Geographic structure in metabolome and herbivore community co-occurs with genetic structure in plant defence genes

LETTER Geographic structure in metabolome and herbivore community

co-occurs with genetic structure in plant defence genes

Carolina Bernhardsson,1,4 Kathryn

M. Robinson,2,4 Ilka N. Abreu,2,4

Stefan Jansson,2 Benedicte R.

Albrectsen2,3,* and P€ar K.

Ingvarsson1*

AbstractPlant–herbivore interactions vary across the landscape and have been hypothesised to promote local adap-

tion in plants to the prevailing herbivore regime. Herbivores that feed on European aspen (Populus tremula)

change across regional scales and selection on host defence genes may thus change at comparable scales.

We have previously observed strong population differentiation in a set of inducible defence genes in Swed-

ish P. tremula. Here, we study the geographic patterns of abundance and diversity of herbivorous insects,

the untargeted metabolome of the foliage and genetic variation in a set of wound-induced genes and show

that the geographic structure co-occurs in all three data sets. In response to this structure, we observe local

maladaptation of herbivores, with fewer herbivores on local trees than on trees originated from more dis-

tant localities. Finally, we also identify 28 significant associations between single nucleotide polymorphisms

SNPs from defence genes and a number of the herbivore traits and metabolic profiles.

KeywordsGeographic structure association, herbivore community traits, maladaptation of herbivores, phenotypic

association, plant–herbivore interaction, Populus tremula, untargeted metabolic profiling.

Ecology Letters (2013) 16: 791–798

INTRODUCTION

The geographic mosaic theory of co-evolution (Thompson 2005) pre-

dicts that plant–herbivore interactions, and thus selection pressures,

vary across a heterogenous landscape, promoting local adaption of

plants to the prevailing herbivore regime (Linhart & Grant 1996;

Hoeksema & Forde 2008; Z€ust et al. 2012). Furthermore, plant resis-

tance traits exert a strong influence on the community structure of

herbivorous insects, which has been documented under different

terms including ‘community’ ‘extended’ or ‘projected’ phenotypes

(Shuster et al. 2006; Whitham et al. 2006; Robinson et al. 2012). These

genetic effects manifest themselves both at the local (within popula-

tions) and the regional scale (between populations) and are hypothes-

ised to be most pronounced in ‘keystone’ or ‘foundation’ species

(Wimp et al. 2005; Bangert et al. 2006; Barbour et al. 2009).

Plants defend themselves against natural enemies by the use of sev-

eral constitutive and inducible mechanisms, where constitutive

defences are continuously expressed, while induced defences are trig-

gered by herbivores (Karban & Baldwin 1997). Plant defences are

considered costly and therefore we expect the trade-off between dif-

ferent defence mechanisms to mirror the frequency and magnitude of

potential enemies (Karban & Baldwin 1997; Kempel et al. 2011) and

to find intraspecific variation in defence expressions across spatial or

temporal scales (e.g. Bangert et al. 2006; Whitham et al. 2006). For

instance, consistently high herbivore pressure should favour constitu-

tive defences, whereas inducible defences are favoured if herbivore

pressure varies across time (Stamp 2003; Karban 2011).

European aspen (Populus tremula) is obligately outbreeding and

has substantial interpopulation gene flow and consequently low

genetic differentiation over both regional (Hall et al. 2007; Ma et al.

2010) and continental scales (De Carvalho et al. 2010). Although

adaptive phenotypic differentiation can be maintained under high

levels of gene flow when selection is strong enough (Yeaman &

Otto 2011), genetic differentiation at the underlying genes that con-

trol these traits is expected to be low, especially for traits con-

trolled by many genes (Le Corre & Kremer 2003). For example,

phenology traits that confer local adaptation to photoperiod show

strong evidence for adaptive population differentiation in P. tremula

(Hall et al. 2007), whereas genetic differentiation at genes that con-

trol these traits is comparable to neutral markers (Hall et al. 2007;

Ma et al. 2010). Contrary to this, genetic variation in genes that are

strongly up-regulated in response to wounding show a strong geo-

graphic structure, unlike that observed in neutral markers

(Bernhardsson & Ingvarsson 2012). Single nucleotide polymor-

phisms (SNPs) from these genes subdivide the Swedish populations

of P. tremula into three distinct clusters roughly corresponding to a

Southern, Central and Northern cluster (Bernhardsson & Ingvars-

son 2012). Across Sweden, the leaf-mining moth (Phyllocnistis laby-

rinthella) is a severe herbivore on aspen, but the severity of

infestation from this species varies considerably across regional

scales (Albrectsen et al. 2010b). Here, we hypothesise that genetic

variation in defence genes varies over comparable scales and that

the strong population differentiation we observe in defence genes

is linked to geographic variation in herbivore community structure,

1Ume�a Plant Science Centre, Department of Ecology and Environmental

Science, Ume�a University, SE-901 87, Ume�a, Sweden2Ume�a Plant Science Centre, Department of Plant Physiology, Ume�a

University, Ume�a, SE-901 87, Ume�a, Sweden3Department of Plant and Environmental Sciences, Section for Plant

Biochemistry, University of Copenhagen, Thorvaldsensvej 40, DK 1871,

Frederiksberg, Denmark

4 Present address: Department of Forest Genetics and Plant Physiology, Swedish

University of Agricultural Science, Ume�a, SE-901 83, Sweden

*Correspondence: E-mail: [email protected],

[email protected]

© 2013 John Wiley & Sons Ltd/CNRS

Ecology Letters, (2013) 16: 791–798 doi: 10.1111/ele.12114

which has also been shown to vary considerably among popula-

tions (Robinson et al. 2012).

Here, we expand previous studies (Bernhardsson & Ingvarsson

2012; Robinson et al. 2012) using common garden data on abun-

dance and diversity of arthropods from trees that we also screen for

variation in the foliage untargeted metabolome. Earlier studies have

established correlations between metabolite concentrations and adap-

tive, whole-plant phenotypes such as disease resistance (Witzell &

Martin 2008), interactions with herbivores (Jansen et al. 2009) or

between fungi and herbivores (Albrectsen et al. 2010a). In this study,

we were particularly interested in associations between the genetic

structure of plants, the community of associated herbivorous insects,

and the metabolic profiles. Finally, we extended our candidate

approach (Bernhardsson & Ingvarsson 2012) to also look for pheno-

typic associations between genetic variation in putative defence genes

and herbivore abundance, herbivore diversity and metabolic profiles.

Our ultimate goal is to understand the forces shaping natural varia-

tion in traits related to herbivore resistance in P. tremula and how

these traits vary across different spatial scales.

MATERIALS AND METHODS

Analysed genes and SNP scoring

As described in Bernhardsson & Ingvarsson (2012), SNPs were

scored in seven defence-related genes, either through cleaved ampli-

fied polymorphic sequence markers (Konieczny & Ausubel 1993) or

by direct sequencing. SNPs that were present in three or more copies

were kept for further analysis, resulting in 8, 8, 4, 8, 17, 12, 14 SNPs

for polyphenol oxidases PPO1, PPO2, PPO3, and trypsin inhibitors

TI2, TI3, TI4 and TI5 respectively (see also Bernhardsson & Ingvars-

son 2012). All SNPs were scored in 96–112 individuals from the

SwAsp collection, a set of trees collected throughout Sweden

spanning approximately 12 latitude degrees (Luquez et al. 2008).

Arthropod surveys

Arthropod field surveys were carried out in the S€avar common gar-

den (63°54′ N, 20°36′ E), where 116 unique genotypes have been

planted in at least four replicated ramets (Luquez et al. 2008).

Arthropod surveys were conducted at regular intervals throughout

the growing seasons from 2005 to 2009. Each ramet was surveyed

by systematically examining all the leaves on every branch, and

recording the number (2005–2009) and morphospecies (2008–2009)of all observed arthropods. We were primarily interested in arthro-

pods that interact directly with leaf tissue because of their sensitivity

to the underlying morphology (Peeters 2002) and genetic structure

of the host plant (Wimp et al. 2005). We classified leaf-feeding

insects into the five guilds: gallers, chewers, rollers, miners and

suckers, based on their mode of utilising host plant leaf tissue (for

details see Robinson et al. 2012).

For each of five seasons, surveys from different time points were

initially compared and the time point with the largest median value

was selected to represent that season in further analyses. Surveys from

2008 to 2009 were further categorised by dividing observations of

herbivorous insects into different guilds (defined above), based on

how herbivores utilise leaf tissue. Species known to feed exclusively

on one plant genus (Populus) were further classified as specialists while

all other species were considered generalists (Robinson et al. 2012).

Untargeted metabolite profiling

To get an unbiased search for candidate metabolites, we used an

untargeted metabolomic approach using undamaged leaves. One

undamaged fully expanded mature leaf was collected from each of

78 clones, corresponding to 63 unique genotypes, on July 4 2007

from the SwAsp common garden in S€avar and flash-frozen in liquid

nitrogen on site. For the GC/MS analyses, 10 mg of frozen sample

powder was extracted in 1 mL of a solution containing chloroform

(20%, v/v), methanol (60% v/v), water (20%, v/v) and 11 isotope-

labelled internal standards ([13C5]proline, succinic acid-D4, salicylic

acid–D6, [13C4] a-ketoglutarate, [

13C5,15N]glutamic acid, putrescine-

D4, [13C3]myristic acid, [13C6]glucose, [13C4]hexadecanoic, [13C12]

sucrose and cholesterol-D7), derivatised and analysed by GC/TOF-

MS (Leco Corp., St. Joseph, MI, USA), according to Jonsson et al.

(2005). The non-processed MS files from the metabolomic analysis

were exported from the Leco software in NetCDF format to MAT-

LAB 7.0 (Mathworks, Natick, MA, USA). Data pre-treatment such

as base-line correction, peak alignment and hierarchical multivariate

curve resolution (H-MCR) were performed using custom scripts

according to Jonsson et al. (2005). The H-MCR deconvoluted GC/

MS data allow the mass spectra of all detected compounds to be

compared with spectra in the National Institute of Standards and

Technology library and in-house databases. The resulting data

matrix was normalised using the concentrations of 11 added internal

standards (Jonsson et al. 2005). Two hundred and seventy-three

unique metabolites were extracted from the data. For genotypes that

were represented by multiple clones, least square means were calcu-

lated across all clones for all observed metabolites and average val-

ues were subsequently used to represent that specific genotype in

further analyses. Finally, to reduce the dimensionality of the metab-

olome data we used a principal component analysis to extract the

main axes of variation in the total data set, and the first seven axes

of this PCA were used in subsequent analyses (cumulative variation

explained 52.6%).

Heritabilities and genetic correlations of herbivore data

To quantify genetic variation in and the co-variation among differ-

ent herbivore communities we calculated heritabilities and genetic

correlations using either of two different measures of herbivore

community structure: (1) herbivores classified into different guilds

or (2) herbivores classified depending on whether they can be con-

sidered generalist or specialists on Populus. If different herbivores

co-occur on the same genotype we expect the genetic correlation

between these traits to be positive. Likewise, if there is some form

of avoidance between two guilds, the genetic correlation is expected

to be negative.

To estimate genetic correlations and heritabilities we used a fully

Bayesian approach to fit a multivariate linear model to the data.

The model included fixed effects for year, block and height as pre-

vious studies have indicated that height is an important variable

determining arthropod abundance on individual ramets (Robinson

et al. 2012). To estimate the complete variance-covariance matrix

for the guilds and specialists/generalists data we implemented the

model in MCMCglmm (Hadfield et al. 2010; code for fitting the

model is given in Appendix S1). From the estimated variance-

covariance matrix, we calculated broad sense (i.e. genotypic) herit-

abilities and genetic correlations between all traits using standard


792 C. Bernhardsson et al. Letter

methods (Lynch & Walsh 1998). Two MCMC chains were run for

102 500 steps, with the first 2500 steps discarded as burn-in sam-

ples. The chain was further thinned by sampling every 200 steps for

a total of 500 samples per chain. The priors of the variances and

covariances of the random effects in the model were chosen to be

uninformative and are given in Appendix S1.

Convergence of the MCMC chain was assessed using the coda

package in R (Plummer et al. 2006) and convergence diagnostics are

summarised in Appendix S1 and Table S1. In a Bayesian analysis,

the full posterior distribution of the statistics of interest is estimated

and it is therefore straightforward to calculate point estimates of

heritability and genetic correlations as the mode of the posterior

densities; the posterior mode is comparable to the maximum likeli-

hood estimate of the statistic of interest (Gelman et al. 2004). Credi-

bility intervals (Bayesian confidence intervals) were calculated from

95% highest posterior density regions, that is, the region of the pos-

terior density that contains 95% of the posterior probability mass

(Gelman et al. 2004).

Associations between genetic structure and phenotypic traits

A multivariate linear model (Zapala & Schork 2006) was used to

test for associations between the underlying genetic structure of

three distinct clusters (Bernhardsson & Ingvarsson 2012), and herbi-

vore community structure or metabolic profile. This method parti-

tions a distance matrix, describing the similarities between different

individuals to determine whether individuals show greater similari-

ties within compared to between clusters. For community structure,

we used data on the number of individuals from each of five guilds

of herbivores and used the Bray–Curtis similarity coefficient (Bray

& Curtis 1957; Whitham et al. 2006) to express similarity between

the genotypes:

BC = 2W/(A + B)

where W is the sum of the minimum abundances between samples

A and B, divided by the total abundance of species on the two trees

so that the BC index scales between 0 and 1. The BC index mea-

sures proportional similarity and can therefore be used to judge

how similar the arthropod communities are on two different indi-

viduals. For the metabolome data, we used the original metabolic

profiles consisting of 273 individual metabolites and calculated pair-

wise similarities between genotypes using the Euclidian distance:

D ¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiX

xik � xjk� �2q

where xik and xjk represent measurements of metabolite k in indi-

viduals i and j. The predictor variable used in both analyses was the

clusters identified using the SNP data from seven defence genes

(Bernhardsson & Ingvarsson 2012), the population of origin, indi-

vidual genotype and tree height. The full model was implemented

using the adonis function available in the package vegan in R

(Oksanen et al. 2012). Testing of predictors were done using stan-

dard F-tests based on sequential sums of squares but with signifi-

cance evaluated from 104 random permutations of the raw data.

Because genotyping and phenotyping was not performed on all

genotypes in the SwAsp collection, the total number of individuals

used for the analyses differ. For herbivore community structure,

data from 94 unique genotypes were used for the analyses, while

for the metabolic profiling we used data from 52 unique genotypes.

Phenotypic associations

To link SNP variation in the seven candidate genes to the nine dif-

ferent herbivore community traits (guilds data and specialists/gener-

alists data), and seven metabolome traits (first seven axes of a PCA

on the total metabolic profiles), we performed association mapping,

using single marker association tests assuming a model of additive

effects of alleles at a locus. We used the mixed-model method of

Yu et al. (2006) to control for possible spurious associations caused

by genetic structure in the sample. Although previous studies show

little population structure at neutral markers in the SwAsp collection

(Ingvarsson et al. 2008; Ma et al. 2010), we included these effects in

our association analyses.

The model we used was as follows:

y ¼ Xbþ Zuþ e

where y is a vector of observed phenotypes, b is a vector on

parameters for fixed effects (including SNP effects) and u and e are

vectors that capture (random) variation due to unmeasured poly-

genic contributions (u) or environmental effects (e). X and Z are

design matrices relating fixed and random effects to individual

observations. The variance of polygenic (genotype) effects equals V

(u) = 2KVG, where K is the matrix of kinship coefficients. Finally,

the variance of the residual effects is V(e) = RVR, where R is a

matrix with zeros for all off-diagonal elements and the reciprocal of

the number of observations along the diagonal. VG and VR measure

the genetic and residual variances for the trait. Estimates of popula-

tion structure were taken from Ingvarsson et al. (2008) and Ma et al.

(2010) and are based on data from 25 putatively neutral SSR mark-

ers. Earlier results have shown that including a kinship matrix in

the association analyses has little or no effect (Ingvarsson et al.

2008), and we therefore substituted K with the identity matrix I,

implicitly assuming that all genotypes are unrelated within popula-

tions.

For the herbivore community traits, we included fixed effects for

block and year in the mixed-model and these variables were

included together with variables controlling for population structure

into the vector of fixed effects (b). For the metabolome data, we

used the first two principal components as our dependent variables.

The association analyses of the metabolome data only included

fixed effects for block and population structure, as data were only

collected in 2007.

As the data consist of counts of individual herbivores (or units,

such as egg clusters), we used a generalised linear mixed-model with

a poisson link function to analyse the data, and analyses were imple-

mented using the glmer function from the lme4 library in R (Pinheiro

& Bates 2000). We used a rather liberal false discovery rate (FDR,

Storey & Tibshirani 2003) of 10% to control for multiple testing

across SNPs since we view these analyses as exploratory and are

more concerned with false negatives compared to false positives at

this stage. The proportion of phenotypic variation explained by the

associated SNPs was calculated using the likelihood-ratio-based R2

model by Magee (1990). This version of R2 considers the change in

likelihood between nested models with and without individual

SNPs. Sun et al. (2010) compared several models for calculating the

variation explained in mixed-model association mapping, and

showed that this version of R2 is the most suitable to use in mixed-

model settings.


Letter Patterns of geographic and genetic structure 793

RESULTS

Herbivore community structure was summarised in nine traits: spe-

cies richness and abundance of specialists and generalists, respec-

tively, and abundance of five feeding guilds: chewers, suckers,

gallers, rollers and miners (also see Robinson et al. 2012). Broad-

sense heritabilities (H2) for traits were in many cases large, suggest-

ing strong genetic control in these traits (Table 1). The highest heri-

tability (0.81) was observed for the leaf galler guild (Table 1), the

second highest heritability (0.78) was found for species richness of

specialist insects (Table 1). Genetic correlations between traits were

overall positive, ranging from 0.01 to 0.25 (Table 1, see Table S2

for full posterior distributions of genetic correlations). The highest

correlations were between abundance and species richness of spe-

cialist herbivores (0.248), and between abundance and species rich-

ness of generalist herbivores (0.211). Metabolic profiles were

conducted on a selected set of aspen genotypes identifying 273

peaks in the chromatograms. We do not attempt to discuss the indi-

vidual metabolites – many of which are yet unidentified – but rather

view them as potentially genotype-specific ‘bio-markers’ with poten-

tial importance for associated herbivore communities. These effects

may range from direct – if metabolites attract or deter certain insect

species – to indirect effects – that is, if metabolites are correlated

with the nutritional value of the leaves. The first two axes from a

principal component analysis explain 18.8 and 9.2% of the total var-

iation in the metabolome data, and with the first seven axes

accounting for 52.6% of the variation (see Table S3 for vector load-

ings on the top seven PCA axes).

Significant associations were found between the genetic structure

of defence genes (Bernhardsson & Ingvarsson 2012) and patterns

of variation in both herbivore community structure and metabolic

profiles (Table 2, Fig. 1). For both traits, height was highly signifi-

cant, verifying previous observations that tree height is an important

factor influencing herbivore community structure in P. tremula

(Robinson et al. 2012). Moreover, variation in metabolic profiles

also differed among populations, even after accounting for variation

associated with large-scale population structure observed in defence

genes, suggesting that genetic variation in these traits is structured

at multiple geographic levels. The greatest difference in both herbi-

vore community structure and metabolic profile was observed

between the Northern and South/Central clusters (Fig. 1)

(Bernhardsson & Ingvarsson 2012). The main drivers that caused

clustering of the metabolome included flavonoids, lipids and ascor-

bic acid, which were more abundant in the Southern and Central

cluster, whereas salicin-like compounds, hydroxyl-benzoic acid and

benzoic acid distinguished the Northern cluster. For guilds and

specialist abundance, the Northern cluster was less attacked com-

pared to the Southern and Central clusters (Fig. 2).

Seventy-one candidate SNPs from seven defence genes were anal-

ysed for association with the sixteen traits describing different

aspects of herbivore community structure and variation in the meta-

bolic profile. After multiple test correction, using a FDR of 10%,

we observed a total of 28 significant associations for six out of the

sixteen phenotypic traits tested. These associations corresponded to

19 unique SNPs from five of the seven genes analysed, and consist

of both synonymous and non-synonymous polymorphisms

(Table 3). The amount of variation explained by individual SNPs

ranged between 6.4 and 28%, while the total amount of variation

explained by all the associated SNPs ranged between 13 and 45%

for different traits (Table 3).

DISCUSSION

Genetic structure in aspen drives variation in arthropod

community structure

We found that variation in both numbers of species and individuals

among genotypes for both specialist and generalist arthropod herbi-

vores have large genetic components, with heritabilities ranging

from 0.12 to 0.81. Although the overall genetic correlations among

specialists and generalists were positive both for species richness

and total abundance, genetic correlations were overall low (< 0.25),

suggesting that genetic variation is largely independent for these

traits. The same general trends were also observed when herbivores

were classified into guilds based on feeding styles and utilisation of

leaf tissues. Heritability for different guilds ranged from 0.12 for

suckers to 0.82 for gallers. In general, genetic correlations between

guilds were low (< 0.2), which is comparable with the community

composition of plant-associated arthropods found in other studies

(e.g. Keith et al. 2010). In summary, these results confirm that there

is ample genetic variation influencing the community structure of

foliar herbivores in P. tremula. Both herbivore related and metabolic

traits grouped into a distinct northern cluster clearly separated from

the southern/central clusters, agreeing with the distribution of

genetic variation we previously found in induced defence genes

(Bernhardsson & Ingvarsson 2012). The genetic structure of these

defence genes differs considerably from that observed for either

neutral genetic markers (Hall et al. 2007) or photoperiod SNPs (Ma

et al. 2010), with population structure in defence genes, as well as

herbivore and metabolome traits, consisting of a northern cluster

that appears to be both genetically and phenotypically quite distinct

from a southern and central cluster (Table 2, Fig. 1, Bernhardsson

Table 1 Genetic correlation (off diagonal) between herbivore traits and the corresponding broad-sense heritability (H2, diagonal) for each herbivore trait

Richness,

specialists

Abundance,

specialists

Richness,

generalists

Abundance,

generalists Chewers Suckers Gallers Rollers Miners

Richness, specialists 0.779 0.248 0.144 0.139 0.124 0.177 0.041 0.255 0.129

Abundance, specialists 0.349 0.191 0.150 0.090 0.040 0.048 0.034 0.081

Richness, generalists 0.264 0.211 0.087 0.146 0.078 0.123 0.077

Abundance, generalists 0.407 0.182 0.255 0.043 0.078 0.058

Chewers 0.144 0.190 0.101 0.252 0.121

Suckers 0.122 0.012 0.141 0.106

Gallers 0.809 0.058 0.069

Rollers 0.164 0.011

Miners 0.618



& Ingvarsson 2012). From this we conclude that the distribution of

genetic variation in both defence genes and the associated herbivore

and metabolic traits cannot be explained by either the immigration

history of P. tremula or as a by-product of adaption to latitudinal

variation in photoperiod.

Are herbivores maladapted?

Another interesting result is the apparent maladaptation of herbi-

vores we observe in the data, with fewer herbivores on genotypes

from local trees compared to trees that originate from more distant

localities. This pattern was repeated across all herbivore guilds and

apparent for both specialist and generalist herbivores. Under the

arms-race model of co-evolution, defensive and counter-defensive

traits evolve in a never-ending escalating cycle, and herbivores are

generally expected to be locally adapted to their host population due

to their much shorter generation time (Mopper & Strauss 1997; Kaltz

& Shykoff 1998). However, several hypotheses have been proposed

to explain a pattern of local maladaptation of herbivores to their host

plants. For instance, the mosaic theory of co-evolution predicts that

temporal maladaptation may arise when co-evolving traits are tempo-

rarily mismatched in a population that exists in a complex geographic

landscape with varying demographics (Thompson 2005). Other pos-

sible explanatory factors include differences in the spatial structure of

host and enemy populations and/or long enemy generation time

(Gandon et al. 1996; Mopper & Strauss 1997). None of these hypoth-

eses can explain our results, since the same pattern of local maladap-

tation is found across all different herbivore guilds. In fact, it is most

pronounced for the abundance of specialist arthropods, whereas

under an arms-race model we would expect to see the strongest evi-

dence for local adaptation of herbivores (Fig. 2). To account for such

observations of local maladaptation, an alternative explanation has

been proposed by Kniskern & Rausher (2001) under the name ‘infor-

mation co-evolution’. This model predicts the existence of an elici-

tor–receptor informational exchange between host and enemy, where

the host plant recognises enemies by elicitors they produce and are

then able to mount a defensive response. The enemy, on the other

hand, tries to avoid being recognised by the host by altering these

elicitors. This model thus predicts that after one round of co-evolu-

tion, the local host population will be resistant to the herbivores,

while non-local host plants will be susceptible due to a lack of elicitor

recognition. This model has received some empirical support by

Parker et al. (2006) who showed in a meta-analysis of over 100 plant

species that plants are especially susceptible to novel, generalist herbi-

vores that they have not been selected to resist. Apparent herbivore

‘maladaption’ may therefore be the norm in many plant–herbivoreinteractions and especially in cases where interactions between plants

and enemies involve mismatches in recognition-based defences (Ve-

rhoeven et al. 2009).

Underlying structure may confound phenotypic associations

We observed 28 significant SNP-trait associations, which included

SNPs from five genes that associated with six of sixteen herbivore

and metabolome traits. We have statistically controlled for tree

height in our association analyses, which is an important trait

Table 2 Association between genotypic and phenotypic population structure

based on an analysis of variance, partitioning sums of squares using semimetric

and metric distance matrices (Oksanen et al. 2012)

d.f. SS MS F P

Herbivore guilds

Height 1 0.463 0.463 12.80 0.0005

Genetic cluster 2 0.274 0.137 3.80 0.0258

Population 15 0.447 0.030 0.83 0.6379

Clone 75 3.517 0.047 1.30 0.0814

Residuals 286 10.332 0.036

Metabolomics data

Height 1 1196.2 1196.2 21.37 0.0001

Genetic cluster 2 345.1 172.6 3.08 0.0105

Population 15 1965 131.0 2.34 0.0020

Clone 33 2682 81.3 1.45 0.0784

Residuals 14 783.7 56.0

−2

−1

0

1

2

Herbivore guilds

PC1

PC

2

−10 −5 0 5 10

−10

−5

0

5

10

15

Metabolites

PC

2

−4 −2 2 40

PC1

(a)

(b)

Figure 1 Visualisation of the phenotypic distance between population clusters

for (a) herbivore guilds and (b) metabolites. Each individual is colonised

according to the cluster recognition found for the defence genes (Bernhardsson

& Ingvarsson 2012), with a geographic structure categorising a Southern cluster

(black), Central cluster (grey) and Northern cluster (white).



known to influence herbivore community structure (Campos et al.

2006; Robinson et al. 2012). A potentially more serious confounding

problem is the strong population structure observed in both

defence genes and herbivore-related traits (Bernhardsson & Ingvars-

son 2012). We controlled for population structure using estimates

based on data from putatively neutral SSRs and SNPs (Hall et al.

2007; Ma et al. 2010), but as both defence genes and herbivore traits

vary across similar spatial scales we would expect to see associations

between SNPs and traits simply due to the underlying population

structure at these genes. The population structure would also induce

strong linkage disequilibrium between SNPs, something that we

indeed observe (Bernhardsson & Ingvarsson 2012). It is therefore

likely that many of these associations are not directly causal, but

instead are driven by the underlying structure that co-occurs

between defence genes and phenotypic traits. This is especially true

for PC1, the first axis of the PCA for the metabolome data, for

which 11 significant associations were found with SNPs from four

genes involved in the inducible defence response. However, correct-

ing for population stratification using the genetic structure observed

in defence genes will not yield appropriate results since the structure

is most likely adaptive and therefore corrections for this underlying

structure will increase the risk of incurring false negatives, that is,

not identifying potentially important associations.

These issues highlight some of the problems of inferring associa-

tion in structured samples and in particular when variation in adaptive

traits of interest is aligned along the same axes as genetic variation in

putative candidate genes. Our data illustrate how the same population

can lack structure in some aspects and yet be substantially structured

in others. These are factors that may need to be considered as gen-

ome-wide association studies are being pursued in many species.

2008

0

2

4

6

8

10

** *

Southern clusterCentral clusterNorthern cluster

2009

0

2

4

6

8

10

* *

*

*

*

Abundancespecialists

Abundancegeneralists

Richnessspecialists

Richnessgeneralists

2009

0

5

10

15

*

**

*

Chewers Suckers Gallers Rollers Miners

Chewers Suckers Gallers Rollers Miners



(a)

(b)

(c)

Figure 2 Mean number of herbivores (with standard errors) on Populus tremula for the three different genetic clusters for different guilds of herbivores (a) 2008, (b) 2009,

and for generalist or specialist herbivores (c) 2009. Stars indicate significant differences between the Northern and Southern/Central clusters after correction for height.



Community phenotypes can be viewed as heritable plant traits

and will likely affect genetic differentiation and ultimately even spe-

ciation in some species in the community (Evans et al. 2008). The

results of this study complement and extend earlier studies on

genetic variation in traits related to arthropod community composi-

tion in P. tremula (Albrectsen et al. 2010b; Robinson et al. 2012) and

add to a number of recent studies that have documented effects of

genetic structure in foundation species on their associated foliage

communities (Bangert et al. 2006; Barbour et al. 2009).

Natural enemies have long been thought to play important roles

in maintaining diversity both within and among species, but only

recently has empirical evidence started to appear in the literature

(e.g. Z€ust et al. 2012). In European aspen, Scandinavia was colon-

ised from two different post-glacial refugia resulting in a current-day

admixture zone in central Sweden (De Carvalho et al. 2010). While

this event has only had weak effects on genetic differentiation at

neutral markers (Hall et al. 2007), it could have profound effects on

adaptive traits by influencing patterns of genetic variation within

and among different populations (De Carvalho et al. 2010). The

strong north–south geographic structure reported here for herbi-

vore-related plant traits, defence genes, and associated arthropod

communities could therefore be a consequence of historical adapta-

tion matching the glacial refugia. This would also explain our obser-

vation of higher herbivore abundance on non-local trees (the

Southern/Central clusters) compared to local trees (Northern clus-

ter). However, to further study to what extent these results are dri-

ven by historical processes, we would need to know more also

about the current-day geographic distribution of different herbivore

species and their post-glacial colonisation history, which could

potentially be studied by assessing the genetic structure also of spe-

cific herbivores across space.

ACKNOWLEDGEMENTS

This study has been funded by grants from the Swedish Research

Council, Carl Tryggers Stiftelse for Vetenskaplig Forskning, Swedish

Foundation of Strategic Research, Formas, The Kempe Founda-

tions.

AUTHORSHIP

CB and PKI planned the association study; BA planned metabolo-

mics analyses and IA conducted them, BA initiated herbivory sur-

veys supported by SJ, and KR continued and developed them; CB

and PKI performed analyses; CB wrote the first draft of the manu-

script and PKI and BA contributed substantially to revisions; KR

and SJ commented on the text.

REFERENCES

Albrectsen, B.R., Bj€ork�en, L., Varad, A., Hagner, �A., Wedin, M., Karlsson, J.

et al. (2010a). Endophytic fungi in European aspen (Populus tremula) leaves —diversity, detection, and a suggested correlation with herbivory resistance.

Fungal Divers., 41, 17–28.Albrectsen, B.R., Witzell, J., Robinson, K.M., Wulff, S., Luquez, V.M.C., �Agren,

R. et al. (2010b). Large scale geographic clines of parasite damage to Populus

tremula. Ecography, 33, 483–493.Bangert, R.K., Allan, G.J., Turek, R.J., Wimp, G.M., Meneses, N., Martinsen,

G.D. et al. (2006). From genes to geography: a genetic similarity rule for

arthropod community structure at multiple geographic scales. Mol. Ecol., 15,

4215–4228.Barbour, R.C., O’Reilly-Wapstra, J.M., De Little, D.W., Jordan, G.J., Steane,

D.A., Humphreys, J.R. et al. (2009). A geographic mosaic of genetic variation

within a foundation tree species and its community-level consequences.

Ecology, 90, 1762–1772.

Table 3 Significant associations of defence SNPs to phenotypic traits. SNP – gene name and SNP number; Site – the SNP’s nucleotide position in the gene; Type – type

of substitution

Gallers Rollers Miners

Specialist

abundance

Generalist

richness PC1

SNP Site Type P-value* R2† P-value* R2† P-value* R2† P-value* R2† P-value* R2† P-value* R2†

PPO2.1 41 NS – – – – <0.001 0.099 <0.001 0.114 – – 0.002 0.125

PPO2.2 112 N ? D – – – – – – – – – – 0.005 0.106

PPO2.5 657 syn – – – – 0.008 0.066 – – – – – –TI2.4 450 syn – – – – – – – – – – 0.001 0.157

TI2.8 545 indel – – – – – – 0.001 0.102 <0.001 0.114 – –TI3.15 474 syn – – – – 0.01 0.064 – – – – – –TI4.4 220 syn – – – – – – – – – – <0.001 0.169

TI4.5 261 syn – – – – – – – – – – 0.005 0.123

TI4.7 335 H ? R – – 0.001 0.104 0.002 0.089 0.004 0.08 – – <0.001 0.28

TI4.8 339 syn – – – – 0.009 0.067 – – – – <0.001 0.259

TI4.10 429 syn – – – – – – – – – – <0.001 0.19

TI4.11 463 syn – – <0.001 0.074 – – – – – – – –TI5.1 258 syn – – – – 0.005 0.079 – – – – – –TI5.2 259 I ? V <0.001 0.147 – – – – – – – – – –TI5.4 270 syn – – – – – – 0.003 0.083 0.002 0.098 – –TI5.6 278 E ? G – – – – 0.002 0.093 – – – – 0.007 0.113

TI5.8 366 syn – – – – 0.005 0.075 – – – – – –TI5.12 494 L ? P – – – – – – – – – – 0.009 0.104

TI5.13 516 syn – – – – – – – – – – 0.011 0.099

Total 0.147 0.13 0.341 0.213 0.188 0.454

*Significance level after correction for multiple testing using the false discovery rate (FDR).

†Percentage of variation explained.



Bernhardsson, C. & Ingvarsson, P.K. (2012). Geographical structure and

adaptive population differentiation in herbivore defence genes in European

aspen (Populus tremula L., Salicaceae). Mol. Ecol., 21, 2197–2207.Bray, J. & Curtis, J. (1957). An ordination of the upland forest communities of

southern Wisconsin. Ecol. Monogr., 27, 326–349.Campos, R.I., Vasconcelos, H.L., Ribeiro, S.P., Neves, F.S. & Soares, J.P. (2006).

Relationship between tree size and insect assemblages associated with

Anadenanthera macrocarpa. Ecography, 29, 442–450.De Carvalho, D.D., Ingvarsson, P.K., Joseph, J., Suter, L., Sedivy, C., Macaya-

Sanz, D. et al. (2010). Admixture facilitates adaptation from standing variation

in the European aspen (Populus tremula L.), a widespread forest tree. Mol. Ecol.,

19, 1638–1650.Evans, L.M., Allan, G.J., Shuster, S.M., Woolbright, S.A. & Whitham, T.G.

(2008). Tree hybridization and genotypic variation drive cryptic speciation of a

specialist mite herbivore. Evolution, 62, 3027–3040.Gandon, S., Capowiez, Y., Dubois, Y., Michalakis, Y. & Olivieri, I. (1996). Local

adaptation and gene-for-gene coevolution in a metapopulation model. Proc. R.

Soc. Lond. B, 263, 1003–1009.Gelman, A., Carlin, J.B., Stern, H.S. & Rubin, D.B. (2004). Bayesian Data Analysis.

Chapman & Hall/CRC, Boca Raton, FL.

Hadfield, J.D., Wilson, A.J., Garant, D., Sheldon, B.C. & Kruuk, L.E.B. (2010).

The misuse of BLUP in ecology and evolution. Am. Nat., 175, 116–125.Hall, D., Luquez, V., Garcia, V.M., Onge, K.R.S., Jansson, S. & Ingvarsson, P.K.

(2007). Adaptive population differentiation in phenology across a latitudinal

gradient in European aspen (Populus tremula, L.): a comparison of neutral

markers, candidate genes and phenotypic traits. Evolution, 61, 2849–2860.Hoeksema, J.D. & Forde, S.E. (2008). A meta-analysis of factors affecting local

adaptation between interacting species. Am. Nat., 171, 275–290.Ingvarsson, P.K., Garcia, M.V., Luquez, V., Hall, D. & Jansson, S. (2008).

Nucleotide polymorphism and phenotypic associations within and around the

phytochrome B2 locus in European aspen (Populus tremula, Salicaceae).

Genetics, 178, 2217–2226.Jansen, J.J., Allwood, J.W., Marsden-Edwards, E., van der Putten, W.H.,

Goodacre, R. & van Dam, N.M. (2009). Metabolomic analysis of the

interaction between plants and herbivores RID C-3707-2011 RID B-3281-

2008. Metabolomics, 5, 150–161.Jonsson, P., Johansson, A., Gullberg, J., Trygg, J., Jiye, A., Grung, B. et al. (2005).

High-throughput data analysis for detecting and identifying differences between

samples in GC/MS-based metabolomic analyses. Anal. Chem., 77, 5635–5642.Kaltz, O. & Shykoff, J.A. (1998). Local adaptation in host–parasite systems.

Heredity, 81, 361–370.Karban, R. (2011). The ecology and evolution of induced resistance against

herbivores. Funct. Ecol., 25, 339–347.Karban, R. & Baldwin, I.T. (1997). Induced Responses to Herbivory. University of

Chicago Press, Chicago, IL.

Keith, A.R., Bailey, J.K. & Whitham, T.G. (2010). A genetic basis to community

repeatability and stability. Ecology, 91, 3398–3406.Kempel, A., Sch€adler, M., Chrobock, T., Fischer, M. & van Kleunen, M. (2011).

Tradeoffs associated with constitutive and induced plant resistance against

herbivory. PNAS, 108, 5685–5689.Kniskern, J. & Rausher, M. (2001). Two modes of host-enemy coevolution.

Popul. Ecol., 43, 3–14.Konieczny, A. & Ausubel, F. (1993). A procedure for mapping arabidopsis mutations

using codominant ecotype-specific PCR-based markers. Plant J., 4, 403–410.Le Corre, V. & Kremer, A. (2003). Genetic variability at neutral markers,

quantitative trait loci and trait in a subdivided population under selection.

Genetics, 164, 1205–1219.Linhart, Y.B. & Grant, M.C. (1996). Evolutionary significance of local genetic

differentiation in plants. Annu. Rev. Ecol. Syst., 27, 237–277.Luquez, V., Hall, D., Albrectsen, B., Karlsson, J., Ingvarsson, P. & Jansson, S.

(2008). Natural phenological variation in aspen (Populus tremula): the SwAsp

collection. Tree Genet. Genomes, 4, 279–292.Lynch, M. & Walsh, B. (1998). Genetics and Analysis of Quantitative Traits. Sinauer

Associates, Sunderland, MA.

Ma, X-F., Hall, D., St. Onge, K.R., Jansson, S. & Ingvarsson, P.K. (2010). Genetic

differentiation, clinal variation and phenotypic associations with growth cessation

across the Populus tremula photoperiodic pathway. Genetics, 186, 1033–1044.

Magee, L. (1990). R2 measures based on wald and likelihood ratio joint

significance tests. Am. Stat., 44, 250–253.Mopper, S. & Strauss, S.Y. (eds). (1997). Genetic Structure and Local Adaptation in

Natural Insect Populations: Effects Of Ecology, Life History, and Behavior, 1st edn.

Springer, New York, NY.

Oksanen, J., Blanchet, F.G., Kindt, R., Legendre, P., Minchin, P.R., O’Hara, R.B.

et al. (2012). vegan: community ecology package. R, package version, 2 0–2.Available at: http://cran.r-project.org/web/packages/vegan/index.html

Parker, J., Burkepile, D. & Hay, M. (2006). Opposing effects of native and

exotic herbivores on plant invasions. Science, 311, 1459–1461.Peeters, P.J. (2002). Correlations between leaf structural traits and the densities

of herbivorous insect guilds. Biol. J. Linn. Soc., 77, 43–65.Pinheiro, J.C. & Bates, D.M. (2000). Mixed Effects Models in S and S-Plus. Springer,

New York, NY.

Plummer, M., Best, N., Cowles, K. & Vines, K. (2006). CODA: convergence

diagnosis and output analysis for MCMC. R News, 6, 7–11.Robinson, K.M., Ingvarsson, P.K., Jansson, S. & Albrectsen, B.R. (2012).

Genetic variation in functional traits influences arthropod community

composition in aspen (Populus tremula L.). PLoS ONE, 7, e37679.

Shuster, S.M., Lonsdorf, E.V., Wimp, G.M., Bailey, J.K. & Whitham, T.G.

(2006). Community heritability measures the evolutionary consequences of

indirect genetic effects on community structure. Evolution, 60, 991–1003.Stamp, N. (2003). Out of the quagmire of plant defense hypotheses. Q. Rev.

Biol., 78, 23–55.Storey, J.D. & Tibshirani, R. (2003). Statistical significance for genomewide

studies. Proc. Natl Acad. Sci. USA, 100, 9440–9445.Sun, G., Zhu, C., Kramer, M.H., Yang, S.-S., Song, W., Piepho, H.-P. et al.

(2010). Variation explained in mixed-model association mapping. Heredity, 105,

333–340.Thompson, J.N. (2005). The Geographic Mosaic of Coevolution. University of Chicago

Press, Chicago, IL.

Verhoeven, K.J.F., Biere, A., Harvey, J.A. & Van Der Putten, W.H. (2009). Plant

invaders and their novel natural enemies: who is na€ıve? Ecol. Lett., 12, 107–117.Whitham, T.G., Bailey, J.K., Schweitzer, J.A., Shuster, S.M., Bangert, R.K.,

LeRoy, C.J. et al. (2006). A framework for community and ecosystem genetics:

from genes to ecosystems. Nat. Rev. Genet., 7, 510–523.Wimp, G., Martinsen, G., Floate, K., Bangert, R. & Whitham, T. (2005). Plant

genetic determinants of arthropod community structure and diversity.

Evolution, 59, 61–69.Witzell, J. & Martin, J.A. (2008). Phenolic metabolites in the resistance of

northern forest trees to pathogens – past experiences and future prospects.

Can. J. For. Res.-Rev. Can. Rech. For., 38, 2711–2727.Yeaman, S. & Otto, S.P. (2011). Establishment and maintenance of adaptive genetic

divergence under migration, selection, and drift. Evolution, 65, 2123–2129.Yu, J., Pressoir, G., Briggs, W.H., Vroh Bi, I., Yamasaki, M., Doebley, J.F. et al.

(2006). A unified mixed-model method for association mapping that accounts

for multiple levels of relatedness. Nat. Genet., 38, 203–208.Zapala, M.A. & Schork, N.J. (2006). Multivariate regression analysis of distance

matrices for testing associations between gene expression patterns and related

variables. PNAS, 103, 19430–19435.Z€ust, T., Heichinger, C., Grossniklaus, U., Harrington, R., Kliebenstein, D.J. &

Turnbul, L.A. (2012). atural enemies drive geographic variation in plant

defenses. Science, 338, 116–119.

SUPPORTING INFORMATION

Additional Supporting Information may be downloaded via the online

versionof this article atWileyOnlineLibrary (www.ecologyletters.com).

Editor, Micky Eubanks

Manuscript received 26 December 2012

First decision made 26 January 2013

Manuscript accepted 18 March 2013



Documents

Geographic structure in metabolome and herbivore community co-occurs with genetic structure in plant defence genes