Selection for chemical trait remixing in an invasive weed after reassociation with a coevolved specialist

Abstract

The interaction between Depressaria pastinacella (parsnip webworm) and wild parsnip (Pastinaca sativa), in its native Europe and in its longstanding nonindigenous range in the midwestern United States, is characterized by chemical phenotype matching, ostensibly mediated by reciprocal selective responses. The first appearance of D. pastinacella on P. sativa in New Zealand in 2004 provided an opportunity to quantify selective impacts of a coevolved herbivore and calibrate rates of phytochemical response in its host plant. Webworms in 2006 reduced seed production up to 75% in New Zealand populations, and in 2007 infestations increased in severity in all populations except one. Most New Zealand populations fall into a furanocoumarin phenotype cluster distinct from European and U.S. phenotypes, although one heavily attacked population clusters with two U.S. populations and one European population long associated with webworms. Multivariate selection analysis substituting realized fitness (with webworms present) for potential fitness (absent webworms) as the dependent variable revealed that reassociation with a coevolved specialist in a nonindigenous area profoundly altered the selection regime, favoring trait remixing and rapid chemical changes in parsnip populations, as predicted by the geographic mosaic theory. That uninfested populations of New Zealand parsnips contain higher amounts of octyl acetate, a floral volatile used by webworms for orientation, suggests that plants that escape from specialized enemies may also experience selection to increase kairomones, as well as to reduce allomones.

Although the ability of herbivorous insects to act as selective agents on the chemistry of their host plants is an essential component of coevolutionary theory (1, 2), experimental demonstrations of such impacts are few (3–5). Accordingly, some skepticism remains as to the efficacy of herbivores as selective agents (6), and alternative explanations, including top-down selection from natural enemies of herbivores, have proliferated (albeit also without an abundance of experimental demonstrations). In addition to its importance in a theoretical context, quantifying selection and rates of response is critical for evaluating the stability of host-plant resistance traits in agricultural crops as well as the sustainability of classical biocontrol of weeds.

One way to examine the rate and magnitude of evolutionary change in phytochemical profiles in response to insect herbivory is to examine an interaction involving coevolved species that are separated and then, by sequential range extensions, reassociated. Reassociation with a specialized coevolved enemy in an area of nonindigeneity is likely to have a profound and more predictable effect on host-plant chemistry. Although little quantitative and ecologically relevant information is available on phytochemical changes in plants that occur after introduction into a nonindigenous area and release from interactions with longtime insect associates (7–10), even less information is available on phytochemical changes that ensue when coevolved enemies that are demonstrated reciprocal selective agents resume interacting with a host plant in an area of invasion. Such changes should be more predictable because the decoupling of defense and herbivory that has been observed in interactions between invasive plants and indigenous herbivores (11) is less likely to occur upon reassociation with a coevolved specialist. Whereas compelling evidence for rapid contemporary evolution of size, fecundity, and leaf area of invasive plant species has been found (12), the potential for evolutionary change of chemical traits, upon which resistance may be based, upon reassociation with a coevolved enemy has yet to be explicitly evaluated. The geographic mosaic theory of coevolution (2) specifically predicts rapid reciprocal responses, including trait remixing, to result from intense selection exerted by a reassociated coevolved specialist, and historical evidence (13) is suggestive of such responses.

A system in which the chemical consequences of reassociation with a coevolved herbivore may be examined involves the introduced Eurasian weed Pastinaca sativa, wild parsnip, and its European insect associate Depressaria pastinacella, the parsnip webworm. This interaction has been well characterized within its native range in Europe and in its introduced range in North America, and the ecological effects of a diversity of plant chemicals on the principal specialized herbivore have been determined. Brought to America by the earliest colonists, the parsnip was “common” by 1630 (14). Escaped from cultivation, it is regarded as noxious because all aerial parts produce furanocoumarins, phototoxic allomones that, on contact with human skin, cause blistering and hyperpigmentation (15). Very few native North American insects have colonized P. sativa (16); currently, the principal (and occasionally only) herbivore throughout its range in North America is D. pastinacella, accidentally introduced from Europe and first reported in 1869 in Ontario, Canada (17). D. pastinacella webs together and feeds on the reproductive structures of species in the closely allied genera Pastinaca and Heracleum; since their introduction, webworms have become established widely in North America (18).

Certain furanocoumarins in wild parsnip function as resistance factors against webworms in North America (19, 20) and Europe (21). Webworm damage in the Midwest United States selects for increased concentrations of three furanocoumarins: xanthotoxin, bergapten, and sphondin (3, 20). Genotypes with high levels of these furanocoumarins experience lower fitness in the absence of herbivores than genotypes with lower furanocoumarin content (22), indicative of a cost of producing these compounds. These three furanocoumarins act as resistance factors in part because webworms metabolize them less efficiently by cytochrome P450 monooxygenases (23). In both Europe and North America, interactions between parsnip webworms and wild parsnips are characterized by chemical phenotype matching: close correspondence between furanocoumarin defense profile of the plants and detoxification capacity of the webworms.

Before the appearance of webworms in North America, parsnips exhibited reduced furanocoumarin defenses in comparison with contemporaneous European parsnips, as determined by examination of herbarium specimens collected in North America over 152 years spanning the period before introduction of webworms to the present, and European specimens from the 19th century to the present, for webworm damage and furanocoumarin content (24). Concomitant with the rise in webworm infestation between 1890 and 1909, levels of all five furanocoumarins (those previously associated with resistance as well as imperatorin and isopimpinellin) increased significantly and continued to increase thereafter. Although it is likely that webworm herbivory contributed to selection for the phytochemical shifts, it is impossible to assign responsibility conclusively a century after the fact.

The discovery of webworms for the first time attacking wild parsnip populations in New Zealand (N.Z.) in 2004 (25, 26) provided an extraordinary opportunity to quantify selection by a coevolved herbivore on host-plant chemistry during the earliest phase of reassociation. As in North America, parsnips were introduced to New Zealand as a food source by European colonists (English and Scottish), escaped from cultivation, and became naturalized, as noted by Hooker in 1867 (27). Consequently, P. sativa has in New Zealand enjoyed at least 140 generations free of webworms. Based on characteristics of this interaction both in its native Europe and in its 150-year tenure in North America, we predicted that N.Z. populations newly infested by webworms should show rapid changes in furanocoumarin chemistry after infestation, with their chemistry converging on that of U.S. plants with long histories of webworm association—i.e., plants in infested populations should be selected for increased amounts of furanocoumarins associated with resistance (xanthotoxin, bergapten, sphondin). In contrast, N.Z. populations that have not been infested by webworms should be in equilibrium and should not exhibit differences in chemical traits over the time period during which webworms have expanded their range elsewhere in New Zealand.

Results

Two webworm host plants were present within the survey area: P. sativa and hogweed, Heracleum sphondylium. In 2006, we found 18 locations of hogweed populations ranging in size from 1 to hundreds and 11 populations of wild parsnip ranging in size from 17 to hundreds [supporting information (SI) Table 3]. Three wild parsnip populations that were far from hogweeds and near the center of the outbreak in the city of Dunedin were heavily infested in 2006: Crimp, Lumber, and Townley. A fourth population, Warrington, at the northern edge of the range and ≈15 km from the presumed point of introduction [Port Chalmers (25)], was lightly infested. By 2007, the northernmost edge of the range extended 5 km to include two additional populations of wild parsnip (Rock and Ocean).

Webworms in 2006 removed half or more of the total fitness of ≈50% of the plants in the Crimp and Townley populations and eliminated seed production by >75% of the plants in the Lumber population (Fig. 1). In 2007, infestation severity increased in all of these populations except Crimp; 75% of the plants in the Lumber and Townley populations failed to produce seed, and nearly half of the Warrington plants, only lightly infested in 2006, failed to produce seeds.

Fig. 1.

Realized fitness of wild parsnips in six N.Z. populations after consumption of reproductive parts by parsnip webworms. Average potential fitness (number of male flowers) of plants within a population is given below each pie chart. Estimates of potential fitness for many plants was not possible because webworms in addition to consuming all of the umbels had as well consumed the stalks bearing the umbels.

Ripe seeds of N.Z. wild parsnips differ substantially in chemical composition from those of European and U.S. parsnips (Fig. 2). Almost all of the N.Z. populations fall into a furanocoumarin phenotype cluster distinct from the phenotypes of Europe and the United States, characterized by lower levels of imperatorin, bergapten, and isopimpinellin and slightly above-average levels of xanthotoxin and bergapten. Occupying its own cluster is population BB, low in all of the linear furanocoumarins and above-average in sphondin. One heavily attacked N.Z. population, Lumber, clusters with two U.S. populations and one European population long associated with webworms (21); this cluster is characterized by high levels of imperatorin, bergapten, and xanthotoxin and low levels of isopimpinellin and sphondin. Populations of N.Z. parsnips were no less phenotypically variable than their North American relatives; in fact, standard deviations of linear furanocoumarin and angular furanocoumarin concentrations ranged higher in New Zealand than in North America (8.13 vs. 5.11 μg/mg for linear furanocoumarins, 0.47 vs. 0.36 μg/mg for angular furanocoumarins).

Fig. 2.

Clustering of 33 wild parsnip populations (involving 1,117 individuals) on the basis of average phenotypic furanocoumarin seed content; North American and European data are from Berenbaum and Zangerl (21). Populations in red have had either long-term associations with webworms [in the cases of North American, prefixed “na,” or European populations (green) from Austria, The Netherlands, or Germany, prefixed “aust,” “neth,” or “germ”] or only a recent association (between 2004 and 2006 in New Zealand, “nz”). Populations in black have had no, or only rare, interaction with webworms. Clustering is based on squared Euclidean distances (SPSS 14, Chicago, IL). Bar graphs show percent deviations (+ or −) from mean furanocoumarin content of each of six clusters for five furanocoumarins (left to right: imperatorin, bergapten, isopimpinellin, xanthotoxin, and sphondin). With one exception (nzLUMB), N.Z. populations are clustered separately from all others. Abbreviated N.Z. population names are LUMB, Lumber; TOWN, Townley; WAR, Warrington; COTT, Cottage; and OCEA, Ocean.

Because the number of traits used in the selection analyses was limited in certain populations by the number of plants available within a population (the number of independent variables in the regressions must be less than the number of samples), comparisons among populations and years are constrained. More informative are the within-population differences in selection coefficients that occur as result of substituting realized fitness (in the presence of webworms) for potential fitness (estimated to have occurred had webworms not been present) as the dependent variable. Changes in coefficients are attributable only to the effect of webworms. For all but the Warrington population in 2006, which was minimally infested that year (average fitness was reduced by only 3.5%), webworms altered selection patterns (Table 1). Populations affording larger sample sizes permitted specification of greater numbers of traits and second-order interactions. In these populations, complicated patterns of nonlinear or correlational selection were evident. In the Crimp 2006 population, selection for potential fitness favored a decline in isopimpinellin (directional selection), but selection incorporating the influence of webworms favored a reduction in sphondin (directional selection) and positive or negative correlations among isopimpinellin, xanthotoxin, imperatorin, bergapten, and octyl butyrate (Table 1). In 2007, in a sample-size-limited model for the Crimp population, webworms intensified selection for an uncoupling of xanthotoxin from sphondin. In the Lumber population in both years, with sample-size-limited models, there were no significant selection patterns for potential fitness, but with webworms there was selection against isopimpinellin (Table 1). Webworms also caused intense nonlinear positive selection for isopimpinellin in the Warrington 2007 population as well as selection for positive coupling of octyl butyrate and sphondin. In the Rock population, webworms exerted disruptive selection on sphondin as well as negative correlational selection for octyl acetate and bergapten. Patterns of nonlinear negative selection on isopimpinellin in the Ocean 2007 population and stabilizing selection for isopimpinellin in the Townley 2006 population were diminished by webworms. In both of these populations, a variety of significant coefficients involving correlational selection and nonlinear selection were diminished by webworms (Table 1). In the case of the Townley 2006 population, negative nonlinear selection on octyl butyrate, known to act as a webworm feeding deterrent (28), was replaced by positive directional selection for the trait after fitness was adjusted by webworms. Finally, in the Townley 2007 population, again from a sample-size-limited model, selection by webworms against xanthotoxin was the only significant coefficient detected. Despite the limitations of regression, particularly when important variables are likely omitted, these results nevertheless suggest an extraordinary degree of selection for trait remixing. In some of these cases, webworms intensified (made significant) like-signed but nonsignificant coefficients detected by potential fitness. We suspect that, rather than amplifying the same selection pattern, our estimates of potential fitness may have been influenced by infestation; parsnips do attempt to compensate for losses to webworms by converting a larger percentage of flowers to seeds in higher-order umbels and also producing more higher-order umbels (29). To the extent that the response involves increased umbel stalk diameter and greater numbers of umbels, potential fitness, as measured, could have been influenced by infestation. Even so, the changes in significance and/or sign between fitness estimates for coefficients indicate that webworms are exerting significant selection on the chemistry of the populations.

Table 1.

Selection in 2006 and 2007 on chemical traits of seeds of wild parsnip in N.Z. populations after reassociation with parsnip webworms

Population (Model)	N	%	Directional selection (β) Nonlinear or correlational selection (γ) Potential fitness/realized fitness
Crimp 2006 (B, OB, IM, IS, S, X)	85	47.3	Sphondin β = +0.02/−0.25
			Isopimpinellin β = −0.38/−0.19
			Isopimpinellin/xanthotoxin γ = −2.20/ −4.46
			Octyl butyrate/xanthotoxin γ = +2.60/+5.28
			Imperatorin/bergapten γ = +2.16/+5.67
			Imperatorin/xanthotoxin γ = −2.91/−5.33
			Octyl butyrate/bergapten γ = −2.42/−4.98
Crimp 2007 (B, OB, S, X)	19	85.4	Xanthotoxin β = +0.80/+0.82
			Xanthotoxin/sphondin γ = −4.79/−10.83
Lumber 2006 (OB, IM, IS)	21	34.9	Isopimpinellin β = −0.14/ −0.58
Lumber 2007 (OA, IS, S)	12	18.4	Isopimpinellin β = +0.18/ −0.94
Ocean 2007 (IM, IS, B, S, X)	26	82.7	Isopimpinellin γ = −6.18/−3.87 (negative)
			Isopimpinellin/sphondin γ = −8.0/−4.04
			Isopimpinellin/bergapten γ = +39.0/+10.86
Rock 2007 (B, IM, OA, OB, S)	74	85.2	Sphondin γ = +1.16/+1.52 (disruptive)
			Octyl butyrate/sphondin γ = +1.81/+2.057
			Octyl acetate/bergapten γ = −0.11/−2.01
Townley 2006 (B, IM, IS, OA, OB, S)	50	52.4	Octyl butyrate β = +0.24/+0.38
			Octyl butyrate γ = −2.91/+0.30 (negative)
			Isopimpinellin γ = −3.152/−2.233 (stabilizing)
			Octyl acetate γ = +2.272/+1.461 (disruptive)
			Isopimpinellin/sphondin γ = +5.928/+0.506
			Octyl acetate/octyl butyrate γ = −3.104/−1.514
			Octyl acetate/sphondin γ = −4.42/−2.998
			Octyl acetate/isopimpinellin γ = +3.4/+2.21
			Imperatorin/octyl butyrate γ = +11.6/+3.62
			Imperatorin/isopimpinellin γ = −11.2/−1.34
			Bergapten/isopimpinellin γ = +3.86/+1.51
Townley 2007 (OB, IM, S, XB)	18	17.9	Xanthotoxin β = +0.57/−1.65
Warrington 2006 (B, IM, IS, OB, S, X)	29	96.5	Not significant
Warrington 2007 (IS, OB, S, X)	19	38.4	Isopimpinellin γ = +5.88/+26.25 (positive)
			Octyl butyrate/sphondin γ = +9.55/+20.369

Multivariate selection analysis according to Lande and Arnold (45) was performed separately for potential fitness (had webworms been absent) and for realized fitness (after consumption of reproductive parts by webworms). N gives the sample size for the multiple regression, and % refers to the average realized fitness as a percentage of potential fitness. Standardized coefficients in boldface were different from zero at P < 0.05. For nonlinear coefficients, the nature of the selection is shown in parentheses. In parentheses to the right of each population name are the traits included in the model: B, bergapten; I, isopimpinellin; S, sphondin; IS, isopimpinellin; IM, imperatorin; X, xanthotoxin; OA, octyl acetate; OB, octyl butyrate.

A high percentage (r² = 0.87) of the variation in pupal size of N.Z. webworms was accounted for by seven constituents of developing seeds (Table 2). Increases in the angular furanocoumarins sphondin and angelicin (the latter of which is the precursor of sphondin and typically not found in ripe seed) together with the linear furanocoumarin imperatorin and the octyl butyrate ester were associated with decreased pupal size, whereas increases in the octyl acetate ester, isopimpinellin, and xanthotoxin were associated with increased pupal size. Addition of bergapten to the model (results not shown) diminished these relationships dramatically, likely due to high correlation with xanthotoxin in all of these populations, and substitution of bergapten for xanthotoxin failed to produce a significant model.

Table 2.

Regression of mean pupal mass per stem for parsnip webworms from 11 N.Z. plants on chemical composition of wild parsnip (r² = 0.874)

Model	df	Mean squares	F	P
Regression	7	8.661	10.937	0.038
Residual	3	0.792
Independents	B	t	P
Angelicin	−4.723	4.19	0.025
Imperatorin	−0.476	4.69	0.018
Isopimpinellin	6.392	6.85	0.006
Xanthotoxin	0.247	3.7	0.034
Sphondin	−3.422	5.64	0.011
Octyl acetate	4.333	4.31	0.023
Octyl butyrate	−2.05	4.27	0.024

In addition to differences in furanocoumarin composition of ripe seeds among N.Z. and U.S. populations (Fig. 2), all non-furanocoumarin components of buds and male flowers, except the ocimenes, differed substantially between countries (Fig. 3). Midwest U.S. populations were >2-fold higher in octyl butyrate and myristicin, higher in the sesquiterpenes caryophellene, α-trans-bergamotene, cis-β-farnesene, and β-cubebene, and higher in 2-methylhexyl propanoate but lower in 1-octanol, octyl acetate, γ-palmitolactone, and α-farnesene.

Fig. 3.

Differences between N.Z. and Midwest U.S. parsnips in chemical composition of buds and male flowers (means ± SE). All differences but those for cis- and trans-ocimene between countries were significant (P < 0.05) in a two-way ANOVA with country and plant part (bud or flower) as main effects. There were no significant interactions and only three compounds differed between buds and flowers: flowers were higher in octyl butyrate and lower in myristicin and cis-ocimene (data not shown).

Discussion

Wild parsnip populations in New Zealand newly infested by a coevolved specialist experienced major fitness reductions, possibly for the first time in the region, and experienced substantial selection pressure for phytochemical trait remixing. It is too early to tell whether selection resulting from reassociation with the coevolved specialist D. pastinacella will favor convergence with North American and European parsnip populations long subjected to webworm herbivory. Some trends are consistent with that direction, most clearly the selection for octyl butyrate, which not only is low in N.Z. parsnip but also is associated with reduced webworm pupal size. The trajectory of furanocoumarins is more difficult to assess, as there are three branches in the biosynthetic pathway, all of which could be targeted by correlational selection. Potentially constraining genetic correlations among furanocoumarins and differences in the availability of additive genetic variance for selection to act on could limit the eventual outcomes, and, indeed, different response trajectories are predicted by the geographic mosaic theory (2). Analysis comparisons of subsequent generations of plants in infested and noninfested populations will be necessary to determine the ultimate phenotype configuration resulting from webworm selection.

Rapid reacquisition of chemical defenses because of selection pressure from a coevolved specialist may account for findings that appear inconsistent with ecological theories of invasion. The evolution of increased competitive ability (EICA) hypothesis, e.g., predicts that introduced plants, by leaving coevolved associates behind, should have reduced investments in defense in the area of introduction (30); however, common garden experiments with Senecio jacobaea, a weed native to Europe, failed to reveal a decline in pyrrolizidine alkaloid production in areas of invasion (31). Moreover, invasive populations possessed greater alkaloid content and greater resistance to Tyria jacobaeae, the cinnabar moth, a European specialist (31). Stastny and Schaffner (32) also found that plants in areas of invasion produced greater quantities of alkaloids than European populations and suggested that the EICA hypothesis fails to account for their findings. The North American populations in these studies, however, include localities in Oregon and British Columbia, where T. jacobaeae has been established as a biocontrol agent for >30 years (33). Reassociation with a specialist may have selected for enhanced chemical defense and thus account for the pattern found.

The potential for rapid phytochemical response of an invasive weed in an area of introduction to reassociation with a coevolved specialist is instructive in two contexts. Not only does it provide an example of the selective impact of an insect herbivore on plant chemistry, a long-held assumption underlying theories of insect–plant coevolution, it also demonstrates the potential for rapid evolution of enhanced resistance in nonindigenous environments. If wild parsnip is typical of invasive weeds in possessing chemical defenses that are genetically variable in areas of introduction, then long-term control by introduced coevolved specialists may be an elusive goal. Moreover, theories relating to evolutionary changes in invasive plants are generally couched in terms of reduced allocation to allomones—i.e., phytochemicals with adverse effects on herbivores. Release from adapted specialists, however, may have other impacts on plant chemistry. Specialist herbivores generally rely on kairomones, or host-plant recognition cues, to find feeding and oviposition sites. Many phytochemicals that are defenses against generalist herbivores may serve as attractant kairomones for specialists (1, 34). In the absence of specialists, plants may experience selection to increase production of certain volatile compounds that attract not only herbivores but also mutualists such as pollinators (35–37). Enhanced reproductive success of invasive species in areas of invasion, although attributed to greater competitiveness or “vigor” (38), may also result from increased production of volatile compounds mediating interactions with pollinating mutualists. Such responses could account for seemingly paradoxical findings of enhanced production of certain “defense” chemicals in areas of invasion; these defense chemicals may serve other ecological roles. That herbivores and pollinators can both exert selection pressures on plant chemical traits is well established (39). Parsnip webworm larvae orient to octyl acetate to find suitable reproductive tissue to consume (28), but octyl acetate is among the volatile compounds that are attractive to the promiscuous pollinators of this plant and its relatives (40, 41). That N.Z. flowers produce higher amounts of octyl acetate than do U.S. plants is consistent with the suggestion that release from its specialist enemy may free the plant from constraints on production of volatile attractants and enhance its ability to attract pollinators (41). Reassociation with a major flower-feeding herbivore may alter pollination success of P. sativa; infestation by only a single webworm reduces pollinator visitation, likely because of changes in volatile cues used by pollinators (40). Whether, after reassociation with webworms, octyl acetate levels will continue to fall in N.Z. flowers and whether kairomones, like allomones, generally change in response to release from specialist herbivores remain to be seen.

Methods

Whereas P. sativa was first recorded in New Zealand in 1867, and the European import hogweed, H. sphondylium in 1939 (42), the webworm was reported for the first time in Port Chalmers on the South Island in 2004; an extensive survey at that time revealed that webworms were sufficiently numerous around Port Chalmers and nearby Dunedin as to be considered beyond eradication (26). In February 2006, populations of webworms and webworm host plants P. sativa and H. sphondylium were located in and around the presumed location of webworm introduction, Port Chalmers (25), and extending as far north as Christchurch, N.Z. All host populations were surveyed for the prevalence of webworms (SI Table 3). Although most webworms had pupated by the time of the surveys, damage characteristic of webworm feeding, namely, presence of webbed umbellets together with frass and partially or wholly consumed reproductive organs, was recorded for each plant surveyed. For large host-plant populations, a haphazardly selected subset of plants was surveyed.

Relative fitness was quantified for plants within four infested wild parsnip populations in 2006 (Crimp, Lumber, Townley, and Warrington) and in the same four populations as well as two newly infested populations in 2007 (Rock and Ocean). Both realized fitness (after webworm activity) and potential fitness (had webworm not been present) were measured by estimating the proportion (in increments of 0.1) of umbellets in each umbel not destroyed by webworms. The diameter of each umbel stalk, as an approximate predictor of the number of reproductive units (units = 91.6 × umbel stalk diameter + 98.3, n = 10, r² = 0.80, P = 0.0005) was also recorded to the nearest 0.1 mm with Vernier calipers for each primary (stem apex), secondary (branch apex), and tertiary (secondary branch apex) umbel on the plant. Total plant fitness after webworm damage was then the sum of the products of the proportions of umbel undamaged times stalk diameter, or, for potential fitness had webworms not been present, simply the sum of stalk diameters. Because parsnips are temporally dioecious and not all flowers in secondary and higher-order umbels transition to female flowers and fruits (each fruit potentially bearing two seeds), application of the regression model to the sum of stalk diameters provides an estimate of the total number of male flowers. Each such flower in the primary umbel typically turns into a female flower and eventually, if pollination is successful, a fruit (29). Transition of flowers on secondary and tertiary umbels to female flowers and fruits is conditional on plant resources and webworm damage (29, 43). Relative fitness was calculated by dividing plant fitness by average fitness in the population.

Ripe seeds were collected from all wild parsnip populations in 2006 and from most of them in 2007, including every plant for which fitness was estimated. Four seeds from each plant were weighed to the nearest 0.1 mg, cut in half with scissors, and placed in a 4-ml glass autosampler vial. The end of the scissors was held inside the vial as 2 ml of ethyl acetate was added to ensure that residues on the scissors were also added to the vial. After 2 h at room temperature, 200 μl of extract was placed in autosampler vials for analysis by gas chromatography, and the remainder was analyzed by high-pressure liquid chromatography. Sample analysis by gas chromatography (GC) was performed on a Hewlett–Packard 5890 instrument (Agilent) with a flame ionization detector and an Alltech EC-1 capillary column (30 m, 0.32 mm inside diameter, 0.25-μm coating). Two microliters of extract was injected into a 200°C inlet, and separation of compounds was accomplished with the following program: 60°C for 0.5 min, 30°C/min, 170°C, 5 min. The predominant non-furanocoumarin essential oil components in ripe seeds, octyl acetate and octyl butyrate, were quantified. Furanocoumarins were autosampler-injected (15 μl), separated, and quantified on a Waters HPLC. The separation was accomplished isocratically (1.5 ml/min flow, 55% cyclohexane, 42% diisopropyl ether, and 2% butanol) on a 250-mm long, 4.6mm inside diameter, Sperisorb 5-μm silica column (Waters). The furanocoumarins imperatorin, bergapten, isopimpinellin, xanthotoxin, and sphondin were quantified. All chemical concentrations were calculated in μg/mg of dry seed mass. Average furanocoumarin concentrations of N.Z. parsnip populations were compared with published mean values for U.S. and European populations (21) by employing hierarchical analysis (SPSS 14 2006) based on Euclidean distances and average between-groups linkage (33 populations in all).

Estimates of directional, nonlinear, and correlational selection by webworms were obtained by multiple regression according to Lande and Arnold (44). Limitations on sample size in some of the populations (all individuals, even those with as few as two or three ripe seeds, were analyzed in these populations) necessitated reducing the number of plant traits analyzed to accommodate analysis of nonlinear and correlational selection. Models were constructed that explained high amounts of the variation in relative realized fitness. A separate regression was performed for relative potential fitness employing the same traits that were specified in the model for realized fitness. The regressions were run again with the products of the dependent variables added to estimate nonlinear and correlated selection. The shape of the trait–fitness relationship for traits under significant nonlinear selection (coefficients for squared terms in the regressions) were evaluated by plotting relative fitness values, predicted from the linear and quadratic terms by using coefficients from the selection model, against the observed trait values from the population. In this way, simple nonlinear selection for or against the trait, disruptive selection, or stabilizing selection could be differentiated.

To determine whether N.Z. P. sativa vary in suitability to D. pastinacella, we quantified the size of pupae inside stems of 11 substantially infested plants near the Warrington population. Each plant was cut at the base of its stem and split lengthwise with a utility knife. Pupae were removed, placed on a white sheet of paper, and photographed next to a Vernier caliper. Larvae almost always pupate within the stem of the plant on which they develop; however, to reduce the likelihood of larval movement before pupation, pupae were collected from plants at least 2 m from their closest neighbor. A sample of ripe seeds was also collected from each plant. Cross-sectional area (mm²) of the pupae was quantified by using Scion for Windows Imaging freeware. Mean pupal areas were compared among plants by ANOVA and were regressed against furanocoumarin and octyl ester content of seeds to discern whether and how pupal size is related to chemistry.

Although the chemistry of ripe seeds of P. sativa can account for up to 70% of the variation in resistance to D. pastinacella (3), webworms consume buds, flowers, and developing seeds over the course of their development. To investigate whether such reproductive structures of N.Z. wild parsnips differ from those of U.S. plants, we collected male flowers and buds from several N.Z. populations and compared them to flowers and buds from plants in U.S. populations. Eleven male flower and nine bud samples were collected from four N.Z. populations (Ocean, Cottage, Townley, and Rock, one to six plants per population). Sixteen male flower and 12 bud samples were collected from five Midwest U.S. populations (two to five plants per population) in the vicinity of Urbana, Illinois. Entire umbellets were collected, air-dried, weighed, and placed in 2-ml centrifuge tubes. A glass bead was added to the tube and the sample was powdered in a shaker (modified Wig-L-Bug, Crescent Dental). Ethyl acetate (0.5 ml) was added to each tube and after 1 h the particulates were spun down in a centrifuge and the extract was analyzed by GC as described for octyl esters except for the oven program (50°C, 2 min, 5°C/min, 200°C, 3 min). Identity of peaks was confirmed by gas chromatography–mass spectrometry. With the exceptions of 1-octanol and 2-methylhexyl propanoate, all components quantified in this analysis were previously reported in P. sativa (45). Where standards were available (octyl acetate and butyrate, myristicin, ocimenes, caryophellene) component concentrations were expressed in μg/mg. For sesquiterpenes for which standards were not available, concentrations were calculated as caryophellene equivalents, and for all other constituents the concentrations were calculated as hexadecane equivalents.