Content uploaded by Tzach Glasser
Author content
All content in this area was uploaded by Tzach Glasser on Jan 01, 2020
Content may be subject to copyright.
Volatiles and Tannins in Pistacia lentiscus and Their Role in Browsing
Behavior of Goats (Capra hircus)
Shilo Navon
1,2
&Jaime Kigel
2
&Nativ Dudai
3
&Ariela Knaanie
4
&Tzach Aharon Glasser
5
&Alona Shachter
3
&
Eugene David Ungar
1
Received: 17 October 2019 / Revised: 17 October 2019 / Accepted: 11 November 2019
#Springer Science+Business Media, LLC, part of Springer Nature 2019
Abstract
Goat herding is an important tool in the ecologically sound management of Mediterranean shrublands and woodlands, although
effective levels of woody biomass removal by the goats is neither guaranteed nor easy to predict. Preliminary observations
indicated that one reason for this may be poor understanding of plant-herbivore interactions that operate intraspecifically at the
local spatial scale. We asked, whether goats show intraspecific preferences among neighboring plants when foraging a small local
population of Pistacia lentiscus, a dominant tall shrub. First, we characterized and quantified the profile of stored and emitted
volatile organic compounds (VOCs) and the PEG-binding capacity oftannins (a proxy for protein binding capacity) in the foliage
of P. lentiscus shrubs, sampled within an area of 0.9 ha. We then tested goat preference between pairs of these shrubs that differed
in chemical composition. Almost all sampled P. lentiscus shrubs were allocated to one of two distinct VOC chemotypes: one
dominated by germacrene D and limonene (designated chemotype L) and the other by germacrene D and α-pinene (chemotype
P). In contrast, continuous moderate variability was found in the binding capacity of tannins in the foliage. Goats showed
preference for shrubs of chemotype L over those of chemotype P, and their preference was negatively correlated with the binding
capacity of tannins. Possible influences of VOCs on goat preference that may explain the observed patterns are discussed in the
light of possible context-dependent interpretation of plant VOC signals by large mammalian herbivores.
Keywords Essential oil .Foraging .Mediterranean shrubland .Local spatial scale .Terpenes .Intraspecific
Introduction
In recent decades, shrublands and woodlands in the
Mediterranean basin have experienced an extensive land-use tran-
sition. Among other processes, nature conservation legislation
together with a rapid shift from traditional to modern life-style,
resulted in a sharp decline in the removal rate of woody biomass.
These changes could not always be compensated by activities of
natural browsers such as Capreolus capreolus and Dama
mesopotamica, as their populations decreased substantially due
to past anthropogenic influences (Perevolotsky and Seligman
1998). This has profoundly influenced the composition of the
vegetation (Papachristou et al. 2005; Torrano and Valderrábano
2004), leading to a large increase in woody vegetation cover and a
concomitant reduction in herbaceous cover (Carmel and Kadmon
1999), as well as a strong decline in species diversity (Shmida
1981). Excessive brush encroachment is undesirable from the
perspectives of nature conservation, forage resource quality for
domesticated grazers and browsers, and fire hazard mitigation.
Goat herding is one means of combatting such encroachment,
and it offers the advantages of being both inexpensive and eco-
logically friendly. However, its effectiveness is neither guaranteed
nor easy to predict because of the complexity of herbivore-plant
interface (Bartolomé et al. 2000;Hesteretal.2006).
Although foraging ecology and feeding behavior of goats
in wooded rangelands, including Mediterranean landscapes,
were extensively examined (Glasser et al. 2012; Papachristou
Electronic supplementary material The online version of this article
(https://doi.org/10.1007/s10886-019-01124-x) contains supplementary
material, which is available to authorized users.
*Shilo Navon
shilo.navon@mail.huji.ac.il
1
Department of Natural Resources, Institute of Plant Sciences,
Agricultural Research Organization - The Volcani Center, P.O. Box 6,
50250 Bet Dagan, Israel
2
Robert H. Smith Faculty of Agriculture, Food and Environment,
Hebrew University of Jerusalem, 76100 Rehovot, Israel
3
Unit of Aromatic and Medicinal plants, Newe Ya’ar Research Center,
Agricultural Research Organization, 3009500 Ramat-Yishay, Israel
4
Large Numbers, Jerusalem, Israel
5
Ramat Hanadiv Nature Park, 3095202 Zichron Ya’akov, Israel
Journal of Chemical Ecology
https://doi.org/10.1007/s10886-019-01124-x
et al. 2005), herbivore-plant interactions were usually studied
only down to the species level and at the landscape scale
(sensu Lindborg et al. 2017). It is well established that al-
though goats behave as generalist browsers and consume fo-
liage of a wide range of woody species (Glasser et al. 2012;
Rogosic et al. 2006), they can be highly selective in their
species preferences (Castro and Fernández Núñez 2016;
Lussig et al. 2015). For instance, goats from the experimental
herd at the Ramat Hanadiv Nature Park, Israel, the site of the
present study, were reported to consume different plant groups
(shrubs, climbers and herbs) in different ratios. On a yearly
average, the goats ingested only 19.5% of their diet from her-
baceous species. Of the dominant large woody shrubs the
goats allocated, on an annual basis, 8.5% and 21.6% of their
total diet to Pistacia lentiscus and Phillyrea latifolia, respec-
tively, responding to their high and low protein binding capac-
ity of foliage tannins (Glasser et al. 2012). However, although
preference between species at the landscape scale is well
established, it is less clear to what extent selective preferences
in the browsing decisions of goats operate intraspecifically at
the local spatial scale (sensu Lindborg et al. 2017).
Better knowledge on the foraging behavior of goats at the
intraspecific, local-scale resolution is desirable because mech-
anisms underlying observed patterns frequently operate at
scales different from those at which the patterns are observed
(Levin 1992). Thus, intraspecific, local-scale examination of
shrubland dynamics, including herbivore selective prefer-
ences, is an essential step towards a better understanding of
observed vegetation patterns and foraging behavior (Bar
Massada et al. 2008) at both the local and landscape scales.
One of the key factors to consider is fine-scale chemical var-
iability. Many plant species show wide intraspecific variation in
nutritional quality as well as in the composition and concentra-
tion of secondary metabolites (Papachristou et al. 2005; Provenza
et al. 2003), and herbivores may sense and respond to such
variation (Andrew et al. 2007; Baraza et al. 2009;Estelletal.
2014; Markó et al. 2011). However, little is known about the
chemical variability within local populations of woody species
in Mediterranean landscapes. As many such species are spread
by birds (Izhaki and Safriel 1985), there is considerable potential
for high genetic and phenotypic variability within a small area,
with possible implications for herbivore-plant interactions.
Pistacia lentiscus, a prominent tall shrub species in
Mediterranean shrublands, was chosen to study the role of
intraspecific, small-scale chemical variability in goat herbivo-
ry. It is a drought-tolerant, thermophilic, evergreen dioecious
shrub that is widely distributed along the Mediterranean low-
lands in North Africa, the Levant, and southern Europe
(Nahum et al. 2008), and it uses an outcrossing breeding sys-
tem of wind pollination and avian seed dispersal (Izhaki and
Safriel 1985). The leathery leaves of P. lentiscus are generally
consumed more readily by goats than by sheep (Glasser et al.
2012) and cattle (Henkin et al. 2009). Its foliage is rich in
secondary metabolites belonging to several chemical classes
(Landau et al. 2014). The most prominent of these are tannins,
which form roughly 20% of foliage dry matter (DM) and are
known to be important limiting factors in its consumption
(Decandia et al. 2000), along with volatile and non-volatile
oleoresin compounds, such as mono- and sesquiterpenes as
well as di- and triterpenes (Landau et al. 2014). P. lentiscus
stores its volatiles in resin ducts, and the profile of these com-
pounds is highly variable among and within geographic re-
gions (Barra et al. 2007; Castola et al. 2000; Llorens-Molina
et al. 2015;Saidetal.2011; Zrira et al. 2003). However, the
fine-scale chemical variability within small, local populations
of this species and its possible involvement in diet selection of
mammalian herbivores has not yet been examined.
An exploratory field survey conducted at the Ramat
Hanadiv Nature Park, Israel, found that several pairs of neigh-
boring P. lentiscus shrubs showed pronounced evidence of
goat preference for one shrub of each pair (Supplemental
Fig. S1). Initial chemical analysis suggested this could be
related to differences between the chemical profiles of volatile
organic compounds (VOCs). These preliminary results led to
the more structured and thorough investigation reported here.
Field observations suggest that the process by which
browsing goats sense P. lentiscus VOCs involves smell as well
as taste; when a goat dissects a P. lentiscus leaf in the course of
bite formation, there is an immediate, substantial emission of
VOCs, originating from the severed resin ducts. This is de-
tectable as a burst of smell that would be perceptible to the
animal itself as well as to nearby conspecifics. Although quan-
tifying the relative importance of taste and smell in goat intra-
specific browsing decisions is beyond the scope of this study,
we analyzed and quantified both stored and emitted VOCs of
the shrubs in the experimental plot, as both seem to be in-
volved in this process. We also quantified the PEG-binding
capacity of foliage tannins, a proxy for the protein binding
capacity (Landau et al. 2004) and tested goat preference be-
tween shrubs that differed in chemical composition.
Plant VOCs have complex and diverse modes of action
(Baldwin 2010) that are challenging to investigate and re-
solve. The focus here on fine-scale differences between fo-
liage VOC compositions of conspecifics, may contribute to
a better understanding of the mechanisms by which plant
VOCs influence herbivory.
Materials and Methods
Chemical Variability within the Population
The Experimental Plot
The experimental plot was chosen to match our definition of
“local scale”, based on the animals’immediate zone of
JChemEcol
perception and foraging choice. Given the nature of herd an-
imals, and under some reasonable assumptions, this equals
roughly the area over which the entire herdis spread at a given
moment in time. By combining field observations and aerial
photographs, we evaluated the maximum spread of the exper-
imental herd within the shrubland to be 0.9 ha. Accordingly,
an experimental plot of this size was located in an open
Mediterranean shrubland on the southern hillside of Zichron
Ya’akov, Israel (32.560197 N; 34.935708 E), dominated by
P. lentiscus shrubs (Supplemental Fig. S2). Examination of
their leaves and twigs suggested that there had been no graz-
ing for many years. The foliage appeared healthy and showed
almost no marks of insect damage, disease or abiotic stress.
The individual shrubs were scattered across the plot, and a
diverse community of annual and perennial herbaceous spe-
cies covered much of the inter-shrub area, along with several
woody species, primarily Phillyrea latifolia,Rhamnus
lycioides,Olea europaea and Calicotome villosa.
Twenty-four individual P. lentiscus plants were chosen for
chemical profiling; they formed two-thirds of the shrub pop-
ulation in the plot. Eleven of the shrubs were male and eight
female, and there were five individuals whose sex could not
be determined because they did not flower during the study
period. The selected shrubs were marked, and their spatial
coordinates were recorded by GPS (Mobile mapper CX,
Magellan, CA, USA).
Sample collection: At dawn on April 7, 2013 small
branches, approximately 50 cm long, were clipped from the
southern sides of the selected 24 shrubs, in order to minimize
variation in leaf chemistry caused by varied exposure direc-
tion (Estell et al. 2013). All clipped branches appeared to be
healthy; immediately they were placed in a cooler-box and
transported to the Newe-Ya’ar Research Center for analysis
of stored and emitted VOCs.
Profiling of Stored VOCs
Extract Preparation and GC-MS Analysis: For each shrub, 2 g
fresh weight (FW) of leaf material, assembled from at least 3
leaves, taken from different branchlets of the sampled branch,
was used for extraction. The samples were put in 20 ml scin-
tillation vials and immersed in 10 ml of methyl tert-butyl ether
(MTBE) containing 10 ppm of isobutylbenzene (IBB) as in-
ternal standard. The vials were shaken for 24 h at room tem-
perature at 600 rpm using a Vibramax 100 (Heidolph
Instruments, Schwabach, Germany). Subsequently, 1 mL of
the extract was filtered through a small column containing
anhydrous sodium sulfate and silica gel to remove water and
high-molecular-weight polar substances, and then analyzed by
GC-MS. The rest of the sample was put in a ventilated oven at
60 °C for 48 h and then weighed in order to determinate the
DM of the extracted leaves.
Qualitative and quantitative analyses of VOCs were run on
an Agilent 6890 N gas chromatograph coupled to a 5973 N
mass spectrometer (Agilent Technologies, Palo Alto, CA,
USA), equipped with a Combi PAL autosampler (CTC
Analytic, CH-Zwingen) and an Rtx-5Sil MS capillary column
(95% dimethyl / 5% diphenyl polysiloxane, 30 m length ×
0.25 mm inner diameter × 0.25 μm film thickness). Injection
of the samples was splitless. The column temperature started
at 50 °C for 1 min, increased to 200 °C at a rate of 5 °C/min,
and then at 20 °C/min to 280 °C with a hold for 10 min.
Temperatures of the injector and transfer line were 250 °C
and 280 °C, respectively. Helium was used as the carrier gas
at a constant pressure, ranging from 8 to 14 psi (gas-flow
linear velocity 32 cm/s at 8 psi) to achieve a retention time
of 7.5 min for the internal standard on a nonpolar column. The
mass spectrometer was operated in EI mode at 70 eV with a
scan range of 41–350 m/z. Identifications of VOCs were ten-
tatively assigned by matching MS fragmentation patterns with
those of NIST and WILEY libraries and, when available, were
also confirmed by matching with mass spectra and retention
times of authentic standards (Sigma-Aldrich, St. Louis, MO,
USA). In addition, we compared measured Kovats indices
(calculated using mixture of straight-chain alkanes C7-C23)
with published KI values (Adams 2001). Previous reports on
the occurrence of the identified compounds in P. lentiscus
were also listed as supportive verification.
For direct profiling of emitted VOCs using solid phase
micro-extraction (SPME) see Appendix 1 in Online Resource 1.
PEG-Binding Capacity of Tannins
When the samples for VOC profiling were collected, addition-
al branches were clipped from the selected shrubs, and used to
determine the PEG-binding capacity of tannins in the foliage.
Branches were stored in paper bags and dried in the laboratory
for 30 days at room temperature. Leaflets were then sampled
separately from each shrub, ground and passed through a 1-
mm sieve. A 5-g sample was taken for analysis by near infra-
red reflectance spectroscopy (NIRS) using a Foss
NIRSystems 5000 reflectance spectrometer (Foss Tecator,
Hoganas, Sweden) according to Landau et al. (2004). The
WinISI II software (Infrasoft International, State College,
PA, USA) was used to apply the calibration equations devel-
oped by Landau et al. (2004) to the spectra of the ground leaf
samples.
Statistical Analysis
Principal component analysis (PCA) was used to quantify the
internal structure of the VOC variability among the shrubs.
Analyses were implemented in Python, version 3.6
(Anaconda distribution) using Sklearn library. A two-sample,
JChemEcol
unequal-variances t-test was applied to the proportions of each
VOC in the main clusters identified by PCA.
Preference Trials
Experimental Design and Animals
In order to test preference for a specific chemical composition,
choices were offered between two branches from the same
VOC-based PCA cluster as well as between branches from
different clusters. All chemically profiled shrubs of PCA main
clusters were used to construct 15 specific shrub pairs: four
pairs from one cluster (from eight different shrubs); five pairs
from the contrasting cluster (from nine different shrubs); and
six pairs from mixed clusters (from 12 different shrubs).
Because of the nature of the study, which examined a small,
local-scale population, and owing to the small size of many of
the shrubs (as seen in Supplemental Fig. S2), the number of
goats on which the shrub pairs could be tested was limited by
the amount of usable foliage. The 15 shrub pairs were tested in
randomized order on three different goats in 45 preference
trials. The goats were adult, female Damascus goats, drawn
from the resident herd at the Ramat Hanadiv Nature Park. The
three replications (goats) for each shrub pair were performed
consecutively with goat order randomized. The identity of the
left shrub (i.e. the shrub to left of the centerline as viewed
when facing the shrub pair) was randomized, with the con-
straint that the same spatial order should not be repeated for all
three replications. The experiments were approved by the
Animal Experimentation Ethics Committee of the Israeli
Agricultural Research Organization (ARO).
Experiment Protocol
The preference trials were conducted in a (5.0× 2.8)-m
2
pen
alongside which was a holding area into which the test ani-
mals for the day were taken from the general herd pen.
Drinking water was always available in both the holding and
the trial areas. Preliminary trials were conducted over several
days in order to optimize the experimental protocol and to
accustom the participating goats to the procedure. Preference
trials were carried out during 7 days between April 23 and
May 16, 2013.
On each trial day, large branches were cut at dawn from the
(usually) six shrubs used in that day’s trials. The branches
were marked and placed in the shade. Prior to each trial, one
or more smaller branches, as needed to achieve a total foliage
weight of approximately 600 g FW, were clipped from the
parent sample taken from each of the two shrubs in the trial.
This amount of foliage was sufficient to prevent appreciable
depletion before the end of the session. The branches from
each shrub were weighed and then attached with plastic cable
ties to the side of a wire-mesh fence in the trial area, oriented
with the foliage upwards (Fig. 1). The foliage bunches from
the two shrubs were placed 50 cm apart and 40–60 cm above
the ground.
For each preference trial, the test goat was led from the
holding area into the adjoining test area and positioned to face
the shrubs on the centerline between the shrubs of a pair at-
tached to a wire-mesh fence (Fig. 1). The animal was allowed
to consume freely from the foliage of either shrub for 15 min.
If it focused on one shrub uninterruptedly for 2 min without
displaying any behavior, such as turning the head, that indi-
cated awareness of the other shrub, the selection was
reinitialized by slowly leading the animal round to the starting
position and again allowing it to select freely. At the end of the
session, the goat was returned to the holding area, and the
branches were detached from the fence and weighed. Intake
was calculated according to weight change without correction
for water loss, which was found to be negligible. The sessions
were recorded with an MV900 Video Camcorder (Canon,
Tokyo, Japan) for later analysis of the feeding behavior.
Statistical Analysis
Hierarchical linear modeling (HLM) was used to analyze re-
sults of the preference tests (n= 44, because of a data-
recording error in one trial). This method is well suited to
accounting for the fact that individual pairs of shrubs were
tested on several different goats; it takes account of the corre-
lation between observations from a given pair of shrubs and
allows the partitioning of variance components among levels
(Raudenbush and Bryk 2002). There were insufficient degrees
of freedom to additionally account for the within-goat corre-
lation, i.e. the correlation among several measurements with
the same goat. The dependent variable was the right-minus-
left (R-L) intake difference, defined as the proportion of intake
Fig. 1 One of the experimental goats during a preference trial session
JChemEcol
that came from the right shrub minus the proportion that came
from the left shrub - two quantities that sum to unity.
Stored VOCs were chosen to enter the HLM as indepen-
dent variables, because under normal conditions they are also
the source of the emitted VOCs which correspond to the same
PCA clusters as found for the stored VOCs. The proportions
of the many stored VOCs included in the analysis could not
simply be added to the HLM because of a high degree of
multicollinearity. Each shrub was assigned with its first prin-
cipal component (PC1) score as a representative of its stored
VOC composition. Two independent variables were defined
in terms of the difference between right and left shrubs: the R-
L VOC difference, based on the PC1 score of the stored VOC
proportions (defined as PC1 score of right shrub minus that of
the left shrub), and the R-L tannin difference, based on bind-
ing capacity of tannins, expressed as percentage of DM (de-
fined as binding capacity of tannins of right shrub minus that
of the left shrub).
Analyses were conducted within the two-level hierarchical
framework, regardless of the significance of the shrub-pair
random effect, because even low levels of between-shrub-
pair variability might lead to biased estimates. We used the
intra-class correlation coefficient (ICC) to estimate the propor-
tion of total variance to be attributed to the correlation among
observations within a given shrub pair.
The first step of the analysis was to fit an unconditional,
two-level model —the null model —(Model 1 in Table 1), in
which the first level represented the trial serial number and the
second the shrub pair, and which provided preliminary infor-
mation regarding between- and within-shrub-pairs variance in
the R-L intake difference). The null model considers only the
shrub pair identifier (of which there were 15) as a source of
variability in R-L intake difference. To examine whether and
to what extent the R-L intake difference responded to changes
in R-L VOC difference and R-L tannin difference, we evalu-
ated a model –Model 2 –with a random intercept of the
shrub-pair and fixed effect of the VOCs and the tannins. To
this, in Model 3, we added the interaction between the Model
2 variables. In the subsequent models, in place of the interac-
tion term, we sequentially added: a random slope of the VOCs
alone (Model 4); a random slope of the tannins alone (Model
5); and a random slope of the VOCs and the tannins together
(Model 6). We used the Akaike Information Criterion (AIC) to
determine whether the inclusion of each term improved the
model fit: the lower the value the less was the error and the
better the fit (Garson 2017). We used Cohen’sf
2
(Selya et al.
2012)toobtainanindicationofeffectsizes.P-values <0.05
were considered significant. Analyses were conducted using
the MIXED procedure of SAS version 9.04 (SAS Institute,
NC, USA).
Results
Intraspecific Chemical Profiling
Composition of Stored VOCs
VOCs in P. lentiscus leaves had a total concentration in DM of
1500 ± 61 μg/g (Supplemental Table S1), including sesquiter-
pene hydrocarbons (50 ± 1.4%), monoterpene hydrocarbons
(45 ± 1.3%), and several volatiles containing hydroxyl or car-
bonyl groups (5 ± 0.2%). Of the 99 VOCs detected, 76
exceeded a threshold proportion of 0.1% of total VOCs in at
least one shrub. This narrowed to 30 VOCs when the threshold
was raised to 1.0% in at least one shrub, and they accounted for
90 ± 0.4% of all VOCs. Twenty-five of these were identified,
account for 85 ± 0.5% of all VOCs. In any single shrub, the
numbers of VOCs that exceeded the 0.1 and 1.0% thresholds
were 42 ± 1.1 and 19 ± 0.4, respectively. For a detailed compi-
lation of stored VOCs see Table 2and Supplemental Table S2.
For MS fragmentation patterns of five unidentified sesquiter-
penes, see Supplemental Material Fig. S3.
Table 1 Overall structure of the examined models
Model Description Equation
1 null model Y
ij
=β
0
+u
0j
+ϵ
ij
2 V, T; random intercept Y
ij
=β
0
+β
1
volatiles
j
+β
2
tannins
j
+u
0j
+ϵ
ij
3 V, T, V x T; random intercept Y
ij
=β
0
+β
1
volatiles
j
+β
2
tannins
j
+β
3
volatiles
j
×tannins
j
+u
0j
+ϵ
ij
4 V, T; random intercept and random slope for V Y
ij
=β
0
+β
1
volatiles
j
+β
2
tannins
j
+u
0j
+u
1j
volatiles
j
+ϵ
ij
5 V, T; random intercept and random slope for T Y
ij
=β
0
+β
1
volatiles
j
+β
2
tannins
j
+u
0j
+u
2j
tannins
j
+ϵ
ij
6 V, T; random intercept and random slope for Vand T Y
ij
=β
0
+β
1
volatiles
j
+β
2
tannins
j
+u
0j
+u
1j
volatiles
j
+u
2j
tannins
j
+ϵ
ij
In the model description, V=VOCs; T = tannins. In the model equations, Y
ij
is the outcome of goat i(i=1,…, 3) presented with shrub-pair j(j=1, …,
15); βand udenote the fixed and random effects of the model, respectively. u
0j
,u
1j
,u
2j
are, respectively, the intercept, VOC slope and tannins slope of
shrub pair j.ϵ
ij
is the random error. All random effects and errors are independent and normally distributed with zero mean and variances σ2
intercept ;
σ2
volatiles;σ2
tannins and σ2
within−shrub−pair ;respectively
JChemEcol
Variability of Stored VOCs
Principal component analysis found that all but two
P. lentiscus shrubs fell into two distinct main clusters (A,
marked in blue and B, marked in green in Supplemental Fig.
S4) highly separated by their first Principal component scores.
A single shrub (21) was distinct from both main clusters,
highly separated by its second Principal component score,
due to a high proportion of myrcene (27.3% of total VOCs;
with PC2 loading: +0.71; Supplemental Fig. S4;
Supplemental Table S2) and was excluded from further anal-
ysis. The first, second, and third principle components ex-
plained 61.5%, 17% and 9% of the variance respectively.
Differentiation between clusters A and B was due to differ-
ences in the monoterpenes fraction (Table 2; Supplemental
Fig. S4); seven of the nine monoterpenes occurred in differing
proportions (Pvalues ranged from <0.0001 to 0.04). The pro-
portion of the sesquiterpene alcohol elemol, also differed be-
tween clusters (t=2.3; P= 0.03). No differences between
clusters were found for any of the sesquiterpene hydrocar-
bons, which were highly similar in all but one of the sampled
shrubs. The exception, shrub 1, was an outlier of cluster A
(Supplemental Fig. S4), and exceptional in its sesquiterpenes
content both in terms of quantity and quality of the composi-
tion (Supplemental Table S2) Therefore, it was also excluded
from further analysis.
Table 2 Composition of VOCs stored in the leaves of P. lentiscus shrubs in the experimental plot, according to PCA clusters. Clusters A and B
comprised 12 and 10 shrubs, respectively
RT KI KI
L
Match Quality
L
Match Quality
S
P. lentiscus References Cluster A Cluster B
α-Pinene 5.669 935 939 0.94 a, b, c, d 11.3 ± 2.3 17.6 ± 5.3
Camphene 6.070 953 954 0.94 0.984 a, b, c, d 1.5 ± 0.4 1.6 ± 0.9
Sabinene 6.651 978 975 0.94 0.997 a, b, c, d 0.6 ± 1.9 4.1 ± 4.4
β-Pinene 6.783 983 979 0.94 0.996 a, b, c, d 4.4 ± 1 8.0 ± 1.9
Myrcene 7.092 996 991 0.93 a, b, c, d 1.6 ± 0.2 1.1 ± 0.4
α-Phellandrene 7.522 1013 1003 0.91 0.994 a, b, d 1.2 ± 0.8 2.1 ± 0.9
Limonene 8.192 1037 1029 0.97 0.998 a, b, c, d 23.0 ± 3.5 2.3 ± 1.6
β-Phellandrene 8.210 1038 1030 0.93 b, d 0.8 ± 2.4 7.7 ± 1.9
(Ε)-β-Ocimene 8.674 1055 1050 0.95 0.999 a, b, c 0.0 ± 0.0 0.1 ± 0.3
2-Undecanone 15.824 1306 1294 0.94 a, b 0.2 ± 0.5 0.4 ± 0.7
α-Copaene 18.182 1392 1377 0.99 a, c, d 0.2 ± 0.4 0.3 ± 0.5
β-Cubenene 18.509 1404 1387 0.97 d 0.1 ± 0.3 0.2 ± 0.5
β-Elemene 18.546 1406 1391 0.99 c, d 1.1 ± 0.4 1.1 ± 0.6
β-Ylangene 19.344 1436 1421 0.96 c, d 3.3 ± 0.6 3.2 ± 0.8
β- Caryophyllene 19.363 1437 1419 0.99 0.997 a, b, c, d 5.0 ± 2.2 4.3 ± 2.3
β-Copaene 19.635 1447 1432 0.96 c 1.9 ± 0.2 1.8 ± 0.3
unknown 1 20.198 1468 0.6 ± 0.6 0.5 ± 0.6
α-Humulene 20.312 1473 1455 0.98 0.997 a, c, d 2.3 ± 0.3 2.1 ± 0.5
unknown 2 20.440 1477 0.7 ± 0.5 0.8 ± 0.6
γ-Muurolene 20.823 1492 1480 0.96 d 1.2 ± 0.4 1.1 ± 0.6
Germacrene D 20.989 1498 1485 0.98 a, b, c, d 20.0 ± 2.8 20.4 ± 2.9
Bicyclogermacrene 21.353 1512 1500 0.96 c 0.2 ± 0.4 0.3 ± 0.5
α-Muurolene 21.441 1516 1500 0.98 a, c 1.3 ± 0.4 1.4 ± 0.2
δ-Cadinene 21.849 1532 1523 0.93 a, b, c 3.3 ± 0.6 3.2 ± 0.8
unknown 3 21.945 1536 1.9 ± 0.9 2.5 ± 0.7
unknown 4 22.081 1541 0.1 ± 0.3 0.3 ± 0.6
Elemol 22.662 1564 1550 0.91 c 0.7 ± 0.8 0.1 ± 0.3
unknown 5 23.383 1593 0.6 ± 0.6 1.0 ± 0.6
Sum 90.0 ± 1.9 89.5 ± 2.0
Only volatiles that contributed at least 1% of the total VOCs in atleast one shrub are presented. Values are proportions(%) oftotal VOC content(mean ±
std). KI: Kovats indices; KI
L
: Published Kovats indices (Adams 2001); Match Quality
L
: comparison of fragmentation patterns with those of NIST and
WILEY libraries; Match Quality
S
: comparison of fragmentation patterns with those of authentic standards; P. lentiscus References: a. Boelens and
Jimenez 1991. b. Castola et al. 2000. c. Llorens-Molina et al. 2015. d. Zrira et al. 2003
JChemEcol
In the 12 shrubs of cluster A, the two most prominent
VOCs were the monoterpene limonene (23 ± 1.0%) and the
sesquiterpene germacrene D (20 ± 1.0%). The monoterpene
α-pinene formed 11 ± 0.6% of all VOCs. In all 10 shrubs of
cluster B, the two most prominent VOCs were germacrene D
(20 ± 1.0%) and α-pinene (18 ± 1.6%). Limonene was present
in very small proportions (2.3 ± 0.3%). PCA clusters A and B
comprised the two dominant chemotypes in the local popula-
tion and, therefore, were used in the preference trials. Based
on the dominant VOCs, cluster A (dominated by germacrene
D and limonene) was designated Chemotype L and cluster B
(dominated by germacrene D and α-pinene) was designated
Chemotype P. The spatial distribution of the chemotypes in
the field is presented in Supplemental Fig. S2.Althoughspa-
tial grouping of the chemotypes seems apparent, spatial auto-
correlation analyses could not be conducted due to the sample
size.
In this full sample PCA, the second principal component
loadings were dominated and overshadowed by the myrcene
proportion of the excluded shrub 21 (Supplemental Fig. S4).
Thus, to better visualize and understand the internal chemical
structure within and between the shrubs of chemotypes L and
P, a second PCA was conducted, this time only with the 22
shrubs of these two chemotypes.
In the PCA including only shrubs of chemotypes L and
P, the first and second principal components explained
74% and 14.5% of the variance, respectively. The load-
ings of the first principal components (Fig. 2) reveal that
the distinction between the chemotypes lay mainly in the
occurrence or absence of limonene (with PC1 loading
+0.89). Its high proportion in chemotype L was replaced
in chemotype P by larger proportions of several other
monoterpenes alongside α-pinene (PC1 loading −0.26):
β-phellandrene, sabinene and β-pinene, with PC1 load-
ings of: −0.30, −0.15 and −0.15, respectively (Table 2;
Fig. 2). In both chemotypes these five monoterpenes
made up the same proportions of total VOCs (40.1%
and 39.6% for chemotypes L and P, respectively;
Supplemental Fig. S5). The second principal component
scores and loadings revealed that chemotype P splits into
two sub-clusters (Fig. 2), one comprising five shrubs (10,
16, 19, 23, 15) with relatively high proportions of α-
pinene (18–26%) and low proportions of sabinene
(<1%), and five shrubs (6, 13, 14, 17, 18) in which an
increase in the proportion of sabinene (8 ± 0.7%) was off-
set by a reduction of similar magnitude in the proportion
of α-pinene (11–16%). Accordingly, the second principal
component loadings of α-pinene (+0.72) and sabinene
(−0.47) were the most dominant PC2 loadings of all 30
volatiles. For representative chromatograms of
chemotypes L and P see Fig. 3a.
The two chemotypes differed also in total amounts of
VOCs, which were slightly higher in chemotype P than in
chemotype L: 1.68 and 1.31 mg/g DM, respectively
(P=0.013, F= 7.53) (Supplemental Table S1). No difference
was found between chemotypes in the proportions of the
shrubs sex (t= 2.13; P= 0.87; chemotype L: 45% females;
chemotype P: 38% females).
Composition of Emitted VOCs
The VOC composition in the headspace of P. lentiscus shrubs
was highly dominated by monoterpene hydrocarbons (97.7 ±
2.8%), with only a small proportion of sesquiterpene hydro-
carbons (2.1± 2.8%) and a negligible proportion of oxygen
containing volatiles (<0.1%). Fifteen compounds, 13 mono-
terpenes and 2 sesquiterpenes, exceeded a threshold propor-
tion of 1% of total headspace VOCs in at least one shrub, and
they accounted for 99.6 ± 0.4% of all headspace VOCs. For a
detailed composition of headspace VOCs see Table 3and
Supplemental Table S3.
Variability of Emitted VOCs
PCA revealed that the variability between the 24 shrubs in the
composition of emitted VOCs had the same structure as found
for stored VOCs (Supplemental Fig. S6). For detailed descrip-
tion see Supplementary Material, Appendix 2.
In the 12 shrubs of cluster A (marked in blue in
Supplemental Fig. S6), the most prominent emitted VOC
was limonene (51.6 ± 4.2%), while in the 10 shrubs of cluster
B (marked in green in Supplemental Fig. S6), a much lower
proportion of limonene (16.0 ± 4.3%) was substituted by
higher proportions of other monoterpenes, mainly, α-pinene
(33.0 ± 6.6%), β-phellandrene (14.9 ± 2.8%) and β-pinene
(8.0 ± 3.2%), together with a smaller increase in several other
monoterpenes (Supplemental Fig. S7). PCA clusters A and B
of the headspace analyses comprised the two dominant
emitted-VOC chemotypes in the local population and were
in full agreement with the chemotypes defined on the basis
of stored-VOC composition. They too are referred to as
chemotype L (headspace dominated by limonene) and P
(headspace dominated by α-pinene).
To better visualize and understand the internal chemi-
cal profile within and between chemotypes L and P in
relation to headspace VOC composition, a second PCA
was conducted, this time only with the 22 shrubs of these
two chemotypes (Fig. 4). The first and second principal
components explained 84.2% and 7.8% of the variance,
respectively. As was found earlier for the stored VOCs,
the loadings of the first principal components reveal that
the distinction between the chemotypes lay mainly in the
occurrence or absence of limonene (with PC1 loading:
+0.91); its high proportion in the headspace of chemotype
L shrubs was replaced in those of chemotype P by larger
proportions of several other monoterpenes apart from α-
JChemEcol
pinene (with PC1 loading: −0.34): β-phellandrene, γ-
terpinene and β-pinene with PC1 loadings of: −0.16,
−0.1 and −0.09, respectively (Table 3;Fig.
4; Supplemental Fig. S7). For representative SPME chro-
matograms of chemotypes L and P see Fig. 3b.
Binding Capacity of Tannins
In contrast to the composition of VOCs, the PEG binding
capacity of tannins in the foliage of the local population of
shrubs showed a continuous distribution over a range of 19–
25% (Supplemental Table S1). No differences were found
between chemotypes L and P in their tannins PEG-binding
capacity: 21.37 ± 0.41 and 22.19 ± 0.44% of DM, respectively
(t=−1.38; P= 0.18). No significant differences were found in
PEG-binding capacity of tannins according to plant sex (t=
2.3; P= 0.09; Male: 22.2% of DM; Female: 20.7% of DM).
Preference Trials
The mean duration (±SE) of the preference-trial sessions was
15:48 ± 0:23 min:sec of which the goats devoted 88.5 ± 1.3%
to feeding. Total time devoted to feeding and the total intake of
fresh-weight (FW) were linearly correlated (P<0.0001;R
2
=
50.9), and mean total FW intake normalized to 15:00 min was
calculated to be 563 ± 16 g. A correlation was found between
body weight and total FW intake (P<0.0001;R
2
=0.40).The
mean absolute difference between the proportions of time al-
located to each shrub by the respective goats was 32%, rang-
ing from 1% to 88%. Interestingly, the total intake was higher
when both shrubs offered were of chemotype L than when
both were of chemotype P. The 15-min-normalized FW in-
takes were: both L, 637 ± 29 g; both P, 552 ± 32 g (t=1.98,
P= 0.06). As a crude indication of preference, based simply
on the chemotype designation, we note that five of the six
shrub pairs that elicited the largest R-L intake differences
Fig. 2 Stored-VOC PCA scores
(upper) and loadings (lower) for
the 22 P. l e n t is c u s shrubs
comprising chemotypes L (cluster
A; scores in dark gray) and P
(cluster B; scores in light gray).
Each score is labeled by its shrub
number; only loadings with
absolute value above 0.09 in one
of the axes are labeled with their
compound name
JChemEcol
comprised contrasting chemotypes, while five of the six shrub
pairs that elicited the smallest R-L intake differences com-
prised non-contrasting chemotypes. Furthermore, in 15 out
of the 18 shrub-pair trials of contrasting chemotypes, goats
consume more from the shrub of chemotype L than from the
shrub of chemotype P (t=1.74;P= 0.0002; intake proportion
67%/33% [L/P]; n= 18); the intake difference was at least
20% in 13 of these 15 trails.
Hierarchical Linear Modeling
The fixed-effect estimates for both VOCs and tannins
were statistically significant in all but one of the models
in which these factors appeared (Table 4), which indi-
cates a degree of robustness of this result. The exception
was Model 6, which we suspect was overly complex for
the data available, as indicated by our inability to esti-
mate all parameters. By a small margin, the model with
the lowest AIC was Model 5, which was based on fixed
effects of VOCs and tannins and random slope for tan-
nins. The estimated variance of the intercept of the
shrub-pair random effect (σ2
intercept ) was not significant
(P= 0.12), nor was that of the slope of the tannins ran-
dom effect (σ2
tannins;P= 0.15). These findings indicate
that the shrub-pair random effects did not contribute sig-
nificantly to variance in the R-L intake difference. The
ICC for the between-shrubs effect (= ICC
intercept
+
ICC
tannins
) was 30%.
The results indicate that the R-L intake difference was
associated with the R-L VOC difference (β
1=
0.65, SE =
Fig. 3 Chromatograms from representative shrubs of chemotypes L and
P: (a) stored-VOCs extracted by solvent. (b) headspace VOCs obtained
using SPME. In both, a and b,the upper trace represents chemotype P; for
conditions see Materials and Methods. Compound numbers: 1: α-Pinene;
2: Camphene; 3: Sabinene; 4: β-Pinene; 5: Myrcene; 6: α-Phellandrene;
7: Isobutylbenzene (Std); 8: α-Terpinene; 9: o-Cymene; 10: Limonene;
11: β-Phellandrene; 12: (Ε)-β-Ocimene; 13: γ-Terpinene; 14:
Terpinolene; 15: α-Terpineol; 16: 2-Undecanone; 17: α-Copaene; 18:
β-Cubenene; 19: β-Elemene; 20: β-Ylangene; 21: β- Caryophyllene;
22: β-Copaene; 23: α-Humulene; 24: γ-Muurolene; 25: Germacrene D;
26: Bicyclogermacrene; 27: α-Muurolene; 28: γ-Cadinene; 29: δ-
Cadinene; 30: Elemol
JChemEcol
0.3, t(16) = 2.14, P= 0.0478) and with the R-L tannin dif-
ference (β
2
=−8.9, SE = 3.3, t(12) = 2.69, P= 0.0198).
Valu e s o f Coh e n ’sf
2
for the R-L VOC difference (0.16)
and the R-L tannin difference (0.19) were in the range
considered “moderate”, albeit at the low end of that range,
and indicated that the effect of VOCs was slightly weaker
than that of tannins.
Linear regression analyses confirmed that low-
proportion VOCs, which might be poorly represented in
the PC1 scores, were not involved in determining prefer-
ences. Total concentrations of VOCs, which differed be-
tween the chemotypes, were also regressed and found not
to be correlated with preference. Only two monoterpenes
—limonene and α-pinene —were strongly related to
preference, in terms of both proportions and concentra-
tions, the relationships being positive and negative, re-
spectively (Fig. 5a, b). For both monoterpenes the corre-
lations were slightly better for proportions (limonene: P=
0.0002; R
2
= 0.27; α-pinene: P= 0.0015; R
2
= 0.20) than
for concentrations (limonene: P=0.0003; R
2
=0.25; α-
pinene: P=0.003; R
2
= 0.17). Likewise, the relation of
tannins to preference was clearly apparent from linear
regression analysis (Fig. 5c). Correlation between shrub
sex and preference could not be analyzed, as in 53% of
the shrub pairs, at least one of the shrub sexes was not
known.
Discussion
General
The study of the intraspecific chemical variability of the
P. lentiscus population in the experimental plot, focused on
quantifying the stored and emitted VOC profiles of the shrubs.
However, P. lentiscus contains additional compounds belong-
ing to different chemical classes (Landau et al. 2014)thatmay
also be involved in herbivore preference. From these, we
quantified the variability in binding-capacity of tannins in
the foliage, which is known to be an important limiting factor
in goat foliage consumption (Decandia et al. 2000). Less vol-
atile resin components such as di- and triterpenes were beyond
the scope of this study, and their profiling and evaluation of a
possible role on herbivore preference will require complemen-
tary studies.
Despite the small area of the experimental plot (0.9 ha), the
chemical variability of VOCs was not continuous; almost all
sampled shrubs fell into one of two distinct and well-defined
chemotypes: L, with stored VOCs dominated by germacrene
D and limonene and emitted VOCs dominated by limonene;
and P, with stored VOCs dominated by germacrene D and α-
pinene and emitted VOCs dominated by α-pinene. In tests for
preference between these chemotypes, the stored VOC com-
position of each shrub was represented in the HLM statistical
Table 3 Composition of emitted
VOCs in the headspace of
P. l e n t is c u s shrubs in the
experimental plot, according to
PCA clusters
RT KI KI
L
Match
Quality
L
Match
Quality
S
P. l e n t is c u s
References
Cluster A Cluster B
α-Pinene 5.669 935 939 0.94 a, b, c, d 20.2 ± 3.3 33.0 ± 6.6
Camphene 6.07 953 954 0.94 0.984 a, b, c, d 1.0 ± 0.3 1.0 ± 0.5
Sabinene 6.651 978 975 0.94 0.997 a, b, c, d 0.6 ± 1.5 1.4 ± 1.8
β-Pinene 6.783 983 979 0.94 0.996 a, b, c, d 4.6 ± 1.0 8.0 ± 3.2
Myrcene 7.092 996 991 0.93 a, b, c, d 2.3 ± 1.0 1.3 ± 1.0
α-Phellandrene 7.522 1013 1003 0.91 0.994 a, b, d 3.4 ± 1.5 6.7 ± 2.2
α-Terpinene 7.75 1021 1017 0.94 a, b, c, d 0.8 ± 1.4 2.8 ± 2.6
o-Cymene 7.967 1029 1026 0.94 c, d 1.0 ± 0.5 3.1 ± 1.5
Limonene 8.192 1037 1029 0.97 0.998 a, b, c, d 51.6 ± 4.2 16.0 ± 4.3
β-Phellandrene 8.21 1038 1030 0.93 b, d 7.8 ± 2.3 14.9 ± 2.8
(Ε)-β-Ocimene 8.674 1055 1050 0.95 0.999 a, b, c 2.6 ± 3.9 3.3 ± 3.7
γ-Terpinene 8.931 1064 1060 0.94 a, b, c, d 1.3 ± 1.8 4.9 ± 4.0
Terpinolene 9.719 1093 1089 0.94 a, b, c, d 0.5 ± 0.3 1.3 ± 0.7
β-Caryophyllene 19.363 1437 1419 0.99 0.997 a, b, c, d 0.3 ± 0.7 0.4 ± 0.7
Germacrene D 20.989 1498 1485 0.98 a, b, c, d 1.4 ± 1.6 1.3 ± 1.9
Sum 99.5 ± 0.4 99.2 ± 0.9
Clusters A and B comprised 12 and 10 shrubs, respectively. Only volatiles that contributed at least 1% of the total
headspace VOCs of at least one shrub are presented. Values are proportions (%) of total headspaceVOCs (mean ±
std). KI: Kovats indices; KI
L
: Published Kovats indices (Adams 2001); Match Quality
L
: comparison of MS
fragmentation patterns with those of NIST and WILEY libraries; Match Quality
S
: comparison of MS fragmen-
tation patterns with those of authentic standards; P. lentiscus References: a. Boelens and Jimenez 1991. b. Castola
et al. 2000.c.Llorens-Molinaetal.2015. d. Zrira et al. 2003
JChemEcol
model by its first principal component score, which proved to
be an effective approach in that the chemotypic identities of
the shrubs were fully defined by PC1 score clusters (Fig. 2).
This enabled us to evaluate the influence of chemotype on
preference while still accounting for the underlying effects
of individual VOCs, via the PCA loadings. The HLM analy-
ses showed that the goats displayed a clear preference for
P. lentiscus shrubs of chemotype L over those of chemotype
P. The influence of protein binding capacity of tannins on goat
preference between shrub species was already well established
(Decandia et al. 2000; Papachristou et al. 2005); our HLM
analyses demonstrated that this pattern also affects the intra-
specific resolution.
The present findings indicate that goats foraging
Mediterranean shrublands sense differences in the composi-
tion of secondary metabolites of nearby P. lentiscus shrubs,
and that this perception is reflected in their foraging behavior.
It seems reasonable to expect that there might be significant
chemical variability within small local populations also of
other shrub and tree species inMediterranean landscapes, with
possible implications for mammalian herbivore foraging be-
havior and diet selection as well as the imprinted patterns of
the browsing on the plant species local populations.
The Underlying Mechanism
Secondary metabolites may defend plants against herbivory
via two distinct pathways: direct defense, in which a chemical
serves as a toxin (Makkar 2003;Utsumietal.2009; Villalba
et al. 2006); and indirect defense, in which a chemical serves
as a signaling agent in a more complex interaction involving
herbivores, plants and frequently other organisms in the local
habitat (Dicke and Baldwin 2010;Masseietal.2007;
Provenza et al. 2003). Our present study suggested that in
P. lentiscus both modes of action influence the dietary prefer-
ence of goats.
Fig. 4 The emitted-VOC PCA
scores (upper) and loadings
(lower) for the 22 P. lentiscus
shrubs comprising chemotypes L
(cluster A; scores in dark gray)
and P (cluster B; scores in light
gray). Each score is labeled by its
shrub number; only loadings with
absolute value above 0.09 in one
of the axes are labeled with their
compound name
JChemEcol
Tannins, when present in high concentration in woody spe-
cies, represent a good example of direct defense, and their
toxic potential is well established (Decandia et al. 2000;
Makkar 2003). However, the mechanism by which
P. lentiscus VOCs influence the reported preference is less
clear; VOCs could have direct toxic effects on herbivores
(Bedoya-Pérez et al. 2014; Illius and Jessop 1995;Jochetal.
2016; Villalba et al. 2006), but also could act as a signature of
the nutritional quality of the plant, via perception through taste
and smell (Bedoya-Pérez et al. 2014; Goff and Klee 2006;
Lawler et al. 1999; Massei et al. 2007; Moore et al. 2004).
The monoterpene hydrocarbons limonene and α-pinene,
which were found here to be related to goat preference (Fig.
5a, b), were previously reported to be involved in almost every
possible type of interaction that involved plant VOCs in eco-
logical systems (Langenheim 1994). However, as pointed out
by Liu et al. (2011) for insects–and by Estell et al. (2014)for
ruminant -, some studies found α-pinene to act as a herbivore
repellent, whereas others reported the opposite effect. In addi-
tion to other possible explanations for such inconsistencies
(Estell et al. 2014) we emphasize here the importance of clear-
ly and precisely distinguishing between the direct and indirect
modes of VOC action.
We argue that it is highly unlikely that the influence of
VOCs on the intraspecific preference of goats reported here
could be attributed to direct toxicity. First, the total VOC
content in the P. lentiscus population in the experimental plot
was relatively low, averaging 0.15% of DM (Supplemental
Tab le S1). Second, the simple terpene hydrocarbons that form
the volatile fraction of the oleoresin are regarded as having
relatively low toxicity against herbivores and pathogens
(Estell et al. 1998). Third, according to the calculation method
of Nagy and Tengerdy (1968), the low VOC content in
P. lentiscus could not accumulate in the goat rumen to a con-
centration that might harm the bacterial flora. Fourth, mam-
mals have effective detoxification mechanisms to neutralize
terpene toxicity (Makkar 2003). These considerations suggest
that the observed patterns could not be attributed to toxicity
effects, but rather to the VOCs operating indirectly as a sig-
naling agent in the herbivore-plant interaction.
Taste and olfactory systems are highly effective signal re-
ceivers for mammalian herbivores; they facilitate associative
learning of the nutritive quality of plants in complex and ever-
changing environments (Bedoya-Pérez 2014; Massei et al.
2007; Provenza et al. 2003; Schmitt et al. 2018; Stutz et al.
2016). With this in mind we asked, whether goat preference
for chemotype L over chemotype P of P. lentiscus indicated a
nutritive advantage; we re-sampled foliage of the L and P
chemotypes and analyzed it by NIRS for several nutritional
factors (Appendix S3 in Online Resource 1). We found no
differences between the chemotypes in digestibility or in the
contents of crude protein, minerals, NDF, ADF and ADL
Table 4 Results of the models examined (see Table 1 for model definitions), including fixed effect estimates, variance estimates of random effects,
intraclass correlation coefficients (ICC; %) and Akaike's information criterion (AIC)
Model
123456
Fixed effect estimates
β
0
(intercept) -2.8836 ns -1.1263 ns 0.9194 ns -0.5101 ns -0.7206 ns -0.4065 ns
β
1
(volatiles) 0.7574 * 0.7333 * 0.8227 * 0.647 * 0.6225 ns
β
2
(tannins) -9.9317 *** -9.7821 *** -9.1874 ** -8.9279 * -9.2046 ^
β
3
(volatiles × tannins) 0.1661 ns
Variance estimates of random effects
σ2
intercept 0 ^ 162.06 ns 141.32 ns 200.75 ns 161.71 ns 189.64 ns
σ2
volatiles 0.2578 ns 0.1521 ns
σ2
tannins 51.1354 ns 48.5942 ns
σ2
within−shrub−pair 1444.58 *** 662.15 *** 664.02 *** 572.34 *** 496.94 *** 452.51 **
ICC (%)
ICC
intercept
0 19.7 17.5 25.96 22.8 27.45
ICC
volatiles
0.03 0.02
ICC
tannins
7.2 7.03
ICC
within−shrub −pair
100 80.3 82.5 74.01 70.0 65.5
AIC 449 428.8 430.1 430.2 427.8 429.5
*p<0.05,**p< 0.01, *** p< 0.001, ^ no estimate available
Terms marked "within-shrub-pair" relate to the fact that the same shrub pair was tested on each goat; therefore, the correlation between the results for each
shrub pair could be included as a random effect
JChemEcol
(Supplemental Table S4). Moreover, the chemotypes did not
differ in their binding capacity of tannins, which constitute the
most prominent herbivore deterrent trait in P. lenticus. Thus,
we found no apparent nutritional advantage of chemotype L
over P that could account for the observed preference.
An alternative explanation, which we propose here, de-
pends on the fact that herbivore-plant interactions are never
isolated “one-on-one games”; a herbivore encounters many
plant species while foraging. The basic monoterpene hydro-
carbons are common to many plant species, and although
plants synthesize a vast array of VOCs, only a small subset
of these compounds generates the “flavor fingerprint”that
helps animals choose appropriate food sources (Goff and
Klee 2006). Thus, a possible explanation is that the goats
browsing P. lentiscus carried a “prejudice”in the form of a
preexisting, conditioned flavor aversion towards α-pinene and
its accompanying subset of monoterpenes associated with
chemotype P. This aversion might be acquired through feed-
ing experience with other plant species that have low nutritive
quality and are rich in chemotype P-related monoterpenes
(Lawler et al. 1999; Massei et al. 2007). Such species might
be Pinus halepensis and Cupressus sempervirens, which were
present in the foraging environment of the experimental goats.
Another possible source of the aversion might be inherited
preferences or maternal effects (Arviv et al. 2016; Biquand
and Biquand-Guyot 1992; Provenza et al. 2003).
In conclusion, it seems that the presently reported goat
preference was context-dependent. This hypothesis could be
formally tested by constructing artificial diets in which
chemotype P-related monoterpenes are associated with a pos-
itive nutritional reward and repeating the preference experi-
ment to compare conditioned versus unconditioned animals.
Although speculative, herbivore preference between similar
plants that seems not to relate to plant nutritional attributes,
calls our attention to a less considered aspect of the herbivore-
plant interface, which is the decoding of VOC-mediated envi-
ronmental information by the herbivore while browsing. The
multipurpose roles, both signaling and non-signaling, of plant
VOCs in ecological systems (Baldwin 2010)makesthem
valuable information carriers, but also a potential random
noise that the herbivores might interpret incorrectly as a mean-
ingful signal. This might influence food choice, even if VOC
variability were not truly correlated with any apparent plant
attributes or states. Since the development of information the-
ory by Claude Shannon in 1948, limitations in information
decoding in communication systems are well understood
and have a solid mathematical basis (Shannon 1948). It will
be no surprise if similar effects are also involved in the com-
plex herbivore-plant interface in ecological systems.
Acknowledgements The authors are grateful to the Ramat Hanadiv
Nature Park for hosting this research and partially funding the chemical
analyses. The authors thank Avi Perevolotsky, Yan Landau, Alexander
Fig. 5 The three plant constituents that were found to be linearly
correlated to preference among P. lentiscus shrubs (n= 44 tests): (a)
limonene, in positive correlation (P= 0.0002; R
2
=0.27); (b)α-pinene,
in negative correlation (P= 0.0015; R
2
=0.20); (c)tannins,innegative
correlation (P< 0.0001; R
2
= 0.34). The right-minus-left (R-L) difference
in measured components between the two shrubs is percentage of total
VOCs for limonene and α-pinene, and percentage of DM for tannins. The
R-L intake difference is the proportion (%) of intake that came from the
right shrub minus the proportion that came from the left shrub (which sum
to unity). LL: both shrubs of chemotype L; PP: both shrubs P; LP one
shrub of each chemotype
JChemEcol
Weinstein, Hillary Voet, Levana Devash, Tania Masci, Ben Spitzer-
Rimon, Uzi Ravid and Fred Provenza for their valuable help and advice.
Compliance with Ethical Standards
The preference experiments were approved by the Animal
Experimentation Ethics Committee of the Israeli Agricultural Research
Organization (ARO).
Conflict of Interest The authors declare that they have no conflict of
interest.
References
Adams RP (2001) Identification of essential oil components by gas
chromatography/quadrupole mass spectroscopy. Allured
Publishing Corporation, Carol Stream
Andrew RL, Peakall R, Wallis IR, Foley WJ (2007) Spatial distribution of
defense chemicals and markers and the maintenance of chemical
variation. Ecology 88:716–728
Arviv A, Muklada H, Kigel J, Voet H, Glasser T, Dvash L, Ungar ED,
Landau SY (2016) Targeted grazing of milk thistle (Silybum
marianum) and Syrian thistle (Notobasis syriaca) by goats: prefer-
ence following preconditioning, generational transfer, and toxicity.
Appl Anim Behav Sci 179:53–59
Baldwin IT (2010) Plant volatiles. Curr Biol 20:R392–R397
Bar MassadaA, Gabay O, Perevolotsky A, Carmel Y (2008) Quantifying
the effect of grazing and shrub-clearing on small-scale spatial pattern
of vegetation. Landsc Ecol 23:327–339
Baraza E, Hódar JA, Zamora R (2009) Consequences of plant–chemical
diversityfordomestic goat food preference in Mediterranean forests.
Acta Oecol 35:117–127
Barra A, Coroneo V, Dessi S, Cabras P, Angioni A (2007)
Characterization of the volatile constituents in the essential oil of
Pistacia lentiscus L. from different origins and its antifungal and
antioxidant activity. J Agric Food Chem 55:7093–7098
Bartolomé J, Franch J, Plaixats J, Seligman NG (2000) Grazing alone is
not enough to maintain landscape diversity in the Montseny bio-
sphere reserve. Agric Ecosyst Environ 77:267–273
Bedoya-Pérez MA, Isler I, Banks PB, McArthur C (2014) Roles of the
volatile terpene, 1, 8-cineole, in plant–herbivore interactions: a for-
aging odor cue as well as a toxin? Oecologia 174:827–837
Biquand S, Biquand-Guyot V (1992) The influence of peers, lineage and
environment on food selection of the criollo goat (Capra hircus).
Appl Anim Behav Sci 34:231–245
Boelens MH, Jimenez R (1991) Chemical composition of the essential
oils from the gum and from various parts of Pistacia lentiscus
L.(mastic gum tree). Flav Frag J 6:271–275
Carmel Y, Kadmon R (1999) Effects of grazing and topography on long-
term vegetation changes in a Mediterranean ecosystem in Israel.
Plant Ecol 145:243–254
Castola V, Bighelli A, Casanova J (2000) Intraspecific chemical variabil-
ity of the essential oil ofPistacia lentiscus L. from Corsica. Biochem
Syst Ecol 28:79–88
Castro M, Fernández Núñez E (2016) Seasonal grazing of goats and
sheep on Mediterranean mountain rangelands of Northeast
Portugal. Livestock Res Rural Dev 28:1–13
Decandia M, Sitzia M, Cabiddu A, Kababya D, Molle G (2000) The use
of polyethylene glycol to reduce the anti-nutritional effects of tan-
nins in goats fed woody species. Small Rumin Res 38:157–164
Dicke M, Baldwin IT (2010) The evolutionary context for herbivore-
induced plant volatiles: beyond the ‘cry for help’. Trends Plant Sci
15:167–175
Estell RE, Fredrickson EL, Tellez MR, Havstad KM, Shupe WL,
Anderson DM, Remmenga MD (1998) Effects of volatile com-
pounds on consumption of alfalfa pellets by sheep. J Anim Sci 76:
228–233
Estell RE, James DK, Fredrickson EL, Anderson DM (2013) Within-
plant distribution of volatile compounds on the leaf surface of
Flourensia cernua. Biochem Syst Ecol 48:144–150
Estell RE, Utsumi SA, Cibils AF, Anderson DM (2014) Is differential use
of Juniperus monosperma by small ruminants driven by terpenoid
concentration? J Chem Ecol 40:285–293
Garson GD (2017) Multilevel mixed modeling. Kindle edn. Statistical
Associates Publishers, NC
Glasser TA, Landau SY, Ungar ED, Perevolotsky A, Dvash L, Muklada
H, Walker JW (2012) Foraging selectivity of three goat breeds in a
Mediterranean shrubland. Small Rumin Res 102:7–12
Goff SA, Klee HJ (2006) Plant volatile compounds: sensory cues for
health and nutritional value? Science 311:815–819
Henkin Z, Perevolotsky A, Rosenfeld A, Brosh A, Provenza F,
Silanikove N (2009) The effect of polyethylene glycol on browsing
behaviour of beef cattle in a tanniferous shrubby Mediterranean
range. Livest Sci 126:245–251
Hester AJ, Scogings PF, Trollope WS (2006) Long-term impacts of goat
browsing on bush-clump dynamics in a semi-arid subtropical savan-
na. Plant Ecol 183:277–290
Illius AW, Jessop NS (1995) Modeling metabolic costs of allelochemical
ingestion by foraging herbivores. J Chem Ecol 21:693–719
Izhaki I, Safriel UN (1985) Why do fleshy-fruit plants of the
Mediterranean scrub intercept fall- but not spring-passage of seed-
dispersing migratory birds? Oecologia 67:40–43
Joch M, Cermak L, Hakl J, Hucko B, Duskova D, Marounek M (2016)
In vitro screening of essential oil active compounds for manipulation
of rumen fermentation and methane mitigation. Asian-Australas J
Anim Sci 29:952–959
Landau S, Dvash L, Decandia M, Cabiddu A, Shapiro F, Molle G,
Silanikove N (2004) Determination of poly (ethylene glycol)-bind-
ing to browse foliage, as an assay of tannin, by near-infrared reflec-
tance spectroscopy. J Agric Food Chem 52:638–642
Landau S, Muklada H, Markovics A, Azaizeh H (2014) Traditional uses
of Pistacia lentiscus in veterinary and human medicine. In: Yaniv Z,
Dudai N (eds) Medicinal and aromatic plants of the middle-east.
Springer, Netherlands, pp 163–180
Langenheim JH (1994) Higher plant terpenoids: a phytocentric overview
of their ecological roles. J Chem Ecol 20:1223–1280
Lawler IR, Stapley J, Foley WJ, Eschler BM (1999) Ecological example
of conditioned flavor aversion in plant–herbivore interactions: effect
of terpenes of Eucalyptus leaves on feeding by common ringtail and
brushtail possums. J Chem Ecol 25:401–415
Levin SA (1992) The problem of pattern and scale in ecology. Ecology
73:1943–1967
Lindborg R, Gordon LJ, Malinga R, Bengtsson J, Peterson G, Bommarco
R, Deutsch L, Gren A, Smith HG (2017) How spatial scale shapes
the generation and management of multiple ecosystem services.
Ecosphere 8:e01741
Liu ZL, Yu M, Li XM, Wan T, Chu SS (2011) Repellent activity of eight
essential oils of Chinese medicinal herbs to Blattella germanica L.
Rec Nat Prod 5:176–183
Llorens-Molina JA, Vacas Gonzales S, Martinez JS (2015) Essential oil
composition of leaves of Pistacia lentiscus L. growing wild in
Valencia (Spain). Nat Volatiles Essent Oils 2:17–26
Lussig G, Probo M, Hodge S, Lombardi G (2015) Plant species selection
by goats foraging on montane semi-natural grasslands and grazable
forestlands in the Italian Alps. Ital J Anim Sci 14:484–494
Makkar HPS (2003) Effect and fate of tannins in ruminant animals, ad-
aptation to tannins and strategies to overcome detrimental effects of
feeding tannin-rich feeds. Small Rumin Res 49:241–256
JChemEcol
Markó G, Novák I, Bernáth J, Altbäcker V (2011) Both gas chromatog-
raphy and an electronic nose reflect chemical polymorphism of ju-
niper shrubs browsed oravoided by sheep. J Chem Ecol 37:705–713
Massei G, Cotterill JV, Coats JC, Bryning G, Cowan DP (2007) Can
Batesian mimicry help plants to deter herbivores? Pest Manag Sci
63:559–563
Moore BD, Wallis IR, Palá-Paúl J, Brophy JJ, Willis RH, Foley WJ
(2004) Antiherbivore chemistry of Eucalyptus —cues and deter-
rents for marsupial folivores. J Chem Ecol 30:1743–1769
Nagy JG, Tengerdy RP (1968) Antibacterial action of essential oils of
Artemisia as an ecological factor, II. Appl Microbiol 16:441–444
Nahum S, Inbar M, Ne’eman G, Ben-Shlomo R (2008) Phenotypic plas-
ticity and gene diversity in Pistacia lentiscus L. along environmental
gradients in Israel. Tree Genet Genomes 4:777–785
Papachristou TG, Dziba LE, Provenza FD (2005) Foraging ecology of
goats and sheep on wooded rangelands. Small Rumin Res 59:141–
156
Perevolotsky A, Seligman NG (1998) Role of grazing in Mediterranean
rangeland ecosystems. BioScience 48:1007–1017
Provenza FD, Villalba JJ, Dziba LE, Atwood SB, Banner RE (2003)
Linking herbivore experience, varied diets and plant biochemistry
diversity. Small Rumin Res 49:257–274
Raudenbush SW, Bryk AS (2002) Hierarchical linear models: applica-
tions and data analysis methods, 2nd edn. Sage, Los Angeles
Rogosic J, Pfister JA, Provenza FD, Grbesa D (2006) Sheep and goat
preference for and nutritional value of Mediterraneanmaquis shrubs.
Small Rumin Res 64:169–179
Said SA, Fernandez C, Greff S, Torre F, Derridj A, Gauquelin T, Mevy JP
(2011) Inter-population variability of terpenoid composition in
leaves of Pistacia lentiscus L. from Algeria: a chemoecological
approach. Molecules 16:2646–2657
Schmitt MH, Shuttleworth A, Ward D, Shrader AM (2018) African ele-
phants use plant odours to make foraging decisions across multiple
spatial scales. Anim Behav 141:17–27
Selya AS, Rose JS, Dierker LC, Hedeker D, Mermelstein RJ (2012) A
practical guide to calculating Cohen’sf
2
, a measure of local effect
size, from PROC MIXED. Front Psychol 3:111
Shannon CE (1948) A mathematical theory of communication. Bell Syst
Tech J 27:379–423
Shmida A (1981) Mediterranean vegetation in California and Israel: sim-
ilarities and differences. Israel J Bot 30:105–123
Stutz RS, Banks PB, Proschogo N, McArthur C (2016)Follow your nose:
leaf odour as an important foraging cue for mammalian herbivores.
Oecologia 182:643–651
Torrano L, Valderrábano J (2004) Impact of grazing on plant communities
in forestry areas. Span J Agric Res 2:93–105
Utsumi SA, Cibils AF, Estell RE, Soto-Navarro SA, van Leeuwen D
(2009) Seasonal change in one seed juniper intake by sheep and
goats in relation to dietary protein and plant secondary metabolites.
Small Rumin Res 81:152–162
Villalba JJ, Provenza FD, Olson KC (2006) Terpenes and carbohydrate
source influence rumen fermentation, digestibility, intake and pref-
erence in sheep. J Anim Sci 84:2463–2473
Zrira S, Elamrani A, Benjilali B (2003) Chemical composition of the
essential oil of Pistacia lentiscus L. from Morocco—a seasonal var-
iation. Flav Frag J 18:475–480
JChemEcol
A preview of this full-text is provided by Springer Nature.
Content available from Journal of Chemical Ecology
This content is subject to copyright. Terms and conditions apply.
