasons. First, plant and animal
populations trapped in small habitat islands are more prone
to random extinction due to demographic and genetic
1
Laboratorio Ecotono-CRUB, Universidad Nacional del Comahue and INI-
BIOMA, Quintral 1250, 8400 San Carlos de Bariloche, R
ıo Negro, Argentina
2
EEA [Estaci
on Experimental Agropecuaria] Balcarce, INTA [Instituto Nacional
de Tecnolog
ıa Agropecuaria], CC 276, 7620 Balcarce, Buenos Aires, Argentina
3
Integrative Ecology Group, Estaci
on Biol
ogica de Do~
nana, CSIC, Calle
Am
erico Vespucio s/n, 41092 Sevilla, Spain
4
Institute of Evolutionary Biology and Environmental Studies, University of
Zurich, Winterthurerstrasse 190, 8057 Zurich, Switzerland
5
Centro de Investigaciones sobre Desertificaci
on (CIDE CSIC-UV-GV), Apartado
Oficial, E-46113 Moncada, Valencia, Spain
*Correspondence: E-mail: maizen@comahue-conicet.gob.ar
©2015 John Wiley & Sons Ltd/CNRS
Ecology Letters, (2016) 19: 29–36 doi: 10.1111/ele.12539
bottlenecks (Young et al. 1996; Tscharntke et al. 2002).
Increasing random population extinction would, in turn,
decrease the strength of the phylogenetic signal in assemblage
composition, because this factor would reduce the determinis-
tic role of trait matching in shaping plant–pollinator interac-
tions (Stang et al. 2006; Schleuning et al. 2015). Second,
meta-community dynamics –characteristic of fragmented
habitats –implies not only the random extinction of small
local populations, but also the colonisation of a focal habitat
island by random samples of non-resident and, probably,
transient species from nearby habitat islands (Leibold et al.
2004). These transient species are expected to establish rather
facultative interactions, particularly with generalist species
(Aizen et al. 2012), which would also weaken the phylogenetic
signal in ecological interactions and the degree of phylogenetic
match between plants and animals. Therefore, the overall phy-
logenetic and co-phylogenetic structure of pollination webs
should be better preserved in large and isolated habitat islands
than in small and central ones, which are subjected to higher
species and interaction turnover (Jamoneau et al. 2012).
Here, we explore the role of habitat size and isolation on
the phylogenetic signal and co-phylogenetic congruence of
plant–pollinator interactions by analysing 12 pollination webs
from the ‘sierras’ of the central Argentine Pampas (Sabatino
et al. 2010; Aizen et al. 2012). These ancient hills, which range
from tens to thousands of hectares, can be considered as
‘habitat islands’ as they were connected by a matrix of natural
grassland prior to European colonisation, but are now sur-
rounded by a matrix devoted to intensive agriculture and iso-
lated from each other by distances ranging from a few
hundred metres to several kilometres (Fig. 1). Because of their
relatively recent isolation and because they remain mostly
untilled, these sierras still preserve many floristic elements that
were formerly common in the surrounding plains and else-
where in southern South America, including several species
from lineages of Gondwanan origin (Cabrera 1994) (see also
Supporting Materials and Methods, Landscape’s human
transformation).
Previous studies in this system (Sabatino et al. 2010; Aizen
et al. 2012; Gilarranz et al. 2015) showed that trends in num-
ber of species and interactions, as well as in overall network
architecture (i.e. nestedness) could be predicted based on
meta-community principles (Leibold et al. 2004). Here, we
move one step forward and ask whether a meta-community
dynamics could also induce changes in the phylogenetic and
co-phylogenetic structure of these pollination webs. In partic-
ular, on the basis of the hypotheses stated above, we test the
predictions that both phylogenetic signal and co-phylogenetic
congruence in interactions between plants and their flower vis-
itors increase with habitat area and isolation. We found sup-
port for these expectations, which also suggests that the
process of habitat fragmentation has community-wide implica-
tions for both the dynamics of mutualisms and the integrity
of co-evolutionary processes.
MATERIALS AND METHODS
Study system and dataset
We surveyed pollination networks from 12 out of a total of
18 sierras located between Mar del Plata (37
o
580S, 57
o
350W)
and Balcarce (37
o
500S, 58
o
150W), Buenos Aires Province,
Argentina (data available from the Dryad Digital Repository:
http://dx.doi.org/10.5061/dryad.cr3ft). These sierras are part
of the Tandilian orographic system dating back to the lower
Paleozoic, a system which comprises about 24 isolated ortho-
quartzitic hills up to c. 500 m in altitude jutting out from the
loessic (Quarternary) Pampas’ plains, lying in the southeast of
the ‘Pampeano Austral’ biogeographic district (Cabrera 1994).
The 12 study sierras represent a wide range in size (from 13
to 2100 ha; Table S1) and isolation (between 0.8 and 9.1 km
from their nearest neighbour), which is typical of the entire
orographic system. The climate of the region is temperate with
warm summers (mean January temperature 20.8 °C) and mild
winters (mean July temperature 5.2 °C).
The sierras have rich vegetation characterised by a gentle
rocky basal slope dominated by shrubs, herbs and geophytes,
a barely vegetated steep scarp and a flat top with a mosaic of
exposed bedrock and loessic patches dominated by grasses.
The most abundant and diverse plant families are insect polli-
nated (i.e. Asteraceae, Apiaceae, Fabaceae and Scrophulari-
aceae). Despite intensive apiculture with European honey bees
(Apis mellifera) in the region, the sierras support a rich
flower-visitor community that comprises mainly insects (Hy-
menoptera, followed by Diptera, Coleoptera and Lepidoptera)
and one species of vertebrate, the hummingbird Chlorostilbon
aureoventris (Aizen et al. 2012). The matrix that nowadays
surrounds the sierras is an intensively managed agroecosystem
devoted principally to the cultivation of soybean (about 70%
of the agriculture area). However, other crops such as sun-
flower, wheat, corn, potato and canola are still cultivated in
the region. Cattle roamed freely and fires were set frequently
Figure 1 The study hilly landscape of the Argentine Pampas. The
photograph shows ‘sierras’ (a and c) and the surrounding agricultural
matrix (b). The conical sierra at the back (a) is ‘Amarante’, a hill of
190 ha and 361 m a.s.l, as seen from ‘La Chata’, a hill of 229 ha and
357 m a.s.l, which is 2.6 km away. The terrain in the foreground (c) at La
Chata is the type of habitat in which we surveyed plants and flower
visitors.
©2015 John Wiley & Sons Ltd/CNRS
30 M. A. Aizen et al. Letter
to promote vegetation re-growth in all the study sierras except
for Difuntito, a small-fenced hill (Table S1). Cattle had been
excluded and fire suppressed at Difuntito for at least the pre-
ceding 18 years (Sabatino et al. 2010).
Field work was conducted during the 2007–2008 Austral
flowering season (October–April). On each sierra, we delim-
ited and sampled an area of 0.5 ha on the north-facing slope,
about 200 m from the edge of the nearest agricultural field.
This sun-exposed slope exhibits the highest plant diversity
among the sierra habitats. Also, while our protocol avoided
edges, proximity to the agricultural fields increased the chance
to sample plant and pollinator species that used to be com-
mon, or at least present, in the surrounding matrix before
agriculturalisation. Within each 0.5 ha area, we set two paral-
lel 100-m transects, 50 m apart and along each transect we
established five permanent 1-m radius plots c. 25 m apart (i.e.
10 plots per sierra). Within each plot, we identified all plants
in flower and recorded all flower visitors that contacted floral
sexual organs during a 15-min period. All plots from a given
sierra were sampled consecutively between 09:00 and 18:00 h
and each sierra was sampled an average of 10 times through-
out the flowering season, once every 2 weeks, recording a
total of 13 174 flower visitors during 318 h of observation dis-
tributed over 127 days. All flower visitors were morphotyped
and identified with the aid of a reference collection and spe-
cialist help at least to the family level. We could identify to
species or genus 52 or 80% of all flower visitors respectively.
Neither the total number of flower visitors recorded in each
sierra (a measure of sampling effort) nor the incidence of alien
species (c. 10% of all species) was influenced by sierra size or
isolation (Sabatino et al. 2010).
Sierra area and connectivity
The 18 sierras from our study area, including the 12 sierras
surveyed, were digitalised from a Google-Earth image. We
measured the area of each sierra to the nearest hectare. Then,
we constructed a spatial network connecting the sierras. Two
patches (in this case two sierras) were connected if the lineal
distance between them (i.e. edge to edge) was smaller than a
threshold distance. From this spatial network map, we esti-
mated the betweenness centrality of each patch as a direct
measure of connectivity or an inverse measure of isolation.
Patch betweenness centrality quantifies how well a given
habitat patch is connected to other patches by taking into
account the spatial configuration of the entire patch network.
For a focal habitat patch, patch betweenness centrality is
proportional to the number of all possible paths connecting
all pairs of habitat patches which pass through that patch
(Freeman 1977). This inverse measure of isolation, which is
based on graph theory, suits to studies concerned with meta-
population and meta-community dynamics (Urban & Keitt
2001; Gilarranz et al. 2015). We chose a threshold distance
of 12 km, a value that maximised the association (i.e. the
coefficient of determination) between phylogenetic structure
and centrality (see Supporting Materials and Methods,
Threshold distance), at least for the two of the three focal
phylogenetic correspondence variables for which centrality
had a significant effect (Table S2). This threshold distance
also maximised the association between nestedness –an
important structural network property –and centrality (Gi-
larranz et al. 2015). From a biological perspective, this
threshold is just above the maximum foraging distances
known for most bees (Greenleaf et al. 2007). We considered
the paths connecting all the 18 sierras in the landscape to
estimate the centrality (hereafter connectivity) of the 12 study
sierras. There was no evidence that the (log) area of a sierra
and its connectivity were significantly associated (r=0.167,
n=12, P=0.60), so both factors estimate independent
aspects of the study habitat islands.
Phylogenies and phylogenetic distances
Phylogenetic distances between all pairs of plant species and
flower-visiting insect species were calculated from recon-
structed phylogenies. Phylogenetic relations among all plant
species recorded in the 12 sierras (i.e. a total of 96 species)
were inferred using the Phylomatic online tool (Webb &
Donoghue 2005), selecting the R20120829 source tree as the
master phylogeny. We further resolved the phylogenetic rela-
tions for plant species in polytomic families using the most
updated intrafamily phylogenies when available (see Support-
ing Materials and Methods, Plant and animal phylogenies).
Phylogenetic relations among all flower-visiting animal spe-
cies (i.e. a total of 171 species) were reconstructed by assem-
bling the information contained in different published
sources. We excluded the one observed vertebrate species, the
hummingbird Chlorostilbon aureoventris, from the final ani-
mal phylogeny because of the long time of divergence
between vertebrates and insects (Blair 2009) and because this
bird was censused only once (Aizen et al. 2012). The phylo-
genetic relations among the main groups of insects were
extracted from Wiegmann et al. (2009). We then added
resolved topologies for each group at the tips of this tem-
plate phylogeny from different source phylogenies (see Sup-
porting Materials and Methods, Plant and animal
phylogenies). On the basis of the time of divergence of major
clades for plants and insects, we estimated branch lengths
using the bladj command in Phylocom (Webb et al. 2008).
Thus, the resulting trees were ultrametric with time-calibrated
branch lengths (Figs S1 and S2). We then obtained matrices
of pair-wise phylogenetic distances among all plant and ani-
mal species pairs using the phydist command in Phylocom
(Webb et al. 2008).
To account for the effect of the incomplete resolution of the
trees, we applied an alternative procedure of phylogenetic
reconstruction that simultaneously resolves polytomies and
adjusts branch lengths using an evolutionary constant rate
birth–death model (see Supporting Materials and Methods,
Plant and animal phylogenies). Mantel correlations between
matrices of pair-wise phylogenetic distances constructed with
the bladj and this alternative procedures were extremely high,
r=0.995 (95% CI =0.994–0.996) and r=0.996 (95%
CI =0.995–0.997) for plants and insects respectively. This
almost perfect congruence implies that polytomies (which were
mostly present at terminal nodes) as well as uncertainty in the
time of divergence of some nodes had minimal influence on
overall phylogenetic structure.
©2015 John Wiley & Sons Ltd/CNRS
Letter Mutualism phylogeny across habitat islands 31
Ecological distances
Following Rezende et al. (2007) and Elias et al. (2013), we
used 1-Sas an estimate of ecological distances, where Sis the
Jaccard index of similarity obtained from qualitative interac-
tion matrices (Legendre & Legendre 1998). Accordingly, the
ecological similarity between two species, iand j, is defined as
S(i, j)=a/(b+ca), where arepresents the number of interact-
ing species shared between species iand j, and band cthe
total number of species interacting with species iand jrespec-
tively. The ecological distance between two plant (animal) spe-
cies ranges from 0, when they interact with the same animal
(plant) species assemblage, to 1, when they interact with com-
pletely different species assemblages. Ecological distances for
all pairs of plants and animals were estimated from the syn-
thetic 96 9171 species interaction meta-web and from the
interaction matrix of each sierra separately.
Phylogenetic signal in species assemblages
We estimated the extent to which phylogenetic proximity is
associated with similarity in the composition of the interacting
species assemblage for the meta-web and the interaction
matrix of each sierra separately. To this aim, we compared
phylogenetic and ecological distance matrices (Rezende et al.
2007; Elias et al. 2013) using Mantel tests performed with the
package vegan v. 2.0-10 (Dixon 2003) for R (R Core Team
2014). Because differences in the number of interacting species
composing each species’ assemblage (i.e. species degree) affect
Jaccard similarity, we performed partial Mantel tests control-
ling for differences in species richness. Degree distance matri-
ces were calculated from the absolute difference in the
number of interacting animal (plant) species for all pairs of
plant (animal) species (Rezende et al. 2007). Therefore, this
test assesses whether phylogeny affects the identity of the
interaction partners independently of differences in the num-
ber of interacting species, which can be unduly influenced by
differences in sampling effort (V
azquez & Aizen 2003).
Plant–animal phylogenetic congruence
We used the ParaFit co-phylogenetic analysis (Legendre et al.
2002), as implemented in the package ape v. 3.1-2 (Paradis
et al. 2004) for R (R Core Team 2014), to assess the degree of
correspondence between plant and animal phylogenetic trees,
for both the meta-web and the interaction web of each sierra
(Fig. S3). This analysis is based on the transformation of plant
and animal phylogenetic trees into matrices of principal coordi-
nates. The ParaFit global estimate is the sum of squares of all
the values of d
i, j
in a matrix D, which is the result of D=P.I
T
.
A, where Pis the plant phylogenetic distance matrix (with prin-
cipal coordinates in rows), Arepresents the animal phylogenetic
distance matrix (with principal coordinates in columns) and I
T
is the transposed plant–animal binary interaction matrix.
Statistical analyses
For significance testing of both partial Mantel correlation
coefficients and ParaFit global estimates, we conducted
permutations of the plant–pollinator interaction matrices. We
preferred this method rather than permuting either distance
matrices (Mantel correlations) or species phylogeny matrices
(ParaFit estimates). On the basis of null model 1 of V
azquez
& Aizen (2003), we used an algorithm, implemented in R (R
Core Team 2014), which shuffled the observed number of
interaction links among the cells of a given matrix with the
only restriction that each species had at least one interaction.
The advantages of this procedure is the preservation of matrix
size, while avoiding misestimated type I errors associated with
the permutation of distance matrices in Mantel correlation
tests (Harmon & Glor 2010). Each estimate was compared
with a distribution of expected values generated from 1000
randomisations of the synthetic plant–pollinator interaction
matrix (i.e. the meta-web) and each sierra’s interaction matrix.
To assess and compare the strength of Mantel associations
and co-phylogenetic signals among binary matrices of differ-
ent shape and completeness, observed estimates were z-trans-
formed according to ðx
xÞ=SD, where
xand SD are the
mean and standard deviations across the 1000 randomisations
(Elias et al. 2013). Because ParaFit estimates tended to differ
by orders of magnitude and have a highly right-skewed distri-
bution, they were log-transformed before calculation of z-
scores. We evaluated differences in the strength of the plant
vs. animal phylogenetic signal (i.e. the z-transformed Mantel
correlations) across the 12 sierras by means of a paired t-test.
Finally, we analysed the effect of (log) sierra area and connec-
tivity on z-scores with a linear multiple regression model.
To discard any confounding effect associated with differ-
ences in web size or phylogenetic diversity among sierras, we
included, alternatively, the total (log) number of species (i.e.
plants and animals), the (log) number of interaction links of
the plant–pollinator web sampled in each sierra and the (log)
Faith’s phylogenetic diversity (i.e. the sum of the lengths of
all phylogeny’s branches; Faith 1992) as predictive variables
in the above regression models. All analyses between each
response variable and sierra area and connectivity yielded the
same directional trends when absolute measurements, rather
than z-scores, were used (results not shown). Spatial autocor-
relation of residuals from the above models was addressed
using Moran’s Iestimates for increasing distance categories
(Fortin et al. 2002; see Supporting Materials and Methods,
Spatial autocorrelation).
Finally, we explored whether patterns of decreasing phylo-
genetic structure in plant–pollinator interaction with decreas-
ing sierra area and increasing connectivity could be linked to
a loss of the phylogenetic signal in specialisation and/or loss
of phylogenetic diversity. To analyse the first possibility, we
considered the number of interaction links per plant or animal
species (i.e. species degree) from the meta-web, corrected by
total observation frequency, as a measure of specialisation/
generalisation. To estimate this parameter, we used the residu-
als of linear regressions between the (log) number of interac-
tion links and the (log) number of observations with
intercepts forced to zero (V
azquez & Aizen 2003), considering
plants and flower visitors separately. Then, we estimated, for
both the meta-phylogenies and each sierra’s phylogenies, the
amount of phylogenetic signal in the adjusted number of
interaction links by means of the Kstatistic (i.e. the amount
©2015 John Wiley & Sons Ltd/CNRS
32 M. A. Aizen et al. Letter
of signal of the real data, expressed as a fraction of that
expected based on a Brownian model of trait evolution) as
detailed in Rezende et al. (2007). To analyse the second possi-
bility, following Eiserhardt et al. (2015), we compared the
observed lengths of the plant and animal phylogenies of each
sierra (i.e. Faith’s phylogenetic diversity) with the lengths of
1000 trees that were sub-sampled randomly from the plant
and animal meta-phylogenies, keeping the number of species
equal to those in the observed trees. For comparisons across
sierras, both the Kstatistic and phylogenetic diversity deficit
were transformed as z-scores.
RESULTS
Plant–pollinator interactions in the study sierras were phylo-
genetically structured. Analysis of the synthetic interaction
meta-web revealed that closely related plants tended to inter-
act with more ecologically similar sets of animal species than
less related plant species (z=5.07, P<0.00l). Similarly,
closely related animals tended to interact with more similar
sets of plant species than less related animal species
(z=7.24, P<0.001). Analysis of these two phylogenetic
signals on species assemblages across the 12 sierras showed
not only significant co-variation between them (Fig. 2a), but
also that the animal phylogeny was more strongly associated
with the identity of interactive plant mutualists than vice
versa (paired t-test: t
11
=5.16, P<0.0005). Overall, there was
some co-phylogenetic congruence between the synthetic plant
and animal meta-phylogenies (z=5.84, P<0.001), which,
at the level of individual sierras, was associated with both
plant and animal phylogenetic effects on the similarity of
the interacting animal and plant assemblages respectively
(Fig. 2b and c).
The strength of all the above three estimates of phylogenetic
structure varied among sierras and was associated with either
sierra area or connectivity. First, the association between ani-
mal pair-wise phylogenetic distances and plant pair-wise eco-
logical distances departed from null expectations in 10 of the
12 individual pollination webs. The strength of this phyloge-
netic effect increased significantly with sierra area (Fig. 3,
Table S3). Second, the association between plant pair-wise
phylogenetic distances and animal pair-wise ecological dis-
tances departed from null expectations in four of the 12 polli-
nation webs. The strength of this phylogenetic effect
decreased with sierra connectivity (Fig. 3, Table S3). Third,
the congruence between the plant and animal phylogenies was
weak in general, exceeding null expectations in only three of
the 12 individual pollination webs (Fig. 4). However, this co-
phylogenetic congruence became stronger as both sierra area
and isolation increased (Fig. 4, Table S4). The relation
between co-phylogenetic congruence and area changed from
marginal to highly significant after excluding Difuntito, a
well-preserved small sierra with an unexpectedly high number
of species and interactions (Table S1). However, Difuntito did
not influence the direction and magnitude of any of the
reported trends to any large extent (Figs 3 and 4). Factors
associated with network size, phylogenetic diversity or any
other spatially autocorrelated factor did not confound the
effects of sierra area and connectivity on the strength of the
web phylogenetic structure (Tables S5–S12). Also, despite an
overall significant phylogenetic signal in the degree of speciali-
sation (i.e. the adjusted number of links; K=0.301,
P=0.008 for the plant meta-phylogeny, and K=0.522,
P<0.001 for the animal meta-phylogeny), there was no evi-
dence of a significant decrease in either the phylogenetic signal
in specialisation or phylogenetic diversity with decreasing
sierra area or increasing connectivity (Figs S4 and S5, Tables
S13–S14).
DISCUSSION
Overall, the most phylogenetically structured pollination webs
were found in large and relatively isolated sierras, supporting
expectations based on meta-community principles. Underlying
these results are the well-sustained assumptions that (1) some
degree of phenotypic matching between co-occurring plant
and pollinator species is needed for interaction establishment
and functional efficiency (Stang et al. 2006; Schleuning et al.
2015), and (2) phenotypic matching involves traits that are
phylogenetically conserved (Rezende et al. 2007; G
omez et al.
2010). Therefore, to the extent that phylogenetic structure of
mutualistic interactions reflects co-evolution, we predict that
Plant phylogenetic effect
Animal phylogenetic effect
r = 0.65, P < 0.05
Plant phylogenetic effect
Cophylogenetic congruence
r = 0.75, P < 0.005
02468
02468
02468
02468
02468
02468
Animal phylogenetic effect
Cophylogenetic congruence
r = 0.62, P < 0.05
(b) (c)(a)
Figure 2 Pearson’s correlations between the three estimates (z-scores) of phylogenetic structure of plant–pollinator interactions across the 12 sierras.
(a) Plant vs. animal phylogenetic effect (i.e. Mantel correlations between phylogenetic relatedness and ecological similarity), (b) co-phylogenetic congruence
(i.e. ParaFit index) vs. plant phylogenetic effect and (c) co-phylogenetic congruence vs. animal phylogenetic effect. The open circle in each panel represents
Difuntito, a well-preserved small sierra with an unexpectedly high number of species and interactions (Table S1). The grey line of intercept =0 and
slope =1 is provided as reference.
©2015 John Wiley & Sons Ltd/CNRS
Letter Mutualism phylogeny across habitat islands 33
the transformation of continuous habitats into a network of
remnant habitat patches will jeopardise co-adaptation.
According to our conceptual framework, species abundance,
which increases with habitat size, represents a key factor
underlying phylogenetic structure of interactions. Although
phenotypic matching indicates the potential for the establish-
ment of an effective interaction (Stang et al. 2006; Schleuning
et al. 2015), species abundance can explain interaction persis-
tence in time and space (Aizen et al. 2012; Carstensen et al.
2014) and, therefore, the potential for co-adaptation. As a
consequence, increasing random extinctions of small popula-
tions of well-matched partner species occurring in small frag-
ments would not only decrease resemblance in species
assemblages between closely related interacting partners
(Figs 3 and 4), but also by disassembling trait matching,
could decrease the fitness of the surviving interaction partners.
Species colonisation, which increases with habitat connectiv-
ity, is the second factor proposed to affect the amount of
underlying phylogenetic structure of interactions in frag-
mented habitats. Central habitat islands are by definition
‘stepping stones’ of a higher number of dispersal routes than
peripheral, isolated islands (Urban & Keitt 2001). Although
populations of resident species in well-connected habitat
patches could benefit from an enhanced rescue effect, popula-
tions of non-resident species are also more likely to colonise
these patches (Leibold et al. 2004). These newcomers could
establish ephemeral and probably loose phenotypically
matched interactions with generalists (Aizen et al. 2012),
increasing, as previously reported, the nested structure of a
pollination web (Gilarranz et al. 2015), but blurring, as found
here, its phylogenetic structure (Figs 3 and 4). As it is pre-
dicted for species losing rare but efficient mutualists in small
habitat islands, a fitness cost is also expected for these tran-
sient species from interacting with ill-matched partners.
Our framework clearly favours a view of random species
extinction and colonisation affecting not only network archi-
tecture, as reported in Gilarranz et al. (2015), but also phylo-
genetic structure (Figs 3 and 4). However, deterministic,
phylogenetic-related factors associated with habitat patch size
and connectivity could provide alternative explanations for
the phylogenetic patterns we found. This could occur if lin-
eages of either plant or pollinator specialists are differentially
pruned in small habitat patches, or if lineages of generalists
proliferate in more connected patches. Both processes would
predict a decrease in the phylogenetic signal in the number of
interaction links per species, and of phylogenetic diversity
compared with random expectations (e.g. Rezende et al.
2007). However, despite significant phylogenetic signals in
−2
Area (ha)
z-score
101102103
Area (ha)
101102103
Plant phylogeny
P = 0.32
−2 2 4 6
Connectivity
z-score
P < 0.05
Animal phylogeny
P < 0.05
0.00 0.10 0.20 0.30 0.00 0.10 0.20 0.30
Connectivity
P = 0.10
Correlation between phylogeny and ecological similarity
0
−2 2 4 60
2460
−2 2460
Figure 3 Correlation between phylogenetic relatedness and ecological
similarity of interacting species as a function of sierra area and
connectivity. Connectivity is measured as betweenness centrality, an
inverse measure of patch isolation. A partial Mantel correlation (z-score)
between phylogenetic and ecological distance matrices was estimated for
each pollination web after accounting for differences in the number of
interacting species composing each species’ assemblage (i.e. species degree)
and plotted against sierra area (upper panels) and connectivity (lower
panels). The grey zone is the region delimited by the 2.5 and 97.5
percentiles from the random distributions of z-scores. The open circle in
each panel represents the correlation estimated for a web sampled at
Difuntito, a well-preserved small sierra with an unexpectedly high number
of species and interactions (Table S1). The corresponding partial
regression linear equation is depicted in each graph (see Table S3).
−2
Area (ha)
z-score
101102103
P = 0.06
P < 0.005
0.00 0.10 0.20 0.30
024−2024
Connectivity
z-score
P < 0.05
P < 0.005
Plant−animal phylogenetic congruence
Figure 4 Congruence between the plant and animal phylogenies as a
function of sierra area and connectivity. Connectivity is measured as
betweenness centrality, an inverse measure of patch isolation. The degree
of congruence between the plant and animal phylogenies (here measured
with the z-transformed ParaFit index) was estimated for each pollination
web and plotted against sierra area (upper panel) and connectivity (lower
panel). The grey zone is the region delimited by the 2.5 and 97.5
percentiles from the random distributions of z-scores. The open circle in
each panel represents the phylogenetic correspondence estimated for the
web sampled at Difuntito, a well-preserved small sierra with an
unexpectedly high number of species and interactions (Table S1). Partial
regression linear equations are depicted in each graph including all
estimates (continuous line) and after excluding the estimate for Difuntito
(dashed line) (see Table S4).
©2015 John Wiley & Sons Ltd/CNRS
34 M. A. Aizen et al. Letter
both plant and pollinator specialisation, none of these pat-
terns were supported by our data (Tables S13 and S14). This
result does not necessarily undermine the importance of
particular life-history species traits as determinants of
interaction disruption (Aizen et al. 2012), but it stresses the
role of a meta-community dynamics influencing phylogenetic
structure.
Although both the plant and animal phylogenetic effects on
the resemblance of their respective animal and plant species
assemblages showed similar trends (Fig. 3), the effect of animal
phylogeny increased significantly with sierra area, whereas the
effect of plant phylogeny increased significantly with sierra iso-
lation. Higher average mobility and shorter generation times of
insects than plants (Men
endez 2007) could be at the core of
these somewhat contrasting results. The animal phylogenetic
effect on plant assemblage composition might decrease in small
habitat islands if insect populations were more vulnerable to
local extinction than plant populations, whereas insects’ higher
vagility could decrease the plant phylogenetic effect on flower-
visitor assemblage composition in more central habitats.
Beyond speculation, our results also showed, as it has been pre-
viously reported (Rezende et al. 2007), that the animal phy-
logeny is more strongly associated with the identity of
interactive mutualists across pollination webs than vice versa.
This stronger animal than plant phylogenetic effect could relate,
among other factors, to (1) a longer evolutionary history of
insects than flowering plants (Hedges & Kumar 2009), (2) con-
vergent character evolution in unrelated plant lineages as a
response to the selection pressure from a suite of phylogeneti-
cally related pollinators (i.e. the floral syndrome hypothesis;
Fenster et al. 2004) and (3) the exploitation of different
resources (e.g. pollen vs. nectar) from a wide range of taxonom-
ically diverse plant species (Engel & Dingemans-Bakels 1980).
Whatever the ultimate cause or causes behind these differences,
both plant and animal phylogenetic effects on the identity of
interaction partners could be contributing to the co-phyloge-
netic structure of plant–pollinator interactions (Fig 2b and c).
The phylogenetic congruence between plants and animals
reported here is basically determined by a correspondence at
internal rather than terminal nodes (see the comparison of poly-
tomic and fully resolved phylogenies in phylogenies and phylo-
genetic distances), which concurs with the view that plant–
pollinator interactions are commonly shaped by co-evolution-
ary processes within species-rich assemblages of low specificity.
According to this perspective, phylogenetic structure of plant–
animal interactions is mostly determined by a correspondence
at high taxonomical (i.e. suprageneric) levels, which means that
guilds (i.e. functional groups) of closely related and unrelated
species, rather than single species, exert symmetric or asymmet-
ric selection pressures on each other (Fenster et al. 2004; Lunau
2004; Strauss & Irwin 2004). Despite its multi-specific nature,
this co-evolutionary process is key in the preservation of evolu-
tionary history, increasing species adaptation (Galetti et al.
2013) and long-term biodiversity maintenance (Thompson
1999). Our results suggest that this co-evolutionary process,
and the co-phylogenetic pattern it generates, is eroded under
scenarios of habitat fragmentation that increase species
turnover (e.g. a network of small but connected habitat
patches), resulting in more erratic selection forces that could
lead to an evolutionary increase in generalisation (Fenster et al.
2004). Accordingly, co-evolutionary processes moulding well-
matched plant–animal mutualisms would be better preserved in
large but relatively isolated habitat islands, expanding to the
level of webs of interactive mutualists Mayr’s popularised pro-
posal that speciation principally occurs in allopatry (Mayr
1976). Of course, in conservation terms the benefits of patch
connectivity for biodiversity maintenance (Leibold et al. 2004;
Gilarranz et al. 2015) will have to be balanced against the
potential costs of co-adaptation loss.
Most natural and semi-natural habitats are now fragmented
to different degrees. Habitat fragmentation has been shown
to be one of the most important drivers of the disruption of
different types of interspecific interactions, principally of
plant–pollinator interactions (Aguilar et al. 2006), which can
represent a proximate cause of species extinctions (Sabatino
et al. 2010). Here, we provide the first empirical demonstra-
tion that this general loss of biodiversity is accompanied by a
loss of phylogenetic structure in pollination mutualistic net-
works. Erosion of this phylogenetic structure can in turn indi-
cate degradation of co-evolutionary processes, which maintain
and boost biodiversity on Earth (Thompson 1999). Our
results suggest that these processes will benefit from setting
aside and preserving large expanses of habitat and isolating
them from much exterior influence.
ACKNOWLEDGEMENTS
The authors thank S. Nuismer, A. Pauw and three anony-
mous reviewers for useful comments and suggestions, A. S
aez
and D. Porrini for field assistance, V. Izpizua and M. Nuciari
for help in plant identification and J. Farina and A. Roig-
Alsina for help in identifying insects. This study was funded
in part by the National Institute of Agricultural Technology
(INTA), Balcarce (PNECO1302), the Argentina National
Research Council for Research (CONICET) (PIP 01623), the
National Fund for Research (PICT 01300), the National
University of Comahue (B152/04) and the European Research
Council (through an Advanced Grant to JB). M.A.A., G.G.
and M.S. are career researchers of CONICET.
AUTHORSHIP
MAA conceived the study with input from all authors,
performed the analyses, and wrote the first draft of the
manuscript; MS collected data; GG and MV built the phylo-
genies and all authors contributed substantially to the final
manuscript.
REFERENCES
Aguilar, R., Ashworth, L., Galetto, L. & Aizen, M.A. (2006). Plant
reproductive susceptibility to habitat fragmentation: review and
synthesis through a meta-analysis. Ecol. Lett., 9, 968–980.
Aizen, M.A., Sabatino, M. & Tylianakis, J.M. (2012). Specialization and
rarity predict nonrandom loss of interactions from mutualist networks.
Science, 335, 1486–1489.
Bascompte, J. & Jordano, P. (2007). Plant-animal mutualistic networks: the
architecture of biodiversity. Annu. Rev. Ecol. Evol. Syst., 38, 567–593.
Blair, J.E. (2009). Animals (Metazoa). In: The Timetree of Life (eds Hedges,
S.B., Kumar, S.). Oxford University Press, Oxford, pp. 223–230.
©2015 John Wiley & Sons Ltd/CNRS
Letter Mutualism phylogeny across habitat islands 35
Cabrera, A.L. (1994). Territorios Fitogeogr
aficos de la Rep
ublica
Argentina. Editorial Acme, Buenos Aires.
Carstensen, D.W., Sabatino, M., Trøjelsgaard, K. & Morellato, L.P.C.
(2014). Beta diversity of plant-pollinator networks and the spatial
turnover of pairwise interactions. PLoS ONE, 9, e112903.
Crandall, K.A., Bininda-Emonds, O.R., Mace, G.M. & Wayne, R.K.
(2000). Considering evolutionary processes in conservation biology.
Trends Ecol. Evol., 15, 290–295.
Dixon, P. (2003). VEGAN, a package of R functions for community
ecology. J. Veg. Sci., 14, 927–930.
Eiserhardt, W.L., Borchsenius, F., Plum, C.M., Ordonez, A. & Svenning,
J.C. (2015). Climate-driven extinctions shape the phylogenetic structure
of temperate tree floras. Ecol. Lett., 18, 263–272.
Elias, M., Fontaine, C. & van Veen, F.J.F. (2013). Evolutionary history
and ecological processes shape a local multilevel antagonistic network.
Curr. Biol., 23, 1355–1359.
Engel, M.S. & Dingemans-Bakels, F. (1980). Nectar and pollen resources
for stingless bees (Meliponinae, Hymenoptera) in Surinam (South
America). Apidologie, 11, 341–350.
Faith, D.P. (1992). Conservation evaluation and phylogenetic diversity.
Biol. Conserv., 61, 1–10.
Fenster, C.B., Armbruster, W.S., Wilson, P., Dudash, M.R. & Thomson,
J.D. (2004). Pollination syndromes and floral specialization. Annu. Rev.
Ecol. Evol. Syst., 35, 375–403.
Fortin, M.J., Dale, M.R. & ver Hoef, J. (2002). Spatial analysis in ecology.
In: Encyclopedia of Environmetrics. Vol. 4 (eds El-Shaarawi, A.H. &
Piegorsch, W.W.). John Wiley & Sons, Chichester, pp. 2051–2058.
Freeman, L.C. (1977). A set of measures of centrality based on
betweenness. Sociometry, 40, 35–41.
Galetti, M., Guevara, R., C^
ortes, M.C., Fadini, R., Von Matter, S., Leite,
A.B. et al. (2013). Functional extinction of birds drives rapid
evolutionary changes in seed size. Science, 340, 1086–1090.
Gilarranz, L.J., Sabatino, M., Aizen, M.A. & Bascompte, J. (2015). Hot
spots of mutualistic networks. J. Anim. Ecol., 84, 407–413.
G
omez, J.M., Verd
u, M. & Perfectti, F. (2010). Ecological interactions
are evolutionarily conserved across the entire tree of life. Nature, 465,
918–922.
Greenleaf, S.S., Williams, N.M., Winfree, R. & Kremen, C. (2007). Bee
foraging ranges and their relationship to body size. Oecologia, 153,
589–596.
Guimar~
aes, P.R., Jordano, P. & Thompson, J.N. (2011). Evolution and
coevolution in mutualistic networks. Ecol. Lett., 14, 877–885.
Harmon, L.J. & Glor, R.E. (2010). Poor statistical performance of the
mantel test in phylogenetic comparative analyses. Evolution, 64, 2173–
2178.
Hedges, S.B. & Kumar, S. (eds.) (2009). The Timetree of Life. Oxford
University Press, Oxford.
Jamoneau, A., Chabrerie, O., Closset-Kopp, D. & Decocq, G. (2012).
Fragmentation alters beta-diversity patterns of habitat specialists within
forest metacommunities. Ecography, 35, 124–133.
Legendre, P. & Legendre, L. (1998). Numerical Ecology, 2nd edn.
Elsevier, Amsterdam.
Legendre, P., Desdevises, Y. & Bazin, E. (2002). A statistical test for
host–parasite coevolution. Syst. Biol., 51, 217–234.
Leibold, M.A., Holyoak, M., Mouquet, N., Amarasekare, P., Chase,
J.M., Hoopes, M.F. et al. (2004). The metacommunity concept:
a framework for multi-scale community ecology. Ecol. Lett.,7,
601–613.
Lunau, K. (2004). Adaptive radiation and coevolution - pollination
biology case studies. Org. Divers. Evol., 4, 207–224.
Marussich, W.A. & Machado, C.A. (2007). Host-specificity and
coevolution among pollinating and nonpollinating new world fig wasps.
Mol. Ecol., 16, 1925–1946.
Mayr, E. (1976). Evolution and the Diversity of Life: Selected Essays.
Cambridge, Belknap.
Men
endez, R. (2007). How are insects responding to global warming?
Tijdschr. voor Entomol., 150, 355–365.
Moritz, C. (2002). Strategies to protect biological diversity and the
evolutionary processes that sustain it. Syst. Biol., 51, 238–254.
Paradis, E., Claude, J. & Strimmer, K. (2004). APE: analyses of
phylogenetics and evolution in R language. Bioinformatics, 20, 289–290.
Pauw, A. (2007). Collapse of a pollination web in small conservation
areas. Ecology, 88, 1759–1769.
Pearse, I.S., Harris, D.J., Karban, R. & Sih, A. (2013). Predicting novel
herbivore–plant interactions. Oikos, 122, 1554–1564.
Peralta, G., Frost, C.M., Didham, R.K., Varsani, A. & Tylianakis, J.M.
(2015). Phylogenetic diversity and co-evolutionary signals among
trophic levels change across a habitat edge. J. Anim. Ecol., 84, 364–372.
R Core Team (2014). R: A Language and Environment for Statistical
Computing. R Foundation for Statistical Computing, Vienna.
Rezende, E.L., Lavabre, J.E., Guimar~
aes, P.R., Jordano, P. &
Bascompte, J. (2007). Non-random coextinctions in phylogenetically
structured mutualistic networks. Nature, 448, 925–928.
Sabatino, M., Maceira, N. & Aizen, M.A. (2010). Direct effects of habitat
area on interaction diversity in pollination webs. Ecol. Appl., 20, 1491–
1497.
Schleuning, M., Fr€
und, J. & Garc
ıa, D. (2015). Predicting ecosystem
functions from biodiversity and mutualistic networks: an extension of
trait-based concepts to plant-animal interactions. Ecography, 38, 1–13.
Segraves, K.A. (2010). Branching out with coevolutionary trees. Evol.
Educ. Outreach,3,62–70.
Stang, M., Klinkhamer, P.G.L. & van der Meijden, E. (2006). Size
constraints and flower abundance determine the number of interactions
in a plant-flower visitor web. Oikos, 112, 111–121.
Strauss, S.Y. & Irwin, R.E. (2004). Ecological and evolutionary
consequences of multispecies plant-animal interactions. Annu. Rev.
Ecol. Evol. Syst., 35, 435–466.
Thompson, J.N. (1999). The evolution of species interactions. Science,
284, 2116–2118.
Tscharntke, T., Steffan-Dewenter, I., Kruess, A. & Thies, C. (2002).
Characteristics of insect populations on habitat fragments: a mini
review. Ecol. Res., 17, 229–239.
Urban, D. & Keitt, T. (2001). Landscape connectivity: a graph-theoretic
perspective. Ecology, 82, 1205–1218.
V
azquez, D.P. & Aizen, M.A. (2003). Null model analyses of
specialization in plant-pollinator interactions. Ecology, 84, 2493–2501.
Webb, C.O. & Donoghue, M.J. (2005). Phylomatic: tree assembly for
applied phylogenetics. Mol. Ecol. Notes, 5, 181–183.
Webb, C.O., Ackerly, D.D. & Kembel, S.W. (2008). Phylocom: software
for the analysis of phylogenetic community structure and trait
evolution. Bioinformatics, 24, 2098–2100.
Wiegmann, B.M., Kim, J. & Trautwein, M.D. (2009). Holometabolous
insects (Holometabola). In: The Timetree of Life (eds Hedges, S.B.,
Kumar, S.). Oxford University Press, Oxford, pp. 260–263.
Young, A., Boyle, T. & Brown, T. (1996). The population genetic
consequences of habitat fragmentation for plants. Trends Ecol. Evol.,
11, 413–418.
SUPPORTING INFORMATION
Additional Supporting Information may be downloaded via
the online version of this article at Wiley Online Library
(www.ecologyletters.com).
Editor, Jennifer Dunne
Manuscript received 15 July 2015
First decision made 30 September 2015
Manuscript accepted 2 October 2015
©2015 John Wiley & Sons Ltd/CNRS
36 M. A. Aizen et al. Letter