Bromeliads affect the interactions and composition of invertebrates on their support tree

Abstract

Individual species can have profound effects on ecological communities, but, in hyperdiverse systems, it can be challenging to determine the underlying ecological mechanisms. Simplifying species’ responses by trophic level or functional group may be useful, but characterizing the trait structure of communities may be better related to niche processes. A largely overlooked trait in such community-level analyses is behaviour. In the Neotropics, epiphytic tank bromeliads (Bromeliaceae) harbour a distinct fauna of terrestrial invertebrates that is mainly composed of predators, such as ants and spiders. As these bromeliad-associated predators tend to forage on the bromeliads’ support tree, they may influence the arboreal invertebrate fauna. We examined how, by increasing associated predator habitat, bromeliads may affect arboreal invertebrates. Specifically, we observed the trophic and functional group composition, and the behaviour and interspecific interactions of arboreal invertebrates in trees with and without bromeliads. Bromeliads modified the functional composition of arboreal invertebrates, but not the overall abundance of predators and herbivores. Bromeliads did not alter the overall behavioural profile of arboreal invertebrates, but did lead to more positive interactions in the day than at night, with a reverse pattern on trees without bromeliads. In particular, tending behaviours were influenced by bromeliad-associated predators. These results indicate that detailed examination of the functional affiliations and behaviour of organisms can reveal complex effects of habitat-forming species like bromeliads, even when total densities of trophic groups are insensitive.

Full text

Vol.:(0123456789)
1 3
Oecologia
https://doi.org/10.1007/s00442-020-04616-w
HIGHLIGHTED STUDENT RESEARCH
Bromeliads aect theinteractions andcomposition ofinvertebrates
ontheir support tree
PierreRogy1 · EddHammill2· M.AlexSmith3· BeatriceRost‑Komiya1· DianeS.Srivastava1
Received: 29 April 2019 / Accepted: 5 February 2020
© Springer-Verlag GmbH Germany, part of Springer Nature 2020
Abstract
Individual species can have profound effects on ecological communities, but, in hyperdiverse systems, it can be challenging
to determine the underlying ecological mechanisms. Simplifying species’ responses by trophic level or functional group
may be useful, but characterizing the trait structure of communities may be better related to niche processes. A largely over-
looked trait in such community-level analyses is behaviour. In the Neotropics, epiphytic tank bromeliads (Bromeliaceae)
harbour a distinct fauna of terrestrial invertebrates that is mainly composed of predators, such as ants and spiders. As these
bromeliad-associated predators tend to forage on the bromeliads’ support tree, they may influence the arboreal invertebrate
fauna. We examined how, by increasing associated predator habitat, bromeliads may affect arboreal invertebrates. Specifi-
cally, we observed the trophic and functional group composition, and the behaviour and interspecific interactions of arboreal
invertebrates in trees with and without bromeliads. Bromeliads modified the functional composition of arboreal invertebrates,
but not the overall abundance of predators and herbivores. Bromeliads did not alter the overall behavioural profile of arbo-
real invertebrates, but did lead to more positive interactions in the day than at night, with a reverse pattern on trees without
bromeliads. In particular, tending behaviours were influenced by bromeliad-associated predators. These results indicate that
detailed examination of the functional affiliations and behaviour of organisms can reveal complex effects of habitat-forming
species like bromeliads, even when total densities of trophic groups are insensitive.
Keywords Non-consumptive effects· Facilitation· Ecosystem engineering· Behaviour· Diel cycles
Introduction
Individual species can have profound effects upon the
interacting network of species in which they are embedded
(Hairston etal. 1960; Fretwell 1987; Abrams 1995). The
importance of species on ecological networks has been dem-
onstrated by either manipulating the density of particular
species (Paine 1980; Peacor and Werner 2001), or by com-
paring communities where one species is naturally absent
(Cox and Ricklefs 1977; Strong 1992). Such studies have
shown that ecological networks have a predominance of
relatively weak interactions, but a few species with dispro-
portionate effects on other species. In some cases, changes in
the density of a single species can ripple through the network
of species interactions, indirectly affecting a large number of
species (Srivastava and Bell 2009). Quantifying the effect
of individual species on an entire ecological network can be
challenging. Simply demonstrating a change in taxonomic
composition may not help in understanding the underlying
ecological mechanisms, because there is no information on
Communicated by Raphael Didham.
By comparing abundance and behavioural patterns, we innovate
in quantifying community shifts. We detect subtle changes that are
contingent on time, and that may be missed by classical trophic
studies.
Electronic supplementary material The online version of this
article (https ://doi.org/10.1007/s0044 2-020-04616 -w) contains
supplementary material, which is available to authorized users.
* Pierre Rogy
pierre.rogy@gmail.com
1 Department ofZoology andBiodiversity Research Centre,
University ofBritish Columbia, 6270 University Boulevard,
Vancouver, BCV6T1Z4, Canada
2 Department ofWatershed Sciences, Utah State University,
5210 Old Main Hill, NR 210, Logan, UT84322-5210, USA
3 Department ofIntegrative Biology, Summerlee Science
Complex, University ofGuelph, Guelph, ONN1G2W1,
Canada

Oecologia
1 3
the type of species most affected. This may be a particular
problem in ecological networks with a large number of spe-
cies, such as those that typify tropical systems.
A classic approach to quantifying effects of species in
entire food webs is to categorize species into trophic levels,
such as predators and herbivores, and to examine changes
in the total abundance or biomass of these trophic levels.
Manipulations of entire trophic levels have revealed the
importance of indirect pathways mediated by multiple spe-
cies, such as trophic cascades: the indirect and alternating
top-down effect of higher trophic levels on lower trophic lev-
els (Fretwell 1987; Schmitz etal. 2000; Ripple etal. 2016).
Such manipulations of trophic levels can be effective in
ecological networks with strong top-down effects of preda-
tor consumption on lower trophic levels, or strong bottom-
up effects of resource production on higher trophic levels.
However, this approach is limited in several ways. First,
predators not only affect their prey through direct consump-
tion, reducing their density, but also by inducing phenotypic
changes in their prey, including morphological or chemical
defenses and behaviour (Jeffries and Lawton 1984; Abrams
1995; Verdolin 2006) as prey attempt to reduce the chances
of being eaten (Schmitz etal. 2004; Bestion etal. 2015;
Buchanan etal. 2017). Second, negative trophic interactions
are not the only important ecological interactions that can
affect predator density. Positive interactions can have far-
reaching impacts on the density and traits of other species
in ecological networks (Boucher 1982; Peacor and Werner
2001; Leclerc etal. 2016). A common example is the facili-
tative ant-homopteran system, where ants tend honeydew-
producing insects, securing this source of energy-rich food
by defending the homopterans against predators, including
other ants (Dejean etal. 1997; Blüthgen etal. 2000; Styr-
sky and Eubanks 2007), resulting in a change in the overall
ecological community (Styrsky and Eubanks 2007). Third,
categorizing species by trophic groups is complicated by the
prevalence of omnivory in many food webs (Thompson etal.
2007), and the specialization of herbivores on different parts
of their host plant (e.g., phloem feeders and leaf chewers
only indirectly compete, Carrillo etal. 2012).
In response to the limitations of the trophic level
approach, ecological networks have often been more
finely characterized in terms of functional groups. Feeding
functional groups aggregate species that exploit a similar
resource in a similar way (e.g., leaf chewer), and provide
a characterization of the community that is both simple
and mechanistic (Blondel 2003). There may be substan-
tial changes in functional group composition even in the
absence of changes in the abundance of a trophic level, for
example when the presence of a predator alters the relative
abundance of edible to non-edible prey (Piovia-Scott etal.
2017). However, functional groups are ultimately based on
categorizing one or two traits of species, and traits may be
better characterized as both continuous and multivariate. A
third approach is therefore to describe ecological networks
of species not in terms of abundances of individuals at all,
but instead in terms of the abundance of traits. The ration-
ale here is that the trait structure of communities should
relate closely to the underlying niche mechanisms, and so
may be more sensitive to any perturbations to the com-
munity than taxonomic composition (McGill etal. 2006).
Even though the role of traits in mediating ecological inter-
actions is increasingly understood, their inclusion in con-
ceptual studies of ecological networks is relatively recent
(e.g., Solé and Bascompte 2006; Mora etal. 2018). To date,
most trait-based studies of community structure have con-
sidered morphological or chemical traits. However, in the
context of examining how an individual species affects an
ecological network of interacting species, behaviour may be
one of the most relevant traits. The behavioural profile of a
community can encapsulate both positive (facilitative) and
negative (consumptive and non-consumptive) interactions
between individuals, allows for interspecific and intraspe-
cific interactions, and can be affected by both individual
decisions (e.g., predator escape behaviour) and species
turnover (e.g., replacement of diurnally active species by
nocturnally active species). Changes in prey behaviour can
indirectly influence other species. For example, epigeous
predators induce burrowing detritivores to move deeper into
the soil, increasing nutrient availability at greater depths and
thus indirectly increasing aboveground plant biomass (Wu
etal. 2015). Nonetheless, surprisingly few studies have con-
sidered how the presence of a particular species may affect
the behavioural profile of the rest of a community (Touchton
and Smith 2011). Here we examine how bromeliad pres-
ence on orange trees affects the invertebrate community on
the tree, comparing the effects on taxonomic composition,
the relative abundance of trophic levels, the composition of
feeding functional groups, and the behavioural profile of the
invertebrate community. We also evaluate if such changes
reflect a shift in the strength of negative (consumptive and
non-consumptive) or positive (facilitative) interspecific
interactions between invertebrates on the support tree.
Epiphytic tank bromeliads, members of the diverse Bro-
meliaceae family, are ubiquitous plants throughout most
of the Neotropics, both in natural and agricultural settings
(Benzing 2000; Toledo-Aceves etal. 2012). Their epiphytic
lifestyle relies on the trapping of water and detritus by their
leaf rosette, and extraction of nutrients by specialised tri-
chomes (Wittman 2000). Bromeliads are considered eco-
system engineers (Linder etal. 2012), as they create terres-
trial microhabitats that are opportunistically occupied by a
diverse array of species (Benzig 2000; Angelini and Silliman
2014). More precisely, the bromeliad leaf rosette is utilized
by a variety of predatory terrestrial arthropods, such as ants
or spiders (Gutierrez Ochoa etal. 1993; Dejean etal. 1995;

Oecologia
1 3
Stuntz etal. 2002; Castaño-Meneses 2016). This increase in
predator microhabitat in trees bearing bromeliads is known
to impact species and functional group densities in support
trees (Cruz-Angón etal. 2009; Yanoviak etal. 2011; Rogy
etal. 2019; Rost-Komiya etal. in press), with some studies
suggesting that, by harbouring predators, epiphytic tank bro-
meliads may provide indirect protection against herbivorous
insects to their support tree (Dejean etal. 1995; Hammill
etal. 2014; but see Rogy etal. 2019 and Rost-Komiya etal.
in press). However, to the best of our knowledge, there has
been no study that has examined if bromeliads, by harbour-
ing predatory insects, impact invertebrates on their support
trees through consumptive or non-consumptive processes,
or if facilitative mechanisms play an important role in com-
munity shifts. In other words, it remains unknown if the
effects of bromeliad-associated predators on the support
tree stems primarily from direct consumption of arboreal
prey, modification of prey behaviour, or, in the case of ants,
through symbiosis with other organisms, such as aphids. If
bromeliad-associated predators affect arboreal communi-
ties through non-consumptive or facilitative mechanisms, it
would highlight a new pathway through which community
structure may be altered.
We hypothesize that: (1) trees with bromeliads will har-
bour higher predator abundances and lower herbivore abun-
dances than trees without bromeliads because bromeliads
provide increased habitat for predators; (2) invertebrate com-
munities in trees with bromeliads will have a different taxo-
nomic and functional group composition, and behavioural
profile, than in trees without bromeliads because of modified
interspecific interactions in the ecological network; (3) there
will be differences in the number of positive and negative
species interactions in trees with bromeliads versus without
bromeliads given that bromeliads harbour ants and ants are
involved in both predation and homopteran honeydew-tend-
ing. We also examine the temporal context dependence of
the above effects, hypothesizing that (4) bromeliad effects on
the arboreal invertebrate communities will differ with time
of the day, as bromeliad inhabitants may exhibit differing
diurnal and nocturnal activity patterns (Way 1963).
Materials andmethods
Site description andobservation design
In this study, we observed the behaviour, abundance, and
composition of invertebrates on the leaves and branches of
orange trees, comparing trees with and without bromeliads.
As predation pressure and facilitative interactions may vary
with diel cycles of organisms (Way 1963; Kohl etal. 2018),
we performed observations both during the day and at night.
These observations took place in June 2017 in two orange
plantations near Santa Cecilia, northern Guanacaste Prov-
ince, Costa Rica (11° 03′51″ N–85° 25′06″ W). The first
plantation, hereafter CP (named after the owners, Calixto
Moraga and Petrona Ríos), consisted of about a hundred
lightly maintained trees in a 10 × 12 matrix, separated by
rows of tall fodder plants, and located in a matrix of human
settlements, pasture, and forest fragments. The other plan-
tation, hereafter DO (named after the owner, the company
Del Oro S.A.), is an intensively managed parcel of many
thousand trees, located at the edge of the Area de Conser-
vación Guanacaste, and isolated from the rest of the com-
pany’s operations by forested areas. Unlike CP, DO was
intensively sprayed with pesticides until September 2016,
but is now used as an experimental parcel to develop sustain-
able agricultural techniques by the company agronomists.
The between-tree rows consisted of a diverse matrix of short
grasses and bushes.
Due to the large differences in the abundance of trees
between our sites, and because of difficulties in traversing
the dense fodder plants at CP, our study design differed
between the two sites (Fig. S1). At CP, where there were a
limited number of trees, we selected three blocks of 12 trees.
Within each block, six trees bore bromeliads and six did not.
Here, blocks did not consist of spatially distinct groups of
trees, but rather a temporally distinct group. Each block con-
sisted of a random set of 12 trees selected across the entire
parcel sampled at the same time. At set hours, two observ-
ers separately conducted either two diurnal or two noctur-
nal observations on each tree of a block. Each observation
consisted of a researcher carefully approaching the edge of
the orange tree—so as to minimize disturbance of behav-
iour—and recording all invertebrate activity in a randomly
selected, eye-level 50*50*50cm volume of leaves and
branches for five minutes. We classified behaviour into one
of 11 categories (Table1b) and recorded the duration of the
activity to the nearest 5seconds. When interspecific inter-
actions were observed during the behavioural observations,
we categorized these as positive or negative (Table1c). We
repeated the same design a second time on each block, but
with time of observation (day or night) switched for observ-
ers to reduce any observer bias. Due to the relative isola-
tion of DO, it was not possible to safely perform nighttime
observations; therefore, the two observers synchronously
performed diurnal observations on opposite sides of each
tree. However, because the abundance of trees was not
limiting in this site, we did not need to repeat observations
on blocks of trees, but rather observed a different random,
relatively close (within a 50m radius) set of 12 trees each
day of observation. Observations lasted for 13days overall:
6days at CP (36 trees, three blocks observed twice) and
7days at DO (84 trees, seven groups observed once). For
both sites, we recorded invertebrate activity four times on
six trees with bromeliads and six trees without bromeliads

Oecologia
1 3
each day (n = 4 observations for each tree; total n = 288 at
CP and total n = 336 at DO). All trees with bromeliads had
1–4 bromeliads of average size for each site, except for three
trees at CP. These threes trees bore 8–11 bromeliads, and are
included in the present analyses because their exclusion did
not significantly change the results.
Table 1 List of a functional and taxonomic groups, b behaviours, and c interspecific interaction types used in the analysis, and their abbrevia-
tions for Figs.2, 3
Trophic level Functional groups Taxonomic groups (or lowest taxonomic unit)
(a)
Predator Predator (Prd)
(includes scavengers (Scv) and omnivores (Omn) with preda-
tory behaviours in univariate analyses)
Ants (excluding leaf-cutter ants), predatory heteropterans
(Pbu), hunting spiders (Hsp), lacewings (Lac), mantids,
predatory beetles (Pbt), predatory flies (Pfl), wasps,
web-weaving spiders (Wsp), cockroaches (Coc), opiliones
(Opl)
Herbivore Leaf chewers (Chw) Atta sp. (leaf-cutter ants), herbivorous beetles (Hbt), her-
bivorous lepidopterans, herbivorous orthopterans (Ort),
herbivorous snails (Snl)
Phloem feeders (Phl) herbivorous heteropterans (Hbu), hoppers (mobile homop-
terans, Hop), scales/aphids (Dew, includes mealybugs,
and the Asian citrus psyllid Diaphorina citri Kuwayama
(Psy))
Leaf miners (Min) All leaf miners (Min), including Phyllocnistis citrella
Others Detritivore (Det) Collembola, Diplopoda, Psocoptera
Granivore (Gra) Rhyparochromidae
Mycophagous (Myc) Lauxaniidae
Non-feeder (Non) Chironomidae, Psychodidae, Sciaridae
Omnivore without predatory behaviours (Omn) Ensifera, earwigs
Parasitoid (Par) Parasitoids (Par)
Palynivore/Nectarivore (Pol) Syprhidae, Apoidea, Lepidoptera
Xylophagous (Xyl) Scolytinae
Unknown feeding behaviour (Unk) Acari, Apocrita, Brachycera, Coleoptera, Diptera, Hemip-
tera, Lepidoptera, Nematocera, Orthoptera, Polyphaga,
Sternorrhyncha, Thysanoptera
Behaviour category Description
(b)
Stationary (Sta) Specimen stationary, including spiders in their web, and phloem feed-
ing herbivores with no evidence of feeding
Detritivory (Det) Detritivore sponging leaf or eating debris
Leaf chewing (Chw) Herbivore chewing leaf
Phloem feeding (Phl) Herbivore feeding on phloem. Includes scale insects and mealybugs
Leaf mining (Min) Leaf miner present inside tunnel
Predation/parasitism (PPr) Predators feeding, parasitism, and attack attempts on other organisms
Defense (Def) Response of an organism to an attack by a predator
Mobile (Mob) Exploring the environment
Tending (Ten) Ants tending honeydew-producers, such as scale insects, mealybugs,
or aphids
Reproduction (Rep) Mating or oviposition, excluding parasitism
Development (Dev) Molting or pupation
Transporting (Tra) Organism carrying food, dead material. Largely concerns ants
Interaction type Description
(c)
Negative Predator or parasitoid attack on an organism
Positive Tending

Oecologia
1 3
Categorization ofinvertebrates
We identified invertebrates to morphospecies, or to ‘near-
morphospecies’ (identification approximate, or slight mor-
phological differences present). Because insitu identification
of arthropods can be challenging, we relied on morphospe-
cies identified in Rogy etal. (2019), a study conducted con-
comitantly at the same sites, in which we compared abun-
dances of invertebrate species within bromeliads versus in
vacuum samples of the surrounding tree leaves. Rogy etal.
(2019) also dissected 117 bromeliads growing on orange
trees in both sites, 50 of which were on trees we observed
for this study, providing information on the local bromeliad
fauna. We categorized specimens into taxonomic groups
(Table1a) and, if identification was certain, we inferred
trophic level from taxonomy (see Rogy etal. 2019). How-
ever, if identification was approximate, or the taxon includes
a range of feeding behaviour, trophic level was considered
unknown. As ants are key arthropod predators in agroeco-
systems (Schmitz etal. 2000), we assigned our ant morphos-
pecies to existing species or taxa with both morphological
and genetic methods (Smith etal. 2014). We also classified
our morphospecies of invertebrates as bromeliad-associated
or not, based on Rogy etal. (2019). Specifically, we defined
bromeliad-associated predators as those that preferentially
occurred in bromeliads, discounting “tourist” species.
In our classification, we did not consider parasitoids as
predators because, as we did not dissect specimens to assess
parasitism rate, their impact can only be detected after the
larvae emerge from hosts. Finally, herbivores were further
classified into functional groups based on their feeding
behaviours, namely leaf chewers, phloem feeders, and leaf
miners.
Statistical analyses
The study design differed between the two sites, so we ana-
lyzed each site separately, using the R programming lan-
guage version 3.5.1 (R Core Team 2018). We separately
examined the effects of bromeliads and of bromeliad-asso-
ciated predators on the invertebrate community.
Community abundance andcomposition
To separately test the associations between bromeliads, all
predators, bromeliad-associated predators and herbivore
functional groups, we used generalized linear models with
Poisson or negative binomial error distributions as appropri-
ate. We pooled all four observations for each tree, as bro-
meliad presence or absence was recorded at the tree level.
At CP, to control for the repeated measures of our block
design, we used generalized linear mixed-effect models with
tree nested within block as random effect, using the ‘glmer’
function of the ‘lme4’ package (Bates etal. 2015). We tested
model outputs with likelihood ratio tests, using the ‘mixed’
function of the ‘afex’ package (Singmann etal. 2018). At
DO, which did not have a random block design, we instead
used generalized linear models, and tested model outputs
with the same method, using the ‘Anova’ function of the
‘MASS’ package (Venables and Ripley 2002). We assessed
fit of all models with the ‘DHARMa’ package (Hartig 2018)
and plotted model outputs using the ‘ggeffects’ package
(Lüdecke 2018). To avoid any circularity, we removed abun-
dances and behaviours of the bromeliad-associated predators
from the response matrix when bromeliad-associated preda-
tors were the explanatory variable. We also removed from
analyses of herbivores one CP tree with > 200 aphids, as this
was the only instance of such infestation, and more than five
times the abundance of herbivores in the next most abundant
quadrat. Finally, leaf miners were not numerous enough to
be analyzed separately in univariate analyses, but were still
included in multivariate analyses.
To analyse the effect of bromeliad and bromeliad-asso-
ciated predators on community composition on orange
trees, we used permutational analysis of variance (PER-
MANOVA), a method to detect changes in community
composition associated with ecological parameters (Ander-
son 2001). We performed PERMANOVA on a Bray–Curtis
dissimilarity matrix with 2000 permutations, which were
either restricted at the block level for CP, or unrestricted
for DO, using the ‘adonis’ function of the ‘vegan’ package
(Oksanen etal. 2018). We first performed this analysis on
a community matrix of abundance within each functional
group (Table1a). We then performed the same taxonomic
analysis on the subset of herbivore and predators that could
be confidently assigned to a taxonomic group (Table1a,
“Restricted taxonomic groups”). To ensure that removing
specimens with low taxonomic resolution did not affect our
analyses, we also repeated the PERMANOVA analyses with
these specimens included (left at order or suborder, “Overall
taxonomic groups”). Inclusion of all taxonomic groups did
not change the results, and can be found in Online Resource
1. In all three matrices, functional or taxonomic groups with
less than five recorded individuals were excluded from the
analyses. Relevant PERMANOVA outputs for this paper
were visualized in ordination space using Principal Coordi-
nate Analysis (PCoA), with the weighted averages score of
relevant groups calculated using the ‘add.spec.scores’ func-
tion in the ‘BiodiversityR’ package (Kindt and Coe 2005).
Behavioural analysis
We conducted further PERMANOVAs with the goal of
assessing the possible impact of bromeliads and bromeliad-
associated predators on invertebrate behaviour, using behav-
iour categories (Table1b) instead of functional or taxonomic

Oecologia
1 3
groups in the community matrix. We examined two aspects
of community behaviour: behaviour duration (in seconds)
in a Hellinger-transformed matrix, and raw behaviour fre-
quency. Here we generated the dissimilarity matrix using
Euclidean distances, rather than Bray–Curtis dissimilarity,
due to the non-integer nature of the data. Looking at these
two aspects of behaviour allows us to understand if behav-
ioural responses to either bromeliads or bromeliad-associ-
ated predators are mediated by strong (change in behaviour
frequency) or weaker behavioural modifications (change in
behaviour duration). The behavioural profile of each tree
was visualized using PCoA analyses, and the vector for the
bromeliad-associated predators was plotted using the ‘envfit’
function in ‘vegan’ (Oksanen etal. 2018).
To assess if bromeliads influence negative or positive
interspecific interactions in their host trees, we used the
same site-dependent model structure as aforementioned.
However, we added total number of observed invertebrates
per tree as a covariate, to account for the dependence of
interaction frequency on invertebrate abundance.
Diurnal analysis
As the impacts of bromeliads on their host tree communities
may be dependent on diurnal patterns of bromeliad-associ-
ated invertebrates, we separated diurnal and nocturnal obser-
vations at CP before repeating the same regression models
and diet, taxonomic and behavioural PERMANOVAs. In
other words, instead of pooling all observations per tree for
each replicate, we only pooled at the time of observation
level (day or night per tree for each replicate). In addition,
we added time of observation (day or night) as an interaction
term in the same CP generalized-mixed effect model and
PERMANOVA structure.
Results
During the study, we observed a total of 3269 individual
invertebrates (1874 at CP and 1395 at DO) and witnessed
174 interactions (99 at CP and 75 at DO; Table1c). At DO,
negative interactions were more numerous than positive
interactions (51 vs. 24, respectively); at CP, positive inter-
actions were more numerous than negative interactions (81
vs. 18, respectively). Three ant species account for most
of these positive interactions (tending): Solenopsis sp. and
Camponotus sp., both bromeliad-associated, and Ectatomma
sp., not bromeliad associated. The bromeliad-associated C.
atriceps and Azteca sp., and the non-bromeliad associated
C. planatus accounted for the remainder.
Predator andherbivore abundance—overall
analysis
There was no effect of bromeliad presence on the abundance
of predators (Fig. S2a for CP), bromeliad-associated preda-
tors (i.e., predators encountered during bromeliad dissection,
Fig. S2c for CP) or herbivore abundances at either of the two
sites (TableS1). Similarly, bromeliads did not affect abun-
dances of either leaf chewers or phloem feeders (TableS1).
Predator andherbivore abundance—diel analysis
On average, predators and bromeliad-associated predators
were, respectively, 30% and 50% more abundant during the
day than at night (TableS4, Fig. S2b, S2d), while herbivores,
including as functional groups, remained unaffected by time
of observation (TableS4). Even when accounting for this
diel pattern, bromeliads still had no effect on abundances of
these three invertebrate groups (TableS4, Fig. S2b, S2d).
Community composition—overall analysis
At CP, bromeliads and their associated predators consist-
ently impacted the taxonomic structure of the invertebrate
community on the support tree when we examined taxo-
nomic structure in a high-confidence but restricted set of
taxa (respectively, F1, 69 = 2.18, P = 0.026, r2 = 0.0311, and
F1, 68 = 4.66, P = 0.065, r2 = 0.0005, Fig. S2b). Bromeliads
and bromeliad-associated predators also modified the func-
tional group structure of invertebrates on orange trees at
CP (respectively, F1, 69 = 3.02, P = 0.007, r2 = 0.0425, and
F1, 68 = 7.81, P = 0.0005, r2 = 0.103, Fig.1a). The effects
of either bromeliads or bromeliad-associated predators on
these compositional groups were minor, explaining between
3.11% and 10.3% of the observed variation. By contrast, at
DO, neither bromeliads nor their associated predators had a
detectable impact on taxonomic (respectively, F1, 82 = 0.393,
P = 0.91, r2 = 0.0048, and F1, 82 = 1.32, P = 0.0161, r2 =
0.022) or functional group composition (respectively, F1,
82 = 0.689, P = 0.66, r2 = 0.0084, and F1, 82 = 1.53, P = 0.15,
r2 = 0.0185).
Community composition—diel analysis
Time of day affected the composition of functional groups,
and taxonomic groups, explaining 1.8–3% of the observed
variation (Table2a; Fig.2a–d, TableS5). However, account-
ing for time of day only lowered the amount of variance
explained (to 2% and less) by bromeliads or bromeliad-asso-
ciated predators, without changing the qualitative results.
Indeed, bromeliads still altered the taxonomic and func-
tional structure of the invertebrate community on orange
trees, and bromeliad-associated predators again altered only

Oecologia
1 3
the functional structure (Table2a; Fig.2a–d). However, the
interaction of bromeliads and time altered the functional
structure of the invertebrate community on orange trees
(Fig.2a–b), but not the taxonomic structure (Table2a)
unless taxa with low taxonomic resolution were included
(TableS5).
Community behaviour—overall analysis
At CP, bromeliad-associated predators altered the frequency
of different behaviours in the rest of the invertebrate com-
munity (F1, 68 = 3.69, P = 0.0025, r2 = 0.0515, Fig. S2c), but
not behaviour duration (F1, 68 = 2.5, P = 0.06, r2 = 0.0354).
This effect was mainly driven by phloem feeding and tend-
ing behaviours (Table1), which are positively associated
with bromeliad-associated predators in the ordination space
(Fig.1c). We note that this association does not reflect tend-
ing of herbivores by the bromeliad-associated predators
themselves, because the bromeliad-associated predators
were removed from the community matrix for this analy-
sis; instead this is a correlational or indirect effect. As with
functional structure, only about 5% of the variation in com-
munity behaviour was explained by bromeliad-associated
predators. On the other hand, there was no effect on behav-
iour frequency and duration on orange trees of bromeliads
at CP (respectively, F1, 69 = 1.78, P = 0.11, r2 = 0.0256 and
F1, 69 = 0.934, P = 0.4, r2 = 0.0136), and, at DO, of brome-
liads (respectively, F1, 82 = 1.07, P = 0.36, r2 = 0.013, and
F1, 82 = 0.586, P = 0.73, r2 = 0.0072), although bromeliad-
associated predators displayed marginally non-signifi-
cant associations (respectively, F1, 82 = 2.01, P = 0.079,
r2 = 0.0242, and F1, 82 = 2.23, P = 0.059, r2 = 0.0268). Bro-
meliads were not associated with the number of positive
or negative interactions in the trees at either site (Table1c,
TableS3).
Community behaviour—diel analysis
Time (either day or night) altered the relative frequency of
community behaviours, with around 2% variation explained
by time of observation (Table2b, Fig.2e, f). There was no
effect of time of observation on behaviour duration, nor any
effect of bromeliads or bromeliad-associated predators, on
the behavioural structure of the community (Table2b).
In diurnal observations, the frequency of positive inter-
specific interactions (Table1) was near-zero in the presence
of bromeliads, while, in the absence of bromeliads, expo-
nentially increased with the number of observed specimens
(TableS6; Fig.3). In nocturnal observations, this pattern
was somewhat reversed: the increase in number of interspe-
cific positive interactions was much stronger in trees with
bromeliads than in trees without bromeliads (TableS6;
Fig.3). Negative interspecific interactions, to the contrary,
remained unaffected by bromeliads, time of day, or number
of observed invertebrates (TableS6).
Discussion
Our study assessed if ecosystem engineering by bromeliads
was associated with an increase in arboreal predator abun-
dances, changing the functional structure and behavioural
profile of arboreal invertebrate communities. By separately
examining the abundance, composition and behaviour of the
community, we can shed light on different ways bromeli-
ads may affect the trait structure of the invertebrate com-
munity on their support trees. In this study, bromeliads did
not alter the overall abundance of predators or herbivores on
orange trees, nor the behavioural structure of the arboreal
invertebrate community. Nonetheless, bromeliads altered
the taxonomic and functional composition of the arboreal
invertebrates, at least at site CP. Although bromeliads did not
affect the behavioural structure of the arboreal invertebrate
community, bromeliad-associated predators did—suggesting
that bromeliad-associated predators are the proximate driver
of altered behavioural structure. The community-wide effect
of bromeliads and their associated predators was detectable
regardless of time of observation (day or night), yet interac-
tive effects between bromeliads and time of observation on
interspecific positive interactions suggest that the effects of
bromeliad may be mediated by invertebrate activity patterns.
In summary, bromeliads tend to modify several aspects of
the invertebrate communities on their support tree, which,
coupled with a lack of numerical impact on broad trophic
groups, suggests subtle responses in invertebrate communi-
ties that may be missed by classic trophic approaches.
There were some stark site-level differences in our results,
with all detected associations of bromeliads occurring at
CP. The qualitative differences in invertebrate responses are
likely a consequence of fundamental differences between
small-scale farming (CP) and intensive commercial opera-
tions (DO). Bromeliads at DO are removed from trees on
a regular basis, as part of the management routine. Even
though DO managers agreed to stop removing bromeliads
7months before the start of the experiment, the long genera-
tion time of bromeliads meant they were generally smaller
and less abundant than at CP (Rogy, pers. obs.). Moreover,
these smaller bromeliads tended to be near or at the top of
the tree crown, hence many of our observation quadrats were
relatively far from the bromeliads. An alternative explana-
tion would be that differences in the experimental designs
between sites led to site differences, however, both the total
numbers of samples and the accumulation of species with
samples were relatively similar between sites.
In terms of trophic levels, bromeliads did not affect
densities of all predators, predators normally associated

Oecologia
1 3
with bromeliads, or herbivores, including herbivore func-
tional groups. We found the opposite pattern in a vacuum-
sampling study conducted concurrently at CP (Rogy etal.
2019), where the presence of bromeliads was associated with
increased abundance of predators during the dry season, and
increased abundance of herbivores in the wet season. As the
current study took place in the transitional phase between
the dry and wet season, our experiment may have coincided
with a seasonal reorganization of invertebrate communities,
resulting in no clear effect on either predators or herbivores.
Alternatively, differences in sampling intensity (and meth-
ods) may contribute to the difference between studies: we
observed in this study only half the individuals that we cap-
tured with vacuum sampling in Rogy etal. (2019). In addi-
tion, while vacuum sampling can collect specimens hiding
in leaf curls, for example, observation is inherently biased
towards conspicuous, active specimens.
Although bromeliads did not affect arboreal predator
and herbivore abundances, the presence of bromeliads in a
tree nonetheless altered the distribution of both functional
and taxonomic groups in the food web. This suggests that
there are subtle changes in invertebrate composition at the
level of diet, coarse taxonomy, and feeding guilds, which
do not appear in the broader categories of trophic levels. In
ordination space, bromeliads tended to be associated with
detritivores, omnivores (mainly opiliones) and scavengers
(cockroaches). In fact, bromeliads can house a substantial
detritivore community (Castaño-Meneses 2016), and are
used by numerous cockroaches and opiliones, including
juveniles (Rogy, pers. obs.). Moreover, subdividing func-
tional groups into predatory and herbivorous feeding guilds
(‘Restricted taxonomic groups’), showed that bromeliads
increased snail and hopper (Homopterans excluding aphids
and scales) presence in their support tree. Snails were com-
monly encountered inside bromeliads, suggesting that bro-
meliads acted as a source habitat or microhabitat refuge
for these species. Hoppers, by contrast, are not known to
be bromeliad-associated (Rogy, pers. obs.), so the mecha-
nism for their increased abundance on bromeliad-bearing
trees remains unclear. In short, bromeliads tend to promote
organisms that are either associated with detrital food webs
within bromeliads, or with the moist microhabitats provided
by bromeliads.
There were minor effects of bromeliad-associated
predators on the community-wide frequency of particular
behaviours, even if we were not able to detect effects of
bromeliads themselves on the entire behavioural profile of
the arboreal community. In particular, phloem-feeding and
tending behaviours were promoted by bromeliad-associated
predators. Predation events are rare and very rapid (Nentwig
1986), unlike tending events, so our study may have under-
estimated their true occurrence. In this experiment, most
observed bromeliad-associated predators were ants, many of
Fig. 1 Principal Coordinate Analysis (PCoA) plots of the effect of
bromeliads on a functional and b restricted taxonomic groups, and c
of bromeliad-associated predators (“Brom. predators”) on behaviour
frequencies at CP. To avoid cluttering of the graphs, the least abun-
dant groups are not plotted, specifically leaf miners, mycophagous
and pollen-feeders in a, and herbivorous and predatory heteropterans,
lacewings, lepidopterans, predatory beetles and flies, and web-weav-
ing spiders in b. Abbreviations as in Table1

Oecologia
1 3
which were attracted to immobile, feeding aphids or scales
(most phloem feeding herbivores) to farm honeydew (tend-
ing on the part of the ants). Honeydew-farming ants tend
to increase densities of their homopteran partners (Styrsky
and Eubanks 2007; Ohgushi 2008), potentially explaining
the association between bromeliad-associated predators
and phloem feeding in the ordination space. These feeding
aphids and scales, in turn, are tended by other ants which
may not necessarily be bromeliad-associated, explaining the
association between phloem feeding and tending behaviours
in ordination space, even when bromeliad-associated preda-
tors were removed from the behavioural matrix.
The influence of bromeliads on the composition of the
arboreal invertebrate community differed between day and
night, a result that we attribute to diel cycles of bromeliad-
associated species. For example, at night we observed large,
herbivorous katydids feeding in the vicinity of bromeliads
whereas we did not see any during the day. In a concomitant
study (Rogy etal. 2019), we found many such individu-
als sheltering within bromeliads (collected during the day),
suggesting that bromeliads can be used as a diurnal refuge
by large herbivores. These diel patterns in bromeliad occu-
pancy were also reflected in bromeliad-associated predators.
In fact, some ant species that nested in bromeliads were
only observed on the orange trees during the day, whereas
other species were only observed at night. This offers an
explanation for why the effects of bromeliads on positive
interspecific interactions differed with time of day: nocturnal
bromeliad-associated ants may rely more on honeydew than
their diurnal counterparts. Similar seasonal differences in
the reliance of ants on honeydew have been documented for
at least half a century (Way 1963), which implies that the
composition of invertebrates in bromeliads may also differ
depending on time of the day, resulting in opposite compo-
sitional shifts in their support tree communities. By increas-
ing energy flow to certain ant species, this kind of positive
interspecific interaction can radically change ant competitive
dynamics, and allow ants that are otherwise subdominant to
exclude competitors, indirectly affecting other trophic levels
(Dejean etal. 1997; Blüthgen etal. 2000).
We found that, in cases where bromeliads positively influ-
enced the interspecific interactions on their host trees, most
of these positive associations were due to the tending activi-
ties of species from three genera of ants; Solenopsis (Myr-
micinae), Camponotus (Formicinae) and Azteca (Dolicho-
derinae). Famously, Hölldobler and Wilson (1990) noted that
most species from these three sub-families attend—at least
to some extent—hemipterans. From Solenopsis, we know of
many examples of tending hemipterans (Vinson 1997) and
in fact, this hemipteran association is widely thought to be
a reason for their successful invasion of habitats across the
globe (Holway etal. 2002). Azteca ants are an arboreal taxon
that includes species of generalists which often tend Hemip-
terans (Longino 2007). Often, Azteca workers tend Hemip-
tera within, or nearby to, their nest (Davidson etal. 2003).
Camponotus is one of the largest ant genera in the world and
exhibit a variety of life-histories (Bolton etal. 2006). How-
ever, one commonly occurring strategy for this group is the
tending of hemipteran (Davidson etal. 2003; Wernegreen
etal.2009). In one example, Camponotus tending resulted in
Table 2 Effects of bromeliad presence (“Bromeliads) and bromeliad-associated predator abundance on a functional or taxonomic composition,
and b behavioural structure of the arboreal invertebrate community, in conjunction with time of observation (day/night)
Only observations from site CP are included here, as nocturnal observations were not performed at site DO
PERMANOVA analysis performed with 2000 permutations restricted at the block level
Significant associations are in boldface
Bromeliads (B) Time of observation (T) B × TBromeliad-asso-
ciated predators
(BP)
Time of observation (T) BP × T
(a)
Functional groups F1, 139 = 2.26
P = 0.025
r2 = 0.0156
F1, 139 = 4.29
P = 0.0005
r2 = 0.0296
F1, 139 = 2.34
P = 0.018
r2 = 0.0162
F1, 138 = 0.945
P = 0.49
R2 = 0.0067
F1, 138 = 3.21
P = 0.0025
r2 = 0.0229
F1, 138 = 1.13
P = 0.32
r2 = 0.0081
Restricted taxonomic
groups
F1, 139 = 2.26
P = 0.0235
r2 = 0.0158
F1, 139 = 4.24
P = 0.0005
r2 = 0.0296
F1, 139 = 0.859
P = 0.55
r2 = 0.006
F1, 138 = 2.88
P = 0.004
r2 = 0.0206
F1, 138 = 2.5
P = 0.0105
R2 = 0.0178
F1, 138 = 1.55
P = 0.15
r2 = 0.011
(b)
Behaviour duration F1, 139 = 0.373
P = 0.81
r2 = 0.0027
F1, 139 = 3.14
P = 0.019
r2 = 0.0224
F1, 139 = 0.642
P = 0.607
r2 = 0.0046
F1, 138 = 0.979
P = 0.38
r2 = 0.007
F1, 138 = 2.09
P = 0.093
r2 = 0.0149
F1, 138 = 1.91
P = 0.15
r2 = 0.0137
Behaviour frequency F1, 139 = 1.38
P = 0.23
r2 = 0.0097
F1, 139 = 3.4
P = 0.008
r2 = 0.0239
F1, 139 = 1.41
P = 0.22
r2 = 0.0099
F1, 138 = 2.25
P = 0.055
r2 = 0.0158
F1, 138 = 3.23
P = 0.0105
r2 = 0.0228
F1, 138 = 1.39
P = 0.24
r2 = 0.0098

Oecologia
1 3
more than a 30% increase in the size of the hemipteran popu-
lation compared to where tending was prevented (Renault
etal. 2005). The presence or absence of ant tending can be
a limiting resource to hemipterans (Holway etal. 2002) and
such interactions are common in general (Delabie 2001) and
in particular in Costa Rica (Espadaler etal. 2012). Such ant
tending could explain the association observed in multivari-
ate space between the presence of bromeliads in a tree, and
the abundance of hoppers (comprising mobile members of
two hemipteran suborders), but confirming this mechanism
would require careful manipulative experiments.
In conclusion, we documented relatively subtle effects of
bromeliads or their associated predators on functional and
taxonomic composition, as well as on the behavioural profile
of the invertebrates on orange trees. This contrasts with the
strong negative effects of bromeliads, especially bromeliads
with ants, on leaf damage reported in the same area of Costa
Rica several years earlier (Hammill etal. 2014). This differ-
ence is likely due to the difference in dominant ant species in
Fig. 2 Mean diurnal (left
panels) and nocturnal (right
panels) abundance of a, c
functional and b, e taxonomic
groups, and c, f frequency of
behaviours on trees with and
without bromeliads. Error bars
represent standard deviation
from the mean. Abbreviations
as in Table1. Ordinate scale dif-
fers among panels

Oecologia
1 3
the two systems: while Hammill etal. (2014) recorded many
predatory ants in the species Odontomachus hastatus Fab-
ricius, the dominant ant species associated with bromeliads
in our study were instead found commonly tending aphids
and scale insects. As a result, our bromeliad-associated ant
species act as much as facilitators as predators, attenuating
potential cascading effects. Our study emphasizes that bro-
meliads can affect invertebrate composition on orange trees,
effects that change with diel patterns in invertebrate occu-
pancy of bromeliads versus tree brancShes, and that appear
contingent on the agricultural practices in different sites. In
conclusion, in line with Paine (1980), detailed observations
on how invertebrates behave, coupled with taxonomical and
functional information, can help deciphering the complex
links between species in ecological communities.
Acknowledgements We would like to thank Francesca Fogliata for
full-time field assistance during this project. We thank Calixto Moraga,
Petrona Ríos, Del Oro S.A. (especially Hugo Segnini) for facilitating
access to sites. We also thank Max Vargas, Cristian Fuentes-Medina
and Eduardo Alvarado for their invaluable help in the field during
this project. In addition, we would like to thank Juli Carrillo for her
input the initial stages of this project. Without these people, this pro-
ject would not have been possible. This project was completed under
MINAE permits ACG-PI-012-2017 and ACG-PI-PC-034-2017.
Author contribution statement PR designed the survey. PR collected
the data and analyzed it with input from DSS. BRK and MAS identified
ant species. PR wrote the first version of the manuscript, and all other
co-authors made contributions to manuscript revision.
Data availability statement Data available from the Open Science
Framework: https ://osf.io/kx7sw /.
Compliance with ethical standards
Conflict of interest The authors declare that they have no conflict of
interest.
References
Abrams PA (1995) Implications of dynamically variable traits for
identifying, classifying, and measuring direct and indirect effects
in ecological communities. Am Nat 146:112–134. https ://doi.
org/10.1086/28578 9
Anderson MJ (2001) A new method for non-parametric multivariate
analysis of variance. Austral Ecol 26:32–46. https ://doi.org/10.1
111/j.1442-9993.2001.01070 .pp.x
Angelini C, Silliman BR (2014) Secondary foundation species
as drivers of trophic and functional diversity: evidence from
a tree-epiphyte system. Ecology 95:185–196. https ://doi.
org/10.1890/13-0496.1
Bates D, Maechler M, Bolker B, Walker S (2015) Fitting linear mixed-
effects models using lme4. J Stat Softw 67:1–48. https ://doi.
org/10.18637 /jss.v067.i01
Benzing DH (2000) Bromeliaceae: profile of an adaptive radiation.
Cambridge University Press, Cambridge
Bestion E, Cucherousset J, Teyssier A, Cote J (2015) Non-consumptive
effects of a top-predator decrease the strength of the trophic cas-
cade in a four-level terrestrial food web. Oikos 124:1597–1602.
https ://doi.org/10.1111/oik.02196
Blondel J (2003) Guilds or functional groups: does it matter? Oikos
100:223–231. https ://doi.org/10.1034/j.1600-0706.2003.12152 .x
Blüthgen N, Verhaagh M, Goitía W, Jaffé K, Morawetz W, Barthlott
W (2000) How plants shape the ant community in the Amazonian
rainforest canopy: the key role of extrafloral nectaries and homop-
teran honeydew. Oecologia 125:229–240. https ://doi.org/10.1007/
s0044 20000 449
Bolton B, Alpert G, Ward PS, Naskrecki P (2006) Bolton’s catalogue
of ants of the world. Harvard University Press, Cambridge
Boucher DH, James S, Keeler KH (1982) The ecology of mutualism.
Annu Rev Ecol Syst 13:315–347. https ://doi.org/10.1146/annur
ev.es.13.11018 2.00153 1
Buchanan AL, Hermann SL, Lund M, Szendrei Z (2017) A meta-
analysis of non consumptive predator effects in arthropods: the
influence of organismal and environmental characteristics. Oikos
126:1233–1240. https ://doi.org/10.1111/oik.04384
Carrillo J, Wang Y, Ding J, Siemann E (2012) Induction of extraflo-
ral nectar depends on herbivore type in invasive and native Chi-
nese tallow seedlings. Basic Appl Ecol 13:449–457. https ://doi.
org/10.1016/j.baae.2012.07.006
Castaño-Meneses G (2016) Seasonal and altitude effects on the struc-
ture of arthropod communities associated with Tillandsia violacea
Baker (Bromeliaceae) in a temperate forest of Mexico. Arthro-
pod-Plant Interact 10:403–417. https ://doi.org/10.1007/s1182
9-016-9451-y
Fig. 3 Estimates of the effect of bromeliad presence, number of
observed specimens, and time of observation (day/night) on positive
interspecific interactions. Error bars represent 95% confidence inter-
val, and do not include variation captured by random effects. Brown
lines represent trees with bromeliads and green lines represent trees
without bromeliads. Opaque confidence intervals of each colour rep-
resent estimations for nocturnal observations, while transparent ones
represent estimates for diurnal observations. The horizontal brown
line represents the negligible estimates for diurnal observations of
positive interactions on trees with bromeliads

Oecologia
1 3
Cox GW, Ricklefs RE (1977) Species diversity and ecological release
in Caribbean land bird faunas. Oikos 28:113–122. https ://doi.
org/10.2307/35433 30
Cruz-Angón A, Baena ML, Greenberg R (2009) The contribution of
epiphytes to the abundance and species richness of canopy insects
in a Mexican coffee plantation. J Trop Ecol 25:453–463. https ://
doi.org/10.1017/S0266 46740 99901 25
Davidson DW, Cook SC, Snelling RR, Chua TH (2003) Explaining the
abundance of ants in lowland tropical rainforest canopies. Science
300:969–972. https ://doi.org/10.1126/scien ce.10820 74
Dejean A, Olmsted I, Snelling RR (1995) Tree-epiphyte-ant rela-
tionships in the low inundated forest of Sian Ka’an Biosphere
Reserve, Quintana Roo, Mexico. Biotropica 27:57–70. https ://doi.
org/10.2307/23889 03
Dejean A, Bourgoin T, Gibernau M (1997) Ant species that pro-
tect figs against other ants: result of territoriality induced by a
mutualistic homopteran. Ecoscience 4:446–453. https ://doi.
org/10.1080/11956 860.1997.11682 422
Delabie JHC (2001) Trophobiosis between Formicidae and Hemip-
tera (Sternorrhyncha and Auchenorrhyncha): an overview. Neo-
trop Entomol 30:501–516. https ://doi.org/10.1590/s1519 -566x2
00100 04000 01
Espadaler X, Hidalgo NP, Muller WV (2012) Ant-aphid relations
in costa rica, central America (Hymenoptera: Formicidae;
Hemiptera: Aphididae). Sociobiology 59:959–970. https ://doi.
org/10.13102 /socio biolo gy.v59i3 .559
Fretwell SD (1987) Food chain dynamics: the central theory of ecol-
ogy? Oikos 50:291–301. https ://doi.org/10.2307/35654 89
Gutierrez Ochoa M, Camino Lavin M, Castrejon Ayala F, Jimenez
Perez A (1993) Arthropods associated with Bromelia hemispha-
erica (Bromeliales: Bromeliaceae) in Morelos, Mexico. Fla Ento-
mol 76:616–621. https ://doi.org/10.2307/34957 95
Hairston NG, Smith FE, Slobodkin LB (1960) Community structure,
population control, and competition. Am Nat 94:421–425. https
://doi.org/10.1086/28214 6
Hammill E, Corvalan P, Srivastava DS (2014) Bromeliad-associated
reductions in host herbivory: do epiphytic bromeliads act as
commensalists or mutualists? Biotropica 46:78–82. https ://doi.
org/10.1111/btp.12073
Hartig F (2018) DHARMa: residual diagnostics for hierarchical (multi-
level/mixed) regression models. R package version 0.1.6. https ://
CRAN.Rproj ect.org/packa ge=DHARM a
Holway DA, Lach L, Suarez AV etal. (2002) The causes and conse-
quences of ant invasions. Annu Rev Environ Resour 33:181–233.
https ://doi.org/10.1146/annur ev.ecols ys.33.01080 2.15044 4
Hölldobler B, Wilson EO (1990) The ants. Harvard University Press,
Cambridge
Jeffries MJ, Lawton JH (1984) Enemy free space and the structure of
ecological communities. Biol J Lin Soc 23:269–286. https ://doi.
org/10.1111/j.1095-8312.1984.tb001 45.x
Kindt R, Coe R (2005) Tree diversity analysis. A manual and software
for common statistical methods for ecological and biodiversity
studies. World Agroforestry Centre (ICRAF), Nairobi, ISBN
92-9059-179-X
Kohl MT, Stahler DR, Metz MC, Forester JD, Kauffman MJ, Varley N,
White PJ, Smith DW, MacNulty DR (2018) Diel predator activ-
ity drives a dynamic landscape of fear. Ecol Monogr. https ://doi.
org/10.1002/ecm.1313
Leclerc J-C, Riera P, Lévêque L, Davoult D (2016) Contrasting tempo-
ral variation in habitat complexity and species abundance distribu-
tions in four kelp forest strata. Hydrobiologia 777:33–54. https ://
doi.org/10.1007/s1075 0-016-2742-6
Linder PH, Bykova O, Dyke J, Etienne RS, Hickler T, Kühn I, Mar-
ion G, Ohlemüller R, Schymanski SJ, Singer A (2012) Biotic
modifiers, environmental modulation and species distribution
models. J Biogeogr 39:2179–2190. https ://doi.org/10.111
1/j.1365-2699.2012.02705 .x
Longino JT (2007) A taxonomic review of the genus Azteca (Hyme-
noptera: Formicidae) in Costa Rica and a global revision of the
aurita group. Zootaxa 1491:1–63
Lüdecke D (2018) ggeffects: Create tidy data frames of marginal effects
for ’ggplot’ from model outputs. R package version 0.3.3. https ://
CRAN.Rproj ect.org/packa ge=ggeff ects
McGill BJ, Enquist BJ, Weiher E, Westoby M (2006) Rebuilding com-
munity ecology from functional traits. Trends Ecol Evol 21:178–
185. https ://doi.org/10.1016/j.tree.2006.02.002
Mora BB, Gravel D, Gilarranz LJ, Poisot T, Stouffer DB (2018) Iden-
tifying a common backbone of interactions underlying food webs
from different ecosystems. Nat Commun 9:2603. https ://doi.
org/10.1038/s4146 7-018-05056 -0
Nentwig W (1986) Non-webbuilding spiders: prey specialists or gen-
eralists? Oecologia 69:571–576. https ://doi.org/10.1007/BF004
10365
Ohgushi T (2008) Herbivore-induced indirect interaction webs on
terrestrial plants: The importance of non-trophic, indirect, and
facilitative interactions. Entomol Exp Appl 128:217–229. https
://doi.org/10.1111/j.1570-7458.2008.00705 .x
Oksanen J, Blanchet GF, Friendly M, Kindt R, Legendre P, McGlinn D,
Minchin PR, O’Hara RB, Simpson GL, Solymos P, Henry M, Ste-
vens H, Szoecs E, Wagner H (2018) Vegan: community ecology
package. R package version 2.4-6. https ://CRAN.R-proje ct.org/
packa ge=vegan
Paine RT (1980) Food webs: linkage, interaction strength and com-
munity infrastructure. J Anim Ecol 49:666–685. https ://doi.
org/10.2307/4220
Peacor SD, Werner EE (2001) The contribution of trait-mediated indi-
rect effects to the net effects of a predator. Proc Natl Acad Sci
98:3904–3908. https ://doi.org/10.1073/pnas.07106 1998
Piovia-Scott J, Yang LH, Wright AN (2017) Temporal variation in
trophic cascades. Annu Rev Ecol Evol Syst 48:281–300. https ://
doi.org/10.1146/annur ev-ecols ys-12141 5-03224 6
R Core Team (2018) R: A language and environment for statistical
computing. R Foundation for Statistical Computing, Vienna
Renault CK, Buffa LM, Delfino MA (2005) An aphid-ant interaction:
effects on different trophic levels. Ecol Res 20:71–74. https ://doi.
org/10.1007/s1128 4-004-0015-8
Ripple WJ, Estes JA, Schmitz OJ, Constant V, Kaylor MJ, Lenz A,
Motley JL, Self KE, Taylor DS, Wolf C (2016) What is a trophic
cascade? Trends Ecol Evol 31:842–849. https ://doi.or g/10.1016/j.
tree.2016.08.010
Rogy P, Hammill E, Srivastava DS (2019) Complex indirect effects
of epiphytic bromeliads on the invertebrate food webs of their
support tree. Biotropica 51:549–561. https ://doi.org/10.1111/
btp.12672
Rost-Komiya B, Smith MA, Rogy P, Srivastava DS (in press) Do bro-
meliads affect the arboreal ant communities on orange trees in
northwestern Costa Rica? J Trop Ecol
Schmitz OJ, Hambäck PA, Beckerman AP (2000) Trophic cascades in
terrestrial systems: a review of the effects of carnivore removals
on plants. Am Nat 155:141–153. https ://doi.org/10.1086/30331 1
Schmitz OJ, Krivan V, Ovadia O (2004) Trophic cascades: the primacy
of trait-mediated indirect interactions. Ecol Lett 7:153–163. https
://doi.org/10.1111/j.1461-0248.2003.00560 .x
Singmann H, Bolker B, Westfall J, Aust F (2018) afex: analysis of
factorial experiments. R package version 0.20-2. https ://CRAN.
Rproj ect.org/packa ge=afex
Smith MA, Hallwachs W, Janzen DH (2014) Diversity and phyloge-
netic community structure of ants along a Costa Rican eleva-
tional gradient. Ecography 37:720–731. https ://doi.org/10.111
1/j.1600-0587.2013.00631 .x

Oecologia
1 3
Solé RV, Bascompte J (2006) Self-organization in complex ecosystems.
Princeton University Press, Princeton
Srivastava DS, Bell T (2009) Reducing horizontal and vertical diversity
in a foodweb triggers extinctions and impacts functions. Ecol Lett
12:1016–1028. https ://doi.org/10.1111/j.1461-0248.2009.01357 .x
Strong DR (1992) Are trophic cascades all wet? Differentiation and
donor-control in speciose ecosystems. Ecology 73:747–754. https
://doi.org/10.2307/19401 54
Stuntz S, Ziegler C, Simon U, Zotz G (2002) Diversity and structure
of the arthropod fauna within three canopy epiphyte species in
central Panama. J Trop Ecol 18:161–176. https ://doi.org/10.1017/
S0266 46740 20021 10
Styrsky JD, Eubanks MD (2007) Ecological consequences of interac-
tions between ants and honeydew-producing insects. Proc R Soc
B 274:151–164. https ://doi.org/10.1098/rspb.2006.3701
Thompson RM, Hemberg M, Starzomski BM, Shurin JB (2007)
Trophic levels and trophic tangles: the prevalence of
omnivory in real food webs. Ecology 88:612–617. https ://doi.
org/10.1890/05-1454
Toledo-Aceves T, García-Franco JG, Hernández-Rojas A, Macmillan
K (2012) Recolonization of vascular epiphytes in a shaded coffee
agroecosystem. Appl Veg Sci 15:99–107. https ://doi.org/10.1111/
j.1654-109X.2011.01140 .x
Touchton JM, Smith JNM (2011) Species loss, delayed numerical
responses, and functional compensation in an antbird guild. Ecol-
ogy 92:1126–1136. https ://doi.org/10.1890/10-1458.1
Venables W, Ripley B (2002) Modern applied statistics with S.
Springer, New York. https ://doi.org/10.1007/978-0-387-21706 -2
Verdolin JL (2006) Meta-analysis of foraging and predation risk trade-
offs in terrestrial systems. Behav Ecol Sociobiol 60:457–464.
https ://doi.org/10.1007/s0026 5-006-0172-6
Vinson SB (1997) Invasion of the red imported fire ant (Hymenoptera:
Formicidae): spread, biology, and impact. Am Entomol 43:23–29.
https ://doi.org/10.1093/ae/43.1.23
Way MJ (1963) Mutualism between ants and honeydew produc-
ing-homoptera. Annu Rev Entomol 8:307–344. https ://doi.
org/10.1146/annur ev.en.08.01016 3.00151 5
Wernegreen JJ, Kauppinen SN, Brady SG, Ward PS (2009) One
nutritional symbiosis begat another: phylogenetic evidence
that the ant tribe Camponotini acquired Blochmannia by tend-
ing sap-feeding insects. BMC Evol Biol 9:1–17. https ://doi.
org/10.1186/1471-2148-9-292
Wittman PK (2000) The animal community associated with canopy
bromeliads of the lowland Peruvian Amazon rain forest. Selbyana
21:48–51. https ://doi.org/10.2307/41760 052
Wu X, Griffin JN, Xi X, Sun S (2015) The sign of cascading predator
effects varies with prey traits in a detrital system. J Anim Ecol
84:1610–1617. https ://doi.org/10.1111/1365-2656.12403
Yanoviak SP, Berghoff SM, Linsenmair KE, Zotz G (2011) Effects
of an epiphytic orchid on arboreal ant community structure
in Panama. Biotropica 43:731–737. https ://doi.org/10.111
1/j.1744-7429.2011.00764 .x