Introduction

Plants have evolved a broad range of defenses against both mammalian and invertebrate herbivores. If a plant secondary metabolite (PSM) confers resistance to multiple herbivores, then the cost of defense can be reduced. However, there are few examples of positively correlated defenses in plants against herbivores of different types, such as generalists and specialists (Leimu and Koricheva 2006) or mammalian and insect herbivores (Rousi et al. 1997; Mutikainen et al. 2002; Puustinen et al. 2004). Because the challenges of multiple herbivores can place constraints on the evolution of plant defenses (Rausher 1996), the chemical and genetic basis of cross-resistance is important.

Formylated phloroglucinol compounds (FPCs) may be responsible for cross-resistance to mammalian and insect herbivores of eucalypts (Eucalyptus l’Héritier and Corymbia Hill and Johnson, Myrtaceae). These compounds are important determinants of palatability to a range of specialist and generalist mammalian herbivores, both captive and free-ranging (Lawler et al. 1998, 2000; Wallis et al. 2002; Moore et al. 2004b; Scrivener et al. 2004; Moore and Foley 2005). The physiological mechanism underlying aversion to FPCs has also been elucidated, making the system one of the best understood of plant–mammal interactions (Lawler et al. 1998). A subgroup of FPCs, known as sideroxylonals, also affects the feeding of insect herbivores in the laboratory (Floyd and Foley 2001), but evidence of deterrence in wild insects is needed to support the hypothesis that FPCs contribute to cross-resistance to mammalian and insect herbivores. We present here the first field demonstration that concentrations of sideroxylonals are related to food plant selection by wild insects, in this case a species of Christmas beetle (Anoplognathus montanus Macleay; Coleoptera: Scarabaeidae), which is a specialist herbivore of eucalypts.

There is substantial variation in the concentrations of FPCs among eucalypt trees within populations (Lawler et al. 2000; Wallis et al. 2002; Moore et al. 2004b). This can create significant habitat patchiness at ecologically relevant scales (Andrew et al. 2007), and understanding how this variation arises requires comprehension of the heritability of the trait. Narrow-sense heritability, the proportion of variation within populations that is due to additive genetic causes, is of crucial importance because it determines the ability of a species to evolve in response to both natural and artificial selection. The heritability of foliar sideroxylonal concentration has been estimated in a natural population of Eucalyptus melliodora using a marker-based method (Andrew et al. 2005), but comparisons have not yet been made with traditional common-garden estimates. Since genetic relatedness is not inferred from molecular markers and sample sizes are less likely to be limited by expense, genetic trial experiments in a common environment (often called common-garden experiments) can offer greater precision in estimates of quantitative genetic parameters. Such estimates from genetic trial experiments are more appropriate than those from natural populations for predicting likely gains from breeding programs, which are aimed at genetic improvement of useful traits in tree plantations. Common-garden experiments also allow variation among populations to be studied without the confounding effects of site differences (O’Reilly-Wapstra et al. 2002; Moore et al. 2004b). Phenotypic and genetic correlations measured in progeny can also provide insight into the evolutionary constraints on sideroxylonal concentrations.

We addressed the relationship between foliar sideroxylonal concentrations and insect herbivory and studied the quantitative genetics of sideroxylonal variation in Eucalyptus tricarpa (LAS Johnson) LAS Johnson and K.D. Hill. This was, until recently, a subspecies of E. sideroxylon (Hill and Johnson 1991), in which mammal feeding studies have been conducted (Lawler et al. 2000). The two species are chemically similar, with 0–23 mg g−1 DW sideroxylonal (mean 10.9 mg g−1 DW) in E. sideroxylon (Lawler et al. 2000; Floyd and Foley 2001) and 0–47 mg g−1 DW sideroxylonal (mean 13.5 mg g−1 DW) in E. tricarpa (this study). Since they also share mammalian and insect herbivores, the influence of foliar defenses on herbivory is likely to be sufficiently similar to provide a reasonable grounds for inferring cross-resistance. Studying a field trial testing open-pollinated progenies from a range of natural provenances, we asked the following questions:

  1. 1.

    Do Christmas beetles preferentially attack E. tricarpa trees that have low foliar concentrations of sideroxylonal?

  2. 2.

    Does significant narrow-sense heritability of variation in foliar sideroxylonal occur in E. tricarpa? Are heritability estimates similar to marker-based estimates from a wild E. melliodora population?

  3. 3.

    Do E. tricarpa populations display genetic differentiation in foliar sideroxylonal concentrations?

  4. 4.

    Do phenotypic and/or genetic correlations occur among foliar sideroxylonal, growth traits, and insect damage?

Materials and methods

Experimental design and sampling

Eucalyptus tricarpa is a widespread species occurring in the low to medium rainfall zone in Victoria and New South Wales (NSW), Australia (Brooker and Kleinig 1999). This species is closely related to E. melliodora, in which sideroxylonal is highly heritable (Andrew et al. 2005). E. tricarpa is being developed for forestry in areas of low rainfall. The E. tricarpa provenance/progeny trial sampled in this study was set up by Forests NSW, near Culcairn, NSW (146°56′E, 35°43′S) in August 2000, as part of the Australian Low Rainfall Tree Improvement Group (ALRTIG) program (Harwood et al. 2001). The trial tested 108 open-pollinated families, representing 15 natural populations in Victoria and NSW (Table 1). Eucalypts have mixed mating systems, with typical outcrossing rates of 70% (Eldridge et al. 1993). Open pollination leads to families consisting of mixtures of half-siblings and selfed or outcrossed full siblings. Trees were arranged in four replicates, with each family represented in each replicate by one five-tree row plot. Trees were planted 1.8 m apart with 4 m between rows and a buffer two trees deep on the outside edges of the trial. The experimental area covered 1.8 ha, and an adjoining 0.6-ha area was planted with surplus stock of some of the families. It was surrounded on all sides by pasture, with the nearest Eucalyptus trees approximately 200 m away.

Table 1 Geographical and climatic variables of populations represented in an E. tricarpa genetic experiment. Longitude, latitude, and approximated altitude were obtained from seedlot collection reports, and climatic variables were predicted using WORLDCLIM data grids
Full size table

Our sampling was conducted in November 2002, when the plants were producing foliage that was intermediate between juvenile and fully mature morphologies. The transition from juvenile to adult foliage is not as dramatic in E. tricarpa as in many other species of Eucalyptus. Leaf shape differs only slightly and foliar chemistry is stable across years (R. L. Andrew, unpublished observations). We selected 9 populations covering the natural distribution of the species and sampled from 3 to 9 families from each population (9 families where possible), giving a total of 67 families selected. In each block, we sampled the two westernmost trees of each plot planted for the selected families. Where trees were stunted or dead, the next suitable tree in the plot was sampled. From the southern side of each tree, approximately 50 g of leaf material was sampled, frozen, and stored at −20°C. We sampled one replicate at a time, so that time of day was not confounded with population or family. Height, diameter at 1.3 m (DBH) of largest stem, and the sum of the squared diameters at 1.3 m of all stems (D 2) were measured for all trees by Forests NSW in January 2004.

Insect damage

At the time of sampling, the experimental plants were under attack by Anoplognathus montanus, a local, native Christmas beetle. A simple presence/absence score was used to assess Christmas beetle attack, as the pattern of damage and beetle presence was very clear: a tree was either heavily infested or free from damage. This is consistent with previously observed patterns of damage by Anoplognathus spp. (Edwards et al. 1993). Paired comparisons were made to test the relationship between beetle infestation and sideroxylonal concentrations. Each pair consisted of one infested tree and one unaffected tree from the same family and replicate (i.e., the same plot), and sideroxylonal was measured using high-performance liquid chromatography (HPLC) for both trees (see below). The 23 pairs sampled were from 16 families, 12 of which were among those included in the quantitative genetic analysis.

In addition to Christmas beetles, defoliation and damage to photosynthetic leaf area by other agents was also apparent. A range of folivorous and sap-sucking insects, such as chrysomelid leaf beetles (Paropsis spp.) and their larvae, sawfly larvae (Perga sp.), leaf-blister sawflies (Phylacteophaga sp.), and Eucalyptus weevils (Gonipterus scutellatus), may have contributed to the observed damage. Each tree that was sampled for chemical analysis was given a score for overall insect damage. Trees were scored from one to four for minimal defoliation, light defoliation, moderate defoliation, and heavy defoliation, respectively. Heavy defoliation (damage score 4) resulted in visible bare branch ends over the whole of the canopy. Our damage classes were similar to those employed by Raymond (1995), but differed in the order of the ranking (from least to most damage) and in the description of overall damage (rather than of that produced by a single insect species). The damage score affected by the presence of Christmas beetle because of their contribution to overall leaf damage, but also describes variation in leaf damage from other sources, which were difficult to distinguish.

Chemical analysis

The leaves were freeze–dried and ground to pass through a 1-mm screen in a Tecator Cyclotec mill. A subset of 105 samples (including 46 paired trees selected for the Christmas beetle test) was assayed for sideroxylonal using the HPLC technique described by Wallis et al. (2003). Sideroxylonal A and sideroxylonal C are found in constant proportions in this solvent system (Moore et al. 2004a) and their concentrations were combined to give the concentration of total sideroxylonals, which will be referred to as sideroxylonal henceforth. We collected the visible and near-infrared spectra (400–2,500 nm) of all samples from dry, ground leaf following Moore et al. (2004b). Calibration equations were developed using modified partial least squares regression using WINISI II (Infrasoft International, Port Matilda, PA, USA) and following standard procedures (American Society for Testing and Materials 1995). The optimal model used the 1,100- to 2,498-nm region, standard normal variate and detrend scatter correction, and a math treatment of 1881 (i.e., first derivative treatment, with a gap width of 8 nm and smoothing over 8 data points; see American Society for Testing and Materials 1995). The model explained 97.8% of foliar sideroxylonal variation (mean 17.9 mg g−1 dm; range 0.0–47.6 mg g−1 dm) in the calibration set and had a standard error of cross-validation of 2.4 mg g−1 dm.

Statistical analysis

Statistical analyses were conducted using GenStat version 8 (VSN International Ltd, Hemel Hempstead, UK). The relationship between sideroxylonal and presence of Christmas beetle was investigated by choosing paired damaged and undamaged trees from the same plot. Beetle-damaged trees were paired with the nearest available undamaged tree within the same plot. This approach not only avoided the large number of zero Christmas beetle damage scores that would result if all trees assessed for sideroxylonal had been included but also controlled for genetic background (family) and location in the trial. The difference in foliar sideroxylonal concentration between the 23 affected trees and their paired unaffected trees was tested using a two-sided, one-sample t-test.

Estimates of narrow-sense heritability were computed for foliar sideroxylonal concentration, damage score, and the growth traits using pooled variance components estimated by restricted maximum likelihood analysis (REML). Variance components were estimated by fitting a mixed model, with replicate and population as fixed effects. Family within population was fitted as a random effect, as well as plot within replicate, to account for the remaining plot-level variance. Variance components estimated for families within populations (σ 2 f ), plots within replicates (σ 2plot ), and residual error (σ 2 e ) were summed to give σ 2 p , the total phenotypic variance within populations and replicates (Williams et al. 2002, p. 103). The use of a single pooled family variance component leads to an estimate of the average heritability among populations. Standard errors were calculated from the family and residual variance components using VFUNCTION (Payne 2005). All available trees were used for heritability estimation for each trait. All 108 families represented in the trial were included for height and DBH, which were measured for 1,772 and 1,692 trees, respectively. In contrast, only 67 families from nine populations were available for estimation of heritability for sideroxylonal and damage (465 trees).

Additive genetic variance in phenotypic traits can be estimated using the expected covariance of half-siblings and adjusted to account for the outcrossing rates, according to the common procedure used for trees with mixed mating systems (Eldridge et al. 1993; Williams et al. 2002). Within-population, narrow-sense heritability (h 2) specific to this environment was calculated using the formula,

$$ h^{2} = \frac{{\raise0.7ex\hbox{$1$} \!\mathord{\left/ {\vphantom {1 r}}\right.\kern-\nulldelimiterspace} \!\lower0.7ex\hbox{$r$} \times \sigma ^{2}_{f} }} {{\sigma ^{2}_{p} }}, $$

where r is the mean relatedness within families, which is typically assumed to be 0.4 in Eucalyptus (Eldridge et al. 1993). Instead of four times the family variance, additive genetic variance is estimated as 2.5 times (or 1/0.4 times) the family variance. As with all half-sibling designs, this model assumes that inbreeding and epistasis are absent. Maternal effects may be important in maternal half-sibling families, but they are not commonly observed in Eucalyptus, based on controlled cross experiments (Lopez et al. 2003). In addition, this design cannot fully control for dominance effects, which may inflate the within-family covariance in plants with mixed mating systems.

Additive genetic correlations (subject to the same assumptions as for heritability estimates) between pairs of traits were estimated from plot means using multivariate REML in GenStat 8 and tested using likelihood ratio tests (Payne 2005). Replicate and population were fixed effects, while family and plot were random effects in the model. Standard errors of genetic correlations were estimated using a standard Taylor expansion formula, implemented in GenStat. Genetic correlations were estimated using only populations for which both growth and sideroxylonal had been measured.

The effects of geographic and climatic variables on phenotypic traits were tested using the best linear unbiased estimates (BLUEs) of population means obtained from the REML mixed model analysis. Geographic variables considered were latitude, longitude, and approximate altitude. Annual means of temperature and precipitation for each population were obtained using the WORLDCLIM climatic grids (Hijmans et al. 2005) in ArcView GIS 3.3 (Environmental Systems Research Institute, Redlands, CA, USA). Since minimum temperatures have been found to be related to foliar sideroxylonal concentrations in Eucalyptus microcorys (Moore et al. 2004b), this variable was also included. Univariate regression and Spearman’s rank correlation were used to test the relationships between geographic or climatic variables and the BLUEs for each trait. Since multiple tests were carried out, sequential Bonferroni corrections were used to adjust P values (Rice 1989).

Results

At the time of investigation, Christmas beetles occurred on 7.7% of trees sampled. The proportion of trees affected by these beetles varied from 0% to 34% among populations (Table 2), with 0–100% of trees within families under attack. Four populations were unaffected by Christmas beetles, including both NSW populations and the southernmost population, from Lorne (Fig. 1). Comparison of paired affected infested with unaffected trees showed that foliar sideroxylonal concentration was, on average, 10.4 mg g−1 dry weight lower in trees under attack (t = 3.81 on 22 df; P < 0.001).

Table 2 Population characteristics and traits measured in an E. tricarpa genetic experiment. Best linear unbiased estimates (BLUEs) of population effects for sideroxylonal, damage score, diameter at 1.3 m (DBH) and the sum of the squared diameters of all stems (D 2) were estimated using REML mixed model estimation, accounting for replicate (fixed), plot within replicate (random) and family within population (random). The grand mean and average standard error of differences among population means were estimated using the same model
Full size table
Fig. 1
figure 1

Percentage of progeny from nine E. tricarpa populations attacked by Christmas beetles during an outbreak in a common garden experiment. Sample sizes (the number of trees from each population) are shown next to each bar

Full size image

The mean concentrations of foliar sideroxylonal in populations, predicted by the REML analysis, ranged from 12.8 to 33.8 mg g−1 dry weight (Table 2). Leaves from the Clunes, Mt Nowa Nowa and Bodalla populations were all consistently high in sideroxylonal, showing that high-sideroxylonal populations can be found throughout the range of the species. Significant differences were also observed among populations for damage score (P < 0.001), and growth traits also varied among populations (P < 0.001). Populations from Gippsland and NSW generally grew faster than those from Central and Goldfields regions (Table 2).

Genetic analysis revealed significant heritability for each trait (Table 3). Sideroxylonal and damage demonstrated heritabilities of 0.60 and 0.34, respectively. The heritabilities of height (0.44) and DBH (0.30) were high for eucalypts and higher still when estimated only for the populations in which sideroxylonal was measured (0.53 and 0.34, respectively; Table 3). Replicates differed significantly for damage and growth traits, but not for sideroxylonal concentration. The two growth traits were highly phenotypically correlated, and damage was negatively correlated with both growth and foliar sideroxylonal concentrations (Table 4). However, there was no phenotypic correlation of foliar sideroxylonal with either growth trait. The additive genetic correlation of height and DBH was high, but sideroxylonal concentration and damage score were not significantly genetically correlated with any trait (Table 4). Predicted population means were negatively correlated for insect damage and plant height (r = −0.734, P = 0.024). The correlation of predicted population means for sideroxylonal and damage score was marginally significant (r = −0.659, P = 0.054).

Table 3 Summary of quantitative genetic analysis using univariate REML
Full size table
Table 4 Phenotypic (below diagonal) and additive genetic (above diagonal, with standard errors in parentheses) correlations for defense and growth traits in E. tricarpa. Average narrow-sense heritability (with standard errors in parentheses), estimated using the same provenances are shown italicized and on the diagonal. The genetic correlation of height and DBH was estimated using Fisher optimisation to obtain model convergence
Full size table

When the effects of source population geographic or climatic variables on traits were tested, no rank correlations and only one univariate regression was significant at the 0.05 level prior to Bonferroni corrections. However, none was significant or close to significance (P < 0.1) after corrections were made for multiple comparisons.

Discussion

Cross-resistance of insect and mammalian herbivores

This study provides field evidence that cross-resistance to insect and mammalian herbivores of Eucalyptus may be conferred by foliar sideroxylonals at typical concentrations. Trees with low concentrations of foliar sideroxylonals were chosen by Christmas beetles over nearby, genetically related trees with higher concentrations. At concentrations similar to those in the present study, sideroxylonals and other FPCs have also been shown to reduce intake by a range of mammalian herbivores (Lawler et al. 2000; Wallis et al. 2002; Moore et al. 2005). The congruence of these results with those demonstrating sideroxylonal-related resistance to herbivory by common ringtail possums (Pseudocheirus peregrinus) in a closely related and chemically similar species (E. sideroxylon, Lawler et al. 2000) suggests that cross-resistance is likely in at least these two eucalypt species, but probably also in others. The variation observed in E. tricarpa foliar sideroxylonal concentrations is most likely constitutive, as there is no evidence of induced defenses in Eucalyptus (Henery 2006). While previous tests of foliar defense induction may have failed to detect a small response, a large change in concentration would be needed to produce the 10.4-mg g−1 difference between the trees affected and unaffected by beetles in the present study.

In contrast to studies of annuals and herbaceous perennials, the chemical basis of defense has been identified in few tree-herbivore systems and can be complex (Haukioja 2003). That the same chemicals are responsible for cross-resistance to both mammalian and insect herbivores at concentrations commonly found in leaves is ecologically significant. Although broad classes of secondary chemicals, such as phenolic glycosides and terpenes, are active against both mammalian and arthropod herbivores, we are not aware of any other examples of specific chemicals known to confer resistance to both mammalian and insect herbivores. The strength of natural selection, both historical and contemporary, on FPCs in eucalypts is still unknown, but their evolution may have been favored by their effectiveness as defensive chemicals against diverse specialist and generalist herbivores.

Adult Christmas beetles are specialist herbivores of eucalypts and can cause severe damage to trees in plantations (Carne et al. 1974). They appear to have increased in numbers in rural areas, probably because their larvae feed on roots underground and can benefit from the increased productivity of pasture and crops (Carne et al. 1974). Christmas beetle numbers are extremely variable and severe outbreaks can result in nearly 100% defoliation of some trees (Pryor 1953; Edwards et al. 1993). Laboratory bioassays have shown that variation in sideroxylonal concentration in leaves can determine the feeding preferences of Christmas beetles (Floyd and Foley 2001) and this study provides evidence that wild Christmas beetles select trees on the basis of foliar sideroxylonal concentration.

In the past, Christmas beetle herbivory has been related to essential oil concentration and in particular the foliar concentration of 1,8-cineole (Edwards et al. 1993). However, resistant E. grandis clones were no higher in 1,8-cineole concentration than susceptible clones (Johns et al. 2004). Furthermore, experiments manipulating the concentration of both 1,8-cineole and sideroxylonal in leaves have suggested that FPCs have a larger effect on Christmas beetles (Floyd and Foley 2001). Foliar concentrations of cineole and sideroxylonal tend to be correlated in natural populations (Moore et al. 2004a; Andrew et al. 2005), which explains why previous studies have identified cineole as the primary deterrent. This association allows mammalian herbivores to use cineole, which is volatile and has a strong odor, as a guide to sideroxylonal concentrations (Lawler et al. 1999). It appears that terpenes may have a similar role as an olfactory cue in feeding by Christmas beetles (Floyd and Foley 2001).

In addition to resistance, plants may employ alternative or additional strategies to limit the costs of herbivory, such as tolerance (Fornoni et al. 2004) or escaping herbivory through faster growth (Kursar and Coley 2003). We found some support for the latter in E. tricarpa, as overall insect damage was related to both sideroxylonal concentration and growth. However, the relationship of damage with growth may also be due to the costs imposed by herbivore damage, since the growth measurements were made a year after the observed damage was done. A negative effect of defoliation on subsequent growth rates was observed by Raymond (1995) in trees damaged by chrysomelid leaf beetles (Chrysoptharta bimaculata), which also chose smaller trees. We were unable to separate these effects, as growth measurements were made once only and insects were not experimentally excluded.

Heritability of sideroxylonal and herbivore defense

Foliar sideroxylonal concentration is highly variable in E. tricarpa, and our estimate of the narrow-sense heritability of this trait is high. Previous progeny experiments in Eucalyptus have significant genetic differentiation in FPCs among populations, but within-population genetic variation in sideroxylonal has not yet been shown (O’Reilly-Wapstra et al. 2002, 2004). The present study supports the finding of significant additive genetic variation of foliar sideroxylonal, with narrow-sense heritability estimates between 0.55 and 0.89, in E. melliodora using a genetic marker-based approach (Andrew et al. 2005). Common-garden experiments are useful because they provide greater precision than those in natural populations. However, the coefficient of relationship of 0.4 among the individuals of open-pollinated families used for estimating heritability in this study is only an approximation (Griffin and Cotterill 1988; Eldridge et al. 1993, p. 174). The degree of relatedness within open-pollinated Eucalyptus families is affected by the levels of selfing and correlated paternity. Studies of progeny arrays using molecular markers have shown that these vary widely both among species (McDonald et al. 2003; Jones et al. 2005) and within species (Burgess et al. 1996; Butcher and Williams 2002). Direct estimates are not yet available for E. tricarpa, and the relationships among progeny within and among families may be influenced by additional factors (Squillace 1974). The true level of relatedness in field-grown progeny may therefore differ substantially from the estimate we have used, but it is clear that heritabilities of sideroxylonal concentration and insect damage in our trial are moderate to high. Our estimates of heritability, especially for growth traits, could be inflated by inbreeding depression, since the open-pollinated design does not control for differences in mating system parameters among families. The heritabilities of height and DBH are surprisingly high, as values of about 0.2 for these traits are more typical of Eucalyptus (Eldridge et al. 1993, p. 175). Estimates of heritability based on a single site may be upwardly biased due to the inclusion of a genotype × environment interaction component in the estimate of additive genetic variance (Tibbits and Hodge 1998; Costa e Silva et al. 2006). There have been few studies on chemical defenses carried out across sites (Bowers et al. 1992; Rosner and Hannrup 2004) and replicating this study at multiple sites will be a high priority for future research on FPCs.

The resistance of eucalypts to insect herbivores is often highly heritable (Raymond 1995; Jones et al. 2002; Jordan et al. 2002; Rapley et al. 2004). Insect damage was heritable in E. tricarpa and phenotypically correlated with sideroxylonal concentration, although the genetic correlation was not significant. The lack of additive genetic correlation between these traits may imply independence, and that different genes are controlling the traits or may have been due to insufficient statistical power. Genetic variation in FPC production may also explain the difference in susceptibility to Christmas beetles observed between interspecific hybrids and pure E. grandis (Shepherd 2000).

Implications for coevolution and genetic improvement

The heritability of a trait is important for understanding the evolutionary dynamics of the trait. While selection pressures are difficult to measure in long-lived trees, we now have some evidence of the evolutionary potential of foliar sideroxylonal concentrations in E. tricarpa. Our results suggest that such concentrations may be capable of responding to human-induced changes in herbivore pressures, such as increasing Christmas beetle or mammalian herbivore abundance in agricultural landscapes. As yet, it is unclear which groups of herbivores might act as the strongest selective agent on sideroxylonal, because the ability of plants to tolerate herbivory by different agents may vary (Kotanen and Rosenthal 2000). The response to breeding for increased concentrations of sideroxylonal will depend on the heritability and phenotypic variation of this trait, which our results suggest are high in E. tricarpa. Increasing leaf sideroxylonal concentration through breeding may increase the economic viability of plantation forestry by reducing the impact of insect and mammalian herbivores.

Population differentiation in heritable traits is of interest because it may result from different selection regimes, among other causes (Merila and Crnokrak 2001; McKay and Latta 2002). High foliar sideroxylonal concentrations and low variability would be expected in populations with high herbivore pressure, but other environmental factors, such as nutrient availability and rainfall, could also affect the costs and benefits of defense. Moore et al. (2004b) related differences in sideroxylonal concentrations among natural populations of E. microcorys to temperature regimes, but could not separate the effects of genetic differentiation from those of environmental differences between the sites, as experiments in controlled conditions were lacking. E. tricarpa populations displayed genetic differentiation in mean sideroxylonal concentrations, although there was no consistent pattern in differences among geographical regions, suggesting that population-level processes are important in this species, rather than broad-scale climatic variation. The lack of any strong relationship between climatic variables at the source populations and sideroxylonal or damage in this species contrasts with those present in E. microcorys (Moore et al. 2004b). This may suggest that the effects observed in the previous study may have been the result of plasticity rather than genetic differentiation driven by environmental variation, although the results of the present study are based on fewer populations and a smaller range of environmental variation. However, our results are congruent with those of Warren et al. (2005), who found only weak relationships between physiological variables and seed-source rainfall in E. tricarpa; the weak tendency of families from wetter population locations.

The long-term stability of population differences in foliar sideroxylonal concentrations is unclear. Seedling establishment and development to the sapling stage, which is likely to occur following fire in most E. tricarpa woodlands (Gill 1997), may be the stage in the life of a tree during which herbivores can inflict the most damage. Christmas beetle densities during recruitment may have a strong influence on the sideroxylonal concentrations of the next generation of E. tricarpa. The impact of Christmas beetles on eucalypts varies considerably from year to year, as populations fluctuate in response to rainfall and temperature patterns. Given the high heritability and phenotypic variance of sideroxylonal concentration, a population may respond quickly to a transient change in herbivore pressure and may not reflect the selection pressures facing the trees at present or further in the past.

Additive genetic correlations may act as evolutionary constraints, but are also important in animal and plant breeding (Lynch and Walsh 1998; Williams et al. 2002). In Eucalyptus, they could limit the potential benefits of breeding for both growth and defense. The lack of correlation between sideroxylonal and growth suggests that breeding for high sideroxylonal production need not interfere with selection gains in growth traits under plantation conditions, although genetic analysis on additional sites would be needed to draw this conclusion.

Physiological trade-offs may constrain the evolution of defense traits in plants, although evidence for the costs of plant defense has been contradictory (Koricheva 2002). No attempts have yet been made to test this hypothesis for foliar FPCs such as sideroxylonal. We found no relationship between foliar sideroxylonal concentrations and growth, implying that the physiological cost of producing sideroxylonal is not great or that the costs and benefits are balanced under the conditions of the experiment. Experiments excluding herbivores are required to measure the true physiological cost of sideroxylonal production, which may be difficult for long-lived trees with many natural enemies. Defenses have been shown to be costly in many plants, but the expression of physiological costs can depend heavily on the environmental conditions in which plants are grown (Siemens et al. 2003). Care must be taken when extrapolating from single common-garden trials such as these to natural populations.

Conclusion

We have provided field confirmation that sideroxylonal is related to resistance to an insect, as well as to mammalian herbivores. Our results support the estimate of narrow-sense heritability made for sideroxylonal using a marker-based method in a natural population of E. melliodora. We found no evidence of a strong growth-defense trade-off in E. tricarpa; however, this only highlights the need for investigation of sideroxylonal production and growth in the absence of herbivory, and of the ability of eucalypts to escape or tolerate damage by different herbivores. We have shown that differences in sideroxylonal concentration among populations have a genetic basis, although the importance of environmental factors should also be investigated. Further understanding of the inheritance of sideroxylonal could be gained from similar experiments in a range of sites aimed at detecting interactions between genotype and environment. These may be critical in applying our results to the genetic improvement of the insect resistance of eucalypts.