Herbivore Responses to Plant Secondary ... - Semantic Scholar

vol. 161, no. 4

the american naturalist

april 2003

Herbivore Responses to Plant Secondary Compounds: A Test of Phytochemical Coevolution Theory

Howard V. Cornell1,* and Bradford A. Hawkins2,†

1. Department of Biological Sciences, University of Delaware, Newark, Delaware 19711; 2. Department of Ecology and Evolutionary Biology, University of California, Irvine, California 92697 Submitted November 26, 2001; Accepted September 13, 2002; Electronically published March 21, 2003

abstract: Literature data were collected on the floristic distribution and toxicity of phytochemicals to herbivores and on herbivore specialization in order to test phytochemical coevolution theory. The theory makes four predictions that can be tested with this information. Herbivores can adapt to novel, more toxic chemicals by becoming specialists, or they can become generalists but at the cost of lower feeding success on any particular host. Thus, the first two predictions are as follows: herbivores should do better on chemicals that are present in their normal host, and this pattern should be stronger for specialists than for generalists. The “escape and radiation” aspect of the theory holds that if a plant taxon with a novel defense chemical diversifies, the chemical will become widespread. Eventually, herbivores will adapt to and disarm it. So the third prediction is that more widespread chemicals are less toxic than more narrowly distributed ones. Because generalists should not do as well as specialists on chemicals disarmed by the latter, the fourth prediction is that the third prediction should be more true for generalists than specialists and should depend on presence/absence of the chemical in the normal host. Multiple regressions of toxicity (herbivore mortality and final weight) on three predictor variables (chemical presence/absence in the normal host, specialism, and chemical floristic distribution) and relevant interactions were used to test these predictions. Chemical presence/absence in the normal host, the interaction between this variable and specialism, and chemical floristic distribution had significant effects on both measures of toxicity, supporting the first three predictions of the model. Support for the fourth prediction (a three-way interaction among all predictor variables) was evident for final weight but not mortality, perhaps because growth is more responsive to toxicity differences than survival. In short, the phytochemistry literature provides broad support for the phytochemical coevolution model. * E-mail: [email protected]. †

E-mail: [email protected].

Am. Nat. 2003. Vol. 161, pp. 507–522. 䉷 2003 by The University of Chicago. 0003-0147/2003/16104-010411$15.00. All rights reserved.

Keywords: apparency, bioassay, coevolution, herbivore, phytochemical, specialization. On-line enhancements: appendix tables.

Insects evolved to feed on terrestrial plants at least 400 million years ago, and herbivores and their host plants now comprise some of the richest assemblages in terrestrial communities (Price 1980; Strong et al. 1984). Investigations into how such assemblages evolved have been guided by two well-founded observations. First, most herbivorous insects are specialized to one or a few host species (Jaenike 1990; Bernays and Chapman 1994; Thompson 1994), yet there are always some generalized feeders. Second, plant secondary chemicals, which play an important role in restricting herbivore diets, are toxic to some herbivores and harmless to others (Dethier 1954; Fraenkel 1959). Attempts to explain these observations eventually begot the theory of phytochemical coevolution (Dethier 1954; Ehrlich and Raven 1964; Berenbaum 1983a). Its essence is that plants are selected to produce secondary chemicals in response to herbivore feeding, whereas herbivores respond by evolving disarming mechanisms. A by-product of this disarmament is the loss of ability to feed on other plant species. Through time, plants thus become more toxic and herbivores become more specialized. Also, plants that develop novel defenses enter a new adaptive zone and undergo phylogenetic diversification (Ehrlich and Raven 1964). The same happens to herbivores that develop novel disarming mechanisms. This so-called escape and radiation (Thompson 1994) may be largely responsible for driving the coevolutionary process. As plant species containing a novel chemical diversify, selection for disarming mechanisms increases. Similarly, diversification of herbivores with disarming mechanisms increases selection for new or stronger defense chemicals. The theory of phytochemical coevolution has been criticized on a number of points (e.g., Jermy 1976, 1984; Strong et al. 1984; Howe and Westley 1988; Futuyma and Keese 1992; Farrell and Mitter 1993; Thompson 1994). The most salient are as follows: first, although insects al-

508

The American Naturalist

most certainly adapt to their host plants, selection on plants by insects is often too weak and variable to drive plant evolution; second, many plant defenses may be against microorganisms rather than herbivores; third, conflicting selection pressures from multiple herbivore and plant species may result in diffuse rather than focused coevolution; fourth, reciprocal adaptive diversifications between insects and plants may derive more from ecology and biogeography than plant chemistry; fifth, the predicted increase of specialization through time does not account for the frequent occurrence of derived generalists. Despite these criticisms, most would agree that herbivore/plant coevolution occurs (e.g., Gilbert 1971, 1975; Berenbaum 1981a, 1981b, 1983a) but that it cannot explain all of the patterns of plant defense and insects’ adaptation to their hosts. Apparency theory, developed somewhat later (Feeny 1976; Rhoades and Cates 1976), addresses a key question left unresolved by coevolution theory: Why are there so many generalists? It posits that “unapparent” plants are defended by toxic chemicals. Some herbivores adapt to and specialize on these just as predicted by coevolution theory. In contrast, “apparent” plants are defended by digestibility-reducing chemicals. Since it is difficult for herbivores to adapt to these, there should be less selection for specialism, and apparent plant faunas should contain more generalists (e.g., Futuyma 1976). Apparency theory thus explains the existence of generalists better than coevolution theory. A corollary, the feeding specialization hypothesis, is that specialists can feed more efficiently on their hosts than generalists that must be adapted to feed on varied defenses (e.g., Scriber 1983, 1984). Coevolution and apparency theory both agree that plant chemistry drives herbivore specialization. Implicit in this process is that specialization occurs because of evolutionary trade-offs; species that excel at one task will lose the ability to perform other tasks well (Futuyma and Moreno 1988; Jaenike 1990; Via 1990; Van Tienderen 1991), probably because of allocation of limited resources or pleiotropy. Specialization is thus the trade-off for the ability to feed on hosts with increasingly toxic chemicals. However, apparency theory and its corollary, the host specialization hypothesis, permit an alternative strategy: retain a generalized feeding habit but at the cost of lower feeding success on any one host. Specialization may be particularly apt for insects with a “parasite” life history. Parasites live their whole life on one plant and therefore should be intensely selected for host specificity. In contrast, generalism may be particularly apt for “grazers” that feed on many host individuals during the course of development (Thompson 1994). If these theories are correct, then we can make two predictions: herbivores should do better on chemicals to which they have become adapted than on

novel chemicals, and this pattern should be stronger for specialists than for generalists. Support for these predictions would imply that specialists pay a cost for but get the benefits of specialism and that generalists pay a cost for but get the benefits of generalism. The third and fourth predictions follow if coevolution drives successive radiations of plant taxa as defenses become increasingly toxic. Such radiations should also occur among herbivores that develop disarming mechanisms, but this issue is less controversial and will not be addressed by this article. If such “escape and radiation” occurs in plants, then defense chemicals that have developed early in the coevolutionary sequence should be older and more widespread and have relatively weak effects on herbivores. That is, defenses that were successful in the past should have produced adaptive plant radiations and therefore should be widespread today. However, since they are widespread, they would have produced intensive selection for herbivore countermeasures. By parallel reasoning, more recently derived chemicals should be more narrowly distributed and toxic except to the few specialists that have adapted to disarm them (Berenbaum 1983a). Synthetic insecticides offer a modern metaphor and provide circumstantial evidence for the speed at which counteradaptation by herbivores can occur. Insecticides that are older and more widely used are less toxic today than they were when first developed, leading to the manufacture of more toxic replacements (Gould 1990; Pimentel et al. 1991). So the third prediction is that more widespread chemicals are less toxic than more narrowly distributed ones. The fourth prediction again derives from the host specialization hypothesis: specialists should respond less to this toxicity gradient than do generalists for two reasons. First, because specialists pay the cost of specialization, any novel chemical may be very toxic regardless of its taxonomic distribution. If, however, the chemical occurs in the specialist’s normal host, it may not be very toxic regardless of its taxonomic distribution. Second, this effect may be enhanced by life-history differences between specialists and generalists. Specialists may disproportionately comprise “parasites,” and generalists may disproportionately comprise “grazers.” So the fourth prediction is that the third prediction should depend on specialism, the effects of which should, in turn, depend on the presence/absence of the chemical in the normal host. In addition to each chemical’s toxicity, information on presence/absence of a chemical in the herbivore’s normal host, herbivore specialism, and floristic distribution of chemicals is needed to test these four predictions. All of this information was collected from the published literature. Scores of laboratory bioassays from the plant defense and “natural insecticide” literature have examined the responses of generalists and specialists to specific plant

Herbivore Response to Plant Chemistry chemicals. The predictions require that toxicity should respond to the presence/absence of the chemical in the herbivore’s normal host (first prediction), the interaction between presence/absence and herbivore specialism (second prediction), the floristic distribution of the chemical (third prediction), and the three-way interaction among all three of these variables (fourth prediction). In order for phytochemical coevolution theory to have explanatory power, the signals of these responses, although they may be weak, must be separable from background variation in chemical toxicity and in the patterns of insect specialization. Methods Data Collection We did systematic searches of books, journals, and gray literature that in our judgment were likely sources for the data sought. We found that intensive perusal and deeper searching of referenced material rather than reliance on abstracting services uncovers the most information. Particularly rich sources of bioassay information included Ecological Entomology, Ecology, Entomologia Experimentalis et Applicata, Journal of Applied Entomology, Journal of Economic Entomology, Journal of Chemical Ecology, Journal of Insect Physiology, Oecologia, and Phytochemistry, but we also drew on many other sources. For the floristic distribution of chemicals, chemical dictionaries, books, and articles on phytochemistry and systematics were the most helpful (e.g., Kjaer 1960; Williamson and Shubert 1961; Bohlmann et al. 1973; Gibbs 1974; Jensen et al. 1975; Ellis 1977; Stumpf and Conn 1981; Murray et al. 1982; Seaman 1982; Harbourne and Baxter 1993). Web pages, books, agricultural experiment station bulletins, and life-history articles as well as some of the bioassay articles themselves provided most of the information in herbivore host ranges (e.g., Oliver 1907; Mulkern et al. 1964; Soo Hoo and Fraenkel 1966a, 1966b; Popov and Ratcliffe 1968; Beckwith 1970; Tietz 1972; Wint 1983; Bowers 1984; Joern 1985; Navon 1985; Montllor et al. 1990). For the bioassay information, all species and chemicals tested in bioassays were included. The species comprised four insect orders (Lepidoptera, Coleoptera, Diptera, Orthoptera; table A1 in the online edition of the American Naturalist), and the chemicals fell into six broad classes including carbohydrates and lipids (organic acids, acetylenes, proteinase inhibitors, thiophenes, saccharides, fatty acids, etc.), alkaloids, other nitrogen-containing compounds (cyanogens, glucosinolates, amino acids, etc.), flavonoid phenolics, other phenolics (chromenes, coumarins, lignans, quinones, tannins, phenolic acids, etc.), and terpenoids (iridoids, saponins, cucurbitacins, sesquiterpene lactones, etc.; table A2 in the online edition of the Amer-

509

ican Naturalist). For uniformity of comparison, we included only studies where single purified chemicals were added to artificial diets or leaf discs/leaves of acceptable hosts. We did not include studies where the chemical was topically applied to or injected into the insect or where the chemical was part of a mix (except in a very few instances where we wanted to include particular herbivores and data on pure chemicals were unavailable) or had not been extracted from the host tissue. To be included in the analysis, the bioassay had to test at least one concentration of the chemical and have a control for feeding success on the diet in the absence of the chemical. We included a very small number of studies that had no controls if there was good circumstantial evidence that the control diet had no negative effect on the insect (e.g., survivorship was 100% on diets with low chemical concentrations). Some insects were tested against many more chemicals than others. This disparity is due to the common practice of choosing “standard species” to test the general effects of secondary phytochemistry against herbivores. The use of standard species has been justified because chemicals are often positively correlated in their toxic effects against many insects (Futuyma and Keese 1992). We found acceptable bioassay data on 85 insect species (table A1 in the online edition of the American Naturalist) and 329 chemicals (table A2 in the online edition of the American Naturalist), making 892 insect/chemical combinations. For information on the floristic distribution of chemicals and herbivore host ranges, we compiled as complete a list of plant species containing each chemical and of hosts fed on by each insect as possible. This sometimes required reference to more than one source. Chemical data were classified into six taxonomic range categories: one family, one genus; one family, two to 15 genera; two families; three to five families; six to 19 families; and 119 families. Host range data were classified into three categories: specialists, one to two families; oligophages, three to nine families; and generalists, greater than nine families. We tried other classifications, but they had no appreciable effects on the results. We were able to classify a large proportion of species and chemicals but not all of them (tables A1 and A2 in the online edition of the American Naturalist). These had to be excluded from analyses that required this information. Nine different performance measures were used by various authors to evaluate chemical toxicity: mortality after a standard feeding period, final weight after a standard feeding period, percent reaching pupation (or some other stage, e.g., adult), time required to reach pupation (or some other stage), relative consumption rate (RCR), relative growth rate (RGR), efficiency of conversion of ingested food (ECI), efficiency of conversion of digested food (ECD), and approximate digestibility (AD). Measures RCR

510


through AD are standard nutritional indices that are fully described elsewhere (Waldbauer 1968; Reese and Beck 1976a). In addition, toxicity was sometimes summarized as an ED50. An ED50 is the chemical dosage necessary to increase mortality or to reduce final weight by 50% from a control value. Mortality and final weight were measured most often and became the focus of the analysis. Since we were primarily interested in toxicity rather than repellency, feeding deterrence by itself was not considered. However, toxins almost always reduce feeding in addition to other effects. Thus, effects on mortality and weight that are apparently toxic might be due to a combination of toxicity and deterrence or, in a few cases, to deterrence exclusively. That is, insects die or do not grow because they are repelled by a particular diet and starve. The issue of deterrence poses little problem in making inferences when toxicity and deterrence are correlated (e.g., Erickson and Feeny 1974), but in some cases they are not correlated (Bernays 1990, 1991). In order to distinguish toxicity from deterrence, we also examined the nutritional index, ECI, which is the ratio of RGR/RCR. It thus permits a distinction to be made between reduced feeding (relative consumption rate) on a chemical and its actual toxic effect on growth (relative growth rate; e.g., Blau et al. 1978). Many fewer data were collected on ECI, limiting the power of the analysis to detect patterns. Nevertheless, it at least permitted us to check some outcomes of the analysis that used final weight as a measure of toxicity.

Data Analysis Toxicity was measured as the rate of mortality increase, final weight reduction, or decreased ECI with increasing chemical dosage. All mortality, weight, and ECI values were first converted to percentages of control values, and chemical dosages were converted into proportions of wet weight of the diet. Percentages were then transformed to standardized normal deviates, and these were regressed on log10-transformed chemical dosage. The transformations

and regression form the basis for probit analysis, a venerable technique for evaluating chemical toxicity in bioassays (Finney 1971; Sokal and Rohlf 1995). The regression was constrained to pass through the control value. A small number (0.0001) was added to each dosage to avoid taking the logarithm of 0 for the control. The resulting regression coefficients (slopes) generated the toxicity values for each of the three toxicity variables (mortality, weight, ECI). The ED50’s were converted to regression coefficients by dividing the standard normal deviate for 50% by the logarithm of the dosage required to increase mortality or reduce weight by 50%. The toxicity variables had to be adjusted before they were used in the main analyses. Although the majority of bioassays tested toxicity on the first larval instar, some tested later instars. Different studies were also run at different temperatures and allowed larvae to feed for different lengths of time. In order to reduce the effects of this variation, each toxicity variable was first regressed against three covariates: larval instar used in the feeding test, duration of larval feeding as a proportion of total larval period in days, and temperature (⬚C). The significance of each covariate was determined. Duration of feeding was significant for mortality, instar tested was significant for weight, and both were significant for ECI. Each toxicity variable was then regressed on its significant covariates, and the residuals from these regressions were used as the dependent variables in the main analyses. Regression was used in the main analyses to assess the effects of three independent variables (specialism, chemical presence/absence in the herbivore’s normal host, and chemical floristic distribution) on the three toxicity variables (mortality, weight, and ECI) generated in the probit analysis. The approach in each main analysis was to include all three independent variables and relevant interaction terms in a multiple regression. Three main analyses were performed, one for each of the three toxicity variables. A model selection procedure was used to determine the best regression on the basis of the largest R2. The results were then used as a guide to relevant patterns that were

Table 1: Multiple regression analysis of herbivore mortality on three independent variables and one interaction term Parameter Constant Chemical in normal host Chemical taxon range Specialism 1 # 3 interaction

Estimate

SE

t statistic

P

2.14448 ⫺1.28724 ⫺.154252 ⫺.330788 .266625

.514541 .355997 .0333899 .207011 .142382

4.16775 ⫺3.61585 ⫺4.61972 ⫺1.59792 1.8726

!.00001

.0003 !.00001

.1109 .0619

Note: For ANOVA, model: SS p 83.9198, df p 4, MS p 20.9799, F ratio p 17.31, P ! .00001; residual: SS p 436.329, df p 360, MS p 1.21202; total: SS p 520.248, df p 364. R2 p 0.152, SE of estimate p 1.10092.

Herbivore Response to Plant Chemistry

511

Table 2: Multiple regression analysis of herbivore final weight on two independent variables and two interaction terms Parameter Constant Chemical in normal host Chemical taxon range Specialism # 1 interaction Specialism # 1 # 2 interaction

Estimate

SE

t statistic

P

⫺3.78257 .87914 .719917 .351778

.40473 .230164 .0987429 .11004

⫺9.34591 3.81963 7.29082 3.19681

!.00001

.023275

⫺4.04203

.0001

⫺.0940782

.0002 !.00001

.0015

Note: For ANOVA, model: SS p 515.668, df p 4, MS p 128.917, F ratio p 38.82, P ! .00001; residual: SS p 1,680.58, df p 506, MS p 3.3213; total: SS p 2,196.24, df p 510. R2 p 0.229, SE of estimate p 1.82244.

examined in more detail with simple regression. Outcomes from the main analyses that would support the four predictions of phytochemical coevolution theory are as follows. First, if chemicals are less toxic to adapted than nonadapted herbivores, then presence/absence of the chemical in the normal host should have a significant impact on toxicity. Second, if this pattern is stronger for specialists than generalists, there should be a significant interaction between specialism and presence/absence of the chemical in the normal host. Third, if more widespread chemicals are less toxic than more narrowly distributed ones, then the floristic distribution of the chemical should have a significant impact on toxicity. Fourth, if the third prediction is more applicable to generalists than specialists—which, in turn, depends on whether the herbivore is or is not adapted to the chemical—then there should be a significant three-way interaction between specialism, the floristic distribution of the chemical, and presence/absence of the chemical in the normal host.

224, n p 918, P ! .00001) because two of the orders (Coleoptera and Hymenoptera) have no oligophages or generalists in the data set. To test for lack of independence, we redid the main analyses after reducing the data set to include just one chemical per insect species. Chemicals were chosen to represent as wide a range of chemical classes as possible. There were still some cases where the same chemical was fed to different insects, but the number of such cases was drastically reduced. This procedure was extremely conservative, since it removed more than 80% of sample size from the analyses. To test for misleading structure in the data, we repeatedly redid the main analyses after removing individual insect taxa and chemicals (e.g., each of the four insect orders, each of the six chemical classes, and each of the four most frequently tested insect species and the three most frequently tested chemicals). Because of the conservative nature of some of these tests, in any case where a main effect or interaction became insignificant, it was retested with a simple regression.

Data Dependencies and Spurious Correlations

Results

We chose parametric regression for the analysis because the variables do not deviate unacceptably from the usual assumptions (normality and homoscedasticity) and because parametric methods are relatively powerful at detecting patterns in noisy data. However, many of the studies examine the impacts of different chemicals on the same or closely related herbivore species and the impacts of the same chemical on different herbivore species. The data are thus not strictly independent. Such dependencies violate the assumptions of any statistical procedure that could be applied to the analysis. In addition, there may particular insect taxa or chemical classes that may be driving the patterns or are otherwise critical to the results, or there might be correlations and misleading structures in the data that have little to do with the variables of interest. For example, specialism correlates with insect order (F p

Main Analyses Three of the four predictions were supported at least marginally by the main analysis with mortality as the dependent variable. The best regression chosen by the model selection procedure included presence/absence of the chemical in the normal host, chemical floristic distribution, specialism, and the 1 # 3 interaction. Just these terms were then run through a multiple regression on mortality. The first two were highly significant, and the fourth was marginally significant (P ! .1; table 1). Specialism was not predicted to be significant by itself, and it was not. The three-way interaction among specialism, chemical floristic distribution, and presence/absence of the chemical in the normal host did not enter the model, giving no indication that generalists and specialists behave differently toward

512

The American Naturalist Table 3: Multiple regression analysis of herbivore ECI on two independent variables and two interaction terms Parameter

Estimate

SE

t statistic

P

Constant Chemical in normal host Chemical taxon range Specialism # 1 interaction Specialism # 1 # 2 interaction

.31475 .660657 ⫺.231095 ⫺.324219

.644551 .28934 .157902 .167376

.488324 2.28332 ⫺1.46354 ⫺1.93707

.6265 .0247 .1467 .0557

.057125

.039164

1.4586

.1480

Note: ECI p efficiency of conversion of ingested food. For ANOVA, model: SS p 9.89723, df p 4, MS p 2.47431, F ratio p 1.97, P p .1049; residual: SS p 117.859, df p 94, MS p 1.25382; total: SS p 127.757, df p 98. R2 p .07747, SE of estimate p 1.11974.

widely versus narrowly distributed chemicals. The model R2 was 0.16, indicating much unexplained variation in the analysis. All four predictions were supported by the main analysis with final weight as the dependent variable. The best regression model included presence/absence of the chemical in the normal host, chemical floristic distribution, the specialism # presence/absence of the chemical in the normal host interaction, and the three-way interaction among specialism, presence/absence of the chemical in the normal host, and chemical floristic distribution. Specialism by itself did not enter the model. The four terms that entered the model were again run through a multiple regression, and all were significant (table 2). The three-way interaction indicates that generalists indeed respond more to differences in the floristic distribution of novel chemicals than specialists. The model R2 was 0.24, somewhat higher than in the mortality analysis. The same four terms from the analysis with final weight were regressed on ECI as a check for bias caused by deterrence rather than toxicity. Trends were all in the right direction, and no strong biases were apparent. However, only two of the four terms were significant (one marginally so), probably in part because of a sample size that was more than 80% smaller than in the analysis using final weight (98 vs. 510 df; table 3). The model R2 was only 0.08, the lowest of the three. Support for the first and third predictions came from two independent variables in the multiple regressions (presence/absence of the chemical in the normal host and floristic distribution of the chemical). Both variables make strong and consistent contributions to the mortality and weight analyses, a contention reiterated by simple regression (figs. 1, 2). Support for the second and fourth predictions came from interactions among these terms, and support was consistent for the second prediction and more ambiguous for the fourth prediction. Because of the less direct nature of this support and its ambiguity for the

fourth prediction, we performed additional analysis to explore these patterns further.

Subsidiary Analyses The second prediction holds that specialism is advantageous because the herbivore will thrive on chemicals to which it has become adapted. The disadvantage is that it will do poorly on others. The advantage of generalism is that feeding is successful on many different plants with varied chemistry. The disadvantage is that success is not as high as for the specialist on chemicals in the specialist’s normal hosts. The significant interactions between specialism and presence/absence of the chemical in the normal host support the prediction. However, it can be tested more directly. The data are first subdivided into two groups. The first group comprises all chemicals that are in the normal hosts of the more specialized species (specialists and oligophages). This is irrelevant for generalists, and so the first group includes chemicals regardless of whether they are in the normal hosts of the generalists. The second group comprises all chemicals that are not in the normal hosts of the more specialized species. Again, the criterion does not matter for generalists, and so the second group also includes chemicals regardless of whether they are in the normal hosts of the generalists. The data sets thus overlap somewhat, but this grouping has the advantage of increasing sample sizes. Separate regressions of toxicity versus specialism are then carried out on the two data groups. If the second prediction is correct, then toxicity should decrease with increasing specialism in the first group and increase with increasing specialism in the second group. Regression coefficients should therefore be positive for the first group and negative for the second group in the mortality analysis and vice versa in the final weight analysis (mortality increases with toxicity, whereas final weight decreases with


513

toxic ones. Chemical floristic distribution should therefore be a more important predictor of toxicity relative to presence/absence of the chemical in the normal host for generalists. These subsidiary predictions were tested in multiple regressions of presence/absence of the chemical in the normal host and chemical floristic distribution on toxicity (mortality and final weight). Separate regressions were run for specialists and generalists. Oligophages were not considered in this analysis. The relative importance of the independent variables was determined by comparing their slopes (regression coefficients) in each multiple regression. All variables were standardized before analysis in order to make the slopes comparable. With mortality as the dependent variable, presence/absence of the chemical in the normal host is more predictive of toxicity than chemical floristic distribution for specialists as predicted. The regression coefficient for the former is more than twice that of the latter (table 4). For gen-

Figure 1: Toxicity versus presence/absence of the chemical in the normal host of the herbivores. Toxicity is measured as (A) mortality and (B) final weight of the herbivores. Toxicity values are raw slopes from the probit analyses uncorrected for duration of feeding or instar tested. Regressions: A, Y p 2.64 ⫺ 1.07X, n p 447, R2 p 0.137, t p ⫺8.49, P ! .00001. B, Y p 1.60X ⫺ 4.09, n p 604, R2 p 0.137, t p 9.84, P ! .00001.

toxicity). This is indeed the case (figs. 3, 4), lending further support for the second prediction. The fourth prediction holds that the effects of chemical floristic distribution on toxicity should be stronger for generalists than specialists and should be stronger for chemicals to which the herbivore is not adapted than for those to which it is adapted. This latter effect should also depend on whether the herbivore is generalist or specialist. The ambiguous results from the main analyses prompted further examination of this prediction. If specialists pay a cost for specializing, then they should do worse on any chemical to which they are not adapted regardless of its floristic distribution. So presence/absence of the chemical in the normal host should be a more important predictor of toxicity relative to chemical floristic distribution for specialists. However, generalists should be able to handle novel chemicals as long as they are widespread and of low toxicity, but they should do poorly on narrowly distributed

Figure 2: Toxicity versus the distributional range of chemicals among plant taxa. Taxonomic range codes are as follows: 1 p one family, one genus; 2 p one family, two to 15 genera; 3 p two families; 4 p three to five families; 5 p six to 19 families; 6 p 119 families. Toxicity is measured as (A) mortality and (B) final weight of the herbivores. Toxicity values are raw slopes from the probit analyses uncorrected for duration of feeding or instar tested. Regressions: A, Y p 1.64 ⫺ 0.20X , n p 603, R2 p 0.087, t p ⫺6.21, P ! .00001. B, Y p 0.47X ⫺ 3.18, n p 525, R2 p 0.180, t p 10.79, P ! .00001.

514

The American Naturalist independent variables shifts with generalism and specialism according to the predictions. The shifts are not always large, which may explain why the tests of the fourth prediction in the main analyses were ambiguous. Moreover, the shift occurs from different relative levels of importance in the morality and weight analyses. Presence/absence of the chemical in the normal host is more predictive overall in the mortality analysis, whereas chemical floristic distribution is more predictive overall in the weight analysis. The reason for this difference is not known, but it explains in part why the independent variables do not switch in rank order but only shift in relative importance between the analyses for generalism and specialism. Tests for Dependence and Structure in the Data After reducing the data set to include just one chemical per species, the P values for each term in the main analysis

Figure 3: Toxicity versus herbivore specialization on host plants. Toxicity is measured as herbivore mortality and is corrected for duration of feeding and instar tested. Specialism is coded as follows: 1 p specialists, one to two families; 2 p oligophages, three to nine families; 3 p generalists, greater than nine families. A, The relationship when chemical is not in the normal host for specialists and oligophages. B, The relationship when chemical is in the normal host for specialists and oligophages. For generalists, all chemicals are included regardless of their presence in the normal host. Regressions: A, Y p 0.93 ⫺ 0.36X, n p 346, R2 p 0.049, t p ⫺4.35, P ! .00001. B, Y p 0.33X ⫺ 1.12, n p 290, R2 p 0.032, t p 3.25, P p .0013.

eralists, chemical floristic distribution increases in importance relative to presence/absence of the chemical in the normal host, which is also as predicted. However, the former does not overtake the latter in importance; both variables are about equally predictive of toxicity (table 4). With weight as the dependent variable, presence/absence of the chemical in the normal host is less predictive of toxicity than chemical floristic distribution for specialists, which is just the opposite of the case for mortality. However, the regression coefficients are actually quite similar, and the two variables have roughly equal predictive power (table 5). For generalists, chemical floristic distribution again increases in importance relative to presence/absence of the chemical in the normal host as predicted, but now the former exceeds the latter in predictive power (table 5). So to summarize, the relative importance of the two

Figure 4: Toxicity versus herbivore specialization on host plants. Toxicity is measured as herbivore final weight and is corrected for duration of feeding and instar tested. Specialism is coded as follows: 1 p specialists; 2 p oligophages; 3 p generalists. A, The relationship when chemical is not in the normal host for specialists and oligophages. B, The relationship when chemical is in the normal host for specialists and oligophages. For generalists, all chemicals are included regardless of their presence in the normal host. Regressions: A, Y p 0.71X ⫺ 1.92, n p 484, R2 p 0.057, t p 5.49, P ! .00001. B, Y p 1.23 ⫺ 0.34X, n p 489, R2 p 0.020, t p ⫺3.29, P p .0011.


515

Table 4: Multiple regression analysis of herbivore mortality on two independent variables for generalists and specialists separately Parameter Generalists: Constant Chemical Chemical Specialists: Constant Chemical Chemical

Estimate

SE

t statistic

P

in normal host taxon range

⫺.0581974 ⫺.207619 ⫺.209313

.0533936 .0593617 .061168

⫺1.08997 ⫺3.49753 ⫺3.42193

.2770 .0006 .0007


⫺.182723 ⫺.436992 ⫺.178759

.0919698 .0903502 .0826146

⫺1.98678 ⫺4.83664 ⫺2.16377

!.00001

.0497 .0329

Note: All variables standardized before analysis. Oligophages not included in this analysis. For generalists ANOVA, model: SS p 25.7048 , df p 2, MS p 12.8524, F ratio p 21.09, P ! .00001; residual: SS p 129.812, df p 213, MS p .609444; total: SS p 155.516, df p 215. R2 p .157, SE of estimate p .780669. For specialists ANOVA, model: SS p 19.3173, df p 2, MS p 9.65864, F ratio p 14.33, P ! .00001; residual: SS p 66.0591, df p 98, MS p .674072; total: SS p 85.3763, df p 100. R2 p .210, SE of estimate p .821019.

for final weight were .172, .006, .099, and .084 for chemical presence/absence in the herbivore’s normal host, chemical floristic distribution, the two-way interaction between specialism and chemical presence/absence in the normal host, and the three-way interaction among specialism, chemical presence/absence in the normal host, and chemical floristic distribution, respectively (n p 50). When final weight was regressed on each term in simple regressions, the P values were .003 (n p 58), .002 (n p 55), .1 (n p 53), and .02 (n p 50), respectively. In the main analysis for mortality, the P values were .058, .206, .790, and .668 for chemical presence/absence in the normal host, chemical floristic distribution, specialism, and the two-way interaction between specialism and chemical presence/absence in the normal host, respectively (n p 62). When mortality was regressed on each term in simple regressions, the P values were .0001 (n p 73), .108 (n p 66), .751 (n p 69), and .026 (n p 69), respectively. Thus, even in this extremely conservative analysis for lack of independence, the trends from the main analyses are still detectable, if not always in the multiple regressions, then at least in the simple ones. Removal of each of the three most frequently tested chemicals (tannin, rutin, gossypol) and the six chemical classes from the two main analyses had no qualitative effect on the results. All significant terms remained significant. Removal of three of the four most frequently tested species (Heliothus zea, Heliothus virescens, Calosobruchus maculatus) also had no qualitative effect on the main analyses. Removal of Pectinophora gossypiella from the final weight analysis reduced the significance of the two interaction terms (two-way interaction: P p .123; three-way interaction: P p .217, n p 459). Both of these terms include specialism, and Pectinophora comprises a substantial proportion of the specialists in this analysis. Nevertheless, both interaction terms were highly significant in simple

regressions without Pectinophora (two-way interaction: P p .0016, n p 521; three-way interaction: P p .00001, n p 459). Removal of each of the four insect orders (Lepidoptera, Orthoptera, Coleoptera, Hymenoptera) again had no qualitative effect on the mortality analysis. This in spite of the fact that most of the insect species are Lepidoptera, and removal of this order reduced the sample size by more than 75%. Removal of three of the four orders had no effect on the final weight analysis, but removal of Lepidoptera resulted in loss of significance of all four terms (presence/absence of chemical in normal host: P p .559; chemical floristic distribution: P p .217; two-way interaction: P p .244; three-way interaction: P p .989; n p 83). Removal of Lepidoptera resulted in over an 80% loss of sample size in this analysis. Nevertheless, all terms were significant in simple regressions except for chemical floristic distribution (P p .375, n p 87). In summary, there is some indication that Lepidoptera at least in part drive the trends in the final weight analysis, which is not surprising considering that most of the insects tested are Lepidoptera. In addition, Pectinophora partially but probably not completely drives some of the patterns attributed to specialism in the final weight analysis. Again, this is not surprising, since Pectinophora comprises a large proportion of specialists in this analysis. In spite of these trends, the main analyses were remarkably stable in these tests for independence and structure in the data. Discussion Several key predictions of the coevolution and apparency theories were supported by the analysis. Specialists pay a cost for but get the benefit of specialization, generalists pay a cost for but get the benefit of generalization, widespread chemicals are less toxic than narrowly distributed

516

The American Naturalist Table 5: Multiple regression analysis of herbivore final weight on two independent variables for generalists and specialists separately Parameter Generalists: Constant Chemical Chemical Specialists: Constant Chemical Chemical

Estimate

SE

t statistic

P


⫺.0273552 .201953 .259487

.0415379 .0464497 .0445482

⫺.658487 ⫺4.34778 ⫺5.82486

!.00001 !.00001


⫺.0532235 .273302 .474956

.104042 .113205 .110377

⫺.511559 2.41422 4.30305

!.0172 !.00001

.5107

.06098

Note: All variables standardized before analysis. Oligophages not included in this analysis. For generalists ANOVA, model: SS p 51.3248, df p 2, MS p 25.6624, F ratio p 46.65, P ! .00001; residual: SS p 185.994, df p 338, MS p .550131; total: SS p 237.269, df p 340. R2 p .212, SE of estimate p .741708. For specialists ANOVA, model: SS p 55.5986, df p 2, MS p 27.7993, F ratio p 19.33, P ! .00001; residual: SS p 188.425, df p 131, MS p 1.43836; total: SS p 244.023, df p 133. R2 p .216, SE of estimate p 1.19931.

ones, and, to some extent, specialists and generalists differ in their responses to chemical floristic distribution and to the presence/absence of chemicals in their normal hosts. However, the three measures of toxicity do not show the same sensitivity to the independent variables. Final weight is most sensitive, mortality is less sensitive, and ECI is the least sensitive. These differences are not surprising. The first effects of a toxin on herbivore fitness are expected to be reductions in growth rate. Only after longer-term feeding and/or higher dosage would individuals actually start to die (e.g., Erickson and Feeny 1974). Similarly, ECI is expected to be less sensitive than final weight independent of sample size. The ECI removes the effects of reduced consumption from the growth response to toxicity. Reduced consumption often works in combination with physiological toxicity to reduce final weight of the developing individual. The fact that ECI trends in the same direction as final weight suggests that physiological toxicity is an important component of herbivore fitness. Nevertheless, the low sensitivity of ECI implicates altered feeding behavior as an additional component. It is clear from the analysis that the four predictions are not deterministic. They are instead somewhat vague because of the noisiness of the data and because for each prediction supported by the analysis, there are ample counterexamples. Specialists often suffer negative effects from chemicals to which they are supposedly adapted (Appel and Martin 1992). Adapted generalists sometimes do as well on the chemicals in their normal hosts as adapted specialists (Blau et al. 1978; Fox and Morrow 1981). Nonadapted generalists sometimes do no better than nonadapted specialists on novel chemicals (this study). Widespread chemicals are not always more benign than narrowly distributed ones (this study). Nevertheless, the predictions are supported on average. In other words,

many processes might explain variation in toxicity, but there is a detectable signal that can be attributed to standard plant defense theories. Noisy data are a common problem with large-scale literature analyses for several reasons, some of which are methodological. Results of bioassays run on different diets by different individuals at different times and places are expected to vary for reasons that have little to do with plant defense. Sometimes the data are incomplete. For the independent variables, some of the lists documenting a chemical’s presence in a plant and plants fed on by particular herbivores are not comprehensive. For the dependent variables, regression coefficients measuring toxicity from converted ED50’s and those generated from only one chemical concentration and a control are based on only two data points. Obviously, such coefficients will vary more widely than those based on more data. Varying degrees of incompleteness will decrease the signal to noise ratio in the analysis, making it more difficult to detect significant patterns. The problem can be ameliorated somewhat by the large amounts of information compiled and by grouping the data, as we have done, into categories. Data should not be left out of the analysis just because they are incomplete. Such data contain valuable information that can be included in tests of broadscale comparisons. Methodological problems are not the only cause of data noisiness. Mechanisms other than those predicted by coevolution and apparency theory might be responsible for differences in toxicity and specialism. For example, resource availability theory predicts that differences in toxicity might depend on the costs versus benefits of defending plant parts (Coley 1983; Coley et al. 1985), regardless of chemical floristic distribution, insect adaptation to the host, or specialism. Another possibility is that

Herbivore Response to Plant Chemistry whole chemical classes (e.g., cyanogens vs. flavonoids) might vary in toxicity, perhaps because of differing evolutionary rates or simply by chance. In a similar vein, specialism might evolve as a response to not only plant chemistry but also natural enemy attacks (Bernays and Graham 1988) or other genetic considerations (Futuyma 1986). At present, variation due to these mechanisms must be absorbed into the error terms of our analyses until it can be thoroughly investigated. One possible objection to the use of laboratory bioassays to test defense theories is that they are too simple to infer the behavior of herbivores in natural settings. The effects of chemistry in nature can be made complicated by the internal state of the insect (Dethier 1982; Ramachandran et al. 1989), by genetic differences among individuals (Lindroth and Weisbrod 1991; Berenbaum and Zangerl 1993), by interactions between chemistry and nutritional level or other factors such as UV light (Berenbaum 1978; Lincoln et al. 1982; Gatehouse and Boulter 1983; Champagne et al. 1986; Bryant et al. 1987; Johnson and Bentley 1988; Berenbaum et al. 1991; Lindroth and Bloomer 1991; Trumble et al. 1991; Joseph et al. 1993; Berenbaum and Zangerl 1994), and by its impact on higher trophic levels (Thurston and Fox 1972; Campbell and Duffy 1979; Wisdom 1985; Faeth and Bultman 1986; McDougall et al. 1988; Barbosa et al. 1991; Trumble et al. 1991; Navon et al. 1993). Moreover, insects in nature are exposed to chemical complexes rather than single chemicals that have undergone various extractive protocols. Chemicals that are part of mixtures in real plant tissues may act in additive or synergistic ways (Lukefahr and Martin 1966; Berenbaum and Neal 1985; Farrar and Kennedy 1987; Berenbaum et al. 1991; Trumble et al. 1991; Berenbaum and Zangerl 1993). Even the extraction method may affect the chemical’s toxicity to an herbivore (Rowell-Rahier and Pasteels 1990). Available information suggests that low nutrition, high UV, and multiple chemicals generally add to or synergistically enhance toxicity. There are likely to be exceptions to these trends, but the general pattern is persuasive. Obviously, the chemical extraction method, the internal state of the insect, and genetic differences among individuals make chemical toxicity more variable (of course, genetic differences also provide the raw material for the coevolutionary process). Plant chemicals can be toxic to the predators, parasites, and diseases that attack herbivores, thus possibly reducing enemy impacts on herbivore mortality. However, chemical effects on the third trophic level are mixed and thus also make the overall effects of toxicity on herbivores more variable. As a result of these various complications, bioassays based on single chemicals in nutritious artificial diets may underestimate toxicity and its variability experienced by herbivores in nature. However, there is no reason to believe that these underestimates will

517

invalidate the qualitative patterns that emerged from this broadscale analysis. Indeed, patterns in nature may be stronger than those detected in the laboratory. Inferences based on single chemicals in nutritious artificial diets are probably conservative with respect to patterns of toxicity predicted by defense theory. Bioassays do have limitations. They are good at testing for herbivore adaptation to defense chemicals, but they can say little about how a particular chemical defense evolved. Jermy (1976, 1984) suggests that insects do not impose significant fitness reductions in plants and thus do not drive secondary chemical evolution. However, the bioassay data do provide circumstantial evidence that insects impose selection on plants. Insects generally become adapted to widespread chemicals and thus, by implication, increase selection on plants to alter their defenses. In addition, there is ample evidence from the plant resistance literature that resistance to herbivores varies among plant individuals and is under genetic control (Marquis 1990, 1992b; Berenbaum and Zangerl 1992), that herbivores can reduce plant fitness and thus impose selection on this resistance (Marquis 1984, 1992a, 1992b; Futuyma and Keese 1992), and that plants respond (Simms and Rausher 1987; Simms 1992; Bergelson and Purrington 1996). The coevolutionary cycle is thus plausible. There is little disagreement that insects undergo adaptive radiations in response to plant defenses (e.g., Strong et al. 1984; Janz and Nylin 1998). But do plants undergo adaptive radiations in response to escape from herbivory? We have shown that widespread chemicals tend to be less toxic than narrowly distributed ones, which suggests that such radiations have taken place in the past. But in order to infer “escape and radiation,” narrowly distributed chemicals must be younger and more derived, and widespread chemicals must be older and more primitive. But some chemicals such as nicotine might be widespread because they arose independently several times in different taxa as a result of their structural simplicity and ease of synthesis, not because they are older. The age-toxicity relationship has been shown for at least two cases where the biosynthetic pathway contains information on the trajectory of coevolution (Berenbaum 1983a; Farrell and Mitter 1998). Unfortunately, chemicals cannot always be organized biosynthetically because precursors for particular compounds are not always known, and if they are, their floristic distribution is often poorly documented. Moreover, defense escalation may not always follow predictable biosynthetic pathways; more toxic chemicals might arise unpredictably from unexpected precursors (e.g., Feeny 1977). Thus, we do not organize chemicals biosynthetically but simply test the relationship between the chemical’s floristic distribution and its toxicity. Available evidence suggests that floristic distribution correlates with age and

518


that examples like nicotine are rare. Plant “escape and radiation” may thus be inferred from the chemical distribution-toxicity relationship. Further support for plant “escape and radiation” comes from phylogenetic analysis. Past radiations leave a signature in the pattern of relatedness among plants containing the same chemical. Since at least the time of Ehrlich and Raven (1964), it has been known that closely related plants tend to share chemistry. Evidence for this pattern is mounting from modern molecular and phylogenetic techniques as well (Farrell et al. 1991; Becerra 1997; Farrell and Mitter 1998). Comparisons among taxa containing and lacking particular chemicals have indicated that adaptive radiations are associated with the acquisition of novel, increasingly more toxic defenses (Farrell et al. 1991; Mitter and Farrell 1991; Farrell and Mitter 1998). Other circumstantial evidence also supports the “escape and radiation” model (Farrell and Mitter 1993; Becerra 1997; Farrell 1998; Janz and Nylin 1998; Kelly and Farrell 1998; Janz et al. 2001). Phylogenetic analysis does not always support escape and radiation (Farrell and Mitter 1993), and flowering plants may have diversified too early to have been affected by Lepidopteran feeding, as was proposed by Ehrlich and Raven (Janz and Nylin 1998). But evidence for the relationship between adaptive diversification and key innovations is persuasive, bolstering our inference that chemicals are widespread because of such diversifications. Conclusion With this analysis, we have come full circle from Ehrlich and Raven’s classic article. They conducted a literature analysis looking for the signal of coevolution in taxonomic and chemical distribution data. Their analysis, in conjunction with studies by Dethier, Fraenkel, and Feeny, stimulated a rash of work testing for coevolution and subsidiary theories with laboratory bioassays. Here we do a literature analysis of the bioassay work, looking for coevolution and apparency effects in the distribution of chemicals and their toxicity to herbivores. Despite past skepticism about the relevance of coevolution to insect/ plant interactions, the bioassay data support the phytochemical coevolution model. Coevolution may be diffuse, and microorganisms may also select for plant defense, but the signal of coevolution still appears in the differential responses of herbivores to toxic chemicals. In addition, numerous phylogenetic analyses support “escape and radiation,” a critical prediction of coevolution. However, at least two key questions are unanswered by our study. First, are the patterns that we found uniquely predicted by standard coevolution and apparency theories, or can they be otherwise explained? This question will have to remain open until other mechanisms are presented and explored

in detail. Second, how much of the unexplained variation in toxicity can be explained by alternative defense theories? More immediate progress can be made on this question. Future studies should take a more comprehensive view of plant defense, testing the relative importance of various processes rather than treating them as mutually exclusive alternatives. Large-scale analyses such as ours may provide a good starting point for teasing apart the effects of these processes once the appropriate hypotheses are formulated. Acknowledgements We wish to thank the following people who generously provided helpful information and who guided us to relevant sources: R. Chapman, E. Connor, M. Crawley, S. Faeth, S. Hartley, A. Herzig, M. Hunter, M. Isman, J. Myers, G. Sword, and D. Tallamy. Thanks are also due to D. Tallamy, J. Thompson, A. Weis, and two anonymous reviewers for critically reading the manuscript and for their helpful suggestions. This work was supported in part by National Science Foundation (NSF) DEB-9628929 and in part by the National Center for Ecological Analysis and Synthesis, a center funded by NSF (DEB-9421535), the University of California, Santa Barbara, and the State of California. Literature Cited Appel, H. M., and M. M. Martin. 1992. Significance of metabolic load in the evolution of host specificity of Manduca sexta. Ecology 73:216–228. Barbosa, P., P. Gross, and J. Kemper. 1991. Influence of plant allelochemicals on the tobacco hornworm and its parasitoid Cotesia congregata. Ecology 72:1567–1575. Becerra, J. X. 1997. Insects on plants: macroevolutionary trends in host use. Science (Washington, D.C.) 276: 253–256. Beckwith, R. C. 1970. Influence of host on larval survival and adult fecundity of Choristoneura conflictana (Lepidoptera: Tortricidae). Canadian Entomologist 102: 1474–1480. Berenbaum, M. 1978. Toxicity of a furanocoumarin to armyworms: a case of biosynthetic escape from insect herbivores. Science (Washington, D.C.) 201:532–533. ———. 1981a. Effect of linear furanocoumarins on an adapted specialist insect (Papilio polyxenes). Ecological Entomology 6:345–351. ———. 1981b. Patterns of furanocoumarin distribution and insect herbivory in the Umbelliferae: plant chemistry and community structure. Ecology 62:1254–1266. ———. 1983. Coumarins and caterpillars: a case for coevolution. Evolution 37:163–179. Berenbaum, M., and J. J. Neal. 1985. Synergism between myristicin and xanthotoxin, a naturally co-occurring

Herbivore Response to Plant Chemistry plant toxicant. Journal of Chemical Ecology 11: 1349–1358. Berenbaum, M., and A. R. Zangerl. 1992. Genetics of secondary metabolism and herbivore resistance in plants. Pages 415–438 in G. R. Rosenthal and M. R. Berenbaum, eds. Herbivores: their interactions with secondary plant metabolites. Vol. 2. Evolutionary and ecological processes. Academic Press, London. ———. 1993. Furanocoumarin metabolism in Papilio polyxenes: biochemistry, genetic variability, and ecological significance. Oecologia (Berlin) 95:370–375. ———. 1994. Costs of inducible defense: protein limitation, growth, and detoxification in parsnip webworms. Ecology 75:2311–2317. Berenbaum, M., J. K. Nitao, and A. R. Zangerl. 1991. Adaptive significance of furanocoumarin diversity in Pastinaca sativa (Apiaceae). Journal of Chemical Ecology 17:207–215. Bergelson, J., and C. B. Purrington. 1996. Surveying patterns in the cost of resistance in plants. American Naturalist 148:536–558. Bernays, E. A. 1990. Plant secondary compounds deterrent but not toxic to the grass specialist acridid Locusta migratoria: implications for the evolution of graminivory. Entomologia Experimentalis et Applicata 54:53–56. ———. 1991. Relationship between deterrence and toxicity of plant secondary compounds for the grasshopper Schistocerca americana. Journal of Chemical Ecology 17: 2519–2526. Bernays, E. A., and R. F. Chapman. 1994. Host-plant selection by phytophagous insects. Chapman & Hall, New York. Bernays, E. A., and M. Graham. 1988. On the evolution of host specificity in phytophagous arthropods. Ecology 69:886–892. Blau, P. A., P. Feeny, L. Contardo, and D. S. Robson. 1978. Allylglucosinolate and herbivorous caterpillars: a contrast in toxicity and tolerance. Science (Washington, D.C.) 200:1296–1298. Bohlmann, F., T. Burkhardt, and C. Zdero. 1973. Naturally occurring acetylenes. Academic Press, London. Bowers, M. D. 1984. Iridoid glycosides and host-plant specificity in larvae of the buckeye butterfly Junonia coenia (Nymphalidae). Journal of Chemical Ecology 10: 1567–1577. Bryant, J. P., T. P. Clausen, P. B. Reichardt, M. C. McCarthy, and R. A. Werner. 1987. Effect of nitrogen fertilization upon secondary chemistry and nutritional value of quaking aspen (Populus tremuloides Michx.) leaves for the large aspen tortrix (Choristoneura conflictana [Walker]). Oecologia (Berlin) 73:513–517. Campbell, B. C., and S. S. Duffy 1979. Tomatine and parasitic wasps: potential incompatibility of plant antibiosis

519

with biological control. Science (Washington, D.C.) 205: 700–702. Champagne, D. E., J. T. Arnason, B. J. R. Philogene, P. Morand, and J. Lam. 1986. Light-mediated allelochemical effects of naturally occurring polyacetylenes and thiophenes from Asteraceae on herbivorous insects. Journal of Chemical Ecology 12:835–857. Coley, P. D. 1983. Herbivory and defensive characteristics of tree species in a lowland tropical forest. Ecological Monographs 53:209–233. Coley, P. D., J. P. Bryant, and F. S. Chapin III. 1985. Resource availability and plant antiherbivore defense. Science (Washington, D.C.) 230:895–899. Dethier, V. G. 1954. Evolution of feeding preferences in phytophagous insects. Evolution 8:33–54. ———. 1982. Mechanism of host plant recognition. Entomologia Experimentalis et Applicata 31:44–56. Ehrlich, P. R., and P. H. Raven. 1964. Butterflies and plants: a study in coevolution. Evolution 18:586–608. Ellis, G. P. 1977. Chromenes, chromanones, and chromones: heterocyclic compounds. Vol. 31. Wiley, New York. Erickson, J. M., and P. P. Feeny. 1974. Sinigrin: a chemical barrier to the black swallowtail butterfly, Papilio polyxenes. Ecology 55:103–111. Faeth, S. H., and T. L. Bultman. 1986. Interacting effects of increased tannin levels on leaf-mining insects. Entomologia Experimentalis et Applicata 40:297–300. Farrar, R. R., and G. G. Kennedy. 1987. 2-undecanone, a constituent of the glandular tricomes of Lycopersicon hirsutum f. glabratum: effects on Heliothus zea and Manduca sexta growth and survival. Entomologia Experimentalis et Applicata 43:17–23. Farrell, B. D. 1998. “Inordinate fondness” explained: why are there so many beetles? Science (Washington, D.C.) 281:555–559. Farrell, B. D., and C. Mitter. 1993. Phylogenetic determinants of insect/plant community diversity. Pages 253–266 in R. E. Ricklefs and D. Schluter, eds. Species diversity in ecological communities: historical and geographic perspectives. University of Chicago Press, Chicago. ———. 1998. The timing of insect/plant diversification: might Tetraopes (Coleoptera: Cerambycidae) and Asclepias (Asclepiadaceae) have coevolved? Biological Journal of the Linnean Society 63:553–577. Farrell, B. D., D. E. Dussourd, and C. Mitter. 1991. Escalation of plant defense: do latex and resin canals spur plant diversification? American Naturalist 138:881–900. Feeny, P. P. 1976. Plant apparency and chemical defense. Pages 1–40 in J. W. Wallace and R. L. Mansell, eds. Biochemical interactions between plants and insects. Plenum, New York.

520


———. 1977. Defense ecology of the Cruciferae. Annals of the Missouri Botanical Garden 64:221–234. Finney, D. J. 1971. Probit analysis. Cambridge University Press, Cambridge. Fox, L. R., and P. A. Morrow. 1981. Specialization: species property or local phenomenon? Science (Washington, D.C.) 211:887–893. Fraenkel G. S. 1959. The raison d’etre of secondary plant substances. Science (Washington, D.C.) 129:473–486. Futuyma, D. J. 1976. Food plant specialization and environmental predictability in Lepidoptera. American Naturalist 110:285–292. ———. 1986. The role of behavior in host-associated divergence in herbivorous insects. Pages 295–302 in M. D. Huettel, ed. Evolutionary genetics of invertebrate behavior. Plenum, New York. Futuyma, D. J., and M. C. Keese. 1992. Evolution and coevolution of plants and phytophagous arthropods. Pages 439–475 in G. R. Rosenthal and M. R. Berenbaum, eds. Herbivores: their interactions with secondary plant metabolites. Vol. 2. Evolutionary and ecological processes. Academic Press, London. Futuyma, D. J., and G. Moreno. 1988. The evolution of ecological specialization. Annual Review of Ecology and Systematics 19:207–233. Gatehouse, A. M. R., and D. Boulter. 1983. Assessment of the antimetabolic effects of trypsin inhibitors from cowpea (Vigna unguiculata) and other legumes on development of the bruchid beetle Callosobruchus maculatus. Journal of the Science of Food and Agriculture 34: 345–350. Gibbs, R. D. 1974. Chemotaxonomy of flowering plants. Vols. 1–4. McGill-Queen’s University Press, Montreal. Gilbert, L. E. 1971. Butterfly-plant coevolution: has Passiflora adenopoda won the selectional race with heliconiine butterflies? Science (Washington, D.C.) 172: 585–586. ———. 1975. Ecological consequences of a coevolved mutualism between butterflies and plants. Pages 210–240 in L. E. Gilbert and P. H. Raven, eds. Coevolution of animals and plants. University of Texas Press, Austin. Gould, F. 1990. Ecological genetics and integrated pest management. Pages 441–458 in C. R. Carroll, J. H. Vandermeer, and P. Rosset, eds. Agroecology. McGraw-Hill, New York. Harbourne, J. B., and H. Baxter, eds. 1993. Phytochemical dictionary: a handbook of bioactive compounds from plants. Taylor & Francis, London. Howe, H. F., and L. C. Westley. 1988. Ecological relationships of plants and animals. Oxford University Press, Oxford. Jaenike, J. 1990. Host specialization in phytophagous in-

sects. Annual Review of Ecology and Systematics 21: 243–273. Janz, N., and S. Nylin. 1998. Butterflies and plants: a phylogenetic study. Evolution 52:486–502. Janz, N., K. Nyblom, and S. Nylin. 2001. Evolutionary dynamics of host-plant specialization: a case study of the tribe Nymphalini. Evolution 55:783–796. Jensen, S. R., B. J. Nielsen, and R. Dahlgren. 1975. Iridoid compounds, their occurrence and systematic importance in the angiosperms. Botaniska Notiser 128: 148–180. Jermy, T. 1976. Insect-host plant relationship: coevolution or sequential evolution? Symposia Biologica Hungarica 16:109–113. ———. 1984. Evolution of insect/host plant relationships. American Naturalist 124:609–630. Joern, A. 1985. Grasshopper dietary (Orthoptera: Acrididae) from a Nebraska sand hills prairie. Transactions of the Nebraska Academy of Sciences 13:21–32. Johnson, N. D., and B. L. Bentley. 1988. Effects of dietary protein and lupine alkaloids on growth and survivorship of Spodoptera eridania. Journal of Chemical Ecology 14: 1391–1403. Joseph, G., R. G. Kelsey, A. F. Moldenke, J. C. Miller, R. E. Berry, and J. G. Wernz. 1993. Effects of nitrogen and Douglas-fir allelochemical on development of gypsy moth, Lymantria dispar. Journal of Chemical Ecology 19:1245–1263. Kelly, S. T., and B. D. Farrell. 1998. Is specialization a dead end? the phylogeny of host use in Dendroctonus bark beetles (Scolytidae). Evolution 52:1731–1743. Kjaer, A. 1960. Naturally derived isothiocyanates (mustard oils) and their parent glucosides. Fortschritte der Chemie Organischer Naturstoffe 18:123–176. Lincoln, D. E., T. S. Newton, P. R. Ehrlich, and K. S. Williams. 1982. Coevolution of the checkerspot butterfly Euphydryas chalcedona and its larval food plant Diplacus aurantiacus: larval response to protein and leaf resin. Oecologia (Berlin) 52:216–223. Lindroth, R. L., and M. S. Bloomer. 1991. Biochemical ecology of the forest tent caterpillar: responses to dietary protein and phenolic glycosides. Oecologia (Berlin) 86: 408–413. Lindroth, R. L., and A. V. Weisbrod. 1991. Genetic variation in response of the gypsy moth to aspen phenolic glycosides. Biochemical Systematics and Ecology 19: 97–103. Lukefahr, M. J., and D. F. Martin. 1966. Cotton-plant pigments as a source of resistance to the bollworm and tobacco budworm. Journal of Economic Entomology 59:176–179. Marquis, R. 1984. Leaf herbivores decrease fitness of a tropical plant. Science (Washington, D.C.) 226:537–539.

Herbivore Response to Plant Chemistry ———. 1990. Genotypic variation in leaf damage in Piper arieianum (Piperaceae) by a multispecies assemblage of herbivores. Evolution 44:104–120. ———. 1992a. A bite is a bite is a bite? constraints on response to folivory in Piper arieianum (Piperaceae). Ecology 73:143–152. ———. 1992b. Selective impact of herbivores. Pages 301–325 in R. S. Fritz and E. L. Simms, eds. Plant resistance to herbivores and pathogens: ecology, evolution, and genetics. University of Chicago Press, Chicago. McDougall, C., B. J. R. Philogene, J. T. Arnason, and N. Donskov. 1988. Comparative effects of two plant secondary metabolites on host-parasitoid association. Journal of Chemical Ecology 14:1239–1252. Mitter, C., and B. D. Farrell. 1991. Macroevolutionary aspects of insect/plant relationships. Pages 35–78 in E. A. Bernays, ed. Insect-plant interactions. Vol. 2. CRC, Boca Raton, Fla. Montllor, C. B., E. A. Bernays, and R. V. Barbehenn. 1990. Importance of quinolizidine alkaloids in the relationship between larvae of Uresiphita reversalis (Lepidoptera: Pyralidae) and a host plant Genista monspessulana. Journal of Chemical Ecology 16:1853–1865. Mulkern, G. B., D. R. Tozcek, and M. A. Brusven. 1964. Biology and ecology of North Dakota grasshoppers. 2. Food habits and preferences of grasshoppers associated with the sand hills prairie. North Dakota Agricultural Experiment Station Research Report no. 11. Murray, R. D. H., J. Mendez, and S. A. Brown. 1982. The natural coumarins: occurrence, chemistry, and biochemistry. Wiley, New York. Navon, A. 1985. Spodoptera littoralis. Pages 469–475 in P. Sigh and R. F. Moore, eds. Handbook of insect rearing. Vol. 2. Elsevier, Amsterdam. Navon, A., J. D. Hare, and B. A. Federici. 1993. Interactions among Heliothis virescens larvae, cotton condensed tannin, and the CryI(c) d-endotoxin of Bacillus thuringiensis. Journal of Chemical Ecology 19:2485–2499. Oliver, G. W. 1907. The mulberry and other silkworm food plants. USDA Bureau of Plant Industry Bulletin no. 119. Government Printing Office, Washington, D.C. Pimentel, D., L. McLaughlin, A. Zepp, B. Lakitan, T. Kraus, P. Kleinman, F. Vancini, et al. 1991. Environmental and economic effects of reducing pesticide use. BioScience 41:402–409. Popov, G., and M. Ratcliffe. 1968. The sahelian tree locust Anacridium melanohodon. Anti-locust Memoir 9. Ministry of Overseas Development, Anti-locust Research Centre, London. Price, P. W. 1980. Evolutionary biology of parasites. Princeton University Press, Princeton, N.J. Ramachandran, R., S. N. Mukherjee, and R. N. Sharma. 1989. Effects of food deprivation and concentration of

521

azadirachtin on the performance of Achoea janata and Spodoptera litura on young and mature leaves of Ricinus communis. Entomologia Experimentalis et Applicata 51: 29–35. Reese, J. C., and S. D. Beck. 1976a. Effects of allelochemics on the black cutworm, Agrotis ipsilon: effects of pbenzoquinone, hydroquinine, and duroquinone on larval growth, development, and utilization of food. Annals of the Entomological Society of America 69:59–67. Rhoades, D. F., and R. G. Cates. 1976. Toward a general theory of plant antiherbivore chemistry. Recent Advances in Phytochemistry 19:168–213. Rowell-Rahier, M., and J. M. Pasteels. 1990. Phenolglucosides and interactions at three trophic levels: Salicaceae-herbivores-predators. Pages 75–94 in E. A. Bernays, ed. Insect-plant interactions. Vol. 2. CRC, Boca Raton, Fla. Scriber, J. M. 1983. Evolution of feeding specialization, physiological efficiency, and host races in selected Papilionidae and Saturnidae. Pages 159–204 in R. F. Denno and M. McClure, eds. Variable plants and herbivores in natural and managed systems. Academic Press, New York. ———. 1984. Host-plant suitability. Pages 159–204 in W. J. Bell and R. T. Carde, eds. Chemical ecology of insects. Chapman & Hall, New York. Seaman, F. C. 1982. Sesquiterpene lactones as taxonomic characters in the Asteraceae. Botanical Review 48: 121–595. Simms, E. L. 1992. Costs of plant resistance to herbivory. Pages 392–425 in R. S. Fritz and E. L. Simms, eds. Plant resistance to herbivores and pathogens: ecology, evolution, and genetics. University of Chicago Press, Chicago. Simms, E. L., and M. D. Rausher. 1987. Costs and benefits of plant resistance to herbivory. American Naturalist 130:570–581. Sokal, R. R., and F. J. Rohlf. 1995. Biometry. 3d ed. W. H. Freeman, New York. Soo Hoo, C. F., and G. Fraenkel. 1966a. The consumption, digestion, and utilization of food plants by a polyphagous insect, Prodenia eridania. Journal of Insect Physiology 12:711–730. ———. 1966b. The selection of food plants in a polyphagous insect, Prodenia eridania. Journal of Insect Physiology 12:693–709. Strong, D. R., J. H. Lawton, and T. R. E. Southwood. 1984. Insects on plants: community patterns and mechanisms. Blackwell Scientific, Oxford. Stumpf, P. K., and E. E. Conn, eds. 1981. The biochemistry of plants: a comprehensive treatise. Vol. 7. Secondary plant products. Academic Press, New York.

522


Thompson, J. N. 1994. The coevolutionary process. University of Chicago Press, Chicago. Thurston, R., and P. M. Fox. 1972. Inhibition by nicotine of Apanteles congregatus from its host the tobacco hornworm. Annals of the Entomological Society of America 65:547–550. Tietz, H. M. 1972. An index to the described life histories, early stages, and hosts of the Macrolepidoptera of the continental United States and Canada. Vols. 1–2. Allyn Museum of Entomology, Sarasota, Fla. Trumble, J. T., W. J. Moar, M. J. Brewer, and W. G. Carson. 1991. Impact of UV radiation on activity of linear furanocoumarins and Bacillus thuringiensis var. kursaki against Spodoptera exigua: implications for tritrophic interactions. Journal of Chemical Ecology 17:973–987. Van Tienderen, P. H. 1991. Evolution of generalists and specialists in heterogeneous environments. Evolution 45:1317–1331. Via, S. 1990. Ecological genetics and host adaptation in

herbivorous insects: the experimental study of evolution in natural and agricultural systems. Annual Review of Entomology 35:421–446. Waldbauer, G. P. 1968. The consumption and utilization of food by insects. Advances in Insect Physiology 5: 229–288. Williamson, J. J., and B. G. Shubert 1961. Alkaloid-bearing plants and their contained alkaloids. USDA Technical Bulletin 1234. Wint, W. 1983. The role of alternative host-plant species in the life of a polyphagous moth, Operophtera brumata (Lepidoptera: Geometridae). Journal of Animal Ecology 52:439–450. Wisdom, C. S. 1985. Use of chemical variation and predation as plant defenses by Encelia farinosa against a specialist herbivore. Journal of Chemical Ecology 11: 1553–1565. Associate Editor: Mark E. Ritchie