You are on page 1of 13

|

Received: 12 February 2021    Accepted: 23 March 2021

DOI: 10.1111/1365-2745.13659

RESEARCH ARTICLE

From the leaf to the community: Distinct dimensions of


phytochemical diversity shape insect–­plant interactions within
and among individual plants

Leandro G. Cosmo1  | Lydia F. Yamaguchi2  | Gabriel M. F. Felix1  |


Massuo J. Kato2  | Rodrigo Cogni3  | Martín Pareja4

1
Programa de Pós-­G raduação em Ecologia,
Institute of Biology, University of Campinas Abstract
–­UNICAMP, Campinas, Brazil 1. Plant secondary chemistry is known to be an important driver of plant–­insect
2
Department of Fundamental Chemistry,
community structure across ecological scales. Recently, the concept of phyto-
Institute of Chemistry, University of São
Paulo, São Paulo-­SP, Brazil chemical diversity (PD) has been introduced to help describe variation in plant
3
Department of Ecology, University of São secondary chemistry and explain how this variation affects community structure.
Paulo, São Paulo, Brazil
4
Previous studies show that PD among individuals and species results in phyto-
Department of Animal Biology, Institute
of Biology, University of Campinas –­ chemical mosaics, known as the phytochemical landscape. However, plant traits
UNICAMP, Campinas, Brazil can vary at finer scales, such as within individuals, and even a single host plant may
Correspondence be perceived as an entire phytochemical landscape by an interacting insect.
Martín Pareja 2. Using the neotropical shrub Piper amalago, we tested and compared how her-
Email: mpareja@unicamp.br
bivory, caterpillar biodiversity and plant–­herbivore network structure are affected
Present address by the compositional (number and concentration of compounds) and structural
Leandro G. Cosmo, Ecology Graduate
School, Biosciences Institute, University of (diversity of distinct chemical structures) dimensions of PD. We analysed variation
São Paulo, São Paulo, 05508-­900, Brazil among individual plants and among-­plant height strata within individual plants.
Funding information This allowed us to decompose PD within and among plant individuals and analyse
FAPESP, Grant/Award Number: how variation at both scales affects the plant–­herbivore network.
2014/50316-­7, 2013/25991-­0, 2019/22146-­
and 3; CAPES, Grant/Award Number: 001; 3. We found that both within and among plants greater structural diversity de-
FAEPEX-­UNICAMP, Grant/Award Number: creased herbivore feeding damage. Furthermore, each dimension of PD has dif-
60412 and PAPDIC 2014/4715; CNPq,
Grant/Award Number: 304607/2019-­3, ferent effects on herbivore biodiversity and network structure depending on the
307015/2015-­7 and 307447/2018-­9; Royal scale of biological organization.
Society
4. Within plants, the compositional dimension, specifically low concentrations of
Handling Editor: Tobias Züst compounds tentatively identified as Piper amides, increased herbivore biodiver-
sity. This dimension also increased the capacity of strata within plants to mediate
ecological cascades through direct and indirect effects on herbivore abundance
in the plant–­herbivore network. In contrast, a greater structural diversity among
plants decreased herbivore biodiversity and the capacity of plants to affect all
other herbivores and plants directly and indirectly in the network.
5. Synthesis. Based on our results we expand the concept of the phytochemical
landscape to multiple scales of biological organization and provide evidence that
PD may be maintained by how its multiple dimensions have distinct roles across
scales of biological organization.

Journal of Ecology. 2021;109:2475–2487. wileyonlinelibrary.com/journal/jec© 2021 British Ecological Society     2475 |


|
2476      
Journal of Ecology COSMO et al.

KEYWORDS

chemodiversity, compositional diversity, herbivory, phytochemical landscape, phytochemical


variation, plant defence, plant–­herbivore interactions, structural diversity

1 |  I NTRO D U C TI O N diversity of insect herbivores and reduced the resulting damage suf-
fered by the plants (Bustos-­Segura et al., 2017; Poelman et al., 2009).
Understanding how plant secondary chemistry and plant diversity Likewise, at this same level of organization, Glassmire et al. (2016,
affect community structure has been a major driver of ecological 2019) found that both the compositional and structural dimensions
and evolutionary research over the last 50  years (Becerra,  2015; of PD increase along elevational gradients and with exposure to UV
Ehrlich & Raven, 1964; Tilman et al., 1996). Although interest in vari- light, which, in turn, reduces caterpillar diversity and can affect the
ation in plant chemistry has a long history, recently chemical varia- genetic structure of specialist insects. Therefore, among plant spe-
tion and biodiversity have been integrated through the concept of cies and individuals, the structural and compositional dimensions
phytochemical diversity (PD) and its cascading ecological effects can produce phytochemical mosaics, recently envisioned as a phy-
(Denno & McClure, 1983; Iason et al., 2005; Kessler & Kalske, 2018; tochemical landscape, which can interact with herbivorous insect
Moore et  al.,  2014; Randlkofer et  al.,  2010). Answering the ques- performance and have community-­wide consequences (Glassmire
tion of how PD affects insect communities remains challenging be- et al., 2019; Hunter, 2016).
cause it involves disentangling two major components of PD. First, Insects, however, can respond to small variation in plant chem-
PD is multidimensional. Phytochemical blends consist of many com- ical traits, and herbivore performance may be affected even when
pounds that differ in their concentration. The richness and abun- leaf quality differs within a single host plant (McCormick et al., 2012;
dance of compounds establish a compositional dimension of PD. Pareja et  al.,  2009; Wetzel et  al.,  2016). Such ‘fine-­grained’ varia-
Each compound, in turn, has a unique structure, composed of dis- tion is widespread in nature, and leaf morphology, phenolic content,
tinct chemical structural features (e.g. a methyl, phenol, amine and flower nectar content, fruit and seed mass have all been shown to
other groups) that are associated with the biological activity of com- vary within plants (Herrera,  2017; Herrera et  al.,  2015; Wetzel &
pounds (Gao et al., 2018; Richards et al., 2018). Similar to the com- Meek, 2019). Hence, from an insect's perspective, a single individ-
positional dimension, the richness and abundance of these structural ual plant may constitute an entire landscape of traits and when PD
features both within and among compounds constitute a structural varies within plants this can be perceived as an entire, fine-­grained,
dimension of PD. Consequently, phytochemical blends can have a phytochemical landscape (Herrera,  2009). This would suggest that
high compositional diversity, but a low structural diversity and vice the phytochemical landscape consists of several layers across scales
versa (Figure 1). Thus, disentangling how each dimension of PD af- which can be perceived by herbivorous insects and give rise to com-
fects plant–­insect interactions is a key step towards understanding plex feedbacks through different biological scales. Thus, to under-
the processes that maintain and drive the evolution of plant chem- stand such feedbacks, it is essential to assess and compare how the
ical traits and patterns of herbivore diversity and herbivory (Dyer many dimensions of PD shape herbivore biodiversity and herbivore
et al., 2003; Glassmire et al., 2019; Richards et al., 2010, 2012, 2016; damage at different scales of biological organization, from the leaf to
Wetzel & Whitehead, 2020). Second, analogous to species diversity, the ecosystem (Levin, 1992; Wetzel & Whitehead, 2020).
PD can vary across different scales of biological organization, from In order to better understand how different dimensions of PD
the individual plant level to the community (Kessler & Kalske, 2018; shape insect–­plant interactions across the levels of biological organi-
Moore et  al.,  2014). At each scale, PD may distinctly shape herbi- zation, we performed a field study in a Brazilian tropical forest using
vore biodiversity and patterns of herbivory (Glassmire et al., 2019; the shrub Piper amalago and its caterpillar herbivores. We quanti-
Hunter,  2016; Massad et  al.,  2017). Therefore a major challenge is fied the structural and compositional dimensions of PD within and
to disentangle how the multiple dimensions of PD vary and affect among plants and its effects on herbivory, caterpillar biodiversity
herbivore biodiversity at different scales of biological organization and network structure. Specifically, our aims were (a) to address how
(Kessler & Kalske, 2018; Wetzel & Whitehead, 2020). much variation in PD, caterpillar biodiversity and herbivory occurs
Over the last few years, several studies have begun to address within plants and among plants; and (b) to test whether the struc-
this challenge. At the species level, Richards et al. (2015) measured tural and compositional dimensions of PD within and among indi-
1
PD based on the number of peaks detected in H nuclear magnetic vidual plants affect herbivore biodiversity, network structure and
resonance (NMR) spectra, which gives a measure of the number herbivory. Addressing these aims will strengthen our understanding
of different structural features present in a blend (Figure  1a). This of PD by separating the effects of composition (compound number
approach captured, to a reasonable extent, the structural dimen- and identity) and structural diversity. Furthermore, focussing on the
sion of PD, which was found to increase herbivore diversity and effects of PD on ecological networks through two different scales
decrease herbivory (Richards et  al.,  2015). At the individual level, of biological organization (within and among plants) will allow us to
increased variation in the foliar glucosinolate profiles of Brassica ol- begin to understand how effects of PD can cascade through the
eracea chemotypes, hence a compositional dimension, increased the community and through ecological scales.
COSMO et al. Journal of Ecology|
      2477

F I G U R E 1   Hypothetical illustration of how the different dimensions of phytochemical diversity can be expressed in plants. (a) Five
different naturally occurring compounds with different structures, which can be reasonably captured by peaks in a 1H NMR spectra. In
an 1H NMR, each peak corresponds to the hydrogen signals present in compounds. These hydrogens resonate with different frequencies
depending on their chemical environment (e.g. a methyl group), which results in different peaks along the spectra. (b) Hypothetical scenarios
of how phytochemical blends can differ in their compositional and structural diversity, and illustrations of how they can be quantified with
the resulting high-­performance liquid chromatograms (HPLC) and 1H NMR spectra of the mixture. In an HPLC, each peak corresponds to at
least one compound that was separated from the mixture. The combinations shown are not exhaustive and there are many ways through
which phytochemical blends can differ in their structural and compositional diversity (e.g. many simple compounds with non-­redundant
structural features can also result in a high compositional and structural phytochemical diversity). The height, shape and position of peaks in
the LC are arbitrary and merely illustrative. 1H NMR spectra presented are not actual spectra but simulations predicted from the structure of
each compound and the hypothetical phytochemical blends using the MestReNova software (version 14.2.0)

2 | M ATE R I A L S A N D M E TH O DS the main Piper herbivores (Cosmo et al., 2019; Dyer & Palmer, 2004)
and are often dominant in herbivore communities. They have low
2.1 | Study site and system dispersal capabilities among plants but can change their feeding be-
haviour within individual plants (Novotny et al., 2006; Pogue, 2009;
This study was carried out in a semi-­deciduous tropical forest at the Singer, 2016). Therefore, Piper species and their herbivorous cater-
Reserva Biológica Municipal da Serra do Japi (Jundiaí, São Paulo, pillars are excellent models for phytochemical and ecological studies
Brazil, 23°14'S 46°58'W; see Supporting Information for details). We in the tropics (Dyer & Palmer, 2004).
used the shrub P. amalago L. (Piperales: Piperaceae) and its herbivo-
rous caterpillars as a model system. Piper amalago is an Atlantic Forest
perennial shrub that grows between 1 and 7  m in height (Cosmo 2.2 | Sampling, data collection and processing
et al., 2019; Guimarães & Valente, 2001). The genus Piper contains
about 1,000 species and is abundant in the Neotropics (Dyer & 2.2.1 | Plant sampling
Palmer, 2004). Piper species leaves are rich in chemical compounds,
which have been found to decrease herbivory and increase herbi- We sampled individual, naturally growing P. amalago plants at the
vore species and genetic diversity (Glassmire et al., 2016; Richards study site. At the beginning of the study, we marked all plants
et al., 2015). Caterpillars (especially of the family Geometridae) are found along three trails in the study site (the trails had a combined
|
2478      
Journal of Ecology COSMO et al.

extension of approximately 6.7  km), that were at least 1  m apart In the binary network, links represent the presence of a specific inter-
from each other with a total height of at least 30 cm. In total, we action between plant strata and caterpillar species, without consider-
marked 128 plants. To control for seasonal effects, we randomly ing the intensity of the interaction (the links do not have weights). In
sampled plants at four different times for 1  year, and each plant the weighted network, we used the abundance of caterpillars in each
was sampled only once (once sampled, the plant was removed from stratum as link weights. Thus, the binary network allows us to under-
the future sampling pool). We sampled 32 individuals in July 2017, stand the capacity of each plant or stratum to propagate direct and
and another 32 individuals in October 2017. Because four of the indirect effects through the identity of herbivores, and in the weighted
remaining 64 marked individuals died, crushed by trees falling dur- network the role of each stratum is mediated by both herbivore iden-
ing the rainy season after October 2017, we sampled 30 individuals tity and abundance. In both the binary and weighted networks, we
in January 2018 and another 30 individuals in April 2018. Thus, we measured the capacity of each plant stratum to propagate direct and
collected data for 62 different individuals in the dry season (July indirect effects through three node-­level metrics: degree, closeness
2017 and April 2018) and another 62 individuals in the rainy season centrality and Katz centrality. The degree quantifies the total number
(October 2017 and January 2018), totalling 124 individuals for the (in the binary web) or total weights (in the quantitative web) of direct
entire study period. We measured plant total height and total num- interactions of a given node. Closeness centrality (Freeman,  1978)
ber of leaves to control for plant size and proximity to the canopy, measures the average shortest distance of a node to all others in
which can affect herbivore biodiversity and herbivory (Glassmire the network and thus, how quickly and efficiently a node can reach
et al., 2019). Each plant was divided into three strata of equal height, all other nodes through either direct and/or indirect paths. Shortest
starting at the first branch with leaves. Stratum height was defined paths were identified via the igraph package (Csardi & Nepusz, 2006)
as: SH = TH − IH ∕3, in which SH is the height of each stratum, TH is and in the weighted network we used the inverse of link weights as
( )

the total height of the plant and IH is the height from the ground to the cost of a given path (i.e. higher abundance indicates that a given
the first branch with leaves. Categorically, each stratum along the path has a low cost). Katz centrality (Katz 1953) quantifies the amount
plant vertical axis was defined as low, medium and high stratum. We of direct and indirect pathways of all lengths that connects to a node
randomly collected 10 leaves within each stratum of each sampled in a network. Hence, in contrast to closeness centrality that uses only
plant (30 leaves per individual plant). These strata within individual the shortest paths, Katz centrality measures the capacity of each node
plants were used as sampling units, totalling 372 units (three strata to reach all others through pathways of all lengths. These three met-
from each of the 124 individual plants). rics, therefore, allow us to quantify the capacity of each plant stratum
to directly affect herbivores (degree), directly and indirectly affect
other herbivores and plant strata through shorter and efficient paths
2.2.2 | Caterpillar biodiversity and network structure (closeness centrality) and affect the entire network through all possible
direct and indirect pathways (Katz centrality). All analyses were per-
To assess caterpillar biodiversity among and within plants and the formed in the R environment v 3.6.0 (R Core Team, 2019).
plant–­caterpillar network structure, we manually collected all cater-
pillars found within each stratum of the sampled plants. In the field,
caterpillars were initially identified as morphospecies and their total 2.2.3 | Leaf area and leaf damage measurements
count in each stratum was recorded. The caterpillars collected were
taken to the laboratory and reared in plastic pots following standard To estimate leaf area and leaf damage, we photographed all collected
techniques (Cosmo et al., 2014). leaves against a white background along with a fixed-­length meas-
We characterized caterpillar biodiversity with three common urement reference scale and used the software ImageJ to measure
quantitative metrics: abundance (total counts), species richness and leaf area (in cm2) from the images. This same software was used to
species diversity, calculated as the Hill number of the Shannon–­ reconstruct leaves that had chewing herbivore damage and estimate
Wiener diversity index (from here on referred to as the caterpillar the total leaf area before herbivore damage. We then used the re-
SW index; Jost, 2006; Oksanen et al., 2019). A high Shannon–­Wiener constructed leaf area to calculate the proportion of total leaf area
diversity indicates samples consisting of a greater number of more lost to chewing herbivores. Finally, to estimate herbivory within and
evenly abundant caterpillar species, while a low value of the index among individual plants we used the average loss of leaf area of the
indicates either a smaller richness of caterpillar species and/or a 10 collected leaves for each stratum and the 30 collected leaves for
highly uneven distribution of abundances. For simplicity, the combi- each individual plant respectively.
nation of these three metrics will be referred to as caterpillar biodi-
versity throughout the text.
Next, we used tools from network analysis to understand how PD 2.2.4 | Structural and compositional dimensions
affects the capacity of each plant to propagate direct and indirect ef- of PD
fects in the plant–­caterpillar network. We constructed a binary qualita-
tive and a weighted quantitative bipartite interaction network in which To assess the structural and compositional dimensions of PD among
each plant stratum (our sampling unit) is linked to a caterpillar species. and within individual plants, we analysed the phytochemical profiles
COSMO et al. Journal of Ecology|
      2479

of crude methanol extracts of the same leaves sampled to assess et al., 1985). For instance, comp-­PC1 had high negative loadings with
1
herbivory. We used two complementary analyses: H NMR spectros- a compound tentatively identified as the major amide of P. amalago,
copy and high-­performance liquid chromatography–­high-­resolution nigrinodine (Jacobs et al., 1999; Yamaguchi et al., 2011), and another
electrospray ionization mass spectrometry (HPLC-­HRESIMS). common Piper amide (2-­methoxy-­4,5-­methylenedioxy-­trans-­cinna
We summarized the structural dimension of PD through the moyl pyrrolidine; Domínguez et al., 1985; Nascimento et al., 2012;
1
processed H NMR spectra (full description provided in Supporting Shah et  al.,  1986) had a high negative loading for both comp-­PC1
Information). The structural dimension was employed here to de- and comp-­PC2. Both principal components also had high positive
fine the variety in the number of the structures and/or structural loadings with other unidentified compounds.
complexity that was represented by the number and intensity of
hydrogen signals (peaks) in the NMR spectra. Following previous
studies, as a measure of the structural dimension of PD, we used 2.2.5 | Decomposing phytochemical diversity into
Hill's number of the Shannon–­Wiener diversity index for each plant within-­and among-­plant components
sample, calculated from the number of peaks and the values of
their integrated areas (Glassmire et  al.,  2019; Jost,  2006; Richards We decomposed our measurements of the structural and compo-
et  al.,  2015; Wetzel & Whitehead,  2020) in the vegan package sitional dimensions of PD into their within-­ and among-­plant com-
(Oksanen et al., 2019). In 1H NMR spectra, each peak represents the ponents. This was a necessary step to avoid incorrect inferences of
signal of hydrogen-­1 nuclei that resonate at different frequencies de- effects that occur at the among-­group level to the within-­subject
pending on their chemical environment. These peaks are associated level (the ‘ecological fallacy’, see Supporting Information for details).
with unique structural features of compounds (e.g. a methyl group), For this we subtracted the means of these variables for individual
and redundant structural features can give rise to peaks in the same plants from each observation, a procedure known as within-­subject
range of frequencies. Thus, the number of peaks in an NMR spec- centring, and a common method used to distinguish within-­ and
trum can be associated with the richness of structural features of the among-­subject effects of predictors (van de Pol & Wright,  2009).
compounds present in phytochemical blends. The relative intensity From this we obtained new variables that vary only within plants
of each peak in the spectra is proportional to the frequency of occur- that were used together with individual plant means as predictors in
rence of the structural feature in a given molecule. Calculating the our statistical analysis.
1
Shannon–­Wiener diversity from H NMR peaks allows us to quantify
the structural diversity of phytochemical blends in terms of the rich-
ness and intensity (evenness) of structural features. Analogous to 2.3 | Hypotheses testing
species diversity, high values of structural diversity represent phy-
tochemical blends whose 1H NMR spectra show a high richness of To address our objectives, we divided our hypothesis testing into
structural features (many peaks) whose abundance (signal intensity) three steps. First, we estimated the proportion of variation in our
is very even. Such high values can be a result, for instance, of a single data that occurs only among individual plants. These estimates were
(or few) compound with many non-­redundant structural features, or calculated through the intra-­class correlation (ICC). The ICC uses the
many simpler compounds that do not share the same structural fea- mixed model framework to partition variance into its within and be-
tures (Figure 1). tween grouping levels components. Such partitioning can be calcu-
The compositional dimension of PD was measured through lated as the ratio:
HPLC-­HRESIMS processed data (full description provided in SI).
The processed peak areas from the total ion chromatograms were VG
ICC = ,
integrated, adjusted by sample dry weight, normalized into propor- VG + VR
tions, clr-­transformed (Brückner & Heethoff, 2017) and submitted to
PCA. The first and second PCs (hereinafter referred to as comp-­PC1 where VG is the among-­group variance and VR is the sample-­level re-
and comp-­PC2) explained approximately 78% of the total variation sidual variance in a mixed model fitted with only a random (grouping)
(Table S2) and were used to represent the compositional dimension. effect as predictor. Hence 1 –­ICC accounts for the variation contained
In contrast to the 1H NMR spectra, this approach allowed us to as- only within groups. We estimated the ICC through the r package rptR
sess the compositional dimension of PD and its effects on caterpil- (Stoffel et al., 2017), which uses a bootstrap resampling approach to
lar biodiversity in terms of the identity of combinations of different calculate uncertainty around ICC estimates (1,000 bootstraps were
compounds in a chemical extract, rather than in terms of structural used). Caterpillar abundance and richness were both fitted with a
features. Although metabolite identification in complex matrices is Poisson GLMM, while Linear Mixed Models (LMM) were used for the
a current major challenge and we could not reliably identify these SW index and herbivory, as well as for our measures of the structural
compounds, for some we provide their molecular formulae, which and compositional dimensions of PD. For all these variables, individual
brings some insight into comp-­PCs 1 and 2 (Table  S1). The two plant ID was used as a grouping effect.
principal components were highly negatively correlated with com- Second, we tested the hypothesis that the structural and compo-
pounds tentatively identified as Piper amides (Table S1, Domínguez sitional dimensions of PD shape the capacity of each plant stratum
|
2480      
Journal of Ecology COSMO et al.

to directly and indirectly affect the entire plant–­caterpillar network. there are missing paths in the specified model and its significance
We fitted LMMs (nlme package, Pinheiro et al., 2020) with the binary is assessed with Fisher's C statistic. Our initial model was a good
and weighted versions of degree, closeness and Katz centrality as fit (Fisher's C = 28.9, p = 0.99, df = 50) and was the one chosen
response variables for the subset of plant strata which had at least as best model. All model assumptions were verified through the
one interaction. In both LMMs, we used plant ID as a random effect. package DHARMa (Hartig, 2017).
In mixed models with a grouping random effect (e.g. individual plant
ID), including a group-­level predictor automatically tests for the
effects of this predictor on the group-­level means of the response 3 | R E S U LT S
variable (Pinheiro et  al.,  2020; van de Pol & Wright,  2009). Thus,
combining mixed models with our within-­subject centring approach 3.1 | Caterpillar biodiversity, herbivory and the
allowed us to simultaneously test for the within-­ and among-­plant dimensions of phytochemical diversity vary within
effects of structural and compositional dimensions of PD in our re- plants
sponse variables. As explanatory variables, we used the within-­plant
and individual plant structural and compositional (comp-­PCs 1 and In total, we collected 273 caterpillars, and, on average, each plant
2) dimensions, within-­plant strata, the individual plant height, season harboured 2.2  ±  3.85 (mean  ±  SD) caterpillars, while we found
and total number of leaves. All predictors were scaled prior to model 0.76 ± 1.47, 0.50 ± 1.01 and 0.97 ± 1.99 (mean ± SD) in the low, me-
fitting to standardize coefficients in terms of standard deviations of dium and high strata respectively (Figure 2a). Caterpillar abundance
the mean xi−x . We tested the significance of predictors with Wald's varied seasonally and was substantially higher in the dry season than
( )
𝜎x
type II test, and model assumptions were visually assessed through in the rainy season (Figure  2a). These caterpillars were distributed
the residuals. among 18 morphospecies and two morphospecies of Eois (Piper spe-
Finally, we tested whether within and among plants the struc- cialists) were the most abundant and accounted for 86.8% of the
tural and compositional dimensions of PD shape caterpillar bio- total (Table  S2). The proportion of leaf area lost to herbivory was
diversity and herbivory. For this, we used piecewise Structural 0.12 ± 0.08, varying from 0.00 to 0.48 and the low, medium and high
Equation Modelling (piecewiseSEM, Lefcheck,  2016), which is strata suffered 0.14 ± 0.08, 0.12 ± 0.08 and 0.11 ± 0.07 (mean ± SD)
a powerful approach to evaluate causal and cascading effects respectively (Figure 2b).
among variables. We built a SEM in which we modelled each re- Our variance partitioning through the intra-­class correlation
sponse variable with LMMs or GLMMs (nlme package, Pinheiro (ICC) revealed that both caterpillar biodiversity and the two dimen-
et  al.,  2020). Based on previous work with Piper insect–­p lant in- sions of PD differed in how much they varied within and among
teractions, we hypothesized the following causal relationships: plants (Figure  3; Table  S3). Caterpillar SW index, herbivory and
(a) individual plant height affects individual plant PD, which, in comp-­PC1, all varied substantially more within plants than among
turn, drives caterpillar biodiversity and herbivory; (b) within-­p lant plants. Richness varied more within plants, but confidence intervals
strata (from low to high) affect within-­p lant PD which also drives were wide. On the other hand, caterpillar abundance and the struc-
caterpillar biodiversity and herbivory; (c) herbivory is affected by tural dimension of PD varied more among plants, while comp-­PC2
caterpillar species diversity; (d) caterpillar abundance affects spe- varied within plants as much as among plants.
cies richness and both drive species diversity; (e) structural and
compositional dimensions of PD are correlated, both within and
among plants (in SEMs this can be specified as correlated errors); 3.2 | PD shapes the direct and indirect influence of
and (f) individual plant height and within-­p lant strata also directly plant strata on the plant–­caterpillar network
affect caterpillar biodiversity and herbivory. Because caterpillar
biodiversity and herbivory in P. amalago is known to vary sea- When herbivores interact with different plants and plant strata,
sonally and with host-­p lant leaf numbers (Cosmo et al., 2019), we these interactions can be depicted as networks. In the network, al-
additionally included season (dry and rainy) and the total num- though plants or plant strata do not directly interact, they can in-
ber of leaves on the plant as fixed effects for all response vari- directly affect each other through shared herbivores. For instance,
ables. Furthermore, plant ID was included as a random effect in one plant may affect one or more herbivores, which in turn affects
all models. Percentage herbivory was logit transformed (Warton other plants and even other herbivores in the network. These ef-
& Hui,  2011). Poisson error distributions were used for caterpil- fects can be mediated solely by the richness of herbivore species
lar abundance and species richness, while Gaussian errors were that interact with a given plant and plant stratum (binary/qualitative
used for the remaining variables. Except for abundance and rich- network) or by the combined richness and abundance of herbivores
ness (because Poisson GLMM accepts only positive integers as (weighted/quantitative network). We found that the influence of
response variables), all predictors were scaled prior to model fit- PD dimension (structural or compositional) and PD scale (among or
ting to standardize path coefficients in terms of standard devia- within plants) on network metrics differed for the binary (herbivore
tions xi−x . We assessed model goodness-­of-­f it through tests of identity) and weighted (herbivore identity and abundance) networks
( )
𝜎x
directed separation (Lefcheck, 2016). This test evaluates whether (Figure 4; Tables S4 and S5).
COSMO et al. Journal of Ecology |
      2481

F I G U R E 2   Variation among seasons and strata of different heights for (a) caterpillar abundance and (b) herbivory (proportion leaf
area lost). The solid line represents the median, the box the 75% quantiles, and the whiskers represents the 95% quantiles. For each
season, data from both sampling dates (July 2017 and April 2018 for the dry season, and October 2017 and January 2018 for the rainy
season) were pooled, representing 62 plants (62 measurements from each height stratum) for each season

F I G U R E 3   Proportion of the total variation that occurred within individual plants for caterpillar abundance, species richness and Hill's
number of the Shannon–­Wiener index (SW index), and the structural and compositional dimensions of phytochemical diversity (PD).
Comp-­PC1 and comp-­PC2 refer to the scores of the two principal components that explained most variation (78.7%) in the compositional
dimension of PD (see Table S1). High values on these principal components represent plants with high concentrations of compounds 1, 2,
10, 12, 13 and 14 (high positive loadings; Table S1) and low concentrations of compounds 3, 4, 5, 6, 7, 8, 9, 11 and 16, some of which were
tentatively identified as Piper amides (high negative loadings; Table S1). Points correspond to mean values, and lines correspond to 95%
confidence intervals estimated via 1,000 bootstraps

In the binary network, the structural dimension at the scale of interactions through shorter and efficient paths) significantly de-
the individual plant was the predominant driver and significantly creased both with increasing the among-­plant structural dimension
decreased the capacity of a plant to directly and indirectly affect (β = −3.24) and increasing comp-­PC1 (β = −2.98) of the compositional
the entire network (Figure 4b; Table S4). Both degree (direct inter- dimension. Plants with low scores on comp-­PC1 were associated
actions) and Katz centrality (direct and indirect interactions through with higher relative amounts of compounds tentatively identified as
pathways of all lengths) significantly decreased with increasing plant Piper amides (Table S1). Therefore, individual plants with low struc-
structural dimension (β = −0.13 and −4.31 for degree and Katz cen- tural PD and higher relative amounts of amides have a greater direct
trality respectively), while closeness centrality (direct and indirect and indirect influence in the network through herbivore identity.
|
2482      
Journal of Ecology COSMO et al.

F I G U R E 4   Representation of the (a) binary plant–­herbivore network and (b) weighted strata–­herbivore network. Red lettered squares
represent caterpillar species. Numbered green circles represent plants in the binary network because only among-­plant metrics were
significant (see panel c) and they represent strata in the weighted network. Green circles with repeated numbers in panel (b) represent
different strata from the same plant. Darker green colours indicate higher levels of the structural dimension of phytochemical diversity
(PD) at the scale of individual plants (a) or strata within plants (b). (c) Coefficient estimates for the mixed models fitted with the binary
(in yellow) and weighted (in blue) versions of degree (direct interactions), closeness (direct and indirect interactions through the shortest
paths) and Katz centrality (direct and indirect interactions through pathways of all lengths) as response variables. Lines represent 95%
confidence intervals. For simplicity, only coefficients of non-­control variables are shown. Comp-­PC1 and comp-­PC2 refer to the scores of
the two principal components that explained most variation (78.7%) in the compositional dimension of PD. High values on these principal
components represent plants with high concentrations of compounds 1,2, 10, 12, 13 and 14 (high positive loadings; Table S1) and low
concentrations of compounds 3, 4, 5, 6, 7, 8, 9, 11 and 16, some of which were tentatively identified as Piper amides (high negative loadings;
Table S1). Please refer to Table S2 for correspondence among letters and caterpillar morphospecies. Tables S4 and S5 contain full model
details

For the weighted network, in contrast, the compositional dimen- (β = 0.38, 5.63 and 13.31 for degree, closeness and Katz centrality
sion at the within-­plant (stratum) scale was the main driver and sig- respectively). On the other hand, comp-­PC1 of the within-­plant com-
nificantly increased the direct and indirect influence of plant stratum positional dimension only increased closeness centrality (β = 6.09),
in the network (Figure 4c; Table S5). Comp-­PC2 of the within-­plant and the within-­plant structural dimension only significantly in-
compositional dimension significantly increased all three metrics creased Katz centrality (β  =  12.13). Strata with higher comp-­PC1
COSMO et al. Journal of Ecology |
      2483

scores are associated with higher concentrations of compounds 10, network metrics. The full models with all coefficients are available in
12, 13 and 14 (Table S1) while strata with higher comp-­PC2 scores the Supporting Information (Tables S4 and S5).
are associated with higher concentrations of compounds 1 and 2
(Table  S1), and these strata are therefore more likely to propagate
direct and indirect effects through both herbivore identity and 3.3 | Within and among plants, phytochemical
abundance in the network. Several compounds tentatively identi- diversity directly and indirectly affects caterpillar
fied as Piper amides (compounds 3, 4 and 8) had negative loadings biodiversity and herbivory
on both comp-­PC1 and comp-­PC2, and therefore within-­plant strata
with high concentrations of these compounds are more isolated in Through the structural equation model (SEM), we again found that
the network and less likely to propagate direct and indirect effects. the direction of the effect of PD depends on both the dimension
The control variables of individual plant height and season (structural or compositional) and the scale (among or within plants)
also significantly affected the binary and weighted version of the of PD (Figure 5). Within plants, caterpillar abundance increased with

F I G U R E 5   Results for the caterpillar biodiversity structural equation model (SEM). For simplicity, only causal and significant (p < 0.05)
paths of non-­control variables are shown (see Table S7 for all model coefficients). Abundance and species richness refer to caterpillars,
and SW index refers to caterpillar Shannon–­Wiener diversity index. Comp-­PC1 and comp-­PC2 refer to the scores of the two principal
components that explained most variation (78.7%) in the compositional dimension of PD. High values on these principal components
represent plants with high concentrations of compounds 1,2, 10, 12, 13 and 14 (high positive loadings; Table S1) and low concentrations of
compounds 3, 4, 5, 6, 7, 8, 9, 11 and 16, some of which were tentatively identified as Piper amides (high negative loadings; Table S1). Red
arrows denote negative effects, while green arrows positive effects. Coefficients are standardized estimates, except for the effects of
abundance on species richness, and the effect of species richness on species diversity (see Table S6 for full SEM results)
|
2484      
Journal of Ecology COSMO et al.

comp-­PC1 (β = 0.19) and comp-­PC2 (β = 0.12). This indicates that dimension decreased herbivore feeding damage, the compositional
higher amounts of Piper amides were associated with lower caterpil- dimension increased herbivore biodiversity. In our analyses, the
lar abundance, since these compounds had large negative loadings on compositional dimension is composed of two principal components,
both PCs (Table S1). Caterpillar abundance among plants decreased and both were negatively correlated with compounds tentatively
with increasing structural PD (β  =  −0.27). Caterpillar richness and identified as P. amalago amides, but positively correlated with other
caterpillar SW index were not directly affected by PD, but richness unidentified compounds. Therefore, within plants it is the structural
increased with abundance (β = 0.18) and richness positively affected diversity of the entire mixture of compounds that regulates the
the SW index (β = 1.27). Therefore, caterpillar richness and SW index amount of damage caused by herbivores, while blends rich in Piper
were positively affected (indirectly) by within-­plant compositional amides indirectly resulted in lower caterpillar richness and diversity.
PD, through its positive effect on abundance. Caterpillar richness These findings suggest that the identity of herbivores on a plant
and SW index were negatively affected (indirectly) by among-­plant may result from the concentration of specific compounds through
structural PD, through its negative effect on abundance. Herbivory, identification of suitable plants and plant tissues (Singer, 2016), but
on the other hand, decreases only with the structural dimension of the structural diversity of compounds in the plant tissue can limit
PD, both within (β = −0.16) and among (β = −0.20) plants (Figure S3). herbivore feeding damage. These results provide an important link
At the among-­plant scale, the structural dimension of PD decreased between the different dimensions of PD and herbivore biodiversity
with individual plant height (β = −0.17). However, within plants, struc- and highlight that both blend composition and structural diversity
tural PD increased when going from low to high strata (β  =  0.39; play important roles in structuring plant–­insect communities. It is
Figure  S2). For the compositional dimension, comp-­PC2 decreased important to note, however, that our field study did not allow us to
when going from low to high strata within plants (β = −0.17), indicating be certain that the sampling location was the result of a consistent
higher strata had higher amounts of Piper amides (Table S2; Figure S2). choice by caterpillars. Furthermore, our design does not allow us
Among plants, comp-­PC1 decreased with individual plant height to disentangle constitutive from induced chemical defences, which
(β = −0.23), indicating that lower plants had higher amounts of Piper could result in a high herbivore biodiversity causing a high PD and not
amides. The structural dimension only weakly negatively correlates the other way around. Nevertheless, because herbivore feeding be-
with comp-­PC2 both within and among plants, and weakly positively haviour and constitutive/inducible chemical diversity of plant tissues
correlates with comp-­PC1 only among plants (Table S6; Figure S4). are heritable traits, our results reinforce previous results showing
All variables were affected by season, except for the structural that plant–­herbivore co-­evolution can be mediated by the variability
and compositional dimensions of PD within plants and herbivory of phytochemicals within plants (Broekgaarden et al., 2010; Utsumi
(Table S6). Furthermore, caterpillar abundance increased with plant et al., 2011). For instance, when herbivores feed or oviposit on dif-
leaf number and individual plant height, while herbivory decreased ferent plant parts and they are unable to distinguish less defended
when going from low to high strata within plants (Table S6). plant parts, they can have longer development times and exposure
Sensitivity analysis showed that these SEM results holds for the to predators, lower pupal mass and reduced offspring fitness. These
entire caterpillar community rather than for only the most abundant fitness costs can select for herbivores to discriminate and selectively
or rare species (Tables S7 and S8). feed on less toxic plant parts. Likewise, herbivores that can detect
less defended plant parts can select for unpredictability and thus,
heterogeneity in plant chemical defences (Herrera, 2009, 2017).
4 |  D I S CU S S I O N Second, the effects of PD at the scale of the individual plant and
the strata within plants have the potential to cascade to the entire
At what levels of organization do the multiple dimensions of PD plant population and herbivore community through direct and indi-
shape insect–­plant interactions? Previously, these interactions rect effects in the plant–­herbivore network. We found that strata
have been shown to be affected at the plant species and individual with higher concentrations of specific compounds (high scores on
levels by both the structural and compositional dimensions of PD comp-­PC1 and comp-­PC2, Table S1), but from individual plants with
(Glassmire et al., 2016; Poelman et al., 2009; Richards et al., 2015). low structural PD, were more likely to directly affect the herbivores
However, we still know little about how these dimensions vary feeding upon them, indirectly affect all other herbivores in the com-
within plants and the consequences of this variation for plant–­insect munity and indirectly affect all the other plants in the population
interactions when compared to variation among plants. Here we (Figure  4). Hence, our results provide evidence that the composi-
used the tropical shrub P. amalago and its associated caterpillars as tional and structural dimensions of PD can have distinct effects on
a model system to disentangle how distinct dimensions of PD drive network structure. Furthermore the scale (individual or plant part)
insect–­plant interactions across the among-­ and within-­plant scales at which each of these dimensions varies influences the network.
of biological organization. Taken together, our results point to three Recent theory shows that in networks, nodes that can propagate
major properties of how the structural and compositional dimen- more direct and indirect effects can drive stronger and more pre-
sions of PD shape insect–­plant interactions among and within plants. dictable extinction cascades (Pires et al., 2020). Our results, there-
First, we show that each dimension of PD can mediate different fore, suggest that variation in chemical traits both among and within
ecological processes within individual plants. While the structural plants is key to understanding and predicting how disturbances and
COSMO et al. Journal of Ecology |
      2485

extinctions affect the associated herbivore community and entire AC K N OW L E D G E M E N T S


plant populations. Another extension of these results, together with The authors are grateful to the Fundação Serra do Japi, in par-
our results that the structural dimension of PD decreases herbivore ticular to Vânia P. Nunes, for logistical support and for allowing
feeding damage, is the possibility that managing PD in agroeco- them to carry out their study at the Reserva Biológica Municipal
systems could be essential for limiting pest damage. For instance, da Serra do Japi. They thank Paulo R. Guimarães Jr, Débora C.
it could be highly beneficial to manage agroecosystem diversity to Rother, André V.L. Freitas, Mariana A. Stanton, Tara J. Massad and
increase overall structural PD throughout the agroecosystem, thus Mathias M. Pires for discussions and suggestions during planning,
reducing damage to crops. It is quite likely that many as yet unknown data analysis and manuscript writing. They are grateful to Luiz H.
mechanisms related to PD are involved in the well-­known benefits Rezende, João V. Suzigan and Juliana Rink for help with the field
of agroecosystem diversification for agroecosystem services and work. They thank two anonymous reviewers whose comments
sustainable crop production, and we believe this area is a promis- greatly improved the manuscript, and Tobias Züst for help during
ing one for future research (Hauri et al., In Press; Silva et al., 2018; the submission process. This study was financed by a FAPESP-­NSF
Vandermeer et al., 2019). Dimensions of Biodiversity-­Biota collaborative grant ‘Chemically
Third, PD may be maintained by how its multiple dimensions mediated multi-­t rophic interaction diversity across tropical gra-
interact with herbivores and mediate ecological and evolutionary dients’ (2014/50316-­7 ) and by CAPES-­F inance Code 001. M.J.K.
processes on different scales of biological organization. Our results has financial support from CNPq (304607/2019-­3). M.P. received
show that the structural dimension varies more among plants than support from FAEPEX-­UNICAMP (grant no 60412 and PAPDIC
within plants. However, within plants, even a small amount of varia- 2014/4715). R.C. is funded by FAPESP (2013/25991-­0), CNPq
tion in this dimension was enough to decrease herbivory. In contrast, (307015/2015-­7 and 307447/2018-­9) and a Newton Advanced
the compositional dimension varied more and affected herbivore Fellowship from the Royal Society. L.G.C. currently receives a
abundance only within plants, while herbivore abundance among FAPESP scholarship (2019/22146-­3). The authors declare no con-
plants decreased only with the structural dimension (Figures 3 and flict of interests.
5). This scale dependence of the effects of PD may be mediated by
how different herbivores respond to variation of the dimensions of AU T H O R S ' C O N T R I B U T I O N S
PD at different scales. Herbivores differ in the scales at which they L.G.C., M.P. and R.C. designed the study; L.G.C. and M.P. collected
make feeding decisions, from the plant part, plant individual to the the field data; L.G.C., L.F.Y. and M.J.K. performed chemical analysis;
plant patch, and PD is known to have different effects on specialists L.G.C., G.M.F.F., M.P. and R.C. conducted statistical analyses; L.G.C.
and generalists (Emerson et  al.,  2012; Massad et  al.,  2017; Moore wrote the first draft of the manuscript. All the authors contributed
et al., 2010). In turn, this interaction of PD between scales is likely substantially to further revisions.
to be a major driver of the evolution of local and systemic induced
responses, through the maximization of chemical noise, as proposed DATA AVA I L A B I L I T Y S TAT E M E N T
by Kessler and Kalske (2018). Data available from the Dryad Digital Repository https://doi.org/​
Our results show that the structural and compositional dimen- 10.5061/dryad.p8cz8​w9mz (Cosmo et al., 2021).
sions of PD may vary and shape insect–­plant interactions on dif-
ferent scales. The question then emerges of how these processes ORCID
transfer from the finer within-­plant scale to the broader among Leandro G. Cosmo  https://orcid.org/0000-0003-2541-2645
individual and species scales. Or, paraphrasing Levin (1992), how Lydia F. Yamaguchi  https://orcid.org/0000-0003-2305-8208
do we scale from leaf chemistry to the plant–­herbivore community Gabriel M. F. Felix  https://orcid.org/0000-0002-3901-4333
to the landscape and beyond? A major development in this direc- Massuo J. Kato  https://orcid.org/0000-0002-3315-2129
tion was the proposal of the concept of a phytochemical landscape Rodrigo Cogni  https://orcid.org/0000-0001-9907-9297
(Hunter, 2016). In this context, we conclude that for herbivorous in- Martín Pareja  https://orcid.org/0000-0002-2615-9947
sects, even a single individual host plant may represent an entire,
fine-­grained phytochemical landscape (Herrera,  2009). Thus, we REFERENCES
expand upon Hunter's (2016) hypothesis and propose that the phy- Becerra, J. X. (2015). On the factors that promote the diversity of her-
tochemical landscape emerges as different layers across scales of bivorous insects and plants in tropical forests. Proceedings of the
National Academy of Sciences of the United States of America, 112,
biological organization, within which the different dimensions of PD
6098–­6103. https://doi.org/10.1073/pnas.14186​43112
can be the dominant components and have distinct scale-­dependent Broekgaarden, C., Poelman, E. H., Voorrips, R. E., Dicke, M., & Vosman,
effects. How information perceived by herbivorous insects is trans- B. (2010). Intraspecific variation in herbivore community composi-
ferred across these layers remains an open and exciting question. tion and transcriptional profiles in field-­grown Brassica oleracea
cultivars. Journal of Experimental Botany, 61(3), 807–­819. https://doi.
Therefore, we argue that to indeed scale from the leaf to the ecosys-
org/10.1093/jxb/erp347
tem, future studies should focus on disentangling how the dimen- Brückner, A., & Heethoff, M. (2017). A chemo-­ecologists’ practical guide
sions of PD and herbivore responses are connected across multiple to compositional data analysis. Chemoecology, 27, 33–­46. https://doi.
scales and synergistically drive insect–­plant interactions. org/10.1007/s0004​9-­016-­0227-­8
|
2486      
Journal of Ecology COSMO et al.

Bustos-­Segura, C., Poelman, E. H., Reichelt, M., Gershenzon, J., & Gols, Herrera, C. M. (2017). The ecology of subindividual variability in plants:
R. (2017). Intraspecific chemical diversity among neighbouring plants Patterns, processes, and prospects. Web Ecology, 17, 51–­6 4. https://
correlates positively with plant size and herbivore load but nega- doi.org/10.5194/we-­17-­51-­2017
tively with herbivore damage. Ecology Letters, 20, 87–­97. https://doi. Herrera, C. M., Medrano, M., & Bazaga, P. (2015). Continuous within-­plant
org/10.1111/ele.12713 variation as a source of intraspecific functional diversity: Patterns,
Cosmo, L. G., Barbosa, E. P., & Freitas, A. V. L. (2014). Biology and morphol- magnitude, and genetic correlates of leaf variability in Helleborous
ogy of the immature stages of Hermeuptychia atalanta (Lepidoptera: foetidus (Ranunculaceae). American Journal of Botany, 102, 225–­232.
Nymphalidae: Satyrinae). Annales de la Societé Entomologique de Hunter, M. D. (2016). The phytochemical landscape: Linking trophic interac-
France, 50, 82–­88. tions and nutrient dynamics. Princeton University Press.
Cosmo, L. G., Nascimento, A. R., Cogni, R., & Freitas, A. V. L. (2019). Iason, G. R., Lennon, J. J., Pakeman, R. J., Thoss, V., Beaton, J. K., Sim,
Temporal distribution in a tri-­trophic system associated with Piper D. A., & Elston, D. A. (2005). Does chemical composition of indi-
amalago L. in a tropical seasonal forest. Arthropod-­Plant Interactions, vidual Scots pine trees determine the biodiversity of their associ-
13, 647–­652. https://doi.org/10.1007/s1182​9-­019-­09687​-­y ated ground vegetation? Ecology Letters, 8, 364–­369. https://doi.
Cosmo, L. G., Yamaguchi, L. F., Felix, G. M. F., Kato, M. J., Cogni, R., & org/10.1111/j.1461-­0248.2005.00732.x
Pareja, M. (2021). Data from: From the leaf to the community: Jacobs, H., Seeram, N. P., Nair, M. G., Reynolds, W. F., & McLean, S.
Distinct dimensions of phytochemical diversity shape insect–­plant (1999). Amides of Piper amalago var. nigrinodum. Journal of the Indian
interactions within and among individual plants. Dryad Digital Chemical Society, 76, 713–­717.
Repository, https://doi.org/10.5061/dryad.p8cz8​w 9mz Jost, L. (2006). Entropy and diversity. Oikos, 113, 363–­375.
Csardi, G., & Nepusz, T. (2006) The igraph software package for complex Kessler, A., & Kalske, A. (2018). Plant secondary metabolite diversity
network research. InterJournal Complex Systems, 1695. and species interactions. Annual Review of Ecology, Evolution, and
Denno, R. F., & McClure, M. S. (1983). Variable plants and herbivores in Systematics, 49, 115–­138. https://doi.org/10.1146/annur​ev-­ecols​ys-­
natural and managed systems (p. 717). Academic Press. 11061​7-­062406
Domínguez, X. A., Verde, J. S., Sucar, S., & Treviño, R. (1985). Two am- Lefcheck, J. S. (2016). piecewiseSEM: Piecewise structural equation
ides from Piper amalago. Phytochemistry, 25, 239–­240. https://doi. modelling in R for ecology, evolution, and systematics. Methods in
org/10.1016/S0031​-­9422(00)94536​-­9 Ecology and Evolution, 7, 573–­579.
Dyer, L. A., Dodson, C. D., Stireman, III J. O., Tobler, M. A., Smilanich, A. M., Levin, S. A. (1992). The problem of pattern and scale in ecology. Ecology,
Fincher, R. M., & Letourneau, D. K. (2003). Synergistic effects of three 73, 1943–­1967.
piper amides on generalist and specialist herbivores. Journal of Chemical Massad, T. J., Martins de Moraes, M., Philbin, C., Oliveira, C., Cebrian
Ecology, 29(11), 2499–­2514. https://doi.org/10.1023/a:10263​10001958 Torrejon, G., Fumiko Yamaguchi, L., Jeffrey, C. S., Dyer, L. A., Richards,
Dyer, L. A., & Palmer, A. D. N. (2004). Piper: A model genus for studies L. A., & Kato, M. J. (2017). Similarity in volatile communities leads to
of phytochemistry, ecology, and evolution (p. 214). Kluwer Academic/ increased herbivory and greater tropical forest diversity. Ecology, 98,
Plenum Publishers. 1750–­1756. https://doi.org/10.1002/ecy.1875
Ehrlich, P. R., & Raven, P. H. (1964). Butterflies and plants: A study in McCormick, A. C., Unsicker, S. B., & Gershenzon, J. (2012). The specific-
coevolution. Evolution, 18, 586–­608. https://doi.org/10.1111/ ity of herbivore-­induced plant volatiles in attracting herbivore ene-
j.1558-­5646.1964.tb016​74.x mies. Trends in Plant Science, 17, 303–­310. https://doi.org/10.1016/j.
Emerson, S. E., Brown, J. S., Whelan, C. J., & Schmidt, K. A. (2012). Scale-­ tplan​t s.2012.03.012
dependent neighborhood effects: Shared doom and associational Moore, B. D., Andrew, R. L., Kulheim, C., & Foley, W. J. (2014). Explaining
refuge. Oecologia, 168, 659–­670. https://doi.org/10.1007/s0044​ intraspecific diversity in plant secondary metabolites in an ecologi-
2-­011-­​2144-­4 cal context. New Phytologist, 201, 733–­750. https://doi.org/10.1111/
Freeman, L. C. (1978). Centrality in social networks: Conceptual clarifi- nph.12526
cation. Social Networks, 1, 215–­239. https://doi.org/10.1016/0378-­​ Moore, B. D., Lawler, I. R., Wallis, I. R., Beale, C. M., & Foley, W. J. (2010).
8733(78)90021​-­7 Palatability mapping: A koala's eye view of spatial variation in habitat
Gao, H., Li, G., & Lou, H.-­X . (2018). Structural diversity and biological ac- quality. Ecology, 91, 3165–­3176. https://doi.org/10.1890/09-­1714.1
tivities of novel secondary metabolites from endophytes. Molecules, Nascimento, J. C. D., Paula, V. F. D., David, J. M., & David, J. P. (2012).
23, 646. https://doi.org/10.3390/molec​ules2​3 030646 Occurrence, biological activities and 13C NMR data of amides from
Glassmire, A. E., Jeffrey, C. S., Forister, M. L., Parchman, T. L., Nice, C. Piper (Piperaceae). Química Nova, 35, 2288–­2311. https://doi.
C., Jahner, J. P., Wilson, J. S., Walla, T. R., Richards, L. A., Smilanich, org/10.1590/S0100​- ­4 0422​01200​1100037
A. M., Leonard, M. D., Morrison, C. R., Simbaña, W., Salagaje, L. A., Novotny, V., Drozd, P., Miller, S. E., Kulfan, M., Janda, M., Basset, Y., &
Dodson, C. D., Miller, J. S., Tepe, E. J., Villamarin-­Cortez, S., & Dyer, Weiblen, G. D. (2006). Why are there so many species of herbivorous
L. A. (2016). Intraspecific phytochemical variation shapes community insects in tropical rainforests? Science, 313, 1115–­1118. https://doi.
and population structure for specialist caterpillars. New Phytologist, org/10.1126/scien​ce.1129237
212, 208–­219. https://doi.org/10.1111/nph.14038 Oksanen, J., Blanchet, F., Friendly, M., Kindt, R., Legendre, P., McGlinn,
Glassmire, A. E., Philbin, C., Richards, L. A., Jeffrey, C. S., Snook, J. S., D., & Solymos, P. (2019). vegan: Community ecology package. R pack-
& Dyer, L. A. (2019). Proximity to canopy mediates changes in the age version 2.5.4.
defensive chemistry and herbivore loads of an understory tropical Pareja, M., Mohib, A., Birkett, M. A., Dufour, S., & Glinwood, R. T. (2009).
shrub, Piper kelleyi. Ecology Letters, 22, 332–­3 41. Multivariate statistics coupled to generalized linear models reveal
Guimarães, E. F., & Valente, C. (2001). Piperaceae –­Piper. In R. Reitz (Ed.), complex use of chemical cues by a parasitoid. Animal Behaviour, 77,
Flora Ilustrada Catarinense (p. 104). Herbário Barbosa Rodrigues. 901–­909. https://doi.org/10.1016/j.anbeh​av.2008.12.016
Hartig, F. (2017). DHARMa: Residual diagnostics for hierarchical (multi-­ Pinheiro, J., Bates, D., DebRoy, S., Sarkar, D., & R Core Team. (2020). nlme:
level/mixed) regression models. R package version 0.1.5. Linear and nonlinear mixed effects models. R package version 3.1-­145.
Hauri, K. C., Glassmire, A. E., & Wetzel, W. C. (In Press). Chemical diver- Pires, M. M., O'Donnell, J. L., Burkle, L. A., Díaz-­C astelazo, C., Hembry,
sity rather than cultivar diversity predicts natural enemy control of D. H., Yeakel, J. D., Newman, E. A., Medeiros, L. P., Aguiar, M. A. M.,
herbivore pests. Ecological Applications, e2289. & Guimarães, P. R. (2020). The indirect paths to cascading effects of
Herrera, C. M. (2009). Multiplicity in unity: Plant subindividual variation extinctions in mutualistic networks. Ecology, 101(7), e03080. https://
and interactions with animals. University of Chicago Press. doi.org/10.1002/ecy.3080
COSMO et al. Journal of Ecology |
      2487

Poelman, E. H., van Dam, N. M., van Loon, J. J. A., Vet, L. E. M., & Dicke, Tilman, D., Wedin, D., & Knops, J. (1996). Productivity and sustainability
M. (2009). Chemical diversity in Brassica oleracea affects biodiversity influenced by biodiversity in grassland ecosystems. Nature, 379, 718–­
of insect herbivores. Ecology, 90, 1863–­1877. 720. https://doi.org/10.1038/379718a0
Pogue, M. G. (2009). Biodiversity of Lepidoptera. In R. G. Foottit & P. Utsumi, S., Ando, Y., Craig, T. P., & Ohgushi, T. (2011). Plant genotypic di-
H. Adler (Eds.), Insect biodiversity: Science and society (pp. 325–­355). versity increases population size of a herbivorous insect. Proceedings
Blackwell Publishing Ltd. of the Royal Society B: Biological Sciences, 278(1721), 3108–­3115.
R Core Team. (2019). R: A language and environment for statistical comput- https://doi.org/10.1098/rspb.2011.0239
ing. R Foundation for Statistical Computing. van de Pol, M. V., & Wright, J. (2009). A simple method for distinguishing
Randlkofer, B., Obermaier, E., Hilker, M., & Meiners, T. (2010). within-­ versus between-­subject effects using mixed models. Animal
Vegetation complexity-­T he influence of plant species diversity Behaviour, 77, 753–­758. https://doi.org/10.1016/j.anbeh​av.2008.​
and plant structures on plant chemical complexity and arthropods. 11.006
Basic and Applied Ecology, 11, 383–­395. https://doi.org/10.1016/j. Vandermeer, J., Armbrecht, I., de la Mora, A., Ennis, K. K., Fitch, G.,
baae.2010.03.003 Gonthier, D. J., Hajian-­Forooshani, Z., Hsieh, H.-­Y., Iverson, A.,
Richards, L. A., Dyer, L. A., Forister, M. L., Smilanich, A. M., Dodson, C. Jackson, D., Jha, S., Jiménez-­Soto, E., Lopez-­Bautista, G., Larsen, A.,
D., Leonard, M. D., & Jeffrey, C. S. (2015). Phytochemical diversity Li, K., Liere, H., MacDonald, A., Marin, L., Mathis, K. A., … Perfecto,
drives plant-­insect community diversity. Proceedings of the National I. (2019). The community ecology of herbivore regulation in an agro-
Academy of Sciences of the United States of America, 112, 10973–­ ecosystem: Lessons from complex systems. BioScience, 69, 974–­996.
10978. https://doi.org/10.1073/pnas.15049​77112 https://doi.org/10.1093/biosc​i/biz127
Richards, L. A., Dyer, L. A., Smilanich, A. M., & Dodson, C. D. (2010). Warton, D. I., & Hui, F. K. C. (2011). The arcsine is asinine: The analysis of
Synergistic effects of amides from two piper species on generalist proportions in ecology. Ecology, 92, 3–­10. https://doi.org/​10.1890/​
and specialist herbivores. Journal of Chemical Ecology, 36(10), 1105–­ 10-­0340.1
1113. https://doi.org/10.1007/s1088​6-­010-­9852-­9 Wetzel, W. C., Kharouba, H. M., Robinson, M., Holyoak, M., & Karban, R.
Richards, L. A., Glassmire, A. E., Ochsenrider, K. M., Smilanich, A. M., (2016). Variability in plant nutrients reduces insect herbivore perfor-
Dodson, C. D., Jeffrey, C. S., & Dyer, L. A. (2016). Phytochemical di- mance. Nature, 539, 425–­427. https://doi.org/10.1038/natur​e20140
versity and synergistic effects on herbivores. Phytochemistry Reviews, Wetzel, W. C., & Meek, M. H. (2019). Physical defenses and herbivory
15(6), 1153–­1166. https://doi.org/10.1007/s1110​1-­016-­9479-­8 vary more within plants than among plants in the tropical understory
Richards, L. A., Lampert, E. C., Bowers, M. D., Dodson, C. D., shrub Piper polytrichum. Botany-­Botanique, 97, 113–­121.
Smilanich, A. M., & Dyer, L. A. (2012). Synergistic Effects of Iridoid Wetzel, W. C., & Whitehead, S. R. (2020). The many dimensions of phy-
Glycosides on the Survival, Development and Immune Response tochemical diversity: Linking theory to practice. Ecology Letters, 23,
of a Specialist Caterpillar, Junonia coenia (Nymphalidae). Journal 16–­32. https://doi.org/10.1111/ele.13422
of Chemical Ecology, 38(10), 1276–­1284. https://doi.org/10.1007/ Yamaguchi, L. F., Freitas, G. C., Yoshida, N. C., Silva, R. A., Gaia, A. M.,
s1088​6 -­012-­0190-­y Silva, A. M., Scotti, M. T., Emerenciano, V. D. P., Guimarães, E. F.,
Richards, L. A., Oliveira, C., Dyer, L. A., Rumbaugh, A., Urbano-­Muñoz, F., Floh, E. I. S., Colombo, C. A., Siqueira, W. J., & Kato, M. J. (2011).
Wallace, I. S., Dodson, C. D., & Jeffrey, C. S. (2018). Shedding light on Chemometric analysis of ESIMS and NMR data from Piper species.
chemically mediated tri-­trophic interactions: A 1H-­NMR network ap- Journal of the Brazilian Chemical Society, 22, 2371–­2382. https://doi.
proach to identify compound structural features and associated bio- org/10.1590/S0103​-­50532​01100​1200019
logical activity. Frontiers in Plant Science, 9. https://doi.org/10.3389/
fpls.2018.01155
Shah, S., Kalla, A. K., & Dhar, K. L. (1986). A cinnamoyl pyrrolidine amide
S U P P O R T I N G I N FO R M AT I O N
from Piper peepuloides. Phytochemistry, 25, 1997–­1998. https://doi.
org/10.1016/S0031​-­9422(00)81196​-­6 Additional supporting information may be found online in the
Silva, R. F., Rabeschini, G. B. P., Peinado, G. L. R., Cosmo, L. G., Rezende, Supporting Information section.
L. H. G., Murayama, R. K., & Pareja, M. (2018). The ecology of
plant chemistry and multi-­species interactions in diversified agro-
ecosystems. Frontiers in Plant Science, 9. https://doi.org/10.3389/
How to cite this article: Cosmo LG, Yamaguchi LF, Felix GMF,
fpls.2018.01713
Singer, M. S. (2016). Behaviorally plastic host-­plant use by larval Kato MJ, Cogni R, Pareja M. From the leaf to the community:
Lepidoptera in tri-­trophic food webs. Current Opinion in Insect Distinct dimensions of phytochemical diversity shape
Science, 14, 56–­60. https://doi.org/10.1016/j.cois.2016.01.005 insect–­plant interactions within and among individual plants.
Stoffel, M. A., Nakagawa, S., Schielzeth, H., & Goslee, S. (2017). rptR:
J Ecol. 2021;109:2475–­2487. https://doi.
Repeatability estimation and variance decomposition by generalized
linear mixed-­effects models. Methods in Ecology and Evolution, 8, org/10.1111/1365-­2745.​13659
1639–­1644. https://doi.org/10.1111/2041-­210X.12797

You might also like