2.1.1 Development

Arabidopsis trichomes are non-glandular, single-cell structures originating from epidermal cells and are found mainly on the adaxial surface of the leaves, but also on the abaxial leaf surface and on stems and sepals, with stem and sepal trichomes usually being unbranched and leaf trichomes having two to four branches depending on accession (Hulskamp and Kirik, 2000). Initiation of trichome formation in Arabidopsis starts and finishes in the leaf primordium stage, with the first trichome forming from an epidermal cell at the tip of the leaf when the primordia are about 100 um long and with trichome initiation being completed when primordia are about 700 um long in the accession Col-0 (Larkin et al., 1996). There is substantial variation between Arabidopsis accessions in trichome density (Mauricio, 2005; Symonds et al., 2005), ranging from completely hairless accessions (such as Mir-0 and Kas-1) to densely covered accessions such as Chi-1 (figure 2). Excluding hairless plants, differences between accessions in trichome density may be reflected in the duration of trichome initiation. For example, trichome initiation in Ler stops when the primordia are about 500 um long and Ler is less-densely covered with trichomes than Col-0 (Larkin et al., 1996). The genetic background of both trichome initiation and development is reasonably well understood, with a large array of mutants available (reviewed by Hulskamp and Kirik, 2000; Marks and Esch, 2003; Szymanski, 2005).

2.1.2 Plant-Insect Interactions

For arthropod herbivores, trichomes can be both a blessing and a curse. As structural features, they may hinder or help landing, impede movement, alter the microclimate, reduce predation rates when they hamper carnivores and can be climbed by small herbivores to avoid detection by predators. On the other hand, trichomes may increase predation rates as some predators lay their eggs in trichome dense areas to avoid intra-guild predation and can use pollen trapped by trichomes as alternative food source (Southwood, 1986; Krips et al., 1999; Roda et al., 2000; Michalska, 2003; Roda et al., 2003; Stavrinides and Skirvin, 2003). In the case of glandular trichomes, they are also part of chemical defenses against herbivores (e.g Chatzivasileiadis and Sabelis, 1997; Maluf et al., 2001), but as trichomes in Arabidopsis are non-glandular, single cells I will not discuss this further. So far, there is no evidence that defense compounds accumulate in Arabidopsis trichomes. Identification of 63 proteins from trichomes revealed no proteins with known defense functions (Wienkoop et al., 2004). On the other hand, some genes with a possible function in plant defense against insects, such as AtBSMT1 involved in benzoid and salicylic acid methylation, are highly expressed at the base of trichomes (Chen et al., 2003a; Wang et al., 2005). Indeed, there are some indications that trichomes are involved in volatile emissions and as such play a role in indirect defense (see §3.21.4.2).

Several studies provide evidence for a role of trichomes in defense of Arabidopsis against herbivores. Reymond et al. (2004) showed that Pieris rapae caterpillars gained more weight on glabrous (=hairless) gl1 mutants compared to wild-type Columbia. Additionally, correlation between trichome density and plant defense was demonstrated in two studies on Arabidopsis populations: In a study using plants collected from various fields in Sweden, Handley et al. (2005) demonstrated that oviposition by Plutella xylostella moths was negatively correlated with trichome density. However, no correlation was found between performance of P. xylostella larvae and trichome density. In a field study in North Carolina, Mauricio (1998) found a negative correlation between feeding damage (likely done for the most part by flea beetles) and trichome density. The latter study also demonstrated that defenses such as trichomes may be costly: in undamaged plants, trichome density was negatively correlated with fruit production, with costs exceeding the benefits in herbivore-damaged plots (see also §4.1.2).

2.1.3 Inducibility

Presumably to minimize costs, plant defenses are often inducible. As trichome density is determined early in leaf development (Mauricio, 2005), a damaged leaf will not start to grow extra trichomes. Nonetheless, wounding of Arabidopsis plants does lead to increased trichome density of newly formed leaves, although this has not been demonstrated as a response to herbivory (Traw and Bergelson, 2003).

2.3.1 Biosynthesis

Synthesis, degradation and function of glucosinolates have been reviewed recently (Wittstock and Halkier, 2002; Kliebenstein, 2004); the following is a short overview. Glucosinolates are amino acid-derived metabolites containing a sulphate and a thioglucose moiety. Depending on the amino acid from which they are derived, glucosinolates can be divided into three classes: aliphatic glucosinolates (in Arabidopsis derived from methionine and some from leucine), aromatic glucosinolates (in Arabidopsis derived from phenylalanine) and indole glucosinolates (derived from tryptophan) (figure 4A). In total, 36 Arabidopsis glucosinolates have been found using various accessions (Reichelt et al., 2002). Additionally, glucosinolate profiles were studied at different developmental stages in all major organs in Col-0 (Brown et al., 2003). Together, these studies reveal that: methionine-derived aliphatic glucosinolates are dominant in all organs at almost all developmental stages; during development, early leaves contain less glucosinolates than later leaves; reproductive organs and especially seeds contain the highest levels of total glucosinolates; seeds also contain the highest diversity of glucosinolates, with aromatic glucosinolates being almost exclusively present in seeds. Between accessions, seeds mainly differ in indole glucosinolate composition and leaves mainly show variation in aliphatic glucosinolate composition. The variation in glucosinolate profiles between accessions may be based on relatively few genes as polymorphism at five loci was sufficient to generate 14 qualitatively different glucosinolate profiles (figure 4B). The dramatic impact of relatively few genes on glucosinolate profiles constitutes an evolutionary flexible system that may be easily adapted to different selective pressures (Kliebenstein et al., 2001b; and see §4.2.1).

2.3.2 Degradation

Although glucosinolates may function as storage compounds for sulphur and nitrogen, their most important role is likely in plant defense. However, not the glucosinolates themselves but rather their degradation products are most active in defense against both herbivores and pathogens (figure 4C). Similar to many other compounds, the glucoside form likely functions as an inactive precursor (Rask et al., 2000). The first step in glucosinolate degradation is the removal of glucose, catalyzed by myrosinase. Myrosinases are compartmentalized in myrosin bodies of specialized myrosin cells, which in Arabidopsis are confined in phloem parenchyma (Andreasson et al., 2001). Glucosinolates can be found in the vacuoles of cells in all plant organs but in the main flower stalk are particularly concentrated in special cell types lining the phloem (Koroleva et al., 2000). Upon tissue-disruption, myrosinases and glucosinolates come into contact, triggering the degradation of glucosinolates. A second step, after removal of glucose, further degrades the glucosinolates to their final products. Arabidopsis accessions can be divided into two classes with respect to the type of final products: those that predominantly produce nitriles and those that predominantly produce isothiocyanates. Accessions having a functional epithiospecifier protein (ESP), such as Ler-0 and Cvi-0 produce nitriles or epithionitriles from alkyl or alkenyl glucosinolates respectively; accessions with a non-functional ESP, such as Col-0 and Ws-0, produce isothiocyanates (Lambrix et al., 2001; Zabala et al., 2005). Recently, another gene involved in glucosinolate degradation has been cloned. This protein, epthiospecifier modifier1 (ESM1), drives glucosinolate degradation towards isothiocyanates. Functional activity of both ESP and ESM1 result in a mixture of isothiocyanates and nitriles (Zhang et al., 2006).

2.3.3 Plant-Herbivore Interactions

That glucosinolate degradation products can be toxic to insects has been demonstrated in many studies, with isothiocyanates being especially toxic (reviewed by Wittstock et al., 2003; and see Lambrix et al., 2001; Rohloff and Bones, 2005 for a list of glucosinolate-degradation products found in Arabidopsis). The mode of action of this toxicity is still unclear, although isothiocyanates are known to react with proteins. Additionally, both nitriles and isothiocyanates may impact cellular respiration through HCN production and other mechanisms (Tsao et al., 2002; Wittstock et al., 2003).

There are several examples of negative correlation between Arabidopsis glucosinolate levels and herbivore performance: (Mauricio, 1998) showed a negative correlation between glucosinolate levels in Arabidopsis and herbivore damage in the same North Carolina field study mentioned before (§ 3.2). Kroymann et al. (2003) demonstrated that the Ler MAM2 gene is responsible for enhanced aliphatic glucosinolate levels and reduced feeding damage by larvae of the generalist moth Spodoptera exigua; Kliebenstein et al. (2005) reported a negative correlation between aliphatic glucosinolate levels and herbivore damaged caused by larvae from S. exigua and another generalist moth, Trychoplusia ni; and Mewis et al., (2005) showed that performance of S. exigua and the specialist aphid B. brassicae and generalist aphid Myzus persicae was negatively correlated with total glucosinolate levels. Additionally, Mauricio (1998) showed a positive correlation between glucosinolate levels and plant fitness in herbivore damaged plants, measured as the number of siliques produced. From these results, it becomes clear that glucosinolates act as defense compounds in interactions between Arabidopsis and herbivores.

However, larvae from the specialist moth P. xylostella do not seem to suffer from the elevated aliphatic glucosinolate levels associated with MAM2 expression. On the contrary, damage by this herbivore was positively correlated with aliphatic glucosinolate levels and herbivore damaged caused by the specialist P. xylostella (Kroymann et al., 2003; Kliebenstein et al., 2005; see also §4.2.1). These examples show that even though glucosinolates are considered defense compounds, several herbivore species have been able to overcome this defense system. Once herbivore species can overcome plant defenses they may specialize on plant species containing these defenses. Ratzka et al. (2002) discovered that larvae of the crucifer specialist P. xylostella can detoxify glucosinolates by desulfating them, with sulfatase activity and gene-expression only found in the gut of the larvae. Other crucifer specialists such as P. rapae and Pieris brassicae caterpillars do not show sulfatase activity, indicating that there must be more strategies to cope with glucosinolates. Indeed it was recently demonstrated that a protein in the gut of P. rapae, a designated nitrile-specifier protein, diverts of degradation of glucosinolates from isothiocyanates to nitriles, when feeding on an accession (Col-0) that normally produces isothiocyanates (Wittstock et al., 2004). This fits well with the observation that P. rapae-infested Col-0 plants emit nitriles and not isothiocyanates (Van Poecke et al., 2001).

When herbivores can overcome plant defenses, they may use these mechanisms to their own advantages. For example, specialist herbivores can use the glucosinolates and/or their degradation products as volatile attractants, and feeding and oviposition stimuli (see review by Rask et al., 2000). Indeed some glucosinolate degradation products can be found in trace amounts in the headspace of undamaged or caterpillar-infested plants (Van Poecke et al., 2001; Vercammen et al., 2001; Rohloff and Bones, 2005). Specialized insects may even sequester glucosinolates for their own defense against predators (Muller and Wittstock, 2005). An intriguing example comes from the specialist aphids B. brassicae and Lipaphis erysimi, that besides sequestering glucosinolates, have myrosinases confined to specialized microbodies, similar to the situation in plants. Possibly, they function as a defense against predators in a similar fashion to plants, such that tissue disruption results in the production of isothiocyanates, that may not only be toxic to predators but additionally act as synergists to the aphid alarm pheromone E-beta-farne-sene (Bridges et al., 2002).

In short, glucosinolates function as defense compounds against generalist herbivores but may be exploited by specialist herbivores. This difference between specialist and generalist herbivores was reflected in QTL analyses performed by Kliebenstein et al. (2001a), where QTL analyses of resistance against generalist T. ni caterpillars resulted in six loci, of which five were involved in either glucosinolate biosynthesis or breakdown (one of these loci corresponds to the ESP locus [Lambrix et al., 2001], which may be identical to the TASTY locus found in a different QTL study on plant resistance using T. ni and Ler x Col recombinant inbred lines [Jander et al., 2001]). In contrast, neither of two QTL affecting resistance against the specialist P. xylostella caterpillars overlapped with glucosinolate biosynthesis or degradation QTL. Besides different effects of glucosinolates on generalists and specialists, these results also indicate that 1) aliphatic glucosinolates deter T. ni herbivory, 2) increased myrosinase activity decreases T. ni herbivory and 3) T. ni prefers nitriles over isothio-cyanates (the latter was also demonstrated by Lambrix et al., 2001). An intriguing question arises from these studies: if isothiocyanates are more effective as defense compounds against generalist herbivores, why do some accessions form nitriles instead of isothiocyanates, which requires an additional enzymatic step? One hypothesis is that as isothiocyanates also function as oviposition stimuli for specialist herbivores, it may be beneficial not to produce these compounds when selection pressure by specialists is high. Another hypothesis is that nitriles may be more effective against generalist herbivores other than T. ni (Lambrix et al., 2001). With respect to the latter, it should be noted that although T. ni is a generalist herbivore, cruciferous plants are among its preferred hosts, hence the common name cabbage looper (Soo Hoo et al., 1984).

2.3.4 Tritrophic Interactions

Besides influencing plant-herbivore interactions directly, glucosinolates also may be important to a third trophic level. Predators or parasitoids that specialize on crucifer feeding herbivores can use volatile glucosinolate degradation products as cues to locate their prey or host. For example, the profile of isothiocyanates produced by different Brassica oleracea near-isogenic lines influences parasitoid behaviour (Bradburne and Mithen, 2000). Van Poecke et al. (2001) demonstrated that the parasitoid wasp C. rubecula distinguishes between mechanically damaged Col-0 and Col-0 infested by it's host P. rapae. One of the differences in the volatile blends of these odor sources was the presence of nitriles in the blend of P. rapae-infested plants. Thus, it may be that C. rubecula wasps use nitriles to locate their host. However, this parasitoid did not distinguish between P. xylostella-infested plants and P. rapae infested plants (Van Poecke et al., 2003). As P. xylostella desulfates glucosinolates (Ratzka et al., 2002) rather than diverting degradation to nitriles, it appears that althought C. rubecula may use nitriles for host-location, it does not use these nitriles for host-discrimination.

2.3.5 Inducibility

The same field study that showed that total glucosinolate level is negatively correlated with feeding damage in natural populations of Arabidopsis, also showed that for undamaged plants, glucosinolate level was negatively correlated with fruit production (Mauricio and Rausher, 1997; Mauricio, 1998). Additionally, undamaged mutant plants that produced less glucosinolates, also showed increased fitness compared to undamaged wild-type plants (Mauricio, 2001). This suggests that there are costs related to glucosinolate defenses. Indeed, glucosinolate levels - especially aliphatic glucosinolates - can be induced by caterpillar and aphid feeding (Mewis et al., 2005), possibly to minimize these costs.

2.4.1 Biosynthesis

Terpenoids are produced in the plant through two distinct pathways: the mevalonate (MVA) pathway is located in the cytosol/endoplasmatic reticulum and produces sesqui-and triterpenoids, the 2-C-methyl-D-erythritol 4-phosphate (MEP) pathway is located in plastids and produces mono-, di- and tetraterpenoids (figure 5). The pathways converge biochemically in the production of isopentenyl diphosphate (IPP) and its isomer dimethylallyl diphosphate (DMAPP). IPP and DMAPP form the building blocks of terpenoids and are linked through the action of prenyltransferases. The products of prenyltranferases are used by terpene synthases (TPS) to produce primary and secondary metabolites (Aubourg et al., 2002; Rodriguez-Concepcion and Boronat, 2002). In Arabidopsis 40 putative TPS genes have been identified of which 32 are apparently intact. Most of these are predicted to be involved in secondary metabolite production (Aubourg et al., 2002). Characterized genes include five monoterpene synthases and two sesquiterpene synthases (Bohlmann et al., 2000; Chen et al., 2003b; Fäldt et al., 2003; Chen et al., 2004; Tholl et al., 2005).

Expression analyses showed that the most diverse array of TPS gene-expression is found in flowers, including two of the five characterized monoterpene synthases and the two characterized sesquiterpene synthases. Of the other three characterized monoterpene synthases, two are expressed throughout the plant and one is almost exclusively expressed in roots (Chen et al., 2003b).

Volatile emissions reflect gene-expression in both quantity and quality, with expression of characterized TPS genes correlating nicely with the terpenoids found: flowers emit the highest amount and diversity of terpenoids, very limited amounts of terpenoids are emitted by the vegetative tissue and only one terpenoid could be detected in root-exudates. (Van Poecke et al., 2001; Aharoni et al., 2003; Chen et al., 2003b; Chen et al., 2004; Steeghs et al., 2004; Tholl et al., 2005). Flowering Arabidopsis show a diurnal rhythm of terpenoid emissions, with a peak in emissions during the day; for non-terpenoid volatiles, no such rhythm could be found (Aharoni et al., 2003). Analyses of the emission of selected mono- and sesquiterpenes from flowers from 37 Arabidopsis accessions showed mainly quantitative and a few qualitative differences (Tholl et al., 2005).

While examining transgenic plants overproducing the monoterpene linalool, Aharoni et al. observed that much of the linalool in these plants was glycosylated and glycosylated linalool could also be detected in wild-type plants, albeit in lesser amounts. This indicates that a significant fraction of the terpenoid pool may be stored in a biologically inactive form.

2.4.2 Plant-Insect Interactions

Terpenoids are known to influence herbivorous insects in various ways: they can be part of direct defenses as repellents, feeding deterrents or toxins, with the toxicity of terpenoids possibly due to neurotoxic effects (Garcia et al., 2005; reviewed by Gershenzon and Croteau, 1991 and Langenheim, 1994). Terpenoids can also be used as attractants and/or feeding or oviposition stimulants by specialist herbivores; they can be part of indirect defense as attractants for carnivores; and they can be part of sexual reproduction and outcrossing as attractants of pollinators (Gershenzon and Croteau, 1991; Langenheim, 1994).

With respect to the latter, it is interesting to note that, although Arabidopsis is a self-pollinating species, Arabidopsis flowers emit by far the largest amount of volatiles, especially terpenoids, compared to other plant parts, with some terpene synthase genes being exclusively expressed in flowers. Based on this result and the fact that in Arabidopsis 1) a low percentage of cross-pollination occurs under natural conditions, 2) the receptive stigma protrudes out of the petals before the stamen are mature, 3) floral nectarines are present at the base of the stamen, and 4) the flowers are visited by small insects, Chen et al. (2003b) argued that pollinators may be involved in Arabidopsis outcrossing. The diurnal rhythm found by Aharoni et al. (2003) would suggest that these pollinators are mainly day-active. However, as terpenoids also readily react with reactive oxygen species and can have antimicrobial properties, it is also possible that floral terpenoids protect the reproductive organs from oxidative stress and pathogens (Chen et al., 2003b).

Undamaged vegetative parts of Arabidopsis do not emit many volatiles, but wounding and herbivory by caterpillars result in an increased emission of volatiles, including terpenoids (Van Poecke et al., 2001; Fäldt et al., 2003). As the increase in terpenoid emissions coincides with increased attraction of parasitoid wasps, terpenoids may be involved in indirect defense of Arabidopsis (Van Poecke et al., 2001). Indeed mutant plants that show reduced emission of terpenoids also showed reduced attraction of parasitoid wasps. However, the role of terpenoids cannot be deduced from this study, as these plants also emit less methyl-salicylate, another known attractant of carnivores (Van Poecke, 2002; Van Poecke and Dicke, 2002; De Boer and Dicke, 2004).

2.4.3 Inducibility

To function as an indirect defense mechanism, volatile emissions likely have to be inducible. In analogy to Aesop's fable ‘The boy who cried wolf’, plants that would call out for help even when not attacked by herbivores would provide little useful information for predators or parasitoids. Indeed, not only the emissions are inducible by wounding and herbivory, this increase in volatile emissions also coincides with elevated expression of genes involved in volatile biosynthesis, including TPS genes, indicating that the increased emissions are likely due to increased synthesis (Van Poecke et al., 2001; Fäldt et al., 2003). It would be interesting to see whether glycoside forms of terpenoids are mobilized upon herbivory.

2.5.1 Proteinase Inhibitors

As their name suggests, proteinase inhibitors are proteins that interfere with the activity of proteinases. This inhibitory activity may be important for both primary processes such as seed dormancy and protein reserve mobilization, but they have been mainly studied as defense mechanisms. In defenses, they function by inhibiting protease activities of pathogens an herbivores, thereby not only decreasing nutrient uptake but also overstimulating production of proteolytic enzymes in the insect gut which could result in depletion of essential amino acids, both with a negative impact on herbivore performance (Ryan, 1990; Lawrence and Koundal, 2002). Two proteinase inhibitor gene-families have been described in Arabidopsis, one encoding trypsin inhibitors (Clauss and Mitchell-Olds, 2004) and one encoding cysteine proteinase inhibitors (cystatins; Martinez et al., 2005). Of the six trypsin inhibitors detected in the Arabidopsis genome, five are transcribed. Expression of these five genes can be induced by P. xylostella feeding. Trypsin inhibitor activity was negatively correlated with P. xylostella activity and larval mass but not with leaf damage by P. xylostella larvae in a study comparing Arabidopsis genotypes varying in, amongst others, trypsin inhibitor activity (Cipollini et al., 2004). For the cystatin gene family, no function in defense against insects has been reported for Arabidopsis, although transferring a cystatin gene from Arabidopsis into white poplar conferred resistance to the crysomelid beetle Crysomela populi (Delledonne et al., 2001).

2.5.2 Polyphenol Oxidase

Other Arabidopsis defenses against insects may include polyphenol oxidase activy (PPOs). Polyphenols act by oxidizing phenolic compounds to reactive quinones, that alkylate essential amino acids and thereby reduce nutritional value. As far as I am aware, the role of PPOs in plant defense against insects has only been demonstrated in vitro (Constabel and Ryan, 1998). In Arabidopsis, PPOs were negatively correlated with P. xylostella activity and larval mass and with leaf damage by P. xylostella larvae in the study by Cipollini et al.(2004).

2.6.1 Transcriptome Analyses

Reymond et al., (2000) were the first to use gene-expression profiling to study Arabidopsis-insect interactions. Together with a follow-up study (Reymond et al., 2004), it showed that caterpillar feeding induces several functional classes of genes, including defense proteins such as a few putative lectins and a cysteine proteinase; phenyl-propanoid pathway genes that may be involved in phytoalexin production, radical scavengers or cell wall fortification; oxidative and abiotic stress related genes; and genes involved in relocation of resources. For none of these it is clear how and whether they function in plant defense against insects, or, in the case of the defense proteins, whether they do so in Arabidopsis. Genes with known functions in Arabidopsis defenses, such as those involved in glucosinolate biosynthesis and degradation and genes involved in signaling (see §4.2), are also up-regulated by caterpillars. Only very few genes were down-regulated by caterpillar feeding, and the function of those are unclear (Reymond et al., 2004). Thus, gene-expression studies do not only confirm the role of already known player but are also helpful to indicate new players. Additional information can be obtained by, for example, studying the effect of different herbivores on gene-expression profiles. Reymond et al. (2000) demonstrated that plants infested by two specialist caterpillars from the same genus (P. rapae and P. brassicae) showed highly similar gene-expression profiles.

They then tested the hypothesis that specialist and generalist herbivores induce different sets of genes, by comparing expression profiles of Arabidopsis plants infested with the specialist herbivore P. rapae and the generalist herbivore Spodoptera littoralis (Reymond et al., 2004). This because specialists may have found ways to attenuate inducible plant defenses. Interestingly, they found hardly any differences in induced gene-expression, indicating the specialist found ways to deal with plant defenses rather than to attenuate plant defenses. Indeed, as mentioned before, P. rapae can divert glucosinolate degradation from harmfull isothiocyanates to less harmful nitriles (Wittstock et al., 2004), something that S. littoralis likely cannot.

In a similar approach, Moran et al. (2002) studied the response of Arabidopsis to aphid feeding. Some of the genes found to be induced by caterpillars in the studies mentioned above were also induced by aphids. These included oxidative stress related genes; a pathogenesis related protein gene (BGL2/PR2); and signaling related genes. Others, such as the pathogenesis related protein PR1, were induced by aphids but not by caterpillars (Moran et al., 2002; Reymond et al., 2004). Aphid feeding also reduced the expression of some genes, including oxidative stress related genes (differential regulation of oxidative stress related genes has been reported before [Kliebenstein et al., 1998]), phenylpropanoid pathway genes and genes involved in signaling (Moran et al., 2002). Again, the function of these aphid-responsive genes in plant-insect interactions is unclear and these studies point out some candidates for further investigation.

Based on the studies above it is clear that inducible Arabidopsis responses to herbivory differ depending on the herbivore species, at least if they differ in feeding mode. However, data collected by different research groups is often hard to compare due to differences in e.g. plant growth conditions. A study by (De Vos et al., 2005) compared gene-expression profiles from plants infested with the caterpillar P. rapae, the thrips Frankliniella occidentalis or the aphid M. perzicae (this experiment also included plants infected with the pathogens Pseudomonas syringae or Alternaria brassicicola). One of the main conclusions from this study is that the responses to different herbivorous attackers are very dissimilar, with caterpillar and thrips induced changes showing the highest degree of overlap, followed by thrips and aphid induced changes, while caterpillar and aphid induced changes hardly overlapped at all. This corresponds well with different types of damage inflicted by these herbivore species: tissue-feeding caterpillars and cell-feeding thrips cause more mechanical damage than phloem-feeding aphids, which may explain the bigger overlap between caterpillars and thrips compared to caterpillars and aphids. On the other hand, both thrips and aphids puncture individual cells whereas caterpillars disrupt tissue more drastically, which may explain the higher overlap between aphids and thrips compared to aphids and caterpillars. Such a correlation between feeding damage and plant responses has been shown before in a study by Van Poecke et al. (2003), where a parasitoid specialized on P. rapae caterpillars was attracted most to volatiles emitted by caterpillar-damaged plants, intermediately to cell-feeding spider mite-infested plants and not at all to aphid infested plants.

2.6.2 Metabolome Analyses

Besides gene-expression profiling, also a metabolomics approach has been used to study Arabidopsis-insect interactions (Van Poecke et al., 2001). This study looked at all metabolites that could be detected in the headspace from caterpillar infested, mechanically damaged or undamaged Arabidopsis plants. Besides terpenoids and glucosinolate-degradation products, several other compounds were detected in the volatile blend emitted by P. rapae-infested Arabidopsis. These include fatty acid derived green leaf volatiles such as (Z)-3-hexenol, which are negatively associated with aphid growth (Hildebrand et al., 1993) and can function as herbivore repellents or attractants (Reddy and Guerrero, 2000; De Moraes et al., 2001), parasitoid attractants (Whitman and Eller, 1990) and in plant defense signaling (Bate and Rothstein, 1998); and other alcohols, aldehydes and ketones, one of which (penten-3-one, a.k.a. ethyl vinyl ketone) has known signaling properties in Arabidopsis (Alméras et al., 2003). However, none of these functions has been tested in Arabidopsis-insect interactions, with the exception of parasitoid attraction. C. rubecula wasps were more attracted to mechanically damaged than to undamaged Arabidopsis. A major difference in the volatile blends of these two odor sources is the abundance of green leaf volatiles in damaged plants, suggesting a role of green leaf volatiles in parasitoid attraction. However, P. rapae infested plants emitted less green leaf volatiles than mechanically damaged plants and were preferred by the wasps, indicating that other compounds, such as nitriles, terpenoids and methyl salicylate, play an additional role in parasitoid attraction. The latter also illustrates the difficulty of determining the role of individual compounds in complex volatile blends (Van Poecke et al., 2001).

2.7.1 Introduction of Proteinase Inhibitors

Besides Arabidopsis proteinase inhibitors being introduced into other species, several publications report foreign proteinase inhibitors introduced into Arabidopsis. De Leo et al., (1998; 2001) introduced the mustard trypsin inhibitor gene MTI-2 and studied its effect on caterpillar performance, which resulted in weight reduction and increased mortality of S. littoralis, increased mortality of Mamestra brassicae and 100% mortality of P. xylostella. Introduction of a modified cystatin from rice resulted in increased mortality of the slug Deroceras reticulatum (Walker et al., 1999).

2.7.2 Alteration of Terpenoid Profile

The function of terpenoids in direct and indirect defense has been demonstrated by using transgenic Arabidopsis plants expressing a strawberry linalool/nerolidol synthase that can produce both the monoterpene linalool from GDP and the sesquiterpene nerolidol from FDP. When the proteins were targeted to the plastids, this resulted in Arabidopsis plants overproducing mainly the monoterpene linalool and a little bit of the sesquiterpene nerolidol. This indicates that the mevalonate and MEP pathway may not be strictly separated. These plants were less attractive to M. perzicae than wild-type plant (Aharoni et al., 2003). When the proteins were targeted to the mitochondria (where FDP is used to generate ubiquinones involved in respiration), this resulted in Arabidopsis plants overproducing the sesquiterpene nerolidol, and to a lesser extend and only in some plants dimethylnonatriene (DMNT). The latter component is a known attractant of Phytoseiulus persimilis mites that prey on the herbivorous mite Tetranychus urticae (Dicke et al., 1990). Plants overproducing nerolidol or both nerolidol and DMNT were more attractive to P. persimilis than wild-type plants, demonstrating that nerolidol can function as a carnivore attractant as well (Kappers et al., 2005).

2.7.3 Alteration of Glucosinolate Profile

Mikkelsen and Halkier (2003) changed the glucosinolate profile of Arabidopsis by introducing CYP79D2 from cas-sava that catalyzes the formation of aldoximes from valine and isoleucine, the first dedicated step towards glucosinolate synthesis. CYP79D2 transgenic Arabidopsis produced valine- and isoleucine-derived glucosinolates not normally found in Arabidopsis. How this affects Arabidopsis-insect interactions was not studied.

2.7.4 Production of a Novel Phytoalexin

Cyanogenic glucosides are related to glucosinolates in that they are also amino-acid derived cyanogenic compounds that are activated upon tissue-disruption by enzymatic removal of glucose. Cyanogenic glucosides are normally not produced by cruciferous plants. Introduction of the three Sorghum bicolor genes required for synthesis of the cyanogenic glycoside dhurrin from tyrosine resulted in dhurrin synthesis in Arabidopsis and conferred resistance to larvae from the flea beetle Phyllotreta nemorum (Tattersall et al., 2001).

2.7.5 Production of a Novel Toxin

Photorhapdus luminescens is a symbiont bacterium of entomopathogenic nematodes (nematodes that infect insects). This bacterium produces several protein toxins that are toxic to caterpillars, including toxin A. Arabidopsis plants producing toxin A by insertion of a modified version of the tcdA gene from P. luminescens showed close to 100% mortality of M. sexta larvae, compared to about 15% on wild-type plants (Liu et al., 2003).

2.7.6 Induction of Anthocyanins

The last example of transgenic plants with enhanced resistance against insect herbivores comes from Arabidopsis plants that show constitutive overexpression of the Arabidopsis PAP1 transcription factor. This transcription factor regulates the biosynthesis of phenyl-propanoids including anthocyanins. PAP1 overexpressing plants showed slightly reduced feeding rates of generalist Spodoptera frugiperda but not of T. ni and no effect on larval weight for either lepidopteran species. Transgenic plants showed reduced fecundity in absence of herbivory, indicating a cost of constitutively enhanced defenses. However, the exact cause of increased resistance and reduced fecundity were unclear (Johnson and Dowd, 2004).

3.1.1 Wounding-Induced Responses

Wounding produces many different signals. First of all, damaged cells release their contents; secondly the damaged cell-wall releases signaling compounds such as oligosaccharides; and thirdly undamaged cells near the wound site will experience stresses such as pressure differences (reviewed by De Bruxelles and Roberts, 2001; León et al., 2001). The involvement of cell-wall components in defense resistance is illustrated by the enhanced resistance against aphid M. persicae in the Arabidopsis cev1 mutant. CEV1 is a cellulose synthase and mutation of CEV1 affects cell-wall formation. Enhanced resistance is possibly caused by a higher concentration of cell-wall derived elicitors triggering plant defenses (Ellis et al., 2002a/b).

The initial elicitors generated upon wounding activate an extensive signaling network triggering a diversity of responses. For example, in Arabidopsis wounding induces 1) signaling compounds such as the plant stress hormones jasmonic acid and ethylene; 2) the expression of a large number of genes; 3) biochemical and structural changes, including trichome density and volatile emissions, and 4) attraction of insects, such as parasitoid wasps (Rojo et al., 1999; Reymond et al., 2000; Van Poecke et al., 2001; Fäldt et al., 2003; Traw and Bergelson, 2003; Delessert et al., 2004; Devoto et al., 2005). Similarly, caterpillar-feeding induces 1) signaling compound such as jasmonates and ethylene; 2) gene-expression of a large number of genes; 3) biochemical and structural changes, including glucosinolate biosynthesis, trypsin inhibitor biosynthesis and volatile emissions; and 4) attraction of insects, such as parasitoid wasps (Van Poecke et al., 2001; Stotz et al., 2002; Clauss and Mitchell-Olds, 2004; Reymond et al., 2004; De Vos et al., 2005; Mewis et al., 2005). As previously mentioned, differences in mechanical wounding inflicted by herbivores with different feeding strategies are correlated with differences in induction of plant hormone levels, gene-expression, and parasitoid attraction (Van Poecke et al., 2001; De Vos et al., 2005). These parallels strongly suggest that wounding can account for many of the responses induced by herbivory. The responses mentioned above also show some differences between wounding and herbivory, which can be partially explained by lack of knowledge. For example, it seems likely that glucosinolate and trypsin inhibitor biosynthesis can also be induced by wounding and trichome density can probably be induced by herbivory but this has not been reported yet. However, some effects that have been studied in parallel for both treatments indicated differences between wound- and herbivory-induced responses of Arabidopsis. For example, Reymond et al. (2000) found that wounding induced the expression of many water stress-related genes, whereas caterpillar feeding did not. Conversely, caterpillar feeding induced the expression of several genes that were not induced by wounding, including defense hormone (jasmonates, see below) and glucosinolate biosynthesis genes, (Reymond et al., 2004). Additionally, caterpillar feeding resulted in the emission of volatile compounds such as methyl salicylate and terpenoids that were not induced by wounding (Van Poecke et al., 2001). These findings suggest that herbivore-derived elicitors are involved in elicitation of plant responses. However, it should be noted that the mechanical damage inflicted by herbivores is hard to mimic (Baldwin, 1990). First of all, the devices commonly used for wounding are much more blunt than the mandibles of a caterpillar and secondly, mechanical damage is often inflicted in a very different spatial and/or temporal pattern compared to herbivory, for example by damaging the plant only at one time point. Herbivore damage on the other hand gradually increases over time. Illustrating the difficulty of mimicking herbivory by mechanical damage, some of the genes reported by Reymond et al. (2004) induced by caterpillar feeding but not by wounding are reported to be induced by wounding in other studies (Reymond and Farmer, 1998; Reymond et al., 2000; Delessert et al., 2004), most likely reflecting differences in mechanical damage.

3.1.2 Herbivore-Derived Elicitors

To get a better understanding of which responses to herbivory are elicited by wounding and which are caused by other elicitors, an often used method is to apply herbivore-derived elicitors, for example in the form of regurgitant, to mechanically damaged plants. In Arabidopsis, application of caterpillar-derived regurgitant resulted in induced gene-expression (Berger et al., 2002; Reymond et al., 2004), parasitoid attraction (Van Poecke and Dicke, 2003) and jasmonate biosynthesis (Van Poecke, unpublished results). However, it is unclear to what extent compounds in the regurgitant come into contact with plant tissue during herbivory and to what extent different plant species are responsive to these compounds.

3.2.1 Jasmonates

3.2.2 Ethylene

3.2.3 Salicylates

3.2.4 Abscisic Acid

In the paragraph on JA-ET interactions, it was mentioned that caterpillar feeding induced the JA/AtMYC pathway (which is either part of or identical to the ‘JA only’ pathway) and represses the JA/ET pathway possibly through concerted actions of JA and ABA. Thus, ABA may play a mediating role in directing signaling downstream of JA in an antagonistic manner to ET. Are there more indications that ABA plays a role in Arabidopsis-insect interactions? The short answer is no, not for Arabidopsis. However, loss of ABA did result in reduced resistance against caterpillars in tomato (Thaler and Bostock, 2004). ABA is mainly known as a dehydration stress hormone and it is also known to interfere with defenses against pathogens (Mauch-Mani and Mauch, 2005). Interestingly enough, feeding by P. rapae suppresses the expression of many dehydration-induced genes compared to wounding (Reymond et al., 2000) From this one might conclude that insect-feeding suppressed ABA induced responses, which is quite the opposite of what has been mentioned before. So far, there have been no reports that ABA levels increase upon herbivory. Whether and, if so, how ABA influences Arabidopsis-insect interactions is therefore unclear.

3.2.5 Gibberellins

Gibberellins are known for their role in developmental processes. So far, only one study demonstrated a possible effect of gibberellins on Arabidopsis defense against insects, as gibberellin induced trichome formation in new leaves Ler synergistically with JA. This induction is inhibited by SA (Traw and Bergelson, 2003). However, as it is unknown whether herbivory induces gibberellin signaling, the relevance to inducible plant defense remains unclear.

3.2.6 Summarizing Signal Transduction

From all the information discussed, the picture arises of a signaling web where JA and other oxylipins play a central role, but where defenses can be fine-tuned by other hormones such as SA, ET and possibly others (figure 7). Jasmonates influence defense-responses to all kinds of insect herbivores, from phloem-feeders to tissue-munch-ers and from specialists to generalists. SA and ET mainly have an attenuating role on the induction of defenses against insect herbivores. Although these generalizations seem to be true for induction of defense responses, they may not be true for the effect on herbivore performance. It has been noted earlier that some defense responses, such as glucosinolate production, are mainly effective against generalists but not specialists. Similarly, ET seems to have a positive effect on the performance of generalists, likely at least partially due to reduced accumulation of JA-inducible glucosinolates caused by the antagonistic effects of ET, but not on the performance of specialists. Studies on the effect of JA and SA on herbivore performance have almost exclusively focused on generalist herbivores. The only exception is a report of increased performance of B. brassicae on coi1 mutants. It seems likely that performance of specialists will be less affected by induced defenses than performance of generalists.

An obvious question is why plants attenuate their inducible defenses by e.g. stimulating ET production upon herbivory? If this is merely a fine-tuning of defenses, one would expect that defenses should be optimized for specific attackers, which is obviously not the case as ET-signaling deficient mutants are more resistant. However, different combinations of signals induce different defenses. For example, although some defense responses will be down-regulated by the combination of JA and ET, others are up-regulated by this combination and yet others are up-regulated by JA irrespective of the presence of ET and so on. This is especially well-demonstrated with respect to induction of different glucosinolates. Thus, different combinations of signaling pathways induce different defense responses and these defense responses are effective against different kinds of attackers. Plants may not be able to defend themselves optimally to one kind of attacker (e.g. insect herbivores) without becoming more susceptible to other kinds of attackers (e.g. biotrophic pathogens). Such a trade-off between defenses could explain why plants are not optimally defended. The interactions between the effects of different attackers on plant defense are discussed next.

3.2.7 Responses to Multiple Attackers: Cross-Talk Among Signaling Pathways

Plant defense responses clearly depend on the type of attacker. These responses can roughly be divided in SA-dependent responses upon attack by biotrophic pathogens, JA/ET-dependent responses upon attack by necrotrophic pathogens, and ‘JA-only’ responses upon attack by insect herbivores. The tight and often negative interactions between different signaling pathways within the signaling web make it likely that the attack of one species, such as a SA inducing pathogen, will affect the plants responses to subsequent attacks by a next species, such as a caterpillar against which mainly JA-inducible defenses are effective. Indeed, several studies have investigated such interactions and found evidence for tradeoffs. Cui et al. (2002) showed that previous infection of Arabidopsis with virulent P. syringae pv maculicola (Psm) resulted in increased performance of T. ni larvae. As virulent P. syringae mainly induces the SA-signaling pathway (Glazebrook, 2005), this seems to fit well with the model of antagonistic SA-JA interactions. Interestingly, although performance of T. ni is generally lower on SA-lacking genotypes such as NahG and npr1, virulent P. syringae still induced increased performance of T. ni in these mutants, indicating that this negative effect on Arabidopsis defenses is actually SA independent. This response was, however, dependent on PAD4. A possible explanation might by that P. syringae, known to induce beside SA, also JA and ET levels, elicits JA/ET dependent signaling. JA/ET signaling is known to work antagonistically to ‘JA only’ responses and is dependent on PAD4. It should be noted that this increased susceptibility seems a rather subtle effect, as researchers of the same group failed to reproduce these results later, possibly due to slightly different environmental conditions (Cui et al., 2005). Another unexpected result was that infection with avirulent Psm induced increased resistance to T. ni and increased susceptibility to Psm (Cui et al., 2002). This was subsequently demonstrated to be caused by the JA mimic coronatine, which is produced by these pathogens as a phytotoxin and likely aimed at reducing SA-dependent defenses which are effective against Psm (Cui et al., 2005). So, apparently there are two, counteracting processes at work. On the one hand, Psm induces a PAD4 dependent pathway that results in increased susceptibility to caterpillars. On the other hand, Psm induces a coronatine-dependent pathway that triggers JA-inducible defenses and results in increased resistance against caterpillars and susceptibility against Psm (Cui et al., 2005). To make matters even more complex, De Vos (2006) demonstrated that previous infestation with P. rapae induces resistance against avirulent Pst. This is rather unexpected as the work from Cui et al. (2005) demonstrated that Psm actually tries to undermine plant defenses by stimulating JA-responses. Similar to the results with virulent Psm and T. ni, the results reported by De Vos (2006) were independent of NPR1, SID2 or EDS5. However, they were also independent of JAR1, COI and EIN2, showing that neither SA, JA nor ET is likely to be involved. Additionally, De Vos (2006) showed that previous infection with P. rapae also induces resistance against turnip crinkle virus and the bacterial pathogen Xanthomonas campestris pv armoraciae, pathogens against which SA-inducible defenses are effective, but not against the fungal pathogen Alternaria brassicicola, against which JA/ET-inducible defenses are effective. It is also induced resistance against subsequent attack of P. rapae itself. The mechanisms behind these findings are unclear, but certainly reward more research.

Cleary, the effect of multiple attacks by different pathogens and pests influence each other in complex ways that may include the plant hormones SA, JA and ET, but also other, largely unknown pathways. This also might indicate why Arabidopsis apparently fails to optimally launch defenses against a certain attacker as discussed in the previous section: as multiple attacks by different species are likely to be common in the field, a plant optimizing its defenses against one pathogen or pest may leave it too vulnerable to attack by others.

4.1.1 Plant Fitness

Only a few studies did look at plant fitness. Examples include the positive effect of attracting parasitoids on fruit number and weight of total seed production in P. rapae infested Arabidopsis, but no evolutionary questions were addressed (van Loon et al., 2000).

In a field trial, Murren et al. (2005) found indications for a selection pressure on tolerance but not on resistance exerted by fungus gnats and aphids. However, there was no control treatment where herbivory was prevented, making these results hard to interpret. For example, herbivores may have had a preference for larger plants that before the onset of herbivory had a greater potential for seed production, resulting in a similar seed production by larger, infested plants and smaller, uninfested plants.

4.1.2 Effect of Glucosinolates, Trichomes and Tolerance on Plant Fitness

In an extensive field study, Mauricio and coworkers addressed several evolutionary questions with respect to the role of tolerance, glucosinolates and trichomes on fitness in natural populations of Arabidopsis in North Carolina. They concluded that in Arabidopsis: tolerance and resistance to insect herbivores are not mutually exclusive - a result confirmed also for herbivory by rabbits, although tolerance and resistance mechanisms are likely to differ between these studies (Weinig et al., 2003). Additionally, they concluded that herbivory exerts a stimulatory selective pressure on glucosinolate levels, trichome densities and tolerance; trichome density and glucosinolate levels are positively correlated; there is no correlation between the defense mechanisms on one hand and tolerance on the other; there are costs associated with both defense mechanisms and with tolerance; and a costs/benefit analyses resulted in a weak balancing selection for glucosinolates in the presence of herbivores, a disruptive selection on tolerance (directed either towards full tolerance or no tolerance, depending on the initial level of tolerance of the genotypes), and a negative selection pressure for trichome density (Mauricio and Rausher, 1997; Mauricio et al., 1997; Mauricio, 1998). The latter raises the question why trichomes are present in these populations. As glucosinolate levels and trichome densities are positively correlated, this may indicate genetic linkage between glucosinolate level and trichome traits, and such linkage disequilibrium can result in positive selection on glucosinolate levels that also increases trichome density. However, it could also be that trichome density is not at an evolutionary equilibrium in those populations and eventually trichomes may be eliminated (Mauricio and Rausher, 1997). The latter explanation demonstrates another problem associated with phenotypic characterization of selection: evolution takes place over long periods of time and a snapshot analyses may not reflect past selection pressures.

4.2.1 Evolution of Glucosinolate Biosynthesis and Degradation Genes

As mentioned in part II, dissection of glucosinolate biosynthesis and degradation pathways revealed a limited set of genes affecting elongation and modification of methionine side-chains (reviewed by Kliebenstein et al., 2005; see figure 4). For example, the GS-ELONG locus, involved in elongation of the methionine side-chains, consists of a three-member gene-family formed by gene-duplication and enzyme activity analysis shows functional divergence. MAM1 catalyzes two cycles of elongation, resulting in C4 side-chains, MAM2 catalyzes one cycle of elongation, resulting in C3 side-chains, and MAML catalyzes multiple cycles of elongation. Additionally, MAM2 activity was associated with higher total aliphatic glucosinolate. Sequence analyses revealed a balancing selection on MAM2, likely due to its impact on aliphatic glucosinolate level. The higher levels of aliphatic glucosinolates associated with MAM2 confer resistance to the generalists S. exigua and T. ni, but not to the specialist P. xylostella. As there were no obvious costs to MAM2 expression, the authors speculated that differences between the effects on generalist herbivores (demonstrated resistance) and specialists (possible attraction/feeding stimulation) may account for this balancing selection (Kroymann et al., 2003).

The GS-AOP locus is involved in side-chain modification. Sequence, gene-expression and biochemical analyses revealed three AOP genes and a pseudogene in three different alleles of the GS-AOP locus found in 21 Arabidopsis accessions (Kliebenstein et al., 2001c). Again, the AOP genes are the result of gene-duplication. Two genes AOP2 and AOP3 have been functionally characterized and produce alkenyl and hydroxyalkyl glucosinolates respectively. Interestingly, AOP2 and AOP3 expression appears to be mutually exclusive, which for AOP2 is associated with polymorphisms in the coding region in a few accessions but with differences in gene-expression in others. Difference in AOP3 activity is caused by differences in gene-expression rather than polymorphisms in the coding region. Variation in the GS-AOP locus thus resulted in three classes of accessions, one class that produces alkenyl glucosinolates associated with AOP2 activity, one class that produces hydroxyalkyl glucosinolates associated with AOP3 activity and one class that shows no side-chain modification associated with inactivity of both AOP2 and AOP3.

A third locus is the eptihiospecifier ESP locus. As mentioned before, plants with a functional ESP gene degrade glucosinolates to nitriles rather than isothiocyanates. Sequence analyses revealed three classes of ESP alleles: functional alleles, non-functional alleles due to differences in gene-expression, and one non-functional allele due to a truncated protein. Nitriles are less effective in resistance against T. ni, and thus EPS activity seems to decrease rather than increase direct defenses (Lambrix et al., 2001). However, nitriles may attract parasitoids and thus contribute to indirect defenses (Van Poecke et al., 2001). Additionally, isothiocyanates are known attractants of specialist herbivores and producing nitriles instead of isothiocyanates may be a way to avoid detection by herbivores (Lambrix et al., 2001)

4.2.2 Evolution of Trichome Development and Proteinase Inhibitor Genes

The studies on glucosinolate production indicate the importance of gene duplication in evolution of plant defense mechanisms against herbivores. The importance of gene duplication is supported by studies on other defense mechanisms, such as proteinase inhibitors. A study by Clauss and Mitchell-Olds (2003) found indications of functional divergence between two paralogues of a trypsin inhibitor gene (ATTI1 and ATTI2); extremely low levels of polymorphism for both; and indications of a selective sweep for ATTI2. An extension of this study found four additional ATTI genes in the same locus, one gene being inactive (Clauss and Mitchell-Olds, 2004). Again, indications of functional divergence was found (based on both amino-acid sequence and on transcriptional differences, which were correlated), although the genetic linkage between the genes is likely to hamper functional divergence (Clauss and Mitchell-Olds, 2004). In these studies, the effect of different paralogues on plant-insect interactions was not studied.

Trichome formation is another Arabidopsis insect-defense mechanism of which evolution was studied. Hauser et al. (2001) found two diverged sequence clades of the GL1 locus. However, sequence diversity did not correlate to trichome densities and only weak evidence for deviation from neutrality was found.

Together, these studies have demonstrated the importance of duplication events on evolution of Arabidopsis defense traits related to plant-herbivore interactions, similar to plant-pathogen interactions. Moreover, indications for balancing selection and selective sweeps have been found. Not surprisingly, variation in traits can be caused by both differences in gene-regulation and in coding-sequence. The studies on evolution of glucosinolate defenses are particularly interesting as they include results on the ecological effects and as such give an indication of which selection pressures may be involved. From this, it appears that specialist and generalist herbivores exert different, sometimes contrasting selection pressures on defense traits.