Content uploaded by Richard Karban
Author content
All content in this area was uploaded by Richard Karban on Oct 17, 2023
Content may be subject to copyright.
LETTER Volatile communication between plants that affects herbivory:
a meta-analysis
Richard Karban,
1
* Louie H. Yang,
1
and Kyle F. Edwards
2
Abstract
Volatile communication between plants causing enhanced defence has been controversial. Early
studies were not replicated, and influential reviews questioned the validity of the phenomenon.
We collected 48 well-replicated studies and found overall support for the hypothesis that resis-
tance increased for individuals with damaged neighbours. Laboratory or greenhouse studies and
those conducted on agricultural crops showed stronger induced resistance than field studies on
undomesticated species, presumably because other variation had been reduced. A cumulative
analysis revealed that early, non-replicated studies were more variable and showed less evidence
for communication. Effects of habitat and plant growth form were undetectable. In most cases,
the mechanisms of resistance and alternative hypotheses were not considered. There was no
indication that some response variables were more likely to produce large effects. These results
indicate that plants of diverse taxonomic affinities and ecological conditions become more resis-
tant to herbivores when exposed to volatiles from damaged neighbours.
Keywords
Eavesdropping, induced resistance, plant behaviour, plant signalling, volatiles.
Ecology Letters (2014) 17: 44–52
INTRODUCTION
When some plants are attacked by herbivores, they release
chemical cues that cause other individuals to change their
traits and become more resistant to herbivory. Communica-
tion between plants was first observed and reported more than
30 years ago (Rhoades 1983; Baldwin & Schultz 1983) and
the number of reported cases has grown rapidly in the recent
past (Heil & Karban 2010). We consider a process to be plant
volatile communication if it involves signalling by a plant that
causes a response in the same or a different individual that
receives the cue. Emission or display of a cue is plastic and
the response of the receiver is conditional on receiving the
cue. We require that emitting the cue could potentially benefit
the emitter, although this has proven difficult or impossible to
establish (Karban 2008). For an interaction to be considered
communication, the responder must have the choice of
responding to the cue or not, a requirement which excludes
allelopathy (Schenk & Seabloom 2010). We make no assump-
tions about the intended target of the cues; many plants use
volatile cues to co-ordinate their own defences against herbi-
vores when one branch is attacked and other branches on the
same individual respond by increasing defences (Karban et al.
2006; Frost et al. 2007; Heil & Silva Bueno 2007; Rodriguez-
Saona et al. 2009). There is no agreed upon definition of
communication; our use of the term is broader than most and
includes phenomena that some authors prefer to call eaves-
dropping or signalling.
Early reports of plant communication met with great inter-
est from scientists and the popular press. David Rhoades
observed that caterpillars placed on willow trees near dam-
aged neighbours grew less well than those placed on trees near
undamaged neighbours (Rhoades 1983). He hypothesised that
the reduction in performance was caused by airborne commu-
nication from the damaged trees that increased resistance in
neighbours. However, he was unable to repeat his initial
results and his experimental design lacked true replication (D.
Rhoades, pers. comm., Fowler & Lawton 1985). In addition,
the poor performance of caterpillars on trees in close proxim-
ity to infested neighbours could have been caused by the
introduction of insect pathogens rather than by communica-
tion between plants. Early experiments conducted on plants in
growth chambers reported that plants exposed to volatiles
coming from chambers containing feeding herbivores became
more resistant, but the experimental designs used in these
studies also lacked true replication (Baldwin & Schultz 1983;
Bruin et al. 1992).
Following an influential review by Simon Fowler and John
Lawton, most ecologists decided that communication between
plants was a phenomenon that had been considered and
debunked and that the phenomenon did not occur in nature
(Fowler & Lawton 1985; Dicke & Bruin 2001). In addition to
the limitations associated with experimental design of the
early studies, communication between plants that benefited
neighbours did not make sense to many ecologists. Early
descriptions of this phenomenon were referred to as ‘talking
trees’ by both the popular press and some scientists in the
field. Natural selection would not be expected to favour the
emission of cues that provided neighbouring competitors with
information about herbivores. Coinciding with the resurgence
1
Department of Entomology, University of California, Davis, CA, 95616, USA
2
Kellogg Biological Station, Michigan State University, 3700 E. Gull Lake Dr.,
Hickory Corners, MI, 49060, USA
*Correspondence: E-mail: rkarban@ucdavis.edu
©2013 John Wiley & Sons Ltd/CNRS
Ecology Letters, (2014) 17: 44–52 doi: 10.1111/ele.12205
in scientific interest in communication has been a resolution
of this apparent contradiction. ‘Talking trees’ is a particularly
poor jargon for these phenomena since ‘listening trees’ may
more accurately reflect the process and should be more in
keeping with how evolution might work. It is well accepted
that plants respond to environmental cues including light and
nutrients to make allocation decisions about the plastic
growth and abscission of shoots and roots (Ballare 1999). It is
easy to imagine selection favouring individuals that adjust
their allocation decisions and their defences based on diverse
sources of reliable information that they acquire from their
environments, including eavesdropping on the wounding
responses of neighbouring plants.
This ecological phenomenon almost got lost because it was
first over-sensationalised by the popular press and then it
became taboo for scientists to consider. D. Rhoades (pers.
comm.) was unable to secure funding to continue his work in
this field and he left science. Interest in communication
between plants revived after the publication of several well-
documented and more carefully controlled cases in 2000
(Arimura et al. 2000; Birkett et al. 2000; Dolch & Tscharntke
2000; Karban et al. 2000) along with more sympathetic
reviews (e.g. Dicke & Bruin 2001). Over the past few years,
communication between plants has been more widely
accepted, and a recent review listed 14 studies (Heil & Karban
2010). However, this literature has recently expanded, and not
all of the studies have found evidence for communication that
affects herbivory (Fig. 2).
In some cases, communication between individuals may be
co-opting a process that originally evolved to allow plants to
co-ordinate their own systemic responses. For many plants,
vascular connections among plant modules (branches, roots,
etc.) are greatly limited, and systemic signals cannot spread
throughout the vasculature of an individual (Waisel et al.
1972; Orians 2005). For individuals of these species, volatile
communication may represent a more effective and rapid
means of signalling among different organs within a single
individual (Farmer 2001; Karban et al. 2006; Heil & Adame-
Alvarez 2010). Furthermore, the relatively short distances over
which volatile signals are effective increases the likelihood that
signals will only provide information to other tissues on the
same individual as the emitter. Many plants reproduce vegeta-
tively or with limited dispersal so that individuals that are
spatially close neighbours are more likely to share genes. Plant
populations that are viscous in this way can more easily
evolve traits that benefit responders since emitters are more
likely to benefit by increasing the inclusive fitness of closely
related neighbours (Hamilton 1964). In addition, if individuals
are able to recognise kin and respond differently to cues from
kin compared to strangers, then cooperative behaviours such
as communication are more likely to evolve. Recent empirical
results supported this hypothesis; communication was more
effective among closely related sagebrush individuals than
among distantly related individuals (Karban et al. 2013).
Many of the volatile chemicals that plants emit when they
have been damaged by herbivores produce a diversity of eco-
logical consequences, any of which could help to explain their
selective advantage (Penuelas & Llusia 2004). For example,
volatile emissions protect plants from heat and oxidative
stresses (Loreto & Velikova 2001; Behnke et al. 2007). The
volatiles released by herbivory are often directly repellent to
herbivores (Bernasconi et al. 1998; De Moraes et al. 2001)
and allelopathic to plant competitors (Karban 2007; Inderjit
et al. 2009). These same volatiles attract the predators and
parasites of herbivores that can decrease levels of damage
inflicted by herbivory (Dicke & Sabelis 1988; Turlings et al.
1990; Thaler 1999; Kessler & Baldwin 2001). However, there
is still surprisingly little convincing evidence that predators
responding to volatile cues released by damaged plants actu-
ally increase the fitness of those plants under natural condi-
tions (Allison & Hare 2009; Kessler & Heil 2011). Theoretical
concerns that plant communication contradicts evolutionary
common sense have largely been addressed in recent years.
We have conducted a meta-analysis of the published and
unpublished studies available to us that have considered com-
munication between plants that affect their interactions with
herbivores. We have addressed these specific questions: (1)
Does volatile plant communication increase resistance of
plants to herbivory, on average? (2) Which response variables
are affected by volatile plant communication? (3) Under what
ecological conditions is volatile plant communication
observed? (4) Is there a publication bias in reporting plant
communication involving volatile cues?
THE DATA SET
The data set was compiled by conducting keyword searches in
the ISI Web of Science up to June 2012, by collecting studies
from recent reviews, and by contacting colleagues who are
actively working in the this field. We included 48 studies in
the meta-analysis that met the following criteria: (1) plants
were subjected to at least two treatments –exposure to a vola-
tile cue and a control; (2) the authors provided means, some
measure of variance, and sample sizes for each treatment
group; (3) there were at least two independent replicates of
each treatment group; (4) the study tested for an effect of
these treatments on herbivores or plant damage caused by
herbivory. We did not require that the treatments be ran-
domly assigned so that we included ‘natural experiments’ in
which the authors observed differences among groups but
could be less confident about cause and effect relationships.
When studies were pseudoreplicated, we reduced the number
of replicates reported by the authors but still included the
study in the meta-analysis, albeit weighted less heavily. How-
ever, studies were excluded if they had only one true replicate
per treatment. We did not include studies documenting
responses to herbivore-induced plant volatiles involving preda-
tors and parasites or indirect plant responses that benefited
these higher trophic levels, as this subject has been reviewed
frequently elsewhere (e.g. Allison & Hare 2009; Kessler & Heil
2011). Studies of associational resistance in which a volatile
cue was not likely to be induced by experimental treatments
were not included. There were many cases in which multiple
experiments conducted at different sites, at different times, or
involving closely related response variables were available. In
these cases, we included only a single pair of means for the
two treatments. We reported mean values over time and space
when these data were presented by the authors or could be
©2013 John Wiley & Sons Ltd/CNRS
Letter Plant communication and herbivory 45
easily calculated. However, publications involving different
emitter species, different receiver species, or different and
unrelated response variables were considered different ‘studies’
and were included in the analysis.
Our final data set consisted of 48 studies published (and
unpublished) between 1983 and 2012. Our literature search
ended in June 2012. Our data set included 33 plant species in
15 different families.
META-ANALYSIS
We performed a meta-analysis using the log response ratio to
quantify effect size across the experiments in our data set. The
log response ratio is defined as log (x
T
/x
C
), where x
T
is the
treatment mean for plants exposed to volatiles and x
C
is the
control mean. To incorporate the fact that uncertainty in
effect size varies across experiments, we calculated the sample
variance for the log response ratio after Hedges et al. (1999).
There are several potential sources of variation in effect size
in the data set: sampling error, quantified using the sample
variance; variation among species, some of which were used
in multiple experiments (‘Species’); variation among studies,
some of which included experiments on multiple species
(‘Study’); variation due to phylogenetic relatedness, which was
approximated using taxonomic family as a predictor (‘Fam-
ily’); variation due to the predictors of interest which we
described below; and variation due to other unknown causes
(‘Other’). Sampling error was accounted for by specifying a
priori sample variances for each experiment, which effectively
weighed effect sizes by their variance. Other between-experi-
ment heterogeneity was accounted for with an experiment-
level random effect and effects of Species, Family and Study
were all accounted for using additional random effects. Thus,
our meta-analysis was structured as a multilevel random
effects model (Gelman & Hill 2006) that accounted for multi-
ple sources of random variation while also testing for hypoth-
esised predictors of effect size. Preliminary analyses indicated
that Species and Family did not explain significant variation
in effect size, and these were excluded from further analyses
in order to reduce the number of parameters in the model.
We were interested in exploring how multiple potential pre-
dictors may affect the responses of plants to volatile cues: lab-
oratory vs. field experiments, damage type, plant growth
form, habitat characteristics, conspecific vs. heterospecific
cues, and the type of response that was measured. However,
this resulted in a large number of potential predictors relative
to the number of experiments in the data set, and several of
these predictors were partially collinear with each other. For
example, most of the studies that used artificial damage to
induce the emission of cues (32 of 35 observations) were
conducted in the field, while most of the studies that used
insect damage were conducted in the laboratory (nine of 13
observations). Consistent with these issues of predictor collin-
earity, a model that included all predictors found that no sin-
gle predictor explained significant variation in effect size, once
the other predictors had been accounted for (see Table S1).
Therefore, we chose to test each of the predictors using ‘uni-
variate’ models in which only the focal predictor was
included, along with the sampling error and random effects
terms described above (Other and Study). Thus, a key caveat
of our results is that the effects of these partially collinear
predictors cannot be fully distinguished by our results. None-
theless, we can describe which factors were associated with
variation in effect sizes, and our results suggest future experi-
ments that can better disentangle these patterns. In these
univariate models, we excluded those studies for which we
could not calculate a log response ratio. Studies with continu-
ous independent factors, studies without replication, and stud-
ies that did not specify the type of damage to plants were all
excluded from these further analyses.
The models were fit using the Bayesian mixed models pack-
age MCMC glmm (Hadfield 2010), in R version 2.15.2 (R
Development Core Team 2012). MCMC chains were sampled
10
6
times, saving every 100th sample to eliminate autocorrela-
tion. For binary predictors (e.g. laboratory vs. field), we
report 95% highest posterior density (HPD) credible intervals
for the difference in effect size between the two categories.
For categorical predictors with more than two levels, we
quantified the variation explained by this factor using the
standard deviation across factor levels (Gelman & Hill 2006),
and we reported the 95% HPD credible interval for this stan-
dard deviation.
RESULTS AND DISCUSSION
Volatile communication between plants is common
Of the 48 observations in this data set, 39 of them showed
evidence for volatile-induced plant resistance to herbivory,
eight showed evidence for induced susceptibility, and one
study showed an effect size of zero. Overall, our MCMC
analysis indicated significant evidence for volatile-induced
plant resistance to herbivory (Fig. 1, posterior mean log
response ratio effect size =0.31, 95% CI =0.44 to 0.18,
pMCMC <0.001). An analysis that controlled for non-inde-
pendence of multiple observations described in a single
citation found the same result (posterior mean log response
ratio effect size =0.44, 95% CI =0.63 to 0.25,
pMCMC <0.001). Plant species and plant family were also
included as random factors in additional models but we did
not find that some plant taxa were more likely to respond to
volatile cues of herbivory than others.
Cumulative meta-analysis allows researchers to determine
whether the mean effect size changes over time (Leimu &
Koricheva 2004). For example, a cumulative meta-analysis of
studies of induced plant responses to herbivory revealed that
early studies reported large effect sizes but as the paradigm of
induced resistance became better accepted by the field, mean
effect size decreased and more studies reported finding no sig-
nificant differences caused by induction (Nykanen & Koriche-
va 2004). A historical view of the data in this meta-analysis
showed that evidence for a general pattern of volatile-induced
plant resistance to herbivory accumulated gradually in the lit-
erature from 1983 to the present (Fig. 1). Observations from
several early studies were not properly replicated but were
influential and controversial (Rhoades 1983; Baldwin &
Schultz 1983; Myers & Williams 1984). While some early
studies suggested induced susceptibility (Rhoades 1983), the
©2013 John Wiley & Sons Ltd/CNRS
46 R. Karban, L. H. Yang and K. F. Edwards Letter
RE Model
−4 −3 −2 −1 0 1 2
Effect size (log response ratio)
Bruin et al. 1992; Gossypium hirstutum
Hildebrand et al. 1993; Solanum lycopersicum
Hildebrand et al. 1993; Nicotiana tabacum
Dolch and Tscharntke 2000; Alnus glutinosa
Dolch and Tscharntke 2000; Alnus glutinosa
Tscharntke et al. 2001; Alnus glutinosa
Tscharntke et al. 2001; Alnus glutinosa
Karban et al. 2003; Nicotiana attenuata
Karban et al. 2003; Nicotiana attenuata
Karban et al. 2003; Nicotiana attenuata
Engelberth et al. 2004; Zea mays
Engelberth et al. 2004; Zea mays
Karban et al. 2004; Lomatium dissectum
Karban et al. 2004; Lupinus polyphyllus
Karban et al. 2004; Valeriana californica
Kessler et al. 2006; Nicotiana attenuata
Heil and Silva Bueno 2007; Phaseolus lunatus
Heil and Silva Bueno 2007; Phaseolus lunatus
Ton et al. 2007; Zea mays
Karban 2007; Wyethia mollis
Frost et al. 2008; Populus deltoides x nigra
Shiojiri and Karban 2008a; Artemisia tr identata
Shiojiri and Karban 2008b; Artemisia cana
Shiojiri and Karban 2008b; Artemisia douglasiana
Rodriguez−Saona et al. 2009; Vaccinium corymbosum
Peng et al. 2010 ; Brassica oleracea
Peng et al. 2010 ; Brassica oleracea
Ramadan et al. 2011; Zea mays
Pearse et al. 2012 ; Achyrachaena mollis
Pearse et al. 2012 ; Achyrachaena mollis
Pearse et al. 2012 ; Lupinus nanus
Pearse et al. 2012 ; Lupinus nanus
Pearse et al. 2012 ; Sinapis arvensis
Pearse et al. 2012 ; Sinapis arvensis
Pearse et al. 2013 ; Salix exigua
Pearse et al. 2013 ; Salix lemmonii
Savchenko el al. 2013; Arabidopsis thaliana
Ishizaki, Shiojiri, Karban unpublished; Wyethia mollis
Karban et al unpublished; Artemisia tridentata
Shiojiri unpublished; Solidago canadensis
Shiojiri unpublished; Cardiocrinum cordatum
Shiojiri unpublished; Pachysandra ter minalis
Shiojiri unpublished; Anthriscus aemula
Shiojiri unpublished; Viola odorata
Shiojiri unpublished; Taraxacum officinale
Shiojiri unpublished; Lathyrus japonica
Shiojiri unpublished; Rosa rugosa
Shiojiri unpublished; Artemisia indica
Myers & Williams 1984; Alnus rubra
Williams & Myers 1984; Alnus rubra
Williams & Myers 1984; Alnus rubra
Baldwin and Schultz 1983; Populus euroamericana
Baldwin and Schultz 1983; Acer saccharum
Rhoades 1983; Salix sitchensis
Rhoades 1983; Salix sitchensis
Rhoades 1983; Salix sitchensis
Figure 1 Log response ratios for studies of volatile communication affecting resistance. Negative effect sizes (log response ratios) indicate induced
resistance, whereas positive values indicate induced susceptibility. Horizontal error bars indicate 95% credible intervals; studies lacking error bars were not
replicated and were not included in the meta-analysis although they were influential in the development of the field. References for these studies are listed
in the Supporting Information.
©2013 John Wiley & Sons Ltd/CNRS
Letter Plant communication and herbivory 47
–2 −1 0 1
Cummulative effect size (log response ratio)
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
Bruin et al. 1992; Gossypium hirstutum
Hildebrand et al. 1993; Solanum lycopersicum
Hildebrand et al. 1993; Nicotiana tabacum
Dolch and Tscharntke 2000; Alnus glutinosa
Dolch and Tscharntke 2000; Alnus glutinosa
Tscharntke et al. 2001; Alnus glutinosa
Tscharntke et al. 2001; Alnus glutinosa
Karban et al. 2003; Nicotiana attenuata
Karban et al. 2003; Nicotiana attenuata
Karban et al. 2003; Nicotiana attenuata
Engelberth et al. 2004; Zea mays
Engelberth et al. 2004; Zea mays
Karban et al. 2004; Lomatium dissectum
Karban et al. 2004; Lupinus polyphyllus
Karban et al. 2004; Valeriana californica
Kessler et al. 2006; Nicotiana attenuata
Heil and Silva Bueno 2007; Phaseolus lunatus
Heil and Silva Bueno 2007; Phaseolus lunatus
Ton et al. 2007; Zea mays
Karban 2007; Wyethia mollis
Frost et al. 2008; Populus deltoides x nigra
Shiojiri and Karban 2008a; Artemisia tridentata
Shiojiri and Karban 2008b; Artemisia cana
Shiojiri and Karban 2008b; Artemisia douglasiana
Rodriguez−Saona et al. 2009; Vaccinium corymbosum
Peng et al. 2010; Brassica oleracea
Peng et al. 2010; Brassica oleracea
Ramadan et al. 2011; Zea mays
Pearse et al. 2012; Achyrachaena mollis
Pearse et al. 2012; Achyrachaena mollis
Pearse et al. 2012; Lupinus nanus
Pearse et al. 2012; Lupinus nanus
Pearse et al. 2012; Sinapis arvensis
Pearse et al. 2012; Sinapis arvensis
Pearse et al. 2013; Salix exigua
Pearse et al. 2013; Salix lemmonii
Savchenko el al. 2013; Arabidopsis thaliana
Ishizaki, Shiojiri, Karban unpublished; Wyethia mollis
Karban et al unpublished; Artemisia tridentata
Shiojiri unpublished; Solidago canadensis
Shiojiri unpublished; Cardiocrinum cordatum
Shiojiri unpublished; Pachysandra terminalis
Shiojiri unpublished; Anthriscus aemula
Shiojiri unpublished; Viola odorata
Shiojiri unpublished; Taraxacum officinale
Shiojiri unpublished; Lathyrus japonica
Shiojiri unpublished; Rosa rugosa
Shiojiri unpublished; Artemisia indica
Figure 2 Cumulative effect size (log response ratios) for all studies that adds cumulative effect sizes chronologically. Error bars show 95% credible intervals
for studies with replication. Negative effect sizes with error bars that do not include 0 indicate significant effects of induced resistance.
©2013 John Wiley & Sons Ltd/CNRS
48 R. Karban, L. H. Yang and K. F. Edwards Letter
overall weight of subsequent studies showed a significant gen-
eral pattern of volatile-induced plant resistance (Fig. 2). Early
critics of this field correctly pointed out that the first studies
were pseudo-replicated (Fowler & Lawton 1985). They specu-
lated that the evidence for communication affecting induced
resistance could have been caused by lack of true replication.
However, a meta-analysis that included only studies with true
replication yielded stronger support for the overall pattern of
volatile communication producing induced resistance than one
that also included the early pseudo-replicated studies (Fig. 1,
analyses not presented).
These results (Fig. 1) indicate that plant communication
cannot be regarded as an aberrant phenomenon that is
observed only occasionally under unusual circumstances.
When ecologists have looked for this phenomenon, they have
found evidence supporting its existence. Some of these
researchers started their experiments expecting to find such
evidence; others were extremely sceptical. This meta-analysis
includes far more studies than previous reviews. However,
communication that affects herbivory is not ubiquitous. Some
studies failed to find it and others found that exposure to
volatile cues from damaged neighbours made plants more
susceptible to their herbivores, rather than more resistant
(positive effects in Fig. 1). Induced susceptibility may result
from ‘inappropriate’ plant responses when tradeoffs exist in
resistance to several different challenges or from herbivores
successfully manipulating plants responses (e.g. gall makers).
Further examination of these examples should provide valu-
able insights into control of plant signalling.
What response variables were measured?
It is useful to clarify exactly what this significant result indi-
cates. Most of the studies included in the meta-analysis mea-
sured plant damage as a response to volatile cues. This
response variable is both relatively easily to measure and
includes a variety of steps –herbivores must locate the host
plant, find it acceptable, choose to consume it, and so on. In
only a few of the studies included in the meta-analysis are
actual plant characteristics known that ultimately influence
plant resistance and even in the best studied cases, our
knowledge of the mechanisms of resistance is incomplete. For
example, tomato plants that were exposed to cues from exper-
imentally wounded sagebrush neighbours increased their
accumulations of proteinase inhibitors, chemicals known to
interfere with herbivore digestion (Farmer & Ryan 1990).
However, accumulation of proteinase inhibitors is only one
part of a much more complex induced response, and the inter-
nal chemical environment of the tomato leaf can largely inac-
tivate proteinase inhibitors (Duffey & Felton 1989). For many
of the plants included in this meta-analysis, we lack evidence
implicating any particular defensive mechanism. In summary,
we know that plants receiving volatile cues experience reduced
levels of damage from herbivores, but we do not understand
the mechanisms leading to this enhanced resistance.
Most (83%) of the studies included in the meta-analysis
reported that herbivory was reduced for plants that were
exposed to volatile cues. In some cases, chemicals implicated
in plant defence have been measured, but rarely have the
researchers evaluated alternative hypotheses. For example,
volatiles emitted by experimentally damaged plants can
directly repel herbivores rather than, or in addition to, causing
changes in resistance levels of neighbours (Bernasconi et al.
1998; De Moraes et al. 2001). In contrast, the volatiles emit-
ted by experimentally clipped sagebrush reduced damage
experienced by neighbours but had no direct repellent effects
on grasshoppers, the herbivores responsible for a lot of this
damage (Karban & Baxter 2001). However, direct repellence
has not been examined in most experiments as an alternative
mechanism.
What conditions favour volatile communication between plants?
In addition to asking whether this phenomenon occurs, we
were interested in understanding the conditions that may have
favoured plants that respond to volatiles cues. In order to
address hypotheses associating resistance mediated by volatiles
with various conditions, we conducted univariate analyses
within the meta-analysis.
We predicted that if variation due to other sources makes it
difficult to detect relatively small effects due to communica-
tion, studies done under uniform environmental conditions or
those involving genetically uniform plants may be more likely
to show significant effects. This line of reasoning led to the
prediction that studies conducted in the laboratory will show
stronger effects of communication than those conducted in
the field. Similarly, we predicted that studies of genetically
more homogeneous agricultural crops will show stronger
effects of communication than those conducted on genetically
diverse natural plants. Laboratory studies generally tended to
report stronger evidence of volatile-induced plant resistance to
herbivory than field studies did (Fig. 3, posterior mean
effect =0.37, 95% CI =0.73 to 0.01, pMCMC =0.05).
Interpretation of this result may be confounded because most
of the studies in the laboratory used real insects to damage
plants (9 of 13), while most of studies in the field used artifi-
cial damage (32 of 35). Studies using actual insects to damage
plants produced stronger effects than those that used artificial
damage (Fig. 3, posterior mean effect =0.27, 95% CI 0.56
to 0.03, pMCMC =0.07). Similarly, agricultural systems
showed a non-significant tendency towards stronger induced
resistance than natural systems (Fig. 3, posterior mean
effect =0.25, 95% CI =0.66 to 0.13, pMCMC =0.21).
Systemic induced resistance among tissues of individuals
can be strongly constrained by the vascular architecture
(‘plumbing’) of the plant (Viswanathan & Thaler 2004; Orians
2005). Plant species vary considerably in the extent and
importance of these vascular constraints (sectoriality). Woody
species that are more constrained have greater xylem vessel
size and lower vessel density than more integrated species
(Zanne et al. 2006). Species that had higher tolerance to
drought and wind were more constrained while those that had
higher tolerance to flooding and shade were less constrained
by vascular architecture. As a result, woody species from arid
environments were predicted to be less able to rely on vascu-
lar signalling and more likely to require volatile communica-
tion to co-ordinate systemic responses (Waisel et al. 1972;
Zanne et al. 2006). We found little support for these predic-
©2013 John Wiley & Sons Ltd/CNRS
Letter Plant communication and herbivory 49
tions based on habitat or plant growth form. Plants from dry
habitats tended to show more induced susceptibility than
those from more mesic habitats (Fig. 3, posterior mean
effect =0.22, 95% CI =0.13 to 0.59, pMCMC =0.22).
Although this effect was not significant, it was in the opposite
direction from our prediction about the distribution of vola-
tile communication. We detected no trend in the likelihood of
woody plants showing volatile-induced resistance compared to
non-woody plants (posterior mean effect =0.10, 95%
CI =0.33 to 0.44, pMCMC =0.41). Similarly, perennial
plants were no more likely to show evidence of volatile-
induced resistance than annuals (posterior mean effect =0.09,
95% CI =0.31 to 0.49, pMCMC =0.65).
There have been some suggestions that communication
between plants may have evolved to allow plants to coordi-
nate their own systemic responses and that the signalling
system allowing for within-plant communication has been
co-opted for between-plant signalling (Karban et al. 2006;
Frost et al. 2007; Heil & Adame-Alvarez 2010). If this has
been the case, we predicted that volatile cues would be more
effective between conspecifics than between heterospecific
plants because conspecifics are more similar genetically.
However, we found no evidence that effects of volatile signal-
ling were stronger between conspecifics than between
heterospecifics (posterior mean effect =0.04, 95%
CI =0.44 to 0.33, pMCMC =0.83). One possible explana-
tion for this surprising result is that some of the volatiles that
plants use to communicate are highly conserved and many
species respond to the same cues.
As mentioned above, this meta-analysis includes studies that
measure effects on several kinds of response variables: perfor-
mance of insect herbivores as well as plant responses, particu-
larly the percentage of leaves that have been damaged. The
effect of volatile cues on induced resistance was not signifi-
cantly different when plant measurements or insect measure-
ments were used to assess induced responses (posterior mean
effect =0.23, 95% CI =0.18 to 0.60, pMCMC =0.24). In
fact, considering the great variety of response variables that
were included in this analysis, it is surprising that any clear
signal was visible through all of the noise.
Is there publication bias?
Researchers conducting meta-analyses often attempt to deter-
mine whether the results that are reported in the literature
present an unbiased sample of the population of results or
whether investigators chose to publish only those results that
were significant or supported their preconceived hypotheses
Figure 3 Univariate tests of various factors that influence effect sizes (log response ratios) of communication. Negative values indicate induced resistance
associated with the first of the paired factors, positive values indicate induced susceptibility, and error bars are 95% highest posterior density (HPD)
credible intervals.
Observed outcome
Standard error
1.1 0.8 0.5 0.3 0.0
−3−2−10 1 2
Figure 4 A funnel plot showing the distribution of effect sizes and
standard errors.
©2013 John Wiley & Sons Ltd/CNRS
50 R. Karban, L. H. Yang and K. F. Edwards Letter
about nature. One way to evaluate this ‘file-drawer problem’
and other reporting biases is to use funnel plots that compare
effect size and SE (Palmer 1999). An unbiased set of observa-
tions should be symmetrical around the true effect size, which
was the case for the data set reported here (Fig. 4). However,
the absence of points in the lower right of the funnel plot sug-
gests the possibility that studies with small sample sizes and
large SE that failed to find induced resistance or found
induced susceptibility may have been under-reported.
CONCLUSIONS
Many studies involving diverse plants reported evidence of
volatile communication resulting in increased resistance to
herbivore attack, indicating that this is a widespread natural
phenomenon. Unlike early studies of this phenomenon, the
studies considered in this review were well-replicated with
independent sampling units.
Many of these studies did not identify mechanisms involved,
even whether a plant response was responsible for the effects.
Alternative hypotheses, such as direct repellency of herbivores
by volatiles, were often not ruled out and would be worth
considering in future studies. Determining the plant responses
involved, particularly the volatile cues that were responsible
will be well worth future effort. As expected, conditions that
minimised background variation, particularly laboratory stud-
ies and studies of genetically homogeneous crop species, were
more likely to detect significant effects of volatile cues on
induced resistance. Future studies are needed to separate
effects due to experimental conditions (laboratory vs. field)
from those caused by different inducing agents (herbivores vs.
artificial damage). Surprisingly, woody plants from arid habi-
tats were no more likely to show evidence of volatile commu-
nication than plants of other growth forms or habitats.
Negative results are difficult to interpret and more studies are
required to evaluate these tentative conclusions. If plants com-
monly use volatile cues to regulate their defences, agricultural-
ists may be able to manipulate plant resistance to pests using
these cues. Future studies should explore the possibility of
using volatile cues in production agriculture as a means of
regulating defences of crops.
Volatile communication between plants is an interesting eco-
logical finding because it suggests that plants share information
with other nearby individuals. Hence, plants are less isolated
and independent than previously assumed. Many plants per-
ceive volatile cues in their environments and respond to those
cues by changing their defences against herbivores.
ACKNOWLEDGEMENTS
We thank Kaori Shiojiri and Satomi Ishizaki for sharing
unpublished results with us and Mikaela Huntzinger, Ian Pe-
arse and anonymous referees for improving the manuscript.
AUTHORSHIP
RK organised the study and wrote the manuscript, LHY and
KFE conducted statistical analyses, prepared the figures and
edited the manuscript.
REFERENCES
Allison, J.D. & Hare, J.D. (2009). Learned and naive natural enemy
responses and the interpretation of volatile organic compounds as cues
or signals. New Phytol., 184, 768–782.
Arimura, G., Ozawa, R., Shimoda, T., Nishioka, T., Boland, W. &
Takabayashi, J. (2000). Herbivory-induced volatiles elicit defence genes
in lima bean leaves. Nature, 406, 512–513.
Baldwin, I.T. & Schultz, J.C. (1983). Rapid changes in tree leaf chemistry
induced by damage: evidence for communication between plants.
Science, 221, 277–279.
Ballare, C.L. (1999). Keeping up with the neighbours: phytochrome
sensing and other signalling mechanisms. Trends Plant Sci.,4,97–102.
Behnke, K., Ehlting, B., Teuber, M., Louis, S., Hansch, R., Polle, A. et al.
(2007). Transgenic, non-isoprene emitting poplars don’t like it hot. Plant
J., 51, 485–499.
Bernasconi, M.L., Turlings, T.C.J., Ambrosetti, L., Bassetti, P. & Dorn,
S. (1998). Herbivore-induced emissions of maize volatiles repel the corn
leaf aphid, Rhopalosiphum maidis. Entomol. Exp. Appl., 87, 133–142.
Birkett, M.A., Campbell, C.A.M., Chamberlin, K., Guereri, E., Hick, A.J.,
Martin, J.L. et al. (2000). New roles for cis-jasmone as an insect
semiochemical and in plant defense. Proc. Natl Acad. Sci., 97, 9329–9334.
Bruin, J., Dicke, M. & Sabelis, M.W. (1992). Plants are better protected
against spider mites after exposure to volatiles from infested
conspecifics. Experientia, 48, 525–529.
De Moraes, C.M., Mescher, M.C. & Tumlinson, J.H. (2001). Caterpillar-
induced nocturnal plant volatiles repel conspecific females. Nature, 410,
577–580.
Dicke, M. & Bruin, J. (2001). Chemical information transfer between
plants: back to the future. Biochem. Syst. Ecol., 29, 981–994.
Dicke, M. & Sabelis, M.W. (1988). How plants obtain predatory mites as
bodyguards. Neth. J. Zool., 38, 148–165.
Dolch, R. & Tscharntke, T. (2000). Defoliation of alders (Alnus glutinosa)
affects herbivory by leaf beetles on undamaged neighbors. Oecologia,
125, 504–511.
Duffey, S.S. & Felton, G.W. (1989). Plant enzymes in resistance to
insects. In: Biocatalysis in Agricultural Biotechnology. (eds Whitaker,
J.R., Sonnet, P.E.). American Chemical Society, Toronto, pp. 289–313.
Farmer, E.E. (2001). Surface-to-air signals. Nature, 411, 854–856.
Farmer, E.E. & Ryan, C.A. (1990). Interplant communication: airborne
methyl jasmonate induces synthesis of proteinase inhibitors. Proc. Natl
Acad. Sci., 87, 7713–7716.
Fowler, S.V. & Lawton, J.H. (1985). Rapidly induced defenses and
talking trees - the devil’s advocate position. Am. Nat., 126, 181–195.
Frost, C.J., Appel, H.M., Carlson, J.E., De Moraes, C.M., Mescher,
M.C. & Schultz, J.C. (2007). Within-plant signalling via volatiles
overcomes vascular constraints on systemic signalling and primes
responses against herbivores. Ecol. Lett., 10, 490–498.
Gelman, A. & Hill, J. (2006). Data Analysis Using Regression and
Multilevel/Hierarchical Models. Cambridge University Press,
Cambridge, UK.
Hadfield, J.D. (2010). MCMC methods for multi-response generalized
linear mixed models: the MCMCglmm R package. Journal of Statistical
Software 33, 1–22.
Hamilton, W.D. (1964). The genetical evolution of social behaviour.
J. Theor. Biol.,7,1–52.
Hedges, L.V., Gurevitch, J. & Curtis, P.S. (1999). The meta-analysis of
response ratios in experimental ecology. Ecology, 80, 1150–1156.
Heil, M. & Adame-Alvarez, R.M. (2010). Short signalling distances make
plant communication a soliloquy. Biol. Lett., 6, 843–845.
Heil, M. & Karban, R. (2010). Explaining evolution of plant
communication by airborne signals. Trends Ecol. Evol., 25, 137–144.
Heil, M. & Silva Bueno, J.C. (2007). Within-plant signaling by volatiles
leads to induction and priming of an indirect plant defense in nature.
Proc. Natl Acad. Sci., 104, 5467–5472.
Inderjit, von Dahl, C.C. & Baldwin, I.T. (2009). Use of silenced plants in
allelopathy bioassays: a novel approach. Planta, 229, 569–575.
©2013 John Wiley & Sons Ltd/CNRS
Letter Plant communication and herbivory 51
Karban, R. (2007). Experimental clipping of sagebrush inhibits seed
germination of neighbors. Ecol. Lett., 10, 791–797.
Karban, R. (2008). Plant behaviour and communication. Ecol. Lett., 11,
727–739.
Karban, R. & Baxter, K.J. (2001). Induced resistance in wild tobacco
with clipped sagebrush neighbors: the role of herbivore behavior. Insect
Behav., 14, 147–156.
Karban, R., Baldwin, I.T., Baxter, K.J., Laue, G. & Felton, G.W. (2000).
Communication between plants: induced resistance in wild tobacco
plants following clipping of neighboring sagebrush. Oecologia, 125, 66–
71.
Karban, R., Shiojiri, K., Huntzinger, M. & McCall, A.C. (2006).
Damage-induced resistance in sagebrush: volatiles are key to intra- and
interplant communication. Ecology, 87, 922–930.
Karban, R., Shiojiri, K., Ishizaki, S., Wetzel, W.C. & Evans, R.Y. (2013).
Kin recognition affects plant communication and defence. Proc. Biol.
Sci., 280, 20123062. DOI: 10.1098/rspb.2012.3062.
Kessler, A. & Baldwin, I.T. (2001). Defensive function of herbivore-
induced plant volatile emissions in nature. Science, 291, 2141–2144.
Kessler, A. & Heil, M. (2011). The multiple faces of indirect defenses and
their agents of natural selection. Funct. Ecol., 25, 348–357.
Leimu, R. & Koricheva, J. (2004). Cumulative meta-analysis: a new tool
for detection of temporal trends and publication bias in ecology. Proc.
Biol. Sci., 271, 1961–1966.
Loreto, F. & Velikova, V. (2001). Isoprene produced by leaves protects
the photosynthetic apparatus against ozone damage, quenches ozone
products, and reduces lipid peroxidation of cellular membranes. Plant
Physiol., 127, 1781–1787.
Myers, J.H. & Williams, K.S. (1984). Does tent caterpillar attack reduce
the food quality for red alder foliage? Oecologia, 62, 74–79.
Nykanen, H. & Koricheva, J. (2004). Damage-induced changes in woody
plants and their effects on insect herbivore performance: a meta-
analysis. Oikos, 104, 247–268.
Orians, C. (2005). Herbivores, vascular pathways, and systemic induction:
facts and artifacts. J. Chem. Ecol., 31, 2231–2242.
Palmer, A.R. (1999). Detecting publication bias in meta-analyses: a case
study of fluctuating asymmetry and sexual selection. Am. Nat., 154,
220–233.
Penuelas, J. & Llusia, J. (2004). Plant VOC emissions: making use of the
unavoidable. Trends Ecol. Evol., 19, 402–404.
Rhoades, D.F. (1983). Responses of alder and willow to attack by tent
caterpillars and webworms: evidence for phenomonal sensitivity of
willows. In: Plant Resistance to Insects, Symposium Series 208. (ed
Hedin, P.A.). American Chemical Society, Washington D. C., pp. 55–68.
Rodriguez-Saona, C.R., Rodirguez-Saona, L.E. & Frost, C.J. (2009).
Herbivore-induced volatiles in the perennial shrub, Vaccinium
corymbosum, and their role in inter-branch signaling. J. Chem. Ecol.,
35, 163–175.
Schenk, H.J. & Seabloom, E.W. (2010). Evolutionary ecology of plant
signals and toxins: a conceptual framework. In: Plant Communication
from an Ecological Perspective. (eds Baluska, F., Ninkovic, V.).
Springer-Verlag, Berlin, pp. 1–19.
Team, R.D.C. (2012). A Language and Environment for Statistical
Computing. R Foundation for Statistical Computing, Vienna, Austria.
Thaler, J.S. (1999). Jasmonate-inducible plant defences cause increased
parasitism of herbivores. Nature, 399, 686–688.
Turlings, T.C.J., Tumlinson, J.H. & Lewis, W.J. (1990). Exploitation of
herbivore-induced plant odors by host-seeking parasitic wasps. Science,
250, 1251–1253.
Viswanathan, D.V. & Thaler, J.S. (2004). Plant vascular architecture and
within-plant spatial patterns in resource quality following herbivory. J.
Chem. Ecol., 30, 531–543.
Waisel, Y., Liphschitz, N. & Kuller, Z. (1972). Patterns of water
movement in trees and shrubs. Ecology, 53, 520–523.
Zanne, A.E., Sweeney, K., Sharma, M. & Orians, C.M. (2006). Patterns
and consequences of differential vascular sectoriality in 18 temperate
tree and shrub species. Funct. Ecol., 20, 200–206.
SUPPORTING INFORMATION
Additional Supporting Information may be downloaded via
the online version of this article at Wiley Online Library
(www.ecologyletters.com).
Editor, Rebecca Irwin
Manuscript received 25 July 2013
First decision made 2 September 2013
Manuscript accepted 3 October 2013
©2013 John Wiley & Sons Ltd/CNRS
52 R. Karban, L. H. Yang and K. F. Edwards Letter