Small-scale indirect effects determine the outcome of a tripartite
plant–disperser–granivore interaction
Raphaël Boulay · Francisco Carro ·
Ramón C. Soriguer · Xim Cerdá
Received: 21 November 2008 / Accepted: 15 June 2009 / Published online: 5 July 2009
© Springer-Verlag 2009
Abstract The microhabitat in which plants grow aVects
the outcome of their interactions with animals, particularly
non-specialist consumers. Nevertheless, as most research
on this topic has dealt with either mutualists or antagonists,
little is known about the indirect eVects of plant microhabitats on the outcome of tripartite interactions involving
plants and both mutualists (e.g. seed dispersers) and antagonists (e.g. granivores). During three consecutive years, we
analysed small-scale variations in the interaction of a
perennial myrmecochore, Helleborus foetidus, with its seed
dispersers and consumers as a function of the intensity of
plant cover. Most seeds were released during the day and
were rapidly removed by ants. Nevertheless, the proportion
of ant-removed seeds was higher for plants located in open
microhabitats than for plants surrounded by dense vegetation and rocky cover. Ant sampling revealed that seed
removers were equally abundant, irrespective of the level of
cover. By contrast, a few tiny ant species that feed on the
reward without transporting the seeds were more abundant
in highly covered microhabitats, irrespective of hellebore
diaspore availability. These “cheaters” decrease the chance
Electronic supplementary material The online version of this
article (doi:10.1007/s00442-009-1404-z) contains supplementary
material, which is available to authorized users.
R. Boulay (&) · F. Carro · R. C. Soriguer · X. Cerdá
Estación Biológica de Doñana,
Consejo Superior de Investigaciones CientíWcas,
Avenida Américo Vespucio s/n, 41092 Sevilla, Spain
e-mail: boulay@ebd.csic.es
R. Boulay
Departamento de Biologia Animal, Universidad de Granada,
Avenida Severo Ochoa s/n, 18071 Granada, Spain
of removal by removers and increase the probability of
seeds remaining on the ground until night, when granivore
mice Apodemus sylvaticus become active. Mice also preferred foraging in covered microhabitats, where they consumed a larger proportion of seeds. Therefore, the density
of cover indirectly increased seed predation risk by attracting more seed predators and cheater ants that contribute to
increase seed availability for seed predators. Our results
emphasize the importance of considering the indirect
eVects of plant microhabitat on their dispersal success.
They highlight the indirect eVect of cheaters that are likely
to interfere in mutualisms and may lead to their collapse
unless external factors such as spatio-temporal heterogeneity in seed availability constrain their eVect.
Keywords Ant–plant mutualism · Seed dispersal ·
Microhabitat · Indirect eVects · Myrmecochory
Introduction
Seed dispersal is an important step in plants’ life cycle with
profound implications for succession, regeneration and
conservation (Wang and Smith 2002). The advantages conferred by seed dispersal include colonisation of remote habitats, transport to favourable microhabitats and reduction of
density-dependent mortality near the mother plant (Herrera
2002; Howe and Smallwood 1982; Janzen 1970). These
advantages are not mutually exclusive and may vary geographically and according to the vector of dispersal. In a
recent review, it was found that a majority of studies supported various selective advantages associated to seed dispersal by ants (Giladi 2006). However, in sclerophyllous
and temperate ecosystems, the “predator-avoidance”
hypothesis was the most often supported hypothesis. Up to
123
89% of the studies concluded that an important beneWt of
myrmecochory was to reduce seed predation, mostly by
rodents, under the mother plant. Ants reduce seed predation
by removing myrmecochorous seeds soon after their
release by the plant, before granivores become active
(Boulay et al. 2007a; Ohkawara et al. 1996; Turnbull and
Culver 1983). Once in their nest, ants feed on the elaiosomes (lipid-rich food bodies; Servigne and Detrain 2008)
but not on the embryos that are protected by a strong coat or
by toxins (Majer and Lamont 1985; Oliveras et al. 2008;
Rodgerson 1998). Then, intact seeds may be abandoned in
the ant nest that constitutes a rodent-proof site favourable to
germination (Manzaneda et al. 2005). Another important
way in which ants reduce seed predation is by modifying
the seed shadow (Gorb and Gorb 2003; Heithaus 1981;
O’Dowd and Hay 1980; Pizo and Oliveira 1998). A nonnegligible fraction of seeds removed by ants are lost during
their transport to the nest (Detrain and Tasse 2000; Gorb
and Gorb 1999; Retana et al. 2004). Moreover, after removing
the elaiosomes, ants sometimes discard the seeds outside
several metres from the nest (Bas et al. 2009; BöhningGaese et al. 1999; Gómez and Espadaler 1998; Hughes and
Westoby 1992). By so doing, they reduce seed aggregation
under the mother plant, which decreases their attractiveness
for granivores that prefer foraging on dense seed patches
(Casper 1987; Hulme 1994; Lortie et al. 2000).
The apparent beneWt obtained by plants and their animal
dispersers may constitute raw material for co-evolutionary
processes. In the last decade, various studies have been conducted to determine whether variations in the abundance
and behaviour of dispersers across plant distribution ranges
give rise to geographic mosaics of co-evolution (Thompson
1994, 2005). Nevertheless, as examples of geographic variations in the degree of plant adaptation to local seed dispersers and seed predators accumulate, it also appears that
local conditions have important indirect eVects on seeds by
favouring certain mutualists and antagonists (Alcántara
et al. 2007; Benkman et al. 2001, 2003; Rey and Manzaneda
2007; Zelikova et al. 2008). For example, fearful rodents
consume more seeds from plants surrounded by dense
rocky or vegetation cover, where they can hide from carnivores (Fedriani and Boulay 2006; García-Castaño et al.
2006; Hulme 1997; Hulme and Borelli 1999; Tarborelli
et al. 2003). Plant location with respect to forest edge (Ness
2004), the presence of additional food around the shoots
(Boulay et al. 2005) and the intensity of Xooding disturbance (Prinzing et al. 2007) were also shown to aVect the
foraging behaviour and abundance of seed-dispersing ants.
Despite these results, our understanding of how animals’
microhabitat preferences aVect plant Wtness is limited by
the fact that too few studies have considered the eVect of
plant microhabitat on both dispersers and seed predators,
simultaneously. An important step towards this goal was
123
made by Hulme (1997), who demonstrated that, because
rodents (predators) and ants (secondary dispersers)
favoured diVerent microhabitats, their net eVect on several
shrubs’ dispersal success depended on where birds (primary dispersers) deposited seeds. When seeds were deposited in open areas, ants rapidly removed them, which
reduced granivory, whereas when they were deposited in
covered microhabitats, dispersal rate decreased and granivory increased.
The aim of the present study was to determine how, in a
heterogeneous environment, microhabitat preferences of
both dispersers and seed predators determine plant dispersal
success. Among the ants that interact with the diaspores of
our model system, Helleborus foetidus (Ranunculaceae),
several small species actually behave as “cheaters” by
detaching the elaiosomes in situ without transporting the
seeds, which does not provide any advantage to the plant
(Boulay et al. 2007a; Giladi 2006; Manzaneda et al. 2005).
Here, we investigated to what degree small-scale variations
in plant environment (namely, the density of cover) aVected
the relative availability of seeds for diurnal ants and for
rodents. In addition, we studied the small-scale assemblages
of ants associated to the level of cover to determine whether
variations in the abundance of dispersers, cheaters or both
determined the relative amount of seeds that could be consumed by rodents. Finally, we analysed rodent abundance
and granivory as a function of the level of plant cover.
Materials and methods
Plant system and study site
Helleborus foetidus is a perennial winter-Xowering herb
whose apocarpous fruits mature from late May to mid-July.
We conducted our study in two populations in the south of
Spain separated by about 100 km and located at Parque Natural de las Sierras Subbeticas (900–1,000 m elevation) near
the city of Cabra (hereafter, Cabra) and at Parque Natural de
la Sierra de Grazalema (700–1,000 m elevation; hereafter,
Grazalema). Both sites are representative of the populations
growing in the Mediterranean habitats. The vegetation at
both sites is composed of sparse trees (Quercus lusitanica,
Q. ilex, Acer monspessulanus) and shrubs (Crataegus
monogyna, Rubus fruticosus, Pistacia lentiscus, Ulex baeticus) separated by gaps of lower grassy vegetation.
Plant microhabitat characterization
Surveys and experiments were conducted in 2004, 2005
and 2006 on a total of 241 focal plants (121 and 120 at
Cabra and Grazalema, respectively), 60 and 30 of which
were studied during two and three consecutive years,
respectively. The remaining 151 plants were studied during
one reproductive season only. Focal plants were separated
by at least 7 m. Each year in early May, the density of cover
in a circle of about 1.5 m around each shoot was characterised visually by one member the research group (R.C.S.). A
total of 89 focal plants grew under the highest cover level
characterised by dense shrub/tree canopy and rocky/grassy
ground oVering rodents maximum protection against both
mammal carnivores (e.g. foxes Vulpes vulpes and martens
Martes foina) and raptors (e.g. barn owls Tyto alba and
tawny owls Strix aluco). Eighty-four focal plants grew in
open microhabitats, that is, on bare ground without tree or
shrub canopy, which oVered little or no hiding place for
rodents. Finally, 68 focal plants had an intermediate cover
with either dense canopy or rocky/grassy cover, but not
both.
Flower, carpel, seed production and seed availability
The total number of Xowers (NXowers) and carpels (Ncarpels)
produced by each focal plant was counted at the beginning
of May each year. During the fruit-ripening season, 4 carpels were randomly selected on each focal plant to evaluate
the average number of seeds per carpel (s) and the total
number of seeds per plant Nseeds = s £ Ncarpels. Then, we
counted the number of closed carpels (ct) on each plant
three times over a 24-h period, at 0800 (c0), 2000 (c12) and
0800 hours (c24) the following day. The proportion of seeds
that were released during the day and could potentially be
removed by ants before rodents became active was deduced
from the rhythm of carpel opening and calculated as
Pseeds(ants) = (c12¡c0)/(c24¡c0). However, this estimate does
not account for small-scale variations in disperser abundance aVecting the proportion of seeds exposed to rodent
predation. To evaluate the proportion of seeds released during the day but not removed by diurnal ants (a), the contents of two to three carpels were deposited at 0800 hours
in a 3-cm Petri dish covered with a wire mesh to allow seed
removal by ants only. The number of seeds in the dishes
was then counted at 2000 hours. The proportion of seeds
exposed to predation by mice was further calculated as
Pseeds(mice) = (1 ¡ Pseeds(ants)) + (a £ Pseeds(ants)), that is, the
proportion of seeds released during the night plus the
proportion of seeds released during the day but not
removed by ants. This equation can be reduced as
Pseeds(mice) = 1¡Pseeds(ants) (1 + a).
Distribution and abundance of ants and rodents
Ants and rodents were sampled twice each year, in early
spring and mid-summer, about 3 weeks before and 3 weeks
after fruit maturation. This delay was suYcient to guarantee
that animal sampling reXected the quality of plant
microhabitat, independently of seed crop. Ants were sampled by placing 2 pitfall traps (white 20-cm3 plastic cups
half-Wlled with soap water) on opposite sides of each focal
plant during 24 h. All sampled material was transferred to
the laboratory to be identiWed. Ant species were separated
in four behavioural categories according to Manzaneda
et al. (2007) and Boulay et al. (2007a): (1) seed removers,
(2) cheaters (=elaiosome predators), (3) species with a
mixed disperser-cheater behaviour, and (4) indiVerent (species that mostly ignore hellebore diaspores).
Rodents were sampled during 2–3 consecutive nights,
using 2 traps per focal plant. Each pair of traps was composed of one Ugglan-type trap that closes by gravity and
one “Hipolito-type” trap (Carro et al. 2007) that closes
when an animal tries to remove the bait (peanut butter in
our case). All traps were checked at dawn and dusk. Captured mice were sexed, marked by toe clipping and released
at the spot of capture.
EVect of plant cover on seed removal by ants and rodents
during the nocturnal phase
Determining the speciWc rate of seed removal by rodents by
excluding ants and allowing seed access to rodents only is a
complicated task that frequently fails or lacks an appropriate control (Fedriani et al. 2004). For that reason, we opted
to estimate the eVect of rodents indirectly by comparing
seed disappearance due to either ant activity only or to both
ants and rodents. In June 2004 and 2005, we placed two
depots (3-cm diameter Petri dishes), each containing 48
elaiosome-bearing seeds on each side of 30 focal plants per
population. For each plant, one depot was covered with a
wire mesh that allowed the passage of ants but prevented
access by mice (rodent-excluded depots). The other one
was open to both ants and mice (rodent-allowed depots).
All depots were set up at 2000 hours and the number of
remaining seeds was monitored at 0800 hours the following
day.
Data analysis
Statistical analyses aimed at testing the eVects of Plant
Cover, Population and Year on plant data (NXowers, Ncarpels,
Nseeds, Nseeds(ants) and Nseeds(mice)), ant data (the occurrence of
seed removers, cheaters, mixed behaviour species and
indiVerent species near the focal plants) and mice data (the
probability of rodent capture). The number of seeds remaining in the depots after one night was also analyzed as a
function of Year, Population, Plant cover plus Rodent
exclusion treatment. To that aim, a series of General Linear
Mixed Models were Wtted using the Non-linear Mixed
EVects Model. All models accounted for random variations
of the response variables between plants within populations.
123
Counts and proportions were log and arcsin transformed,
respectively. For each response variable, the maximal
model (containing all factors and interactions) was Wrst
Wtted to the data. The most parsimonious model was then
obtained by successively eliminating eVects with low
explanatory power (Crawley 2007). In practice, a factor or
an interaction was retained if removing it signiWcantly
enhanced the Akaike’s Information Criterion (AIC). If the
diVerence of AIC between the current model and the previous, more complicated model was not signiWcant, the eVect
was removed. So the model that best Wtted to the data has
the smallest AIC and only contained signiWcant eVects. The
signiWcance of diVerences between levels of signiWcant factors was assessed a posteriori using the same method.
BrieXy, levels of signiWcant factors were recoded to
regroup those with the smallest observed mean diVerences.
Comparison of AIC between the model with all levels and
the model with regrouped levels indicated whether a diVerence was signiWcant or not. The Bonferroni procedure was
employed to control type I error inXation due to the correlation between variables. Thus, for NXowers, Ncarpels, Nseeds,
Pseeds(ants) and Pseeds(mice), the signiWcance threshold to
exclude or retain an eVect in the model was = 0.01. For
the occurrence of disperser, cheater, mixed behaviour and
indiVerent ants, = 0.0125. After model selection, the
“anova” function (F statistics) was used to summarise the
signiWcance of factors retained in the most parsimonious
model. However, because our experimental design was not
orthogonal, the eVects of factors that were not retained in
the most parsimonious model were not tested with the
anova function (Crawley 2007). All estimates are
means § SE.
Results
EVect of plant microhabitat on Xower, carpel and seed
production, seed availability for ants and rodents
in 2005, these three variables were signiWcantly smaller
than in 2004–2006 considered together (NXowers:
30.6 § 2.2; Ncarpels: 66.9 § 5.4 and Nseeds: 726.9 § 64.7,
respectively; anova: Year—F1,118 = 29.3, P < 0.001;
F1,118 = 30.0, P < 0.001 and F1,118 = 26.0, P < 0.001 for
NXowers, Ncarpels and Nseeds, respectively).
For Pseed(ants), successive removal of all eVects did not
cause any signiWcant increase of the AIC, indicating that
none of the factors had a signiWcant eVect on the proportion
of seeds available for ants (ESM appendix 1, model 35).
Hence, focal plants released 90.7% § 0.98% of their seeds
during the diurnal phase (Pseed(ants)), independently of Plant
Cover, Population or Year (Fig. 1). By contrast, the most
parsimonious model for Pseed(mice) included Plant Cover but
no other eVect (ESM appendix 1, model 43). Thus, the proportion of seeds that remained available for rodents at dusk
depended signiWcantly on the level of cover (Fig. 1). The
highly signiWcant eVect of Plant Cover on Pseed(mice)
resulted from the fact that ants removed fewer seeds from
densely covered microhabitats (66%) than from mediumcovered and open microhabitats (87 and 94%, respectively). Deletion of Cover levels indicated that Pseed(mice)
was not signiWcantly diVerent in medium-covered and in
open microhabitats (regrouping both levels did not signiWcantly increase the AIC), but was signiWcantly higher in
densely covered microhabitats than in open and mediumcovered microhabitats considered together (anova: Cover—
F1,238 = 30.0, P < 0.001).
The eVect of plant cover on ants and mice visits
The Cabra and Grazalema ant communities were composed
of typically Mediterranean species, the abundance and
behaviour of which are given in ESM appendix 2. AICbased model simpliWcation indicated that the occurrence of
1
0.9
The most parsimonious models for NXowers, Ncarpels and
Nseeds only retained Year as a signiWcant Wxed factor (Electronic supplementary material, ESM, appendix 1, models 8,
17 and 26, respectively). Therefore, these three variables
diVered signiWcantly between years but not between microhabitats or populations. Stepwise deletion revealed that the
diVerences of NXowers, Ncarpels and Nseeds between 2004 and
2006 were not signiWcant (NXowers: 49.6 § 2.7 vs
49.2 § 3.6, Ncarpels: 113.2 § 6.8 vs 114.8 § 8.0, and Nseeds:
1,298.3 § 89.3 vs 1,113.4 § 82.7 for 2004 and 2006,
respectively). In eVect, regrouping 2004 and 2006 in a single level did not lead to a signiWcant increase of the model
AIC compared with the more complicated 3 year-levels
model (ESM appendix 1, models 7, 16 and 25). However,
123
0.8
Open
0.7
0.6
Medium
0.5
Dense
0.4
b
a
0.3
a
0.2
0.1
0
Ants
Mice
Seed consumers
Fig. 1 Proportion (mean § SE) of seeds available for diurnal ants
Pants and for nocturnal mice Pmice as a function of the level of cover
of the hellebore (Helleborus foetidus) shoots. DiVerent letters denote
signiWcant diVerences between levels of cover
seed removers such as Aphaenogaster senilis and Camponotus cruentatus was signiWcantly higher at Cabra than at
Grazalema (2.39 § 0.07 vs 1.54 § 0.04 occurrences,
respectively; ESM appendix 1, model 51; anova: Population—F1,239 = 69.8, P < 0.0001) but did not diVer signiWcantly between years or levels of cover.
A diVerent pattern was found for the occurrence of
cheater ants such as Crematogaster scutellaris or Tetramorium spp. The occurrence of these species was inXuenced
by Population and Plant Cover, although there was no signiWcant interaction between both factors (ESM appendix 1,
model 71). Cheater ants were more frequent at Grazalema
than at Cabra (anova: Population—F1,238 = 32.0,
P < 0.0001). Although they occurred with a similar frequency in open than in medium-cover microhabitats, they
were more frequent in densely covered microhabitats than
in the other two microhabitats (ESM appendix 1; anova:
Cover—F1,238 = 16.4, P < 0.0001).
The occurrence of species with a mixed behaviour (e.g.
Pheidole pallidula) was not signiWcantly aVected by any of
the tested eVects (Fig. 2; ESM appendix 1, model 79).
Finally, the pattern of occurrence of ant species that
ignored hellebore seeds was more complicated (ESM
appendix 1, model 86). More of these species were encountered in densely covered microhabitats than in open and
medium-covered microhabitats (Fig. 2; anova: Cover—
F2,237 = 5.34, P = 0.005). Moreover, the interaction
between Population and Year was signiWcant (anova:
Year—F1,118 = 1.22, P = 0.2708; Population: F1,237 = 12.5,
P < 0.0001; Year £ Population: F1,118 = 23.12, P < 0.0001).
Thus, in 2004 the number of “indiVerent” ant species
trapped at Grazalema was lower than at Cabra (2004:
0.29 § 0.06 vs. 0.48 § 0.09, for Grazalema and Cabra,
respectively) while, it was the opposite during the following
1.4
b
1.2
1
a
Open
years (2005–2006: 0.23 § 0.05 vs. 0.65 § 0.08, for Grazalema and Cabra, respectively).
Only one species of rodent, the wood mouse Apodemus
sylvaticus, was captured at Cabra and Grazalema (total: 53
females and 64 males). Only 34% of the marked individuals were recaptured, most of them only once. More wood
mice were captured near plants with dense cover than near
those with low or medium cover (Fig. 3; ESM appendix 1,
model 93; anova: Cover—F1, 238 = 8.7, P < 0.0035). Overall, the probability of mice capture per plant and night of
trapping was also much higher in 2004 and 2006
(0.10 § 0.02 on average for both years) than in 2005
(0.01 § 0.02; anova: Year—F2,116 = 13.7, P < 0.001).
Although over the 3 years of study there was no signiWcant
diVerence in the probability of rodent capture between
Cabra and Grazalema, the diVerence between years was
greater at Cabra than at Grazalema (anova:
Year £ Population—F2,116 = 14.3, P < 0.001).
EVect of plant cover on seed removal during the nocturnal
phase
Analysis of the maximal model, containing all interactions
between Rodent Exclusion, Year, Cover and Population
indicated that the latter factor did not signiWcantly aVect
the amount of seeds removed per night. Therefore, Population was not included in further steps of model selection. However, there was a highly signiWcant 3-way
interaction between Rodent Exclusion, Plant Cover and
Year (anova: Exclusion £ Cover £ Year—F1,170 = 11.6,
P < 0.001. Among Plant Cover levels, regrouping mediumcovered and non-covered plants in the same level did not
enhance the model AIC, indicating that signiWcant diVerences were between highly covered and less covered plants
(ESM appendix 1, model 98). Comparison of seed removal
from rodent-allowed and rodent-excluded depots indicated
that, overall, rodents removed a signiWcant amount of seeds
Medium
a
Dense
b
0.8
0.3
b
0.6
a
a
0.4
b
Open
0.2
Medium
Dense
a
0.2
a
0
Seed removers
Cheaters
Mixed
Indifferent
0.1
a
b
Category
Fig. 2 Ant species behaviour occurrences (number of species;
mean § SE) as a function of the level of plant cover. Ant behaviour
was categorized as following: Seed removers transport the entire diaspores; Cheaters consume the elaiosome in situ without transporting the
a
a
a
0
2004
2005
2006
Years
seed; Mixed can either behave as dispersers or as cheaters; IndiVerent
Fig. 3 Probability (mean § SE) of mice capture in diVerent cover
ignore hellebore diaspores, with or without an elaiosome. DiVerent
types during the 3 years of study. DiVerent letters denote signiWcant
diVerences between cover levels (P < 0.05)
letters denote signiWcant diVerences between levels of cover
123
Open
40
Medium
35
a
Dense
ab
30
b
25
20
15
10
5
0
Rodents-excluded depots
Rodents-allowed depots
Treatment
Fig. 4 Number (mean § SE) of hellebore seeds remaining in the
depots after one night as a function of the level of cover and rodent
exclusion treatment. Initially, 48 diaspores were put in each depot.
DiVerent letters denote signiWcant diVerences between levels of cover
for rodent-excluded and rodent-allowed depots (P < 0.05)
(Fig. 4; Rodent Exclusion: F1,170 = 18.2, P < 0.0001). However, the diVerence between rodent-excluded and rodentallowed depots was signiWcantly more pronounced for
plants located in densely covered microhabitats than for
those located in medium-covered and open microhabitats (anova: Cover—F1,170 = 15.70, P < 0.0001, Rodent
Exclusion £ Cover: F1,172 = 14.72, P < 0.0001). Finally,
the interaction between Cover and Year was slightly signiWcant (anova: Cover £ Year—F1,170 = 4.55, P < 0.0343),
indicating that the diVerence of seed removal between
highly covered and less covered microhabitats was greater
in 2004 than in 2005.
Discussion
The results of this 3-year study indicate that the dispersal
success of H. foetidus shoots may largely depend on smallscale environmental conditions, namely the degree of
cover, through its indirect eVect on both ants and rodents.
The level of cover had no signiWcant net eVect on the production of Xowers, carpels and seeds, suggesting that predispersal direct (e.g. through the amount of light received
by the plants) and indirect (through pollinators and herbivores) factors that potentially aVected seed crop in diVerent
microhabitats compensated each other. By contrast, the
proportion of seeds that were exposed to granivory by
rodents depended on the proportion that was not removed
by ants, whose activity was clearly inXuenced by the level
of cover. Most seeds were released during the light phase
and, on average, diurnal ants removed 94% of them in open
microhabitats but only 66% in densely covered microhabitats. Because nocturnal ants harvested few seeds, independently of the level of cover, the net eVect of ants was to
123
indirectly enhance the relative availability of seeds for
rodents in covered microhabitats with respect to open
microhabitats. Rodents also preferred foraging in covered
microhabitats where they removed more seeds, probably
because they were less visible to potential predators (Edut
and Eilam 2003). Apodemus sylvaticus is not known to
cache hellebore seeds and, in all likelihood, consumed most
of the seeds they removed (Fedriani and Boulay 2006).
These results tend to conWrm those of Hulme (1997),
who suggested that seed removal by ants and rodents
depended on micro-environmental factors. A major diVerence between both studies is that, in our study, ants are the
primary dispersers of hellebore seeds, whereas they only
acted as secondary dispersers of the plants studied by
Hulme. Moreover, the assemblages of ants associated with
H. foetidus are complex and involve seed removers, cheaters and species with a mixed behaviour (Manzaneda et al.
2007). Four species of ants (A. senilis, C. cruentatus, Cataglyphis velox and Messor structor) remove a large number
of seeds under the mother plant and carry them to their
nests where they consume the elaiosomes (Boulay et al.
2006). Interestingly, species of the genus Aphaenogaster
are the main transporters of Mediterranean and North
American myrmecochores (Bas et al. 2009; Boulay et al.
2007a; Smallwood and Culver 1979; Zelikova et al. 2008
and references therein). This is due to a conjunction of
traits including their relatively large size, abundance, wide
diet and mode of foraging. Aphaenogaster species are generally subordinates whose scouts rapidly discover food
items but are unable to defend them against smaller dominant species. Therefore, they prefer to transport the elaiosome-bearing seeds to their nest and detach the elaiosomes
there, rather than feeding on them in situ.
Although our study did not investigate seed fate after
removal by ants, the behavioural characteristics of the four
main seed removers and several Weld observations suggest
that they contribute to modify the seed shadow and reduce
granivory. First, the excavation of a few A. senilis nests
suggests that part of the seeds are stored in superWcial
chambers where they are protected from rodents and can
germinate after the ants have abandonned the nest (Howe
and Smallwood 1982; Smallwood and Culver 1979). Second, a large amount of seeds may be scattered over large
distances (as far as 40 m in the case of C. cruentatus;
Boulay et al. 2007b). A proportion of seeds may be lost
during transport to the nest (Detrain and Tasse 2000; Gorb
and Gorb 1999; Retana et al. 2004) or rejected a few metres
away from the nest entrance (Gómez and Espadaler 1998;
Bas et al. 2009). Hence, we have frequently seen both
C. cruentatus and Cataglyphis velox transporting hellebore
seeds more than 10 m from their nests (R.B., personal
observation). Although seeds scattered outside the nest can
still be discovered by rodents, this is much more unlikely
than under the mother plant, where seed density is high
(Casper 1987; Hulme 1994; Lortie et al. 2000). At Cabra,
seed removal by M. structor may be less beneWcial because
this ant is mostly granivore. However, the strong coat of
hellebore seeds may constitute an adaptation against ants’
granivory (Boulay et al. 2005).
Mutualisms in general are likely to generate species that
take up the rewards without returning any beneWts in
exchange (Bronstein 1994; Fedriani and Boulay 2006;
Herre et al. 1999). This lack of reciprocity may disrupt the
cooperation unless cheaters’ eVect is constrained by ecological factors (Segraves et al. 2005; Yu et al. 2001). In our
case study, cheater ants are small species that, unlike
Aphaenogaster and Camponotus, can dominate food items
by recruiting numerous workers. They prefer to detach the
elaiosomes in situ and, by so doing, reduce diaspore
rewarding for seed removers (Boulay et al. 2007a). This
may limit the colonisation of new areas, which in itself
reduces plant Wtness (Andersen 1988; Valverde and
Silvertown 1995). More importantly, cheaters may indirectly increase the proportion of seeds remaining on the
substrate until night and, consequently, the risk of predation
by rodents (Boulay et al. 2007a; Garrido et al. 2002; Ohara
and Higashi 1987). Nevertheless, the occurrence of cheaters was spatially constrained to the most covered microhabitats, maybe because the small size of these ants reduces
their tolerance to high ground temperatures and forces them
to forage in the shade (Cerdá et al. 1997). Moreover, their
indirect eVect on seed granivory depended on spatiotemporal variations in rodent abundance. In 2005, rodent
population dropped down which reduced cheaters impact
on seeds. Such limitations to the eVect of cheaters on plant
Wtness may contribute to maintaining the selective advantage of myrmecochory and prevent the breakdown of this
mutualism.
Indirect eVects have long been suspected to have major
and often underestimated consequences in species communities, and particularly in the various interactions between
plants and consumers (Strauss 1991; Strauss and Irwin
2004). Our results and those of Hulme (1997) suggest that
the strength of the interactions that a plant maintains with
multiple dispersers and granivores depends on the context
(e.g. the level of cover). Such local scale indirect eVects are
likely to blur the potential selection exerted by consumers
on plant dispersal-enhancing traits such as elaiosome quality (Boulay et al. 2006; Mark and Olesen 1996), seed size
(Garrido et al. 2002) and the phenology and timing of seed
release (Boulay et al. 2007a; Oberrath and Böhning-Gaese
2002). This, in turn, may limit the formation of a geographic mosaic of plant adaptations to local dispersers in
spite of the major variations in the assemblages of consumers (Alcántara et al. 2007; Garrido et al. 2002; Manzaneda
et al. 2007). More generally, the guilds of plant enemies
and mutualists are often composed of non-specialist animals that depend less on plants than vice versa. Our study
exempliWes the fact that these consumers’ foraging decisions and their eVect on plant Wtness might be inXuenced by
a multitude of environmental variables that are diYcult to
disentangle. Taking into account such a complexity of
small-scale indirect eVects on species interactions is arduous. However, it is a necessary task in order to determine
the extent to which selection exerted by animals on plants
can generate local adaptations and lead to the formation of
a large-scale geographic mosaic of co-evolution.
Acknowledgments This work was funded by the Spanish “Ministerio de Educación y Ciencia” (project BOS2003-01536 and CGL200604968/BOS to X.C. and R.B.). It was conducted with the authorization
of Parque Natural de la Sierra de Grazalema and Parque Natural de las
Sierras Subbéticas. Accommodation at Cabra was kindly provided by
the Camacho family and at Grazalema by the authority of the park.
Assistance in the Weld was provided by E. Angulo, A. Carvajal,
I. Luque, O. González, E. García Márquez, I. Villalta and M. Vonshak.
We also thank J. Minett for editing assistance and Drs J. Cronin and
T. Payne for constructive comments on a previous version of the man-
uscript. R.B. was funded by an I3P fellowship. This work complies
with current Spanish laws.
References
Alcántara JM, Rey PJ, Manzaneda AJ, Boulay R, Ramírez JM,
Fedriani
JM (2007)
Geographic
variation
in the adaptive
landscape for seed size at dispersal of the myrmecochorous
Helleborus foetidus. Evol Ecol 21:411–430
Andersen AN (1988) Dispersal distance as a beneWt of myrmecochory.
Oecologia 75:507–511
Bas JM, Oliveras J, Goméz C (2009) Myrmecochory and short-term
seed fate in Rhamnus alaternus: ant species and seed characteristics A. Oecology (in press)
Benkman CW, Holimon WC, Smith JW (2001) The inXuence of a
competitor on the geographic mosaic of coevolution between
crossbills and lodgepole pine. Evolution 55:282–294
Benkman CW, Parchman TL, Favis A, Siepielski AM (2003) Recipro-
cal selection causes a coevolutionary arms race between crossbills
and lodgepole pine. Am Nat 162:182–194
Böhning-Gaese K, Gaese BH, Rabemanatsoa SB (1999) Importance of
primary and secondary seed dispersal in the Malagasy tree Com-
niphora guillaumini. Ecology 80:821–832
Boulay R, Fedriani JM, Manzaneda AJ, Cerdá X (2005) Indirect eVects
of alternative food resources in an ant–plant interaction. Oecologia 144:72–79
Boulay R, Coll-Toledano J, Cerdá X (2006) Geographic variations in
Helleborus foetidus elaiosome lipid composition: implications for
dispersal by ants. Chemoecology 16:1–7
Boulay R, Carro F, Soriguer RC, Cerdá X (2007a) Synchrony between
fruit maturation and eVective dispersers’ foraging activity
increases seed protection against seed predators. Proc R Soc Lond
B 274:2515–2522
Boulay R, Cerdá X, Simon T, Roldan M, Hefetz A (2007b) Intraspe-
ciWc competition in the carpenter ant Camponotus cruentatus:
should we expect the Dear Enemy EVect? Anim Behav 74:985–
993
Bronstein JL (1994) Conditional outcomes in mutualistic interactions.
Trends Ecol Evol 9:214–217
123
Carro F, Pérez-Aranda D, Lamosa A, Schmalenberger HP, Pardavila
Manzaneda AJ, Fedriani JM, Rey PJ (2005) Adaptive advantages of
X, Soriguer RC (2007) Indice de capturas y tipo de trampa: ¿qué
myrmecochory: the predator-avoidance hypothesis tested over a
trampa es para capturar micromamiferos? Galemys 19:73–81
Casper BB (1987) Spatial patterns of seed dispersal and postdispersal
wide geographic range. Ecography 28:583–592
Manzaneda AJ, Rey P, Boulay R (2007) Geographic and temporal
seed predation of Cryptantha Xava (Boraginaceae). Am J Bot
74:1646–1655
variations in the ant–seed dispersal assemblage of the perennial
Cerdá X, Retana J, Cros S (1997) Thermal disruption of transitive hierarchies in Mediterranean ant communities. J Anim Ecol 66:363–
374
Crawley MJ (2007) The R book. Wiley, New York
Detrain C, Tasse O (2000) Seed drops and caches by the harvester ant
Messor barbarus: do they contribute to seed dispersal in Mediter-
ranean grasslands? Naturwissenschaften 87:373–376
Edut S, Eilam D (2003) Rodents in open space adjust their behavioral
response to the diVerent risk levels during barn-owl attack. BMC
Ecol 3:1–16
Fedriani JM, Boulay R (2006) Foraging by fearful frugivores: combined eVects of fruit ripening and predation risk. Funct Ecol
20:1070–1079
Fedriani JM, Rey PJ, Garrido JL, Guitián J, Herrera CM, Medrano M,
Sánchez-Lafuente AM, Cerdá X (2004) Geographical variation in
the potential of mice to constrain an ant–seed dispersal mutual-
ism. Oikos 105:181–191
García-Castaño JL, Kollman J, Jordano P (2006) Spatial variation of
post-dispersal seed removal by rodents in highland microhabitats
of Spain and Switzerland. Seed Sci Res 16:213–222
Garrido JL, Rey P, Cerdá X, Herrera C (2002) Geographic variation of
diaspore trati of an ant-dispersed plant (Helleborus foetidus): are
ant community and diaspore trait correlated. J Ecol 90:446–455
Giladi I (2006) Choosing beneWts or partners: a review of the evidence
for the evolution of myrmecochory. Oikos 112:481–492
Gómez C, Espadaler X (1998) Aphaenogaster senilis Mayr (Hymenoptera, Formicidae): a possible parasite in the Myrmecochory of
Euphorbia characias. Sociobiology 32:441–450
Gorb SN, Gorb E (1999) Dropping rates of elaiosome-bearing seeds
during transport by ants (Formica polyctena Foerst.): implica-
tions for distance dispersal. Acta Oecol 20:509–518
Gorb E, Gorb S (2003) Seed dispersal by ants in a deciduous forest
ecosystem. Kluwer, Dordrecht
Heithaus ER (1981) Seed predation by rodents on three ant-dispersed
plants. Ecology 62:136–145
Herre EA, Knowlton N, Mueller U, Rehner S (1999) The evolution of
mutualisms: exploring the paths between conXict and coopera-
herb Helleborus foetidus. Biol J Linn Soc 92:135–150
Mark S, Olesen JM (1996) Importance of elaiosome size to removal of
ant-dispersal seeds. Oecologia 107:95–101
Ness JH (2004) Forest edges and Wre ants alter the seed shadow of an
ant-dispersed plant. Oecologia 138:448–454
O’Dowd DJ, Hay ME (1980) Mutualism between harvester ants and a
desert ephemeral: seed escape from rodents. Ecology 61:531–540
Oberrath R, Böhning-Gaese K (2002) Phenological adaptation of ant-
dispersed plants to seasonal variation in ant activity. Ecology
83:1412–1420
Ohara M, Higashi S (1987) Interference by ground beetles with the dis-
persal by ants of seeds of Trillium species (Liliaceae). J Ecol
75:1091–1098
Ohkawara K, Higashi S, Ohara M (1996) EVects of ants, ground bee-
tles and the seed-fall patterns on myrmecochory of Erythronium
japonicum Decne. (Liliaceae). Oecologia 106:500–506
Oliveras J, Gómez C, Bas JM, Espadaler X (2008) Mechanical defence
in seeds to avoid predation by a granivorous ant. Naturwissenschaften 95:501–506
Pizo MA, Oliveira PS (1998) Interaction between ants and seeds of a
nonmyrmecochorous neotropical tree, Cabralea canjerana (Meliaceae), in the Atlantic forest of southeast Brazil. Am J Bot
85:669–674
Prinzing A, Dauber J, Hammer EC, Hammouti N, Böhning-Gaese K
(2007) Perturbed partners: opposite responses of plant and animal
mutualist guilds to inundation disturbances. Oikos 116:1299–
1310
Retana J, Picó X, Rodrigo A (2004) Dual role of harvesting ants as seed
predators and dispersers of a non-myrmechorous mediterranean
perennial herb. Oikos 105:377–385
Rey PJ, Manzaneda AJ (2007) Geographical variation in the determinants of seed dispersal success of a myrmecochorous herb. J Ecol
95:1381–1393
Rodgerson L (1998) Mechanical defense in seeds adapted for ant
dispersal. Ecology 79:1669–1677
Segraves KA, AlthoV DM, Pellmyr O (2005) Limiting cheaters in
mutualism: evidence from hybridization between mutualist and
cheater yucca moths. Proc R Soc Lond B 272:2195–2201
tion. Trends Ecol Evol 14:49–53
Herrera CM (2002) Seed dispersal by vertebrates. In: Herrera CM,
Servigne P, Detrain C (2008) Ant–seed interactions: combined eVects
Pellmyr O (eds) Plant–animal interactions. Blackwell, Oxford,
Howe HF, Smallwood J (1982) Ecology of seed dispersal. Annu Rev
55:220–230
Smallwood J, Culver DC (1979) Colony movements of some North
American ants. J Anim Ecol 48:373–382
Ecol Syst 13:201–228
Hughes L, Westoby M (1992) Fate of seeds adapted for dispersal by
Strauss SY (1991) Indirect eVects in community ecology: their deWnition, study and importance. Trends Ecol Evol 6:206–210
ants in Australian sclerophyll vegetation. Ecology 73:1285–1299
Hulme PE (1994) Post-dispersal seed predation in grassland: its mag-
Strauss SY, Irwin RE (2004) Ecological and evolutionary consequences of multispecies plant–animal interactions. Annu Rev
nitude and sources of variation. J Ecol 82:645–652
Hulme PE (1997) Post-dispersal seed predation and the establishment
Ecol Evol Syst 35:435–466
Tarborelli PA, Dacar M, Giannoni SM (2003) EVect of plant cover on
seed removal by rodents in the Monte Desert (Mendoza, Argen-
pp 185–208
of vertebrate dispersed plants in Mediterranean scrublands.
Oecologia 111:91–98
Hulme PE, Borelli T (1999) Variability in post-dispersal seed predation in deciduous woodland: relative importance of location, seed
species, burial and density. Plant Ecol 145:149–156
Janzen DH (1970) Herbivores and the number of tree species in tropical forests. Am Nat 104:501–528
Lortie CJ, Ganey DT, Kotler BP (2000) The eVects of gerbil foraging
on the natural seedbank and consequences on the annual plant
community. Oikos 90:399–407
Majer JD, Lamont BB (1985) Removal of seed of Grevillea pteridifolia (Proteaceae) by ants. Austr J Bot 33:611–618
123
of ant and plant species on seed removal patterns. Insect Soc
tina). Aust Ecol 28:651–657
Thompson JN (1994) The coevolutionary process. The University of
Chicago Press, Chicago
Thompson JN (2005) The geographic mosaic of coevolution. The
University of Chicago Press, Chicago
Turnbull CL, Culver DC (1983) The timing of seed dispersal in Viola
nuttallii: attraction of dispersers and avoidance of predators.
Oecologia 59:360–365
Valverde T, Silvertown J (1995) Spatial variation in the seed ecology
of a woodland herb (Primula vulgaris) in relation to light environ-
ment. Funct Ecol 9:942–950
Wang BC, Smith TB (2002) Closing the seed dispersal loop. Trends
Ecol Evol 17:379–385
Yu DW, Wison HB, Pierce NE (2001) An empirical model of species
coexistence in a spatially structured environment. Ecology
Zelikova TJ, Dunn RR, Sanders NJ (2008) Variation in seed dispersal
along an elevational gradient in Great Smoky Mountains National
Park. Acta Oecol 34:155–162
82:1761–1771
123