au pure · web view6ecoinformatics and biodiversity, department of bioscience, aarhus university,...
TRANSCRIPT
Article title: Ecological, historical and evolutionary determinants of modularity in weighted
seed-dispersal networks
Running title: Modularity of weighted ecological networks
Type of article: Letter
Authors: Matthias Schleuning1, Lili Ingmann1, Rouven Strauß2, Susanne A. Fritz1, Bo
Dalsgaard3, D. Matthias Dehling1, Michaela Plein1,4, Francisco Saavedra1,5, Brody Sandel6,
Jens-Christian Svenning6, Katrin Böhning-Gaese1,7, Carsten F. Dormann8
Affiliations: 1Biodiversity and Climate Research Centre (BiK-F) and Senckenberg Gesellschaft für
Naturforschung, Senckenberganlage 25, 60325 Frankfurt am Main, Germany2Department of Computer Science, Technion - Israel Institute of Technology, Haifa 32000,
Israel3Center for Macroecology, Evolution and Climate, Natural History Museum of Denmark,
University of Copenhagen, 2100 Copenhagen Ø, Denmark4School of Botany, The University of Melbourne, Parkville, VIC 3010, Australia5 Institute for Biology/Geobotany and Botanical Garden, Martin-Luther-University
Halle-Wittenberg, Am Kirchtor 1, 06108 Halle (Saale), Germany6Ecoinformatics and Biodiversity, Department of Bioscience, Aarhus University, 8000 Aarhus
C, Denmark7Department of Biological Sciences, Johann Wolfgang Goethe-Universität Frankfurt,
Max-von-Laue-Straße 9, 60438 Frankfurt (Main), Germany8Faculty of Forest and Environmental Science, University of Freiburg, 79106 Freiburg,
Germany
Email contacts (order follows author list): [email protected];
[email protected]; [email protected]; [email protected];
[email protected]; [email protected]; [email protected];
[email protected]; [email protected];
[email protected]; [email protected];
Number of words: 150 (abstract), 4,999 (main text)
Number of references: 50; number of figures: 2; number of tables: 3.
Supporting Information included with this submission (within the same manuscript file):
number of figures: 3, number of tables: 4, number of appendices: 1.1
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
Correspondence: Matthias Schleuning, Biodiversity and Climate Research Centre (BiK-F),
Senckenberganlage 25, 60325 Frankfurt am Main, Germany; phone: +49 6975421892; fax:
+49 6975421801; email: [email protected]
Statement of authorship: MS and CFD conceived the idea. RS and CFD developed the
modularity algorithm. MS, LI, DMD, MP, FS and KBG contributed network data. BD, BS,
and JCS contributed climate models. SAF contributed phylogenetic data. MS, LI, RS, SAF
and CFD performed analyses. MS and CFD drafted the manuscript. All authors contributed to
interpretation and writing.
Keywords: Avian seed dispersal; current and past climate; ecological networks; evolutionary
history; macroecology; modularity; phylogeny; seasonality; traits; weighted bipartite
networks.
Abstract
Modularity is a recurrent property of ecological networks. Although ecological networks
usually describe interaction frequencies between species pairs, modularity has been analysed
only on the basis of binary presence-absence data. We employ a new algorithm to detect
modularity in weighted bipartite networks in a global analysis of avian seed-dispersal
networks. We define roles of species such as connector values for weighted and binary
networks and associate them to avian species traits and phylogeny. The weighted, but not
binary, analysis identified a positive relationship between climatic seasonality and modularity,
whereas past climate stability and the phylogenetic signal in interaction patterns were only
weakly related to modularity. Connector values were associated with foraging behaviour and
were phylogenetically conserved. The weighted modularity analysis demonstrates the
dominating impact of ecological factors on the structure of seed-dispersal networks, but also
underscores the relevance of evolutionary history for the variability in species roles in
ecological communities.
2
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
Introduction
Species evolve and coexist entangled in networks of interacting species (Bascompte &
Jordano 2007). A recurrent structure of many types of ecological networks is modularity,
which describes the existence of sub-communities within networks (Newman & Girvan 2004;
Thébault 2013). The modular structure of ecological networks is a consequence of both
ecological and evolutionary processes and has been suggested to be important for species
coexistence and community stability (Olesen et al. 2007; Thébault & Fontaine 2010).
Interaction frequencies between species pairs are closely associated with the functional
interdependence between species (Vázquez et al. 2005) and define the structure of ecological
networks (Ings et al. 2009). Neglecting the quantitative nature of species interactions can lead
to an incomplete understanding of the processes shaping ecological networks (Scotti et al.
2007; Gilarranz et al. 2012, Staniczenko et al. 2013). In contrast to weighted analysis of other
network properties, such as nestedness (Bascompte & Jordano 2007; Staniczenko et al. 2013),
algorithms for detecting modularity in weighted bipartite networks have not yet been explored
in ecology (Thébault 2013). In consequence, we have an incomplete knowledge of modularity
patterns in bipartite networks, such as those describing reciprocal mutualisms between plants
and animals (Bascompte & Jordano 2007).
Macroecological analyses of ecological networks have begun to examine the impacts
of current ecological and past climatic factors on network structure (Dalsgaard et al. 2011,
2013; Schleuning et al. 2012). First, ecological factors that may influence modularity include
gradients in productivity and resource diversity (Trøjelsgaard & Olesen 2013), including
seasonal resource fluctuations (Bosch et al. 2009). Ecological responses of species to
spatiotemporal resource availability are driven by the ability of consumer species to adapt
their foraging behaviour to current ecological conditions (Wheelwright 1988). This process
may lead to generalized interactions at high productivity and resource diversity (Schleuning et 3
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
al. 2012) and may favour the formation of modules comprising species with overlapping
phenological schedules in seasonal climates (Bosch et al. 2009; Martín Gonzalez et al. 2012).
Second, past climate change may be associated with the disruption of co-adapted species
pairs, especially in periods of high past climatic instability. Hence, a decrease of modularity
with increasing past climatic instability can be expected and has been found for pollination
networks (Dalsgaard et al. 2013). Third, interactions in ecological networks are
phylogenetically conserved (Rezende et al. 2007; Gomez et al. 2010) and related species may
form modules that interact with similar sets of species (Krasnov et al. 2012), potentially
leading to a positive relationship between modularity and phylogenetic signal in a network.
Yet, there are no integrative studies testing whether current ecological factors, past climate
and community history, or evolutionary processes are the main determinants of bipartite
network structure.
In addition to comparisons of modularity among networks, the variability in species
roles within networks is ecologically relevant, because ecological networks are comprised of
individual species that vary in their functional importance (Stouffer et al. 2012). Olesen et al.
(2007) adopted a classification system that assigns species roles based on positions of species
in modular networks, distinguishing between species defining the modules and species linking
different modules (Guimerà & Amaral 2005). Subsequent studies of binary networks have
referred to this classification (e.g. Donatti et al. 2011; Mello et al. 2011), but have rarely
tested explicitly whether species roles were randomly distributed among species or were
associated with species traits or phylogeny (but see Donatti et al. 2011 for a single binary
network).
We employ a new method to detect modularity and to describe species roles in
bipartite weighted networks, and apply this method to a global dataset of 18 seed-dispersal
networks describing interactions between fleshy-fruited plants and frugivorous birds. To test
the effects of current ecological and past climatic factors on modularity, we obtained 4
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
information about mean and variance of current climatic conditions and quantified past (Late
Quaternary) climatic fluctuations at all study locations. To estimate the phylogenetic signal in
plant-frugivore interactions, we obtained a complete phylogeny of avian frugivores and
computed the co-variation between shared phylogenetic history and interaction similarity
across all avian species pairs (Rezende et al. 2007). In addition, we collected information on
five important traits of avian frugivores (Schleuning et al. 2011; Menke et al. 2012), related to
their morphology (body mass), foraging behaviour (degree of frugivory, social foraging
behaviour) and spatiotemporal occurrence (forest dependence, migratory behaviour). We use
this unique set of weighted interaction networks and climatic, phylogenetic and trait data to
test (i) whether biogeographical patterns in modularity are primarily influenced by current
ecological factors, past climatic stability, or evolutionary history, and (ii) whether species
traits and phylogeny influence species roles in ecological networks. The weighted modularity
analysis finds that (i) the degree of modularity in seed-dispersal networks is most closely
related to current ecological factors, and that (ii) foraging behaviour and evolutionary history
contribute to the variability in avian species roles.
Material and methods
Dataset
We compiled a dataset of 18 weighted interaction networks between plants with fleshy fruits
and their avian seed dispersers (see Table S1 in Supporting Information). Other animal seed
dispersers (e.g. monkeys, bats) were not included because a comparative analysis of
phylogenetic and trait effects is not meaningful for non-monophyletic taxonomic groups.
Ecologically, the focus on avian seed dispersers is reasonable because birds are the most
species-rich group of frugivorous animals (Kissling et al. 2009).
All networks in the dataset describe interactions between fleshy-fruited plants and
avian frugivores at the community level, although the extent of sampling varied among
5
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
studies (Table S1). Networks were recorded by observing frugivore visits to plant individuals
in transect walks or focal plant observations, except in two studies that were based on faecal
samples from caught birds (Table S1). All studies provided estimates of interaction
frequencies, i.e. the number of bird individuals observed feeding on a plant species or the
number of bird individuals carrying seeds of a particular plant species. Estimates of
interaction efficiencies, such as differences in seed handling, were not available. Interaction
frequencies are a good proxy for interaction strength, which is more important than
interaction efficiency (Vázquez et al. 2005), and we therefore do not differentiate between
seed-dispersal and plant-frugivore networks in this study.
All studies covered at least the main fruiting period in the study area, and species
richness per network ranged from 24 to 121 plant and bird species (see Table S1 for details).
Overall, networks comprised almost 85,000 interaction events. We defined sampling intensity
for each network as the ratio between the number of observed interaction events (square-root-
transformed) and the geometric mean of the number of plant and animal species (Schleuning
et al. 2012). This measure of sampling effort reflects the number of interaction events
observed per species and accounts for higher observation requirements in species-rich than
species-poor networks. This matters because species richness decreases with absolute latitude
(in our dataset, n = 18 networks; Pearson correlation, r = –0.57, P = 0.01).
Modularity algorithm
We employed a new algorithm (QuanBiMo) to calculate the weighted modularity of bipartite
interaction networks, which is described in detail in Dormann & Strauß (2013). In principle,
this algorithm follows the approach of Clauset et al. (2008). It builds a random binary tree
whose leaves represent the interacting species and associates a structure with each tree
defining the subdivision of species into modules. To define a new subdivision, random swaps
of branches at any level are performed, followed by an evaluation of whether the new
6
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
subdivision has a higher modularity value than the previous one. The algorithm employs a
stochastic hill-climbing approach, i.e. an increase in modularity is always accepted, and a tree
with lower modularity is accepted with a probability inversely proportional to the loss in
modularity. The objective function is the bipartite version of Newman’s quantity of
modularity Q (Barber 2007):
Q= 12 N
∑ij (A ij−K ij ) δ (mi , m j ) ,
where N is the total number of observed interactions in the network and Aij is the normalised
number of interactions between bird species i and plant species j. The term Kij represents the
expected probability of interactions within a module assuming no preferences in the
participating species, which is a suitable null model (Barber 2007). When applying the
algorithm to binary data, the null model does not constrain the number of interactions, but the
number of links per species. The module to which a species i or j is assigned is mi and mj,
respectively. The indicator function δ (mi, mj) is 1 if mi = mj (i.e. when species i and j are in
the same module) and 0 if mi ≠ mj. Q ranges from 0, which means the community has no more
links between species within a module than expected by chance, to 1, which equals the
maximum degree of modularity.
We searched for the best division of a network into modules in five independent runs
of the algorithm. If no further improvement was recorded after 107 swaps, a run was
terminated and the result interpreted as the optimum,. We recorded the degree of modularity
Q, the number of detected modules and the affiliation of each species to a module for the run
with the highest modularity (see Table S1 for the low variability in Q among runs). To
compare the performance of the algorithm in detecting modules in weighted and binary
networks, we calculated Q for each network from the weighted interaction matrices and from
7
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
binary-transformed interaction matrices (n = 18; weighted vs. binary Q, r = 0.49, P = 0.04).
To test whether the classification of plant and bird species into modules was associated with
their niche overlap, we employed a multivariate analysis of variance based on Horn-Morisita
interaction distances among plant and bird species, respectively (Gomez et al. 2010).
Since modularity tends to be overestimated in poorly sampled networks (Dormann &
Strauß 2013) and because sampling intensity varied among studies (Table S1), we corrected
estimates of modularity Q with two alternative null models. First, we randomized interactions
with the Patefield algorithm (null model PA, see Blüthgen et al. 2008), which randomly
redistributes interaction events among all cells of the network while constraining the total
number of interactions per species. It assumes that species interact randomly, without
constraining the degree of specialization in a network. Second, we randomized interactions
with an alternative null model (null model VA), proposed by Vázquez et al. (2007),
constraining the total number of interactions per species and the network connectance. This
null model assumes that network connectance is an inherent network property (e.g. defined by
the number of forbidden links). It redistributes interaction events randomly among species
until the number of filled cells in the matrix equals that in the original matrix; remaining
interactions are then distributed among filled cells (Vázquez et al. 2007). To obtain estimates
of Q for both null models, we used the same settings for the modularity algorithm as for the
real networks (107 swaps, 5 independent runs). For each null model, we obtained ten
randomizations, which were sufficient because Q was very similar among randomizations.
Null-model estimates of Q were closely associated with sampling intensity and were not
confounded with latitude (Figs. S1a-d). For each network, we calculated two null-model
corrected versions of weighted and binary modularity, ΔQPA and ΔQVA, as the difference
between observed modularity Q and mean QNULL.PA and QNULL.VA, respectively.
To identify species roles in modular networks, we followed Guimerà & Amaral (2005)
and Olesen et al. (2007) and defined standardized within-module degree z and among-module 8
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
connectivity c for each species from weighted and binary matrices. Within-module degree z
and among-module connectivity c were characterized for a network with M modules as:
z=( k is−k s
SD ks)
;
c= 1−∑t=1
M ( k it
k i)2
where kis is the number of links of species i to other species in its own module s, k s
is the
average kis of all species in module s, SDks is the standard deviation of kis of all species in
module s, kit is the number of links of species i to module t, and ki is species degree (binary) or
species strength (weighted, Bascompte & Jordano 2007) of species i.
Macroecological patterns
We recorded absolute latitude for each study location (range: 0°–52°) and obtained climate
estimates at a resolution of 2.5 arc-minutes for mean annual temperature (MAT), temperature
seasonality measured as the coefficient of variation of monthly mean temperatures on the
Kelvin scale (CVMAT), mean annual precipitation (MAP), and precipitation seasonality
(CVMAP), i.e. the coefficient of variation of monthly precipitation (Hijmans et al. 2005). MAT
and MAP as well as CVMAT and CVMAP were correlated (n = 18; MAT vs. MAP: r = 0.71, P <
0.01; CVMAT vs. CVMAP: r = –0.59, P = 0.01). Past climate stability was estimated as climate-
change velocity since the Last Glacial Maximum (LGM, 21,000 years ago), which describes
the rate at which climatic conditions have moved over the Earth's surface (Sandel et al. 2011).
Climate change since LGM captures one of the strongest climatic shifts of the Quaternary,
and the spatial pattern of this change is representative for the last several hundred thousand
years (Sandel et al. 2011). We derived climate-change velocities for both changes in
temperature (VELMAT) and precipitation (VELMAP), which were based on 2.5 arc-minute
resolution maps of contemporary climate (Hijmans et al. 2005) and paleo-climate projections
(CCSM3 model; Braconnot et al. 2007). VELMAT and VELMAP were not significantly
9
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
correlated (n = 18, r = 0.20, P = 0.44). Estimates of current climate and past climate stability
were not significantly correlated (n = 18; MAT vs. VELMAT, r = –0.31, P = 0.20; MAP vs.
VELMAP, r = 0.06, P = 0.81).
To assess the degree of shared evolutionary history among birds, we obtained a
phylogeny of avian frugivores from a recently published super-tree (Jetz et al. 2012). We
obtained a sample of 1,000 dated pseudo-posterior trees for the 390 bird species in our
networks. Since these trees did not vary substantially in topology and branch lengths, we
obtained a maximum clade credibility tree across our 1,000 samples. To quantify the
phylogenetic signal in interaction patterns for each network, we calculated phylogenetic
pairwise distances across all species in the phylogenetic tree (standardized to range between 0
and 1). These were related to an inverse measure of niche overlap among avian frugivores, i.e.
the distances in interaction patterns between all avian species pairs, employing the Horn-
Morisita metric (ranging between 0 and 1). The avian phylogenetic signal in interaction
patterns (PHYLO) was then calculated as the correlation coefficient between phylogenetic
and interaction distances (separately for weighted and binary matrices), as obtained from
parametric Mantel tests (Rezende et al. 2007), i.e. high correlations indicate similar
interaction patterns in phylogenetically related species. We also ran partial Mantel tests
accounting for differences in degree among species which resulted in qualitatively identical
results (Table S1).
We tested whether modularity Q and ΔQ of weighted and binary interaction matrices
were associated with absolute latitude and whether sampling effort and species richness
influenced modularity and its relationship with latitude. Then, we fit univariate relationships
of MAT, MAP, CVMAT (log), CVMAP (log), VELMAT (log), VELMAP (log) and PHYLO with
weighted and binary modularity ΔQ with simple linear models and spatial simultaneous
autoregressive error (SAR) models. For spatial analyses, we defined the neighbourhood of
each study location by the four nearest locations. To test model robustness and the influence 10
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
of single, potentially inadequately sampled networks, we sub-sampled studies with a jackknife
procedure and computed standard deviations of r2 across the resulting linear models (n = 18).
Finally, we compared model fit according to the small-sample corrected Akaike Information
Criterion (AICc) among linear models including all combinations of predictor variables and a
null model only including the intercept. We did not include interaction terms between
predictor variables because of the limited sample size. We calculated the relative importance
of each predictor variable by summing Akaike weights across all models including the
respective variable (Burnham & Anderson 2002, p. 168). We fit multi-predictor models for
both weighted and binary modularity ΔQ.
Species roles
We selected avian species traits that were related to bird morphology (body mass), foraging
preferences and behaviour (degree of frugivory, social foraging) or to spatial and temporal
patterns in species occurrence (forest dependence, migratory behaviour). We expected that
species roles in networks could be simultaneously influenced by each of the five traits. Bird
taxonomy followed Clements et al. (2012), except for Chlorophonia cyanea (Thraupidae, not
Fringillidae) and Spindalis portoricensis (bird family undefined). We were able to compile
complete trait information for 345 of the 390 bird species (see Appendix S1 for reference
details). (i) Body mass was recorded as mean body mass for male and female individuals. We
further distinguished between (ii) obligate and partial frugivores feeding on fruit as a major
food source and opportunistic frugivores that use fruit as a minor complementary food source
(according to Kissling et al. 2009); (iii) social foragers that frequently forage in conspecific
and mixed-species flocks, and non-social foragers; (iv) forest-specialists and species that also
inhabit non-forest habitats; (v) short- or long-distance migratory species and residents.
We calculated z values (within-module degree), c values (among-module connectivity)
and species strength and degree for weighted and binary matrices and excluded all species
11
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
with a single observation from these calculations. To ensure comparability among networks,
we standardized z and c values as well as species strength and degree to zero mean and unit
variance for each network. For each bird species that occurred in more than one network (n =
81 species), we calculated an average of standardized z and c values and species strength and
degree across all networks (Krasnov et al. 2012). We tested phylogenetic signals in z and in c
values, both from weighted and binary matrices, and in species strength and degree, by
computing the λ statistic (Pagel 1999), testing it against a random shuffle of values across the
tips of the phylogeny. An analysis of phylogenetic signal based on Blomberg's K and the same
null model yielded qualitatively identical results (Blomberg et al. 2003). To test the influence
of species traits on z and c values and species strength and degree, while accounting for
phylogenetic co-variation among species, we fit phylogenetic generalized linear models
(PGLMs), while optimizing the degree of phylogenetic signal (λ), with a maximum-likelihood
approach (Freckleton et al. 2002). This was necessary because all species traits were
phylogenetically conserved (body mass, lambda=1; all binary traits, 0.16<d<0.44; P < 0.01 in
all cases). Phylogenetic signals in binary traits were tested employing the method proposed by
Fritz and Purvis (2010). We fit PGLMs of weighted and binary z and c values and of species
strength and degree with all combinations of trait variables (but not their interaction terms),
and a null model only including the intercept, and identified the minimal adequate model
based on the lowest AICc. We additionally calculated the relative importance of each trait
across all model combinations according to their summed Akaike weights (Burnham &
Anderson 2002, p. 168).
Results
Different measures of modularity
In almost all seed-dispersal networks, modules reflected interaction distances among both bird
and plant species, especially in weighted analyses (Table S2). Weighted networks were
12
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
significantly more modular than expected from null models (except a single network, P < 0.1
for both null models). By contrast, only 10 of 18 binary networks were significantly modular
according to both null models (Table S1). Weighted and binary Q were positively related to
the number of modules detected in a network, although this relationship was significant only
for binary networks (n = 18; weighted Q, r = 0.39, P = 0.11; binary Q, r = 0.67, P < 0.01).
The number of modules in a network was closely related to sampling intensity (n = 18;
weighted analysis, r = –0.79, P < 0.01; binary Q, r = –0.76, P < 0.01), as was the case for
uncorrected weighted and binary Q (Table S3, Figs. S1e-f). Both null-model corrected
modularities (ΔQPA, ΔQVA) were not influenced by effects related to sampling effort and
species richness (Table S3, Figs. S1e-f) and were used in further analyses.
Macroecological patterns
Weighted network modularity ΔQPA increased with latitude (Table 1a), whereas binary ΔQPA
was also positively, but not significantly related to latitude (Table 1a). In univariate linear
models and in SAR models, weighted modularity ΔQPA was associated with current mean
annual temperature (MAT) and temperature seasonality (CVMAT), but was unrelated to
temperature climate-change velocity (VELMAT) and the phylogenetic signal in avian
interaction patterns (PHYLO) (Table 1, Fig. 1). Effects of temperature variables on weighted
modularity were mostly stronger than those of precipitation variables (Table 1). In multi-
predictor models, all best models of weighted ΔQPA included CVMAT and its importance was
more than twice as high as that of other predictors (Table 1). The proportion of explained
variance was much lower in binary than in weighted modularity ΔQPA (Table 1, Fig. S2),
probably because null-model corrections removed most variability among binary networks
(Fig. S1f). Results based on the second, more constrained null model (ΔQVA) were
qualitatively identical, especially in weighted analyses (Table S4).
Species roles
13
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
Weighted and binary measures of z and c values were positively correlated (n = 312 species; z
values: r = 0.46, P < 0.01; c values: r = 0.63, P < 0.01). However, the positive correlation was
contingent on sampling intensity and decreased exponentially with increasing sampling
intensity (Fig. S3). Species strength was positively related to weighted measures of z and c
values, especially to z values (z values: r = 0.78, P < 0.01; c values: r = 0.30, P < 0.01),
Species degree was correlated with binary measures, especially with c values (z values: r =
0.56, P < 0.01; c values: r = 0.72, P < 0.01). We did not detect a phylogenetic signal in
within-module degree z, either in weighted (λ < 0.01, P = 1) or binary analyses (λ = 0.180, P
= 0.28), or in species strength (λ < 0.001, P = 1). By contrast, among-module connectivity c
exhibited a moderate but significant phylogenetic signal in weighted (λ = 0.371, P < 0.01) and
binary (λ = 0.309, P = 0.02) analyses, corresponding to a significant phylogenetic signal in
species degree (λ = 0.350, P < 0.01). Mostly tropical lineages, e.g. paleotropical families
Pycnonotidae and Lybiidae and neotropical families Pipridae and Thraupidae, showed
consistently high c values and species degree (Fig. 2).
Effects of species traits on z and c values and on species strength and degree were
generally weak (Table 2). Consistent across all species-level metrics and weighted and binary
analyses, values were higher for obligate and partial than for opportunistic frugivores (Table
2). Additionally, frugivorous species with social foraging behaviour showed consistently
higher weighted and binary c values than solitary foragers (Table 2); this difference was only
not detected in c values. Other species traits were not important for explaining the variability
in z and c values and in species strength and degree (Table 3).
Discussion
Ecological factors drive modularity in seed-dispersal networks
Weighted seed-dispersal networks were organised in modules, and they were more modular
than one would expect in randomly associated communities. This is consistent with previous
14
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
studies of modularity of binary seed-dispersal networks (Donatti et al. 2011; Mello et al.
2011). We found that tropical avian seed-dispersal networks were less modular than temperate
networks and that this latitudinal trend in modularity was much stronger in weighted than
binary analyses. Uncorrected modularity of weighted and binary networks was strongly
influenced by spatially varying sampling intensities (Fig. S1). These effects masked
macroecological patterns in our analyses and warn against uncorrected comparisons of
network metrics among studies. Null-model corrections removed the effects related to
differences in sampling intensity, but in binary analysis these corrections erased almost all
variability among networks (Fig. S1). This indicates that null models can separate sampling
artefacts from ecological patterns only in weighted network analyses. Hence, binary analyses
may be less suitable for comparative network analyses because sampling bias may often
superimpose ecological patterns.
The latitudinal pattern of decreasing modularity towards the tropics was also reflected
in underlying climatic drivers, as the degree of modularity in avian seed-dispersal networks
was closely related to climatic seasonality. Consistently, species communities of temperate
seed-dispersal networks vary seasonally (Plein et al. 2013). By contrast, seasonal turnover in
fruiting plant communities is less pronounced in the wet tropics, where fruiting phenologies
are subject to strong inter-annual, but weaker seasonal fluctuations (Howe & Smallwood
1982; Chapman et al. 2005). Thus, the latitudinal gradient of weighted modularity in avian
seed-dispersal networks may arise from higher seasonal partitioning of fruit and frugivore
communities in temperate than tropical ecosystems.
Modularity of avian seed-dispersal networks was only weakly associated with past
climatic fluctuations and the shared evolutionary history of avian seed dispersers. Our
findings contrast with those from binary pollination networks, where modularity is
particularly evident in the climatically stable tropics (Dalsgaard et al. 2013; Trøjelsgaard &
Olesen 2013). While most animal-dispersed plant species address functionally diverse seed-15
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
disperser communities that vary in space and time (Plein et al. 2013), animal-pollinated plants
depend on transfer of conspecific pollen and thus on flower-constant pollinators (Pellmyr
2002). Hence, selective forces on plants in seed-dispersal systems may differ from those in
pollination systems, which could explain the observed differences in latitudinal patterns of
modularity. The strong ecological effects on modularity observed in this study, emphasizes
that ecological factors, such as seasonal fluctuations in ecological communities (Bosch et al.
2009; Martín Gonzalez et al. 2012), should in the future receive the same attention in the
interpretation of modularity as evolutionary processes (Donatti et al. 2011; Krasnov et al.
2012).
Phylogeny and foraging behaviour influence species roles
We found no phylogenetic signal in the within-module degree of bird species, which was also
only weakly related to species traits. This suggests that modules in avian seed-dispersal
networks are mostly formed by temporary associations of bird species feeding on the same
plants (Plein et al. 2013). Species strength was correlated with within-module degree and was
also unrelated to avian phylogeny, highlighting that abundances of species, rather than
phylogenetically conserved traits, were the main determinants of the modular structure of
avian seed-dispersal networks.
In contrast to weak effects on within-module degree, phylogeny and species traits
were significantly related to among-module connector values of frugivores, albeit a high
proportion of variance was unexplained. The capacity of species to connect different modules
was related to its species degree, which was also phylogenetically conserved (see also
Rezende et al. 2007). Accordingly, the ability of a consumer species to generalize its diet
seems to be a conservative trait across the avian phylogeny. In general, traits related to
foraging preference and behaviour were most important for differences in connector values,
whereas avian body size was the trait contributing the least information to weighted analyses. 16
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
This contrasts with the particular importance of body size in food webs (Woodward et al.
2005). It is likely that traits related to foraging behaviour are more important than coarse
morphological variables in studies focusing on specific functional groups of species. The
apparent importance of body size in previous work on binary seed-dispersal networks
(Donatti et al. 2011) could potentially be explained by fundamental differences between
phylogenetic lineages (i.e. mammals vs. birds) rather than by body size per se, calling for
rigorous phylogenetic correction in comparative analyses of species roles.
Our study underscores previous findings that opportunistic frugivores fill marginal
positions in seed-dispersal networks, whereas the core of interactions is contributed by
obligate and partial frugivores (Schleuning et al. 2011). It seems ecologically plausible that
obligate and partial frugivores are generalist connector species within seed-dispersal networks
(Schleuning et al. 2011) because they have to feed on many fruiting plants to balance their
diverse nutritional demands and the spatial and temporal patchiness in fruit availability. The
importance of obligate and partial frugivores is particularly high in the tropics (Kissling et al.
2009), where they contribute more interactions to seed-dispersal networks (mean ± SE: 80.5 ±
2.2%; n = 18 networks) than in temperate systems (57.3 ± 8.1%). Social foraging behaviour
was the other key trait defining avian connector species. Bird species foraging in conspecific
and mixed-species flocks rarely associated with specific modules, and social foraging may
constrain the formation of modules within seed-dispersal networks. Many tropical birds
forage in mixed-species flocks (Saracco et al. 2004). In our study, species that tended to
forage in social flocks contributed more than half of the interactions to tropical networks (56.9
± 6.7%) and significantly fewer interactions to temperate networks (32.1 ± 5.7%).
The results of species-level analyses correspond to biogeographic patterns in
modularity. Low modularity in tropical seed-dispersal systems may be associated with the
importance of social frugivores found in several phylogenetic lineages of tropical birds. The
importance of super-generalist species for the evolution of seed-dispersal systems has been 17
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
noted previously, assuming that these species are primarily large frugivores (e.g. quetzals and
large cotingids, Guimãraes et al. 2011). Our findings suggest that also species-rich lineages of
small tropical frugivores, such as bulbuls (Pycnonotidae) and tanagers (Thraupidae), have
been key nodes in the evolution of these networks. Hence, the low modularity of tropical
seed-dispersal systems may also result from the pervasive connector traits of several lineages
of tropical frugivores.
Conclusions
Networks of any kind are usually poorly represented by binary links (Barrat et al. 2004; Scotti
et al. 2007; Gilarranz et al. 2012), and the robustness and relevance of modularity analysis in
ecology depends on information on interaction frequencies. We employ a weighted
modularity concept for bipartite networks and show that macroecological patterns in seed-
dispersal networks were only detectable in weighted modularity analyses, employing null-
model corrections of sampling bias. Our approach to weighted modularity was also
informative for associating roles of species with phylogeny and species traits, here yielding
similar patterns in weighted and binary analyses. We believe that modularity analyses of
weighted bipartite networks may improve our understanding of the ecological and
evolutionary causes of modularity in different types of bipartite ecological networks. In the
case of seed-dispersal networks, our analysis demonstrates that the modular structure of plant-
frugivore associations is primarily determined by current ecological factors and that
phylogeny and species traits have weak, albeit significant effects on the functional roles of
avian seed disperser species in modular networks.
Acknowledgements
David Weiß and Mathias Templin compiled trait data for frugivorous birds, and Thomas
Hovestadt provided the seed-dispersal network published in his dissertation. MS, LI, DMD,
SAF, MP, FS and KBG were supported by the research funding program Landes-Offensive
18
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
zur Entwicklung Wissenschaftlich-ökonomischer Exzellenz (LOEWE) of Hesse’s Ministry of
Higher Education, Research, and the Arts. DMD was also supported by the German
Academic Exchange Service (DAAD) and FS by the German Research Foundation (DFG).
JCS was supported by the European Research Council (ERC-2012-StG-310886-HISTFUNC).
BD was supported by the Carlsberg Foundation. CFD acknowledges funding by the
Helmholtz Association (VH-NG 247).
References
1. Barber, M. (2007). Modularity and community detection in bipartite networks. Phys. Rev. E, 76, 066102.
2. Barrat, A., Barthélemy, M., Pastor-Satorras, R. & Vespignani, A. (2004). The architecture of complex weighted networks. Proceedings of the National Academy of Sciences of the United States of America, 101, 3747–3752.
3. Bascompte, J. & Jordano, P. (2007). Plant-animal mutualistic networks: the architecture of biodiversity. Ann. Rev. Ecol. Evol. Syst., 38, 567–593.
4. Blomberg, S.P., Garland Jr., T. & Ives, A.R. (2003). Testing for phylogenetic signal in comparative data: behavioral traits are more labile. Evolution, 57, 717–745.
5. Blüthgen, N., Fründ, J., Vázquez, D.P. & Menzel, F. (2008). What do interaction network metrics tell us about specialization and biological traits? Ecology, 89, 3387–3399.
6. Bosch, J., Martín González, A.M., Rodrigo, A. & Navarro, D. (2009). Plant–pollinator networks: adding the pollinator's perspective. Ecol. Lett., 12, 409–419.
7. Braconnot, P., Otto-Bliesner, B., Harrison, S., Joussaume, S., Peterchmitt, J.-Y., Abe-Ouchi, A., et al. (2007). Results of PMIP2 coupled simulations of the Mid-Holocene and Last Glacial Maximum-Part 1: experiments and large-scale features. Clim. Past, 3, 261–277.
8. Burnham, K.P. & Anderson, D.R. (2002). Model selection and multimodel inference. 2nd edn. Springer-Verlag, New York.
9. Chapman, C.A., Chapman, L.J., Struhsaker, T.T., Zanne, A.E., Clark, C.J. & Poulsen, J.R. (2005). A long-term evaluation of fruiting phenology: importance of climate change. J. Trop. Ecol., 21, 31–45.
10. Clauset, A., Moore, C. & Newman, M.E.J. (2008). Hierarchical structure and the prediction of missing links in networks. Nature, 101, 98–101.
11. Clements, J.F., Schulenberg, T.S., Iliff, M.J., Sullivan, B.L., Wood, C.L. & Roberson, D. (2012). The eBird/Clements checklist of birds of the world: Version 6.7. Downloaded from http://www.birds.cornell.edu/clementschecklist/downloadable-clements-checklist [date of download: 19/11/2012].
12. Dalsgaard, B., Magård, E., Fjeldså, J., Martín González, A.M., Rahbek, C., Olesen, J.M., et al. (2011). Specialization in plant-hummingbird networks is associated with species richness, contemporary precipitation and quaternary climate-change velocity. PLoS ONE, 6, e25891.
19
462
463
464
465
466
467
468
469470
471472473
474475
476477
478479
480481
482483484485
486487
488489490
491492
493494495496
497498499500
13. Dalsgaard, B., Trøjelsgaard, K., Martín González, A.M., Nogués-Bravo, D., Ollerton, J., Petanidou, T., et al. (2013). Historical climate-change influences modularity and nestedness of pollination networks. Ecography, in press. DOI: 10.1111/j.1600-0587.2013.00201.x
14. Donatti, C.I., Guimarães, P.R., Galetti, M., Pizo, M.A., Marquitti, F.M.D. & Dirzo, R. (2011). Analysis of a hyper-diverse seed dispersal network: modularity and underlying mechanisms. Ecol. Lett., 14, 773–781.
15. Dormann, C.F. & Strauß, R. (2013). Detecting modules in quantitative bipartite networks: the QuaBiMo algorithm. arXiv:1304.3218v1 [q-bio.QM].
16. Fritz, S.A. & Purvis, A. (2010). Selectivity in mammalian extinction risk and threat types: a new measure of phylogenetic signal strength in binary traits. Cons. Biol., 24,1042–1051.
17. Freckleton, R.P., Harvey, P.H. & Pagel, M. (2002). Phylogenetic analysis and comparative data: A test and review of evidence. Am. Nat., 160, 712–726.
18. Gilarranz, L.J., Pastor, J.M. & Galeano, J. (2012). The architecture of weighted mutualistic networks. Oikos, 121, 1154–1162.
19. Gomez, J.M., Verdú, M. & Perfectti, F. (2010). Ecological interactions are evolutionarily conserved across the entire tree of life. Nature, 465, 918–922.
20. Guimerà, R. & Amaral, L.A.N. (2005). Functional cartography of complex metabolic networks. Nature, 433, 895–900.
21. Guimarães, P.R., Jordano, P. & Thompson, J.N. (2011). Evolution and coevolution in mutualistic networks. Ecol. Lett., 14, 877–885.
22. Hijmans, R.J., Cameron, S.E., Parra, J.L., Jones, P.G. & Jarvis, A. (2005). Very high resolution interpolated climate surfaces for global land areas. Int. J. Climatol., 25, 1965–1978.
23. Howe, H.F. & Smallwood, J. (1982). Ecology of seed dispersal. Annu. Rev. Ecol., 13, 201–228.
24. Ings, T.C., Montoya, J.M., Bascompte, J., Blüthgen, N., Brown, L., Dormann, C.F., et al. (2009). Ecological networks - beyond food webs. J. Anim. Ecol., 78, 253–269.
25. Jetz, W., Thomas, G.H., Joy, J.B., Hartmann, K. & Mooers, A.O. (2012). The global diversity of birds in space and time. Nature, 491, 444–448.
26. Krasnov, B.R., Fortuna, M.A., Mouillot, D., Khokhlova, I.S., Shenbrot, G.I. & Poulin, R. (2012). Phylogenetic signal in module composition and species connectivity in compartmentalized host-parasite networks. Am. Nat., 179, 501–511.
27. Kissling, W.D., Böhning-Gaese, K. & Jetz, W. (2009). The global distribution of frugivory in birds. Global Ecol. Biogeogr., 18, 150–162.
28. Martín González, A.M., Allesina, S., Rodrigo, A. & Bosch, J. (2012). Drivers of compartmentalization in a Mediterranean pollination network. Oikos, 121, 2001–2013.
29. Mello, M.A.R., Marquitti, F.M.D., Guimarães, P.R., Kalko, E.K.V., Jordano, P. & de Aguiar, M.A.M. (2011). The modularity of seed dispersal: differences in structure and robustness between bat- and bird-fruit networks. Oecologia, 167, 131–140.
30. Menke, S., Böhning-Gaese, K. & Schleuning. M. (2012). Plant–frugivore networks are less specialized and more robust at forest–farmland edges than in the interior of a tropical forest. Oikos, 121, 1553–1566.
20
501502503504
505506507
508509
510511512
513514
515516
517518
519520
521522
523524525
526527
528529
530531
532533534
535536
537538
539540541
542543544
31. Newman, M.E.J. & Girvan, M. (2004). Finding and evaluating community structure in networks. Phys.Rev. E, 69, 1–16.
32. Olesen, J.M., Bascompte, J., Dupont, Y.L. & Jordano, P. (2007). The modularity of pollination networks. Proceedings of the National Academy of Sciences of the United States of America, 104, 19891–19896.
33. Pagel, M. (1999). Inferring the historical patterns of biological evolution. Nature, 401, 877–884.
34. Pellmyr, O. (2002). Pollination by animals. In: Plant-Animal Interactions: An Evolutionary Approach (eds. Herrera, C.M. & Pellmyr, O.). Blackwell, Oxford, UK, pp. 157–184.
35. Plein, M., Längsfeld, L., Neuschulz, E.L., Schultheiß, C., Ingmann, L., Töpfer, T., et al. (2013). Constant properties of plant-frugivore networks despite fluctuations in fruit and bird communities in space and time. Ecology, 94, 1296–1306.
36. Rezende, E.L., Lavabre, J.E., Guimarães, P.R., Jordano, P. & Bascompte, J. (2007). Non-random coextinctions in phylogenetically structured mutualistic networks. Nature, 448, 925–928.
37. Sandel, B., Arge, L., Dalsgaard, B., Davies, R.G., Gaston, K.J., Sutherland, W.J., et al. (2011). The influence of late Quaternary climate-change velocity on species endemism. Science, 334, 660–664.
38. Saracco, J.F., Collazo, J.A. & Groom, M.J. (2004). How do frugivores track resources? Insights from spatial analyses of bird foraging in a tropical forest. Oecologia, 139, 235–245.
39. Schleuning, M., Blüthgen, N., Flörchinger, M., Braun, J., Schaefer, H.M. & Böhning-Gaese, K. (2011). Specialization and interaction strength in a tropical plant-frugivore network differ among forest strata. Ecology, 92, 26–36.
40. Schleuning, M., Fründ, J., Klein, A.-M., Abrahamczyk, S., Alarcón, R., Albrecht, M., et al. (2012). Specialization of mutualistic interaction networks decreases toward tropical latitudes. Curr. Biol., 22, 1925–1931.
41. Scotti, M., Podani, J. & Jordán, F. (2007). Weighting, scale dependence and indirect effects in ecological networks: A comparative study. Ecol. Complex., 4, 148–159.
42. Staniczenko, P.P.A., Kopp, J.C. & Allesina, S. (2013). The ghost of nestedness in ecological networks. Nature Communications, 4, 1391.
43. Stouffer, D.B., Sales-Pardo, M., Sirer, M.I. & Bascompte, J. (2012). Evolutionary conservation of species’ roles in food webs. Science, 335, 1489–1492.
44. Thébault, E. (2013). Identifying compartments in presence–absence matrices and bipartite networks: insights into modularity measures. J. Biogeogr., 40, 759–768.
45. Thébault, E. & Fontaine, C. (2010). Stability of ecological communities and the architecture of mutualistic and trophic networks. Science, 329, 853–856.
46. Trøjelsgaard, K. & Olesen, J.M. (2013). Macroecology of pollination networks. Global Ecol. Biogeogr., 22, 149–162.
47. Vázquez, D., Morris, W.F. & Jordano, P. (2005). Interaction frequency as a surrogate for the total effect of animal mutualists on plants. Ecol. Lett., 8, 1088–1094.
48. Vázquez, D.P., Melián, C., Williams, N.M., Blüthgen, N., Krasnov, B.R. & Poulin, R. (2007). Species abundance and asymmetric interaction strength in ecological networks. Oikos, 116, 1120–1127.
21
545546
547548549
550551
552553554
555556557
558559560
561562563
564565566
567568569
570571572
573574
575576
577578
579580
581582
583584
585586
587588589
49. Wheelwright, N.T. (1988). Fruit-eating birds and bird-dispersed plants in the tropics and temperate Zone. Trends Ecol. Evol., 3, 1–5.
50. Woodward, G., Ebenman, B., Emmerson, M., Montoya, J.M., Olesen, J.M., Valido, A., et al. (2005). Body size in ecological networks. Trends Ecol. Evol., 20, 402–409.
22
590591
592593
Table 1. Relationships between weighted and binary modularity ΔQPA and (a) absolute
latitude, (b) mean annual temperature (MAT), (c) temperature seasonality (CVMAT), (d)
temperature climate-change velocity since the Last Glacial Maximum (VELMAT), (e) mean
annual precipitation (MAP), (f) precipitation seasonality (CVMAP), (g) precipitation climate-
change velocity since LGM (VELMAP), and (h) the avian phylogenetic signal in interaction
patterns (PHYLO). Standardized regression coefficients β and their standard errors (SE) are
given for univariate linear models (t values) and SAR models (z values) accounting for spatial
sampling locations. For each predictor variable, we provide r² values with their standard
deviations derived from jackknifing each univariate model, and importance weights (weight)
by summing up the weights of linear models including the respective predictor variable
(computed across all combinations of multi-predictor models that included all predictors
except latitude). Modularity estimates ΔQPA were obtained by correction with a Patefield null
model (Blüthgen et al. 2008), which randomly redistributes interaction events within the
network, constraining the total number of interactions per species.
23
594
595
596
597
598
599
600
601
602
603
604
605
606
607
Weighted analysis Binary analysisβ SE (β) t / z P β SE (β) t / z P
(a) Latitude r² = 0.42 ± 0.045 r² = 0.10 ± 0.0460.649 0.190 3.42 0.004 0.318 0.237 1.34 0.198
SAR model 0.676 0.154 4.39 <0.001 0.248 0.098 2.53 0.011
(b) MAT r² = 0.42 ± 0.047 / weight = 0.23 r² = 0.19 ± 0.059 / weight = 0.51–0.649 0.190 –3.41 0.004 –0.440 0.225 –1.96 0.068
SAR model –0.624 0.193 –0.23 0.001 –0.291 0.111 –0.26 0.009
(c) CVMAT r² = 0.50 ± 0.044 / weight = 0.87 r² = 0.14 ± 0.049 / weight = 0.300.707 0.177 3.99 0.001 0.377 0.232 1.63 0.123
SAR model 0.791 0.090 8.77 <0.001 0.261 0.107 2.44 0.015
(d) VELMAT r² = 0.02 ± 0.023 / weight = 0.25 r² = 0.01 ± 0.011 / weight = 0.220.153 0.247 0.62 0.544 –0.104 0.249 –0.42 0.680
SAR model –0.144 0.212 –0.68 0.497 0.092 0.179 0.52 0.605
(e) MAP r² = 0.29 ± 0.052 / weight = 0.16 r² = 0.10 ± 0.055 / weight = 0.19–0.537 0.211 –2.55 0.021 –0.321 0.237 –1.36 0.194
SAR model –0.482 0.215 –2.24 0.025 –0.215 0.137 –1.57 0.117
(f) CVMAP r² = 0.17 ± 0.076 / weight = 0.14 r² < 0.01 ± 0.007 / weight = 0.24–0.411 0.228 –1.81 0.090 –0.284 0.250 –0.11 0.911
SAR model –0.274 0.226 –1.21 0.227 –0.230 0.127 –1.81 0.070
(g) VELMAP r² = 0.05 ± 0.025 / weight = 0.40 r² < 0.01 ± 0.006 / weight = 0.15–0.232 0.243 –0.95 0.355 –0.101 0.250 –0.04 0.968
SAR model –0.246 0.209 –1.18 0.239 –0.018 0.198 –0.09 0.923
(h) PHYLO r² = 0.07 ± 0.025 / weight = 0.14 r² = 0.18 ± 0.041 / weight = 0.46–0.259 0.242 –1.07 0.299 –0.420 0.227 –1.85 0.083–0.162 0.210 –0. 77 0.442 –0.339 0.216 –1. 57 0.118
24
608
Table 2. Effects of species traits on (a) within-module degree z, (b) among-module
connectivity c and (c) species strength and degree in weighted and binary avian seed-dispersal
networks. Minimal adequate phylogenetic generalized linear models (PGLMs), controlling for
avian phylogeny, are given (according to AICc values), identified in comparisons between
PGLMs containing all possible combinations of five traits of avian frugivores, i.e. degree of
frugivory, social foraging behaviour, migratory behaviour, forest dependence and body mass.
Standardized regression coefficients β with their standard errors, t-statistics and P-values for
each predictor as well as R² and optimized phylogenetic covariation λ for each model are
given.
Weighted analysis Binary analysis
β SE (β) t P β SE (β) t P
(a) z values λ= 0, P = 1; R² = 0.033 λ= 0, P = 1; R² = 0.050
Degree of frugivory 0.324 0.099 3.287 0.001 0.295 0.106 2.776 0.006
Forest dependence - - - - 0.248 0.104 2.375 0.018
(b) c values λ= 0.263, P = 0.056; R² = 0.056 λ= 0.180, P = 0.034; R² = 0.106
Degree of frugivory 0.388 0.109 3.561 <0.001 0.478 0.108 4.443 <0.001
Social foraging 0.354 0.132 2.686 0.008 0.456 0.131 3.476 <0.001
Migratory behaviour - - - - –0.097 0.124 –0.781 0.436
(c) strength / degree λ= 0, P = 1; R² = 0.064 λ= 0.195, P = 0.014; R² = 0.087
Degree of frugivory 0.432 0.094 4.618 <0.001 0.551 0.102 5.421 <0.001
25
609
610
611
612
613
614
615
616
617
Table 3. The importance of bird traits for explaining variation in within-module degree z and
among-module connectivity c in weighted and binary avian seed-dispersal networks. We also
provide the importance of bird traits for variability in species strength and degree. Importance
weights were calculated across phylogenetic generalized linear models (PGLMs) including all
combinations of main effects of trait variables, and the importance for each predictor variable
is given by summing up the Akaike weights of all PGLMs including the respective variable
(Burnham & Anderson 2002, p. 168).
Weighted analysis Binary analysis
z value c value strength
z value c value degree
Degree of frugivory
0.983 0.990 0.999 0.936 0.999 0.999
Social foraging 0.451 0.890 0.465 0.458 0.980 0.596
Migratory behaviour
0.292 0.306 0.268 0.271 0.574 0.277
Forest dependence 0.306 0.337 0.290 0.863 0.191 0.366
Body mass (log) 0.270 0.201 0.288 0.291 0.340 0.382
26
618
619
620
621
622
623
624
625
Fig. 1. Relationships of weighted modularity ΔQVA with ecological factors (a, b), past climate
(c) and phylogenetic signal (d). Shown are (a) current mean annual temperature [MAT in
°Celsius], (b) temperature seasonality [CVMAT in %, computed on the Kelvin scale], (c)
temperature climate-change velocity [VELMAT in m year-1] since the Last Glacial Maximum,
and (d) the avian phylogenetic signal in interaction patterns [PHYLO]. PHYLO is given by
the correlation between interaction and phylogenetic distances of bird species in each
network; significant correlations in (d) are indicated by filled symbols. Standardized
regression coefficients β are given with their P-values; significant relationships (P < 0.05) are
indicated by trend lines derived from linear models (Table 1).27
626
627
628
629
630
631
632
633
634
635
636
Fig. 2. The distribution of weighted (inner circle) and binary (middle circle) c values as well
as species degree (outer circle) across the phylogeny of avian frugivore species. Red tips
indicate high values, blue tips indicate low values. For improved visualization, colour scales
are based on raw c values and standardized species degree (standard deviation units). Branch
lengths are proportional to time (see scale bar), ancestral branches of key taxonomic groups
and bird families are labelled for orientation. Grey branches indicate bird families for which
connector values and species degree were significantly larger than the overall mean in one-
sided t-tests (only families with >5 species tested): Lybiidae, Pipridae and Pycnonotidae for
weighted and binary analyses and species degree, plus Thraupidae for weighted analysis and
species degree, and Turdidae for binary analysis only.
28
637
638
639
640
641
642
643
644
645
646
647
Supplementary Information
Baird USA 40.3 ‐74.7 11.4 3.16 49.9 1155 14.0 6.8 0.26 0.22Carl o Puerto Rico 18.3 ‐66.6 22.2 0.46 0.3 2113 39.0 0.8 0.14 0.18Dehling_ 1 Peru ‐13.1 ‐71.6 19.0 0.29 0.2 1804 51.0 0.0 0.29 0.32Dehling_ 2 Peru ‐13.2 ‐71.6 12.3 0.44 0.3 459 73.0 0.2 0.10 0.12Frost South Africa ‐29.0 31.8 21.6 1.00 3.7 1082 37.0 1.8 0.11 0.32GalettiBrazil ‐22.8 ‐47.1 19.7 0.88 5.5 1311 65.0 4.2 0.35 0.32Gorcho v Peru ‐4.9 ‐73.8 27.0 0.15 9.7 2599 20.0 4.3 0.30 0.30Hovestad t Ivory Coas t 9.0 ‐3.6 27.1 0.51 3.5 1090 70.0 0.0 0.10 0.06Jordan o Spai n 37.6 ‐2.5 13.5 2.30 0.8 462 41.0 3.6 ‐0.06 0.07Kanta k Mexic o 18.5 ‐89.5 24.6 0.81 14.8 1130 58.0 13.4 0.04 0.05Plein German y 50.3 8.7 9.5 2.35 6.2 669 14.0 0.0 0.07 0.18Poulin Panam a 9.2 ‐79.7 26.3 0.26 0.9 2438 60.0 2.0 0.27 0.32Saavedr a Bolivia ‐16.4 ‐67.5 20.0 0.45 0.4 1253 63.0 0.6 ‐0.02 ‐0.02Schleunin g Keny a 0.4 34.9 20.0 0.25 5.5 1903 40.0 1.7 ‐0.01 ‐0.03Snow_ 1 Trinidad 10.7 ‐61.2 23.0 0.24 0.8 2723 40.0 0.0 0.50 0.28Snow_ 2 UK 51.8 ‐0.8 9.8 1.84 81.0 632 12.0 1.0 0.43 0.50Sorense n UK 51.8 ‐1.3 9.7 1.70 19.5 629 14.0 0.3 ‐0.03 0.02Stiebe l German y 51.2 9.0 7.6 2.23 8.1 924 11.0 0.2 0.15 0.09
Network_I D Latitud e Longitude TAMnoitacoL [°C] CVMAT [%]VELMAT [m/year]
binary EVO
weighted EVO
VELMAP [m/year]
MAP [mm] CVMAP [%]
BairdCarloDehling_1Dehling_2FrostGalettiGorcho vHovestadtJordanoKantakPleinPoulinSaavedraSchleuningSnow_1Snow_2SorensenStiebel
Network_ID
direct observat . 180 655 7 21 3 0.37 6 0.00 0 0.298 5 0.32 5 0.00 0 0.139direct observat . 240 949 63 24 6 0.44 2 0.05 8 0.32 2 6 0.35 4 0.00 5 0.099direct observat . 365 5165 49 72 5 0.20 2 0.00 1 0.15 2 7 0.282 0.00 5 0.100direct observat . 365 1563 49 36 5 0.33 7 0.01 9 0.23 4 5 0.319 0.00 3 0.140direct observat . 365 3136 16 8 4 0.31 2 0.00 0 0.27 5 3 0.129 0.00 8 0.108direct observat . 365 397 35 29 6 0.40 1 0.00 5 0.19 5 6 0.37 6 0.00 4 0.127faecal sample s 365 187 91 7 6 0.30 5 0.00 2 0.05 6 5 0.393 0.00 0 0.041direct observat . 365 1720 9 34 39 4 0.23 7 0.01 5 0.21 4 4 0.22 8 0.00 4 0.057direct observat . 365 7010 25 33 3 0.30 2 0.00 9 0.27 5 5 0.325 0.00 7 0.135direct observat . 90 5549 5 27 4 0.24 9 0.00 0 0.220 3 0.18 4 0.00 4 0.142direct observat . 130 3241 27 39 4 0.38 9 0.00 0 0.33 5 5 0.272 0.00 7 0.086faecal sample s 365 492 17 20 5 0.26 3 0.00 0 0.149 4 0.29 4 0.00 8 0.065direct observat . 365 539 40 47 6 0.46 9 0.02 7 0.27 3 7 0.44 4 0.00 6 0.152direct observat . 90 2745 33 83 6 0.32 2 0.00 3 0.24 7 7 0.29 9 0.00 9 0.092direct observat . 365 2144 50 14 6 0.30 0 0.00 3 0.22 7 5 0.273 0.01 2 0.157direct observat . 365 1994 6 29 19 4 0.33 1 0.00 9 0.31 2 4 0.21 7 0.00 5 0.069direct observat . 220 7434 11 14 2 0.23 2 0.00 0 0.22 1 4 0.341 0.00 6 0.152direct observat . 365 6360 29 30 5 0.42 0 0.00 0 0.38 2 5 0.313 0.01 2 0.170
# interaction events
binary ΔQ
Sampling method
Time span [days]
weighted Q
weighted ΔQ
binary Q
weighted # modules
binary # modules
sd (weighted Q )
sd (binary Q )
# bird species
# plant species
BairdCarloDehling_1Dehling_2FrostGalettiGorcho vHovestadtJordanoKantakPleinPoulinSaavedraSchleuningSnow_1Snow_2SorensenStiebel
Network_ID
J. W. Baird, Wilson Bull. 92 , 63 (1980).T. A. Carlo, et al. , Oecologia 134 , 119 (2003).Unpublished data provided by D.M. Dehling, K. Böhning ‐Gaese, M. Schleunin gUnpublished data provided by D.M. Dehling, K. Böhning ‐Gaese, M. Schleunin gP. G. H. Frost, in Acta XVII Congressus Internationalis Ornithologic i , R. Noring, Ed. (Berlin, 1980), pp. 1179 ‐1184.M. Galetti, M. A. Pizo, Ararajuba 4, 71 (1996).D. L. Gorchov, et al. , Oikos 74 , 235 (1995).T. Hovestadt, doctoral thesis, University of Würzburg, Germany (1997).E.L. Rezende, et al. , Nature 448 , 925 (2007).G. E. Kantak, Auk 96 , 183 (1979).M. Plein, et al. , Ecolog y 94 , 1296 (2013).B. Poulin, et al. , J. Trop. Ecol. 15 , 213 (1999).Unpublished data provided by F.V. Saavedra, M. Schleunin gM. Schleuning, et al ., Ecolog y 92 , 26 (2011).B. K. Snow, D. W. Snow, J. Anim. Ecol. 41 , 471 (1972).B. K. Snow, D. W. Snow, Birds and Berrie s (T & AD Poyser, Calton, England, 1988).A. E. Sorensen, Oecologia
29
648
649
Table S1. Detailed information about the 18 avian seed-dispersal networks and their original
sources. We provide sampling locations, current climatic conditions and past climate stability.
For each network, we recorded the avian phylogenetic signal in interactions (raw and partial
Mantel correlations, significant values are indicated with an asterisk), method and time span
of sampling and the number of interaction events, plant and bird species. Modularity measures
are given by the number of detected modules, modularity Q with its standard deviation across
five independent runs of the algorithm and two null-model corrected modularities (ΔQPA,
ΔQVA) for weighted and binary networks (values that were significantly larger than expected
from the respective null models [ΔQNULL.PA, ΔQNULL.VA] are indicated with an asterisk, P < 0.1).
Abbreviations follow those in the main text.
30
650
651
652
653
654
655
656
657
658
659
Table S2. Multivariate analysis of variance testing the relationship between interaction
distances among (a) bird and (b) plant species against the classification of species into
modules. Significant relationships indicate that the categorization of species into modules is
significantly associated with the niche overlap between bird and plant species, respectively.
Interaction distances were derived from weighted and binary interaction matrices with the
Horn-Morisita distance metric. P-values are based on 9,999 permutations of distance
matrices; P-values < 0.1 are printed bold. We provide the number of bird and plant species in
each network because non-significant relationships are more likely to occur in small
networks, due to lack of statistical power. Network abbreviations follow those in Table S1.
(a) Birds
Weighted analysis Binary analysis
# spp. r² F P r² F P
Baird 21 0.277 7.287 0.001 0.248 6.254 0.002
Carlo 24 0.082 1.969 0.067 0.110 2.724 0.006
Dehling_1 72 0.195 16.998 <0.001 0.088 6.747 <0.001
Dehling_2 36 0.133 5.227 <0.001 0.054 1.929 0.043
Frost 8 0.286 2.409 0.089 0.331 2.966 0.037
Galetti 29 0.095 2.845 0.001 0.137 4.270 <0.001
Gorchov 7 0.234 1.524 0.220 0.185 1.137 0.346
Hovestadt 39 0.131 5.599 0.002 0.140 6.017 <0.001
Jordano 33 0.785 113.377 <0.001 0.101 3.469 0.022
Kantak 27 0.171 5.145 0.010 0.167 4.999 0.031
Plein 39 0.313 16.376 <0.001 0.170 7.361 <0.001
Poulin 20 0.301 7.739 <0.001 0.447 14.523 <0.001
Saavedra 47 0.067 3.227 0.001 0.087 4.295 <0.001
Schleuning 83 0.065 5.632 <0.001 0.079 6.910 <0.001
Snow_1 14 0.193 2.863 0.020 0.121 1.655 0.131
Snow_2 19 0.256 5.857 <0.001 0.185 3.848 0.003
Sorensen 14 0.363 6.850 0.011 0.155 2.199 0.067
Stiebel 30 0.215 7.686 <0.001 0.133 4.302 <0.001
31
660
661
662
663
664
665
666
667
668
669
670
(b) Plants
Weighted analysis Binary analysis
# spp. r² F P r² F P
Baird 7 0.515 5.317 0.009 0.168 1.008 0.445
Carlo 63 0.140 9.898 <0.001 0.099 6.708 <0.001
Dehling_1 49 0.088 4.420 <0.001 0.093 4.716 <0.001
Dehling_2 49 0.117 6.230 <0.001 0.077 3.941 <0.001
Frost 16 0.361 7.893 0.001 0.232 4.218 0.003
Galetti 35 0.160 6.297 <0.001 0.153 5.963 <0.001
Gorchov 91 0.143 7.977 <0.001 0.121 6.593 <0.001
Hovestadt 34 0.136 5.041 0.002 0.083 2.894 0.016
Jordano 25 0.228 6.801 <0.001 0.121 3.171 0.004
Kantak 5 0.247 0.982 0.432 0.211 0.801 0.500
Plein 27 0.095 2.617 0.039 0.064 1.697 0.135
Poulin 17 0.108 1.813 0.188 0.244 4.850 0.001
Saavedra 40 0.095 3.992 0.001 0.074 3.035 0.004
Schleuning 33 0.118 4.028 <0.001 0.085 2.785 0.001
Snow_Arima 50 0.132 7.323 <0.001 0.155 8.782 <0.001
Snow_Ayles 29 0.219 7.562 <0.001 0.128 3.963 0.008
Sorensen 11 0.349 4.822 0.018 0.116 1.186 0.324
Stiebel 29 0.167 5.411 0.002 0.133 4.159 0.001
32
671
Table S3. Influence of sampling effort on (a) uncorrected modularity Q and null-model
corrected modularity (b) ΔQPA and (c) ΔQVA of weighted and binary interaction matrices. We
fit linear models that included, in addition to absolute latitude, time span of sampling, species
richness per network (i.e. the sum of plant and animal species) and sampling intensity (i.e. a
measure of the number of observed interaction events per species). We assumed an
exponential decline of modularity with all three measures of sampling effort and included
logarithmic terms of time span, species richness, and sampling intensity in linear models.
Standardized regression coefficients β with their standard errors (SE) and statistics are given.
Effects of sampling intensity were negligible in all analyses of null-model corrected
modularity ΔQPA and ΔQVA. Effects of absolute latitude were significant in all weighted
analyses.
Weighted analysis Binary analysis
β SE (β) t P β SE (β) t P
(a) Modularity Q
Sampling time span –0.112 0.185 –0.60 0.556 0.028 0.167 0.17 0.870
Species richness –0.108 0.260 –0.41 0.686 –0.039 0.236 –0.17 0.871
Sampling intensity –0.960 0.274 –3.45 0.004 –0.987 0.248 –3.98 0.002
Absolute latitude 0.708 0.225 3.14 0.008 0.424 0.204 2.08 0.058
(b) Modularity ΔQPA
Sampling time span –0.190 0.203 –0.94 0.366 –0.051 0.264 –0.19 0.851
Species richness 0.247 0.286 0.86 0.404 –0.229 0.372 –0.62 0.549
Sampling intensity 0.036 0.301 0.12 0.906 –0.195 0.392 –0.50 0.628
Absolute latitude 0.751 0.247 3.04 0.009 0.309 0.322 0.96 0.355
(c) Modularity ΔQVA
Sampling time span –0.205 0.195 –1.05 0.313 0.111 0.219 0.51 0.621
Species richness 0.361 0.275 1.31 0.212 0.785 0.308 2.54 0.025
Sampling intensity 0.210 0.290 0.73 0.481 0.458 0.325 1.41 0.183
Absolute latitude 0.710 0.238 2.98 0.011 0.451 0.267 1.69 0.115
33
672673
674
675
676
677
678
679
680
681
682
683
684
Table S4. Relationships between weighted and binary modularity ΔQVA and (a) absolute
latitude, (b) mean annual temperature (MAT), (c) temperature seasonality (CVMAT), (d)
temperature climate-change velocity since the Last Glacial Maximum (VELMAT), (e) mean
annual precipitation (MAP), (f) precipitation seasonality (CVMAP), (g) precipitation climate-
change velocity since LGM (VELMAP), and (h) the avian phylogenetic signal in interaction
patterns (PHYLO). Standardized regression coefficients β and their standard errors (SE) are
given for univariate linear models (t values) and SAR models (z values) accounting for spatial
sampling locations. For each predictor variable, we provide r² values with their standard
deviations derived from jackknifing each univariate model, and importance weights (weight)
by summing up the weights of linear models including the respective predictor variable
(computed across all combinations of multi-predictor models that included all predictors
except latitude). Modularity estimates ΔQVA were obtained by correction with a Vázquez null
model (Vázquez et al. 2007), which randomly redistributes interaction events within the
network, constraining network connectance and the total number of interactions per species.
34
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
Weighted analysis Binary analysisβ SE (β) t / z P β SE (β) t / z P
(a) Latitude r² = 0.42 ± 0.044 r² = 0.09 ± 0.0410.650 0.190 3.42 0.004 0.302 0.238 1.27 0.223
SAR model 0.656 0.176 3.73 <0.001 0.537 0.115 4.67 <0.001
(b) MAT r² = 0.41 ± 0.046 / weight = 0.23 r² = 0.21 ± 0.060 / weight = 0.60–0.643 0.192 –3.36 0.004 –0.457 0.222 –2.05 0.057
SAR model –0.569 0.198 –2.88 0.004 –0.556 0.153 –3.62 <0.001
(c) CVMAT r² = 0.47 ± 0.046 / weight = 0.86 r² = 0.03 ± 0.032 / weight = 0.210.689 0.181 3.80 0.002 0.172 0.246 0.70 0.495
SAR model 0.808 0.113 7.17 <0.001 0.116 0.244 0.48 0.633
(d) VELMAT r² = 0.02 ± 0.023 / weight = 0.20 r² < 0.01 ± 0.006 / weight = 0.160.150 0.247 0.61 0.553 0.024 0.250 0.10 0.924
SAR model –0.223 0.194 –1.15 0.249 –0.061 0.235 –0.26 0.797
(e) MAP r² = 0.30 ± 0.054 / weight = 0.16 r² < 0.01 ± 0.032 / weight = 0.30–0.545 0.210 –2.60 0.020 –0.068 0.249 –0.27 0.789
SAR model –0.440 0.213 –2.07 0.039 0.069 0.255 0.27 0.786
(f) CVMAP r² = 0.18 ± 0.081 / weight = 0.14 r² = 0.11 ± 0.038 / weight = 0.21–0.427 0.226 –1.89 0.077 –0.329 0.236 –1.39 0.183
SAR model –0.150 0.222 –0.68 0.498 –0.700 0.113 –6.19 <0.001
(g) VELMAP r² = 0.08 ± 0.027 / weight = 0.54 r² = 0.15 ± 0.029 / weight = 0.32–0.284 0.240 –1.19 0.253 –0.386 0.231 –1.67 0.114
SAR model –0.299 0.192 –1.56 0.119 –0.379 0.218 –1.74 0.082
(h) PHYLO r² = 0.09 ± 0.027 /weight = 0.17 r² < 0.01 ± 0.010 / weight = 0.13–0.301 0.238 –1.26 0.225 –0.014 0.250 –0.06 0.957–0.177 0.197 –0.90 0.370 –0.023 0.225 –0.10 0.918
35
36
700
Fig. S1. Relationships between (a, b) sampling intensity (i.e. the number of interaction events
observed per species) and null-model modularity QNULL in weighted and binary analyses.
Shown are relationships for the Patefield null model QNULL.PA (black triangles up) and the
Vázquez null model QNULL.VA (grey triangles down). The Patefield null model randomly
redistributes interaction events within the network, constraining the total number of
interactions per species; the Vázquez null model randomly redistributes interaction events,
constraining network connectance and the total number of interactions per species. Sampling
intensity was closely associated with null-model estimates of QNULL in both null models (raw
values shown in a, b).
Relationships between (c, d) absolute latitude and null-model modularity QNULL in
weighted and binary analyses. Associations with latitude were never significant and were
particularly weak in weighted analyses; values were controlled for differences in sampling
intensity (partial residuals shown in c, d).
Relationships between (e, f) sampling intensity and modularity Q (red symbols) as well
as null-model corrected modularity Q (Q PA, black symbols; Q VA, grey symbols).
Sampling intensity did not influence null-model corrected ΔQ in both null models; values
were controlled for potentially confounding variables, as given in Table S3 (partial residuals
shown in c, d).
37
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
Fig. S2. Relationships of binary modularity ΔQPA with (a) current mean annual temperature
[MAT in °Celsius], (b) temperature seasonality [CVMAT in %, computed on the Kelvin scale],
(c) temperature climate-change velocity [VELMAT in m year-1] since the Last Glacial
Maximum, and (d) the avian phylogenetic signal in interaction patterns [PHYLO]. PHYLO is
given by the correlation between interaction and phylogenetic distances of bird species in
each network; significant correlations in (d) are indicated by filled symbols. Given are
standardized regression coefficients β with their respective P-values; significant relationships
(P < 0.05) are indicated by trend lines derived from linear models (Table 1).
38
720
721
722
723
724
725
726
727
728
729
730
Fig. S3. Influence of sampling intensity on collinearity in (a) z values and (b) c values
between weighted and binary modularity analysis. Pearson correlation coefficients r are
plotted against the sampling intensity in each network. Sampling intensity reflects the number
of interaction events observed per species and was computed as the ratio between the number
of interactions events (square-root-transformed) and the geometric mean of the number of
animal and plant species. Given are standardized regression coefficients β with their
respective P-values for linear models with log-transformed sampling intensity.
39
731
732
733
734
735
736
737
738
739
Appendix S1. Original sources for the compilation of trait and taxonomy information for
avian frugivores. Recent taxonomic information was obtained from Clements et al. (2012).
Bezzel, E. (1993). Kompendium der Vögel Mitteleuropas, Passeres. Aula Verlag, Wiesbaden,
Germany.
Borrow, N. & Demey, R. (2001). Birds of Western Africa. Christopher Helm Publishers,
London, UK.
Clements, J.F., Schulenberg, T.S., Iliff, M.J., Sullivan, B.L., Wood, C.L. & Roberson, D.
(2012). The eBird/Clements checklist of birds of the world: Version 6.7. Downloaded
from http://www.birds.cornell.edu/clementschecklist/downloadable-clements-checklist
[date of download: 19/11/2012].
del Hoyo, J., Elliott, A. & Sargatal, J. (1992). Handbook of the birds of the world, Vol. 1:
Ostrich to Ducks. Lynx Edicions, Barcelona, Spain.
del Hoyo, J., Elliott, A. & Sargatal, J. (1994). Handbook of the birds of the world, Vol. 2: New
World Vultures to Guineafowl. Lynx Edicions, Barcelona, Spain.
del Hoyo, J., Elliott, A. & Sargatal, J. (1996). Handbook of the birds of the world, Vol. 3:
Hoatzin to Auks. Lynx Edicions, Barcelona, Spain.
del Hoyo, J., Elliott, A. & Sargatal, J. (1997). Handbook of the birds of the world, Vol. 4:
Sandgrouse to Cuckoos. Lynx Edicions, Barcelona, Spain.
del Hoyo, J., Elliott, A. & Sargatal, J. (1999). Handbook of the birds of the world, Vol. 5:
Barn Owls to Hummingbirds. Lynx Edicions, Barcelona, Spain.
del Hoyo, J., Elliott, A. & Sargatal, J. (2001). Handbook of the birds of the world, Vol. 6:
Mousebirds to Hornbills. Lynx Edicions, Barcelona, Spain.
del Hoyo, J., Elliott, A. & Sargatal, J. (2002). Handbook of the birds of the world, Vol. 7:
Jacamars to Woodpeckers. Lynx Edicions, Barcelona, Spain.
del Hoyo, J., Elliott, A. & Christie, D.A. (2003). Handbook of the birds of the world, Vol. 8:
Broadbills to Tapaculos. Lynx Edicions, Barcelona, Spain.
del Hoyo, J., Elliott, A. & Christie, D.A. (2004). Handbook of the birds of the world, Vol. 9:
Cotingas to Pipits and Wagtails. Lynx Edicions, Barcelona, Spain.
40
740741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
del Hoyo, J., Elliott, A. & Christie, D.A. (2005). Handbook of the birds of the world, Vol. 10:
Cuckoo-shrikes to Thrushes. Lynx Edicions, Barcelona, Spain.
del Hoyo, J., Elliott, A. & Christie, D. A. (2006). Handbook of the birds of the world. Vol. 11:
Old World Flycatchers to Old World Warblers. Lynx Edicions, Barcelona, Spain.
del Hoyo, J., Elliott, A. & Christie, D. A. (2007). Handbook of the birds of the world. Vol. 12:
Picathartes to Tits and Chickadees. Lynx Edicions, Barcelona, Spain.
del Hoyo, J., Elliott, A. & Christie, D. A. (2008). Handbook of the birds of the world, Vol. 13:
Penduline-tits to Shrikes. Lynx Edicions, Barcelona, Spain.
del Hoyo, J., Elliott, A. & Christie, D. A. (2009). Handbook of the birds of the world, Vol. 14:
Bush-shrikes to Old World Sparrows. Lynx Edicions, Barcelona, Spain.
del Hoyo, J., Elliott, A. & Christie, D. A. (2010). Handbook of the birds of the world, Vol. 15:
Weavers to New World Warblers. Lynx Edicions, Barcelona, Spain.
Dunning, J.B. (2008). CRC handbook of avian body masses. 2nd edn. CRC Press, Boca
Raton, FL.
Ehrlich, P.R., Dobkin, D.S. & Wheye, D. (1988). The Birder’s Handbook. A Field Guide to
the Natural History of North American Birds. Simon & Schuster, New York, NY.
Howell, S.N.G. & Webb, S. (1995). A Guide to the Birds of Mexico and Northern Central
America. Oxford University Press, New York, NY.
Isler, M. L. & Isle, P.R. (1987). The tanagers: natural history, distribution, and identification.
Smithsonian Institution Press, Washington, D.C.
Jaramillo, A. & Burke, P. (1999). New World Blackbirds: The Icterids. Christopher Helm
Publishers, London, UK.
Ridgely, R.S. & Tudor, G. (2009). Birds of South America, Passerines. Christopher Helm
Publishers, London, UK.
Schulenberg, T.S., Stotz, D.F., Lane, D.F., O’Neill, J.P. & Parker III, T.A. (2007). Birds of
Peru. Princeton University Press, Princeton, NJ.
Sibley, D. (2000). The North American Bird Guide. Pica Press, Mountfield, East Sussex, UK.
Sinclair, I. & Ryan, P. (2003). Birds of Africa south of the Sahara. Struik Publishers, Cape
Town, South Africa.
41
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797