Natural Enemies Drive Geographic Variation in Plant Defenses

See allHide authors and affiliations

Science  05 Oct 2012:
Vol. 338, Issue 6103, pp. 116-119
DOI: 10.1126/science.1226397

Plant Anti-Insect Armaments

Because individual plants are unable to relocate, they are subject to extreme selection by the insects feeding upon them. One means by which plants suppress herbivory is to produce toxic compounds to deter feeding (see the Perspective by Hare). Agrawal et al. (p. 113) compared pesticide–treated or untreated evening primroses. Over 5 years of pesticide treatment, the production of defensive chemicals in the fruit reduced and flowering times shifted, and the primrose's competitive ability against dandelions improved. Züst et al. (p. 116) examined large-scale geographic patterns in a polymorphic chemical defense locus in the model plant Arabidopsis thaliana and found that it is matched by changes in the relative abundance of two specialist aphids. Thus, herbivory has strong and immediate effects on the local genotypic composition of plants and traits associated with herbivore resistance.


Plants defend themselves against attack by natural enemies, and these defenses vary widely across populations. However, whether communities of natural enemies are a sufficiently potent force to maintain polymorphisms in defensive traits is largely unknown. Here, we exploit the genetic resources of Arabidopsis thaliana, coupled with 39 years of field data on aphid abundance, to (i) demonstrate that geographic patterns in a polymorphic defense locus (GS-ELONG) are strongly correlated with changes in the relative abundance of two specialist aphids; and (ii) demonstrate differential selection by the two aphids on GS-ELONG, using a multigeneration selection experiment. We thereby show a causal link between variation in abundance of the two specialist aphids and the geographic pattern at GS-ELONG, which highlights the potency of natural enemies as selective forces.

Intraspecific genetic variation is essential in enabling species to respond rapidly to evolutionary challenges such as changing environmental conditions (1) or the emergence of novel pests and pathogens (2). This diversity often reflects the balance between the strength of local selection and the current and historical levels of population substructure and gene flow (3, 4). Geographic analyses of genetic variation in several plant species have revealed clear genetic signals of local adaptation (5), caused by differences in the selective regime among locations. These analyses are further supported by reciprocal transplant experiments, in which home genotypes generally outperform those transplanted from other populations (6, 7). Although the drivers of local adaptation often remain unidentified, there is evidence that climate and soil can exert strong local selective pressures and play important roles in shaping large-scale genetic patterns (8, 9).

In contrast to the clear role of abiotic factors, there is little direct evidence that biotic forces, such as herbivory or competition, can lead to the maintenance of genetic variation across large geographic scales, despite the exceptional levels of polymorphism associated with genes involved in defense (10, 11). In theory, interactions between organisms and their natural enemies can lead to differences in the local selective regime because of geographic variation in the abundance or species composition of the enemy community (3). This spatial variation can affect defense if it is costly; e.g., if the average level of herbivory varies across populations, defended genotypes might dominate in heavily attacked populations whereas undefended genotypes would prosper when enemies are absent or rare (12). Another less studied effect is how defense might vary if plants are attacked by diverse collections of herbivore species that differ in feeding style and specialization. This could lead to higher levels of polymorphism in defense genes due to selection for specific defensive profiles matched to the predominant local herbivore or herbivore community [e.g., (13)]. However, there is no direct evidence that variation in local herbivore communities represents a sufficiently strong selective pressure to favor specific defensive traits and maintain polymorphisms in defense-related genes.

The unparalleled genetic and molecular resources available for the model plant Arabidopsis thaliana make this species an ideal candidate to study the process of local adaptation to herbivores. The primary defensive trait in A. thaliana is a series of indolic and aliphatic glucosinolates, which are secondary plant metabolites with antiherbivore properties (14). The accumulation and structure of aliphatic glucosinolates are mechanistically determined by alleles at the GS-ELONG locus that regulate the carbon side-chain elongation (3C or 4C) (15) and by alleles at the GS-AOP locus that modify the functional group of the biologically active glucosinolate side chain (ALK, OH, or NULL). The combination of these alleles yields six distinct chemotypes present in natural populations in varying proportions (16). Both individual glucosinolate compounds and full chemical profiles affect the susceptibility of a plant to specific herbivores (17, 18); hence, the aliphatic chemotype is likely under differential, qualitative selection by herbivores. By contrast, accumulation of the main indolic glucosinolates in A. thaliana is highly plastic and modulated by a large number of small-effect genetic loci, which are therefore less likely to show clear signatures of selection (19).

We mapped geographic variation in the abundance of the six chemotypes within Europe from a set of 96 accessions (75 European) (20) with known chemical profiles (16) (Fig. 1). There was no apparent pattern in the distribution of the GS-AOP chemotypes, but for GS-ELONG, the frequency of 3C to 4C chemotypes increases with both latitude and longitude (Fig. 1 and fig. S1). If this pattern results from geographical variation in herbivore feeding pressure, we would expect it to be closely matched by variation in herbivore abundance patterns. Although A. thaliana is attacked by a range of invertebrate herbivores, many of which preferentially feed on specific chemotypes (17), we hypothesized that the aphid species Brevicoryne brassicae and Lipaphis erysimi are likely drivers of these patterns as they are both abundant, mobile Brassicaceae specialists, yet differentially sensitive to environmental conditions (21). Fluctuations in aphid populations have been monitored since 1964 through the EXAMINE network ( with suction traps that operate throughout the aphid flight season (22). We retrieved data on the two aphid species from 61 traps in eight European countries. These data revealed that the abundance of L. erysimi is usually lower than that of B. brassicae, but that the geographic pattern in the relative abundances of L. erysimi and B. brassicae closely mirrors the pattern at GS-ELONG (Fig. 1 and fig. S1). Variation in the relative abundance of these two specialist aphids could therefore underlie variation in the predominant GS-ELONG chemotype found in natural populations.

Fig. 1

Location of European A. thaliana accessions with known chemical profile. Symbol color indicates the GS-ELONG chemotype (orange: 3C; green: 4C) and symbol shape indicates the GS-AOP chemotype (square: ALK; circle: OH; triangle: NULL). For GS-ELONG, the probability of finding 3C populations increases strongly with longitude (binomial GLM: t = 5.11, df = 85, P < 0.001) and weakly with latitude (t = 1.75, df = 85, P = 0.084). Countries with available aphid data are colored in blue. The shade of blue corresponds to the relative frequency of L. erysimi based on model predictions from a binomial GLM using data from 61 aphid suction traps. The relative frequency of L. erysimi increases strongly with longitude (t = 5.03, P < 0.001) and weakly with latitude (t = 1.89, P = 0.060). Pie charts indicate the observed average relative abundance of B. brassicae (white) and L. erysimi (blue) in each country.

Because causal inferences are impossible from such correlative data, we tested the causality of aphid selection on GS-ELONG, carrying out a multigenerational selection experiment on populations of A. thaliana (22). We assembled 30 replicate populations from equal numbers of seeds from each of 27 natural accessions, including 6 of the 75 European accessions mapped above. Accessions were chosen to maximize variation in defense traits while including all six glucosinolate chemotypes in a range of genetic backgrounds (table S1 and fig. S2). Over five generations, we consistently exposed populations to replicate (n = 6) treatments of a single specialist aphid species: either B. brassicae or L. erysimi; a single generalist aphid, Myzus persicae; a mixture of all three aphid species; and a no-aphid treatment. The generalist aphid was included as a negative control, because M. persicae is unresponsive to aliphatic glucosinolates (23) and we therefore would not expect it to exert directional selection on plant chemotype. The no-aphid treatment served as a control for other selective forces that were likely to affect the outcome of the experiment, such as intraspecific competition among accessions. Seeds were collected at the end of each generation with no mixing among populations, and a subset was used to establish the next generation at a constant density. After five generations of repeated herbivore treatments, we sampled 24 individuals from each population in generation 5 (144 individuals per treatment) and determined their genotype. To have a marker for changes in genotypic composition through time, we also measured leaf trichome density, a trait under strong genetic control (fig. S3), on a representative sample of plants in all generations.

Rapid adaptation occurred in the selection experiment, as evidenced by a progressive reduction in the effects of aphid feeding on final plant biomass in each generation (Fig. 2A). In line with the expected severity of aphid feeding based on previously reported population growth rates (21), L. erysimi caused the strongest reduction in plant biomass, whereas M. persicae had an intermediate effect and B. brassicae had the least effect. The mixture treatment caused a reduction similar to that produced by L. erysimi alone, probably because aphid mixtures were dominated by this fast-growing aphid species. With each generation, trichome density decreased in the no-aphid treatment, whereas it remained at significantly higher levels in all aphid treatments (Fig. 2B). Adaptation to herbivore feeding was accompanied by considerable changes in the genotypic composition of populations, including the complete loss of nine genotypes (Fig. 3). There was a nonspecific aphid effect on total indolic glucosinolates [linear mixed effects model: F1,28 = 10.66, P = 0.003], with plants in the no-aphid treatment producing on average 0.98 (±0.03, SEM) μmol g−1, and plants in aphid treatments producing 0.87 (±0.03, SEM) μmol g−1. By contrast, the different aphid treatments had a marked effect on the dominant aliphatic chemotypes within experimental populations. Notably, the relative proportions of 3C and 4C chemotypes differed strongly among aphid treatments (Fig. 3 and fig. S4). After selection, populations of the no-aphid treatment consisted of approximately two-thirds 3C and one-third 4C chemotypes. Specialist aphids selected for different chemotypes at GS-ELONG: The 4C chemotypes strongly dominated in B. brassicae treatments [binomial generalized linear model (GLM), t = 3.08, df = 25, P = 0.002] and the 3C chemotypes strongly dominated in both L. erysimi (t = 2.01, df = 25, P = 0.045) and the aphid mixture treatments (t = 2.21, df = 25, P = 0.027). The relative proportions of 3C to 4C chemotypes in populations exposed to the generalist aphid M. persicae did not differ from the no-aphid treatment (t = 0.18, df = 25, P = 0.858). Despite this similarity, the identity of the successful genotypes differed among the two treatments, with accession Sap-0 accounting for a large fraction of plants in the no-aphid treatment but being absent from all other treatments (Fig. 3). The genotypic composition of plant populations with L. erysimi and aphid mixtures was near-identical, confirming that L. erysimi dominated the mixture treatments and suggesting that in cofounded populations, L. erysimi is the most important selective force. Most successful genotypes had either a 3C-OH or a 4C-NULL chemotype, and we found no individuals belonging to either alkenyl chemotype (3C-ALK or 4C-ALK) in any treatment. Alkenyl chemotypes were common in generation 1 of the selection experiment (fig. S2), and simulations of random sampling on the basis of observed population sizes revealed that their loss could not be due to drift alone (fig. S5) but rather was a consequence of selection.

Fig. 2

(A) Change in the negative impact of aphid treatments on final plant biomass over five generations, displayed as the log-difference to final plant biomass in the no-aphid treatment (dashed line): B. brassicae (light blue); M. persicae (light green); L. erysimi (orange); and aphid mixture (yellow). Stars denote significantly less damage after five generations of selection (table S2). (B) Mean number of trichomes on the fourth leaf of 50 plants per population. Stars denote significant difference from the no-aphid treatment (black line) after five generations of selection.

Fig. 3

Change in the composition of A. thaliana accessions, from equal proportions of 27 genotypes in the ancestral population to treatment-specific compositions after five generations of selection. Each chart gives mean genotype frequencies based on n = 6 replicate populations. 3C chemotypes are indicated by solid, orange colors, and 4C chemotypes by hatched, green colors.

To identify potential causes for the loss of particular genotypes, we measured size-standardized growth rate as a measure of fitness, together with total aliphatic glucosinolate content and trichome density in a separate experiment on all 27 ancestral accessions. Alkenyl chemotypes produced the highest amount of glucosinolates and were among the slowest growing genotypes overall (fig. S6A). Alkenyl glucosinolates are an effective defense against leaf-chewing herbivores such as caterpillars (24), but their efficiency against specialist aphids remains largely unknown, and they have little effect on M. persicae (23). The loss of the alkenyl chemotypes therefore probably resulted from selection against a costly defense trait that provided insufficient benefits in our experiment. This cost-benefit balance is also the most likely reason for the difference in dominant genotypes between the no-aphid treatment and the aphid treatments (Fig. 3). The dominant genotype in no-aphid populations, Sap-0, was completely absent from all aphid treatments, indicating low fitness in the presence of herbivores. The Sap-0 genotype had the lowest trichome density of all nonglabrous accessions, and as trichome production had a growth cost (fig. S6B), its success can explain the observed decrease in trichome density in the no-aphid populations over time (Fig. 2B). Compared to other 3C-OH chemotypes, Sap-0 also produces low amounts of glucosinolates, an additional indication that in the absence of herbivores, undefended, fast-growing genotypes will prosper.

Despite known epistatic interactions between GS-ELONG and GS-AOP (19), our data suggest that aphid selection acts independently on the two loci. The magnitude and direction of selection exerted by the two specialist aphids on GS-ELONG in our experiment suggest a causal link between the observed cline in GS-ELONG across Europe and the changes in the relative abundance of the same aphids. Although B. brassicae is numerically dominant across most of Europe, the faster-growing L. erysimi can inflict greater damage on plants and quickly dominates populations that are cofounded by both aphid species; thus, even a modest change in relative abundance could cause the loss of C3 populations. All plants in the selection experiment experienced strong intraspecific competition, and because growth rate is a good predictor of competitive ability (25), it is unsurprising that fast-growing plant genotypes were generally selected, whereas the slowest-growing alkenyl chemotypes were lost. Alkenyl chemotypes are, however, very common in natural populations and could be maintained by other herbivores, such as leaf-chewing caterpillars (24).

Ecological theory has consistently emphasized the role of natural enemies in maintaining diversity both within and among species, but convincing empirical evidence has been lacking. Here, we demonstrate that even functionally similar herbivores such as different species of aphid have the potential to select for specific chemotypes and drive large-scale geographic patterns in plant defense profiles. It therefore seems likely that natural herbivore communities, with their greater variety of feeding styles and specializations, play a major role in shaping and refining the plant defenses observed in natural communities.

Supplementary Materials

Materials and Methods

Figs. S1 to S7

Tables S1 to S3

References (2639)

Data Files S1 to S6

References and Notes

  1. Materials and methods are available as supplementary materials on Science Online.
  2. Acknowledgments: This work was supported by the Forschungskredit of the University of Zürich (T.Z.) and by Swiss National Science Foundation grant 31-107531 (L.A.T.), by U.S. NSF grant DBI0820580 (D.J.K.), an Advanced Grant of the European Research Council (U.G.), and a special grant of the University of Zürich in commemoration of Prof. Christine B. Müller. The Rothamsted Insect Survey is supported by a Biotechnology and Biological Sciences Research Council National Capability Grant and the Lawes Agricultural Trust. We thank members of the EU EXAMINE Project (EVK2-1999-00151) for use of data and A. A. Agrawal, S. West, J. Levine, and two anonymous reviewers for comments that improved the manuscript. All data presented in this paper and R code of all analyses are provided in the supplementary materials.
View Abstract

Stay Connected to Science

Navigate This Article