Oikos 120: 1509–1518, 2011 doi: 10.1111/j.1600-0706.2011.19053.x © 2011 The Authors. Oikos © 2011 Nordic Society Oikos Subject Editor: Matt Ayres. Accepted 10 February 2011
Abundance trumps quality: bi-trophic performance and parasitism risk fail to explain host use in the fall webworm Peri A. Mason, Steven R. Wilkes, John T. Lill and Michael S. Singer P. A. Mason (
[email protected]) and M. S. Singer, Biology Dept, Wesleyan Univ., Middletown, CT 06459, USA. – S. R. Wilkes and J. T. Lill, Dept of Biological Sciences, The George Washington Univ., Washington, DC 20052, USA.
Factors including host abundance, quality, and the degree to which hosts provide enemy-free space (EFS) may drive host plant choice by phytophagous insects. Herbivores may also experience fitness tradeoffs among hosts, promoting polyphagy. The fall webworm Hyphantria cunea is a dietary generalist that feeds on a broad array of trees across its geographic range. Here, we investigate the drivers of host tree use by the fall webworm in Connecticut (CT) and Maryland (MD). Neither caterpillar performance nor EFS was associated with the frequency with which host trees were used, and no tradeoff between host quality and EFS was identified. Vegetation surveys adjacent to host trees showed that at both localities, host use was non-random with respect to tree species, and that the main predictor of use among suitable host trees was host tree abundance. This suggests that webworms are under selection to reduce search time for oviposition sites. Although we did not detect a tradeoff between host plant quality and availability in MD, we did identify that tradeoff in CT. This disparity amid otherwise similar patterns of host use between CT and MD may be explained by the relative rarity of high quality hosts in CT compared to MD. Our results illustrate that geographic mosaics in patterns of host use may arise in the absence of local adaptation if host use is based upon availability rather than host plant attributes.
Tri-trophic studies provide unique opportunities to improve our understanding of patterns of host use in phytophagous insects. Although enemy-free space (EFS) hypotheses have been invoked most frequently when bi-trophic hypotheses (e.g. physiological efficiency) fail to explain host specificity (Denno et al. 1990, Murphy 2004), recent work extends this reasoning to studies of host use in generalist herbivores (Singer et al. 2004). Polyphagy is predicted when environments vary such that natural selection favors different hosts under different circumstances (Via 1984), or if aspects of a particular host’s suitability vary with respect to one-another so that alternative hosts provide the herbivore equivalent fitness (Strong et al. 1984, Michaud 1990). It is clear then, that the explanatory power of bi- and tri-trophic hypotheses in explaining host choice is limited when they are viewed as mutually exclusive. A productive alternative is to assess how various needs of the insect may act jointly to optimize fitness (Bernays and Graham 1988, Strong 1988). This includes the tri-trophic sacrifice of food quality for EFS (Murphy 2004, Singer et al. 2004), but is by no means limited to that exchange (Singer 2008). For example, when there are temporal limitations on oviposition such as host plant phenology or the longevity of the insect itself, females may become less choosy, preferring to deposit some eggs onto hosts of intermediate quality over losing the opportunity to deposit the full complement of eggs (Singer et al. 1992). Trading off host quality for host availability in this way seems particularly likely when
high quality hosts are rare or their availability unpredictable (Jaenike 1978, Doak et al. 2006). The ecological drivers of host use are expected to vary geographically if the herbivore encounters different selection environments across its geographic range (Thompson 2005, Schluter 2009). Diet breadth itself may vary geographically if, for example, only a small subset of usable hosts is found at a given locality, or if natural selection favors a greater degree of dietary specificity at some sites than others (Fox and Morrow 1981, Thompson 1998). Though coevolutionary mosaic theory has largely been applied to specialists to date, it also has applications for taxa involved in more diffuse co-evolutionary relationships, such as the relationship between generalist herbivores and their host plants (Thompson 2005). If gene flow is limited and natural selection strong, a generalist species may become locally adapted to vegetative communities and natural enemy regimes that differ across its geographic range, causing it to form site-specific, host-plant preference hierarchies (Wiklund 1981, Thompson 1988a, 1993, Feeny 1995). The covariance of bi- and tri-trophic performance parameters with host preference at a given locality would constitute evidence for local adaptation (Egan and Ott 2007). Although the bi-trophic, preference-performance relationship has been investigated for oligophagous species over different parts of their range (Thompson 1993), local adaptation has not been studied as extensively either in herbivores that are broad dietary generalists, or from a tri-trophic 1509
perspective. Here we investigate the drivers of host use in the fall webworm, Hyphantria cunea, a dietary generalist with a large geographic distribution, in Connecticut (CT) and Maryland (MD) populations. In particular we ask whether: 1) H. cunea shows differential performance on various host species, as indicated by survival, feeding efficiency, developmental time, and pupal mass; 2) H. cunea experiences variable threat of mortality from parasitoids across hosts; 3) host availability, quality, or EFS account for observed patterns of host use; and 4) tradeoffs among these variables account for these patterns.
Study system The fall webworm Hyphantria cunea (Arctiidae) is a gregarious caterpillar that feeds and builds silken webs on the terminal branches of a wide range of deciduous forest, understory, and fruit tree species. The moth is native to North America, ranging from Canada to Mexico, and has been introduced into other continents where it has become an invasive pest (Gomi and Takeda 1996). Although the host range of this widely distributed species is estimated to encompass as many as 636 plants (Warren and Tadic 1970), regular host use at a given locality typically includes a dozen or fewer tree species (Fig. 1), and preference hierarchies appear to vary regionally in both its native and introduced ranges (Berger 1906, Oliver 1964, Greenblatt et al. 1978). Because H. cunea caterpillars experience high levels of parasitism (Morris 1976, Wagner 2005), and relatively low larval predation by other invertebrate predators and birds (Morris 1972a, b), estimating EFS in terms of parasitism is reasonable. Red-headed and black-headed morphs of H. cunea are found in both CT and MD, which may represent different lineages (Jaenike and Selander 1980). However, only black-headed morphs, which are bivoltine with a summer and fall generation in both MD and CT, were included in this study.
Methods Natural patterns of host plant use To assess patterns of host use, field surveys were conducted at several sites in Maryland and Connecticut, USA. H. cunea webs are found almost exclusively along roads, rivers or other edge-habitats which may act as flyways for adult moths (Barbosa and Greenblatt 1979). In MD, nine surveys were conducted, each averaging about four hours in the field at two field locations in the greater Washington DC area: Patuxent National Refuge (PNWR, hereafter) located in Laurel, MD, and the Chesapeake and Ohio (C&O hereafter) Canal between Locks 9 and 15 (Montgomery County, MD). These field sites represent distinct ecological habitats, PNWR being a mix of bottomland forest and upland beech/oak/ hickory forest, and C&O being riparian strips of lowland tree species. Surveys of MD sites began on 13 May 2006, and continued through the first week of June, corresponding to H. cunea’s first seasonal generation. Because H. cunea webs are sparse during the summer generation in Middlesex County, CT, surveys there were 1510
Figure 1. The distributions of host use by H. cunea in CT (A) and at the two MD sites (B).
conducted from mid-August to the end of September, corresponding with the second fall webworm generation. Four main sites were surveyed from 2006–2008. Surveys in Middlesex County took place at River Road/Maromas, Middletown (2006–2008), Haddam Meadows, Haddam (2006–2008), and the west bank of the Connecticut River on the south side of Cromwell (2007, 2008). In Hartford County, surveys were conducted along Route 189 in Granby (2008). Additional data were gathered opportunistically from roadsides in Portland and Cromwell, CT. All CT sites are in close proximity to the Connecticut River, except for Granby, which is also in Connecticut’s Central Valley but represents a slightly more upland habitat. In an effort to assess the choices available to female moths, all trees within a 20 m transect of each web (10 m from the host plant in each direction along the edge) were identified and recorded. At the MD sites, all trees and shrubs smaller than 1 cm DBH were excluded from the survey. However, since webworms at some CT sites did not seem to discriminate among hosts based on size (e.g. many hosts chosen at the River Road site were less than 1 cm DBH), small trees and shrubs were included in surveys at those localities.
CT surveys also differed in that trees were recorded in transects on a presence/absence basis only, rather than counting the number of stems of each species. To estimate the relative abundance of potential host tree species, the total number of individuals of a given species was divided by the sum of all individuals of all tree species encountered on transects at that site, and expressed as a percentage. Relative abundance in the CT populations was calculated as the number of times a host tree species was present on a transect, divided by the total number of times any tree species was present on a transect, expressed as a percentage (pooling all transects). Occasionally webs were located on adjacent trees, or on the same tree, resulting in double counting of some trees. Abundance rankings with and without double-counted trees revealed no difference between the two measures in determining host plant rankings, and therefore alleviated the concern that double counting biased results (unpubl.). Relative host use was calculated as the number of times a species was used as a host divided by the total number of webs encountered at a given site, expressed as a percentage. Because fewer webs were encountered at CT sites overall, statistical power was insufficient to treat sites separately. However, pooling the data from all sites seemed prudent given the close geographic proximity and similarity in habitat type of all localities. Data from the more distant Granby site were shown to comply with trends observed in the Middlesex County sites (likelihood ratio test and regression analyses yielded qualitatively similar results) before being included. Lab rearing Maryland (MD)
Larvae reared from eggs collected at PNWR were reared on a series of hosts to identify diet-based differences in fall webworm performance. Since fall webworms deposit all eggs into a single clutch (JTL unpubl.), the nine broods of eggs collected at PNWR represented nine distinct families. Eight focal tree species were selected based on previously gathered field data and reports in the literature: boxelder Acer negundo; persimmon Diospyros virginiana; white mulberry Morus alba, sweetgum Liquidambar styraciflua; redbud Cercis canadensis; black cherry Prunus serotina; pignut hickory Carya glabra; and green ash Fraxinus pennsylvanica. These are native to North America with the exception of M. alba, which was introduced to the region in colonial times and has replaced the native M. rubra in much of its range (Uva et al. 1997). Prior to hatching, egg masses were divided into eight equal parts using a straight-edged razor. A pin was placed through each of the egg mass sections, securing it to a leaf of each of the eight host plant species. Rearing chambers were constructed using deli containers and lids. A circular hole was cut into the lids, where aquapics were inserted. Upon hatching, caterpillars were allowed to feed collectively with their siblings for 14 days on their treatment host plant, which preliminary trials had indicated was necessary for establishment. After these two weeks, 15 caterpillars from each host plant and each brood were randomly selected, weighed to the nearest 0.1 mg, and separated into individual rearing chambers (as described above). The diet of a particular larva remained the same over the entire course of its development and was
limited to a single tree species. Larvae from each brood were distributed across all eight host plant species. Leaves were changed weekly, with fresh leaves gathered in small sprigs the morning prior to leaf change. Leaves were collected from designated ‘feeding trees’ each week. Fresh leaves were cut from mature trees, using terminal branches in direct sunlight, the natural habitat of H. cunea. Four to eight ‘feeding trees’ were used per species and were assigned randomly to individuals in the respective host plant treatments. In order to measure one aspect of feeding efficiency, fecal matter was collected once per week until pupation. Due to the slight variation and microclimates in the laboratory, larvae were haphazardly shuffled around the room, so that no particular subset of larvae was exposed to consistently different environmental factors. After pupation, when pupae were fully hardened and dark, pupae were weighed, sexed and placed on shelves exposed to ambient daylight until adult emergence. Since the mass of pupae changes slightly as metamorphosis occurs, masses were taken three days after pupation first occurred. The fecal matter of an individual caterpillar was collected over the course of development (individual rearing to pupation), dried, and weighed to the nearest mg. To evaluate feeding efficiency, we plotted fecal mass against body mass gain for each host plant, and compared the slopes. We also used these slopes as our measure of feeding efficiency in subsequent regression and correlation analyses. Steeper slopes indicate that more growth was obtained per the amount of food eaten, which may be due to differences in nutrition, secondary compounds or fiber content between hosts (Bernays and Chapman 1994, Singer 2001). Development time (hatching to pupation) was recorded for each larva, and the fate of each larva was recorded as pupated, parasitized, or died of unknown causes. Only one parasitoid species, Cotesia hyphantriae, an egglarval parasitoid, infected any of the larvae reared in the lab. Due to the efficiency of C. hyphantriae as a parasitoid that attacks entire broods before hatching, a large percentage of individuals of one family were infected, and subsequently excluded from all further analyses. Larvae that died during the rearing process but were not parasitized were excluded from all analyses because the cause of death could not be determined. Connecticut (CT)
When webs encountered on the CT transects were inhabited and accessible to researchers, 5–12 H. cunea larvae were collected and brought back to the lab for rearing. Webs containing larvae of less than 1 cm in length were flagged for later sampling (1–2 weeks later) to ensure that there had been opportunity for parasitoid attack. Caterpillars were housed singly in 5 oz. plastic cups and fed foliage from the host species upon which they were collected until pupation, parasitoid emergence, or death by other causes. New leaves were supplied each day from branches stored in sealed bags under refrigeration. To assess performance on various hosts, pupae were weighed and sex was determined. The frequency of parasitism for each family (proportion of individuals parasitized of the total number of individuals collected from each web) was calculated and expressed as a percentage, and EFS was calculated as 100 – % parasitism for each family. 1511
Statistical analyses Likelihood ratio tests were used to assess whether host use in H. cunea was non-random with respect to host identity. Non-hosts (trees never chosen in our surveys) were excluded from subsequent analyses to avoid inflation of variation in the data due to factors that may not be ecologically relevant. Linear regression analyses were used to test whether hostplant abundance, quality, or EFS were predictive of relative host use (data from each site were examined separately). These factors were examined individually rather than jointly because low statistical power precluded examination of more complex models that included interactions. We tested the ecological null hypothesis that host plants would be used in equal proportion to their abundance using a G-test for goodness of fit, where trees with expected values smaller than 3 were pooled into an ‘other trees’ category in each population, and a Williams correction was applied to the G statistics to reduce the likelihood of type I errors (Sokal and Rohlf 1995). A χ2-test was used in our analysis of host-associated survivorship in the MD study. Variance in parasitism frequency and performance variables was analyzed using ANOVA. For the MD data, the main factors used in initial models included host plant, family and sex. Initial models included all two-way interactions, and the results were used to develop reduced models including main factors and any significant interactions. The factors host plant and sex were considered fixed effects, whereas family was considered a random effect. Host plant and sex were considered fixed effects in analyses of pupal mass in CT and the interaction term was included in models when it was significant. Since our sampling units for EFS were field-collected families, a given family was represented on only one host, so we used a one-way ANOVA with host as the main effect to analyze EFS. When main effects of host plant were significant, we used Tukey’s HSD tests to compare mean fitness measures among host plant species. To evaluate the effect of host plant on the relationship between mass gain and fecal mass (feeding efficiency) we used ANCOVA, with host plant as the main effect, fecal mass as a covariate, and mass gain as the response variable. A significant host plant ⫻ fecal mass interaction would indicate differences in feeding efficiency when caterpillars are fed different host plants. Tradeoffs (host quality/EFS and host quality/host availability) were evaluated using correlation analysis. Spearman’s rank correlation was used to assess trade-offs in MD due to non-normality of the data, and all variables expressed as percentages in both data sets (survival, abundance, use and parasitism) were angularly transformed to improve normality of residuals (Zar 2010). Statistical tests were done using JMP statistical software (SAS Inst. 2007).
but supported no webs (e.g. oaks and conifers). However, among trees that were acceptable to H. cunea, abundance was a strong predictor of host use in CT and at both MD sites (Fig. 6). Although host plants were used in proportion to their abundance in the PNWR population in MD, some hosts were used with greater or lesser frequency than predicted by their abundance in the C&O population, and especially in CT. For example, at C&O, Diospyros was used significantly more, and Fagus less, than expected, and in CT, Salix was chosen significantly more and Rhus less, than expected (CT: G ⫽ 55.22, DF ⫽ 16, p ⬍ 0.001; MDC&O: G ⫽ 18.83, DF ⫽ 7, p ⬍ 0.01; MDPNWR: G ⫽ 6.77, DF ⫽ 7, p ⬎ 0.10)(Fig. 6). It is clear that some of the variation in host use by H. cunea remains unexplained by our measures of host abundance. Performance and parasitism rates Differences in larval performance on different host species were identified for every measure of performance considered in this study. Even in CT, where sample sizes limited statistical power, H. cunea showed differences in pupal mass, based on sex and host (Fig. 2). For example, caterpillars reared on Morus were twice as heavy as those reared on Prunus, and overall, females were 36.7% heavier than males. No interaction between sex and host was identified in the CT rearing data, and family is absent as a factor in the CT model since all instars were reared on their natal host (i.e., no family was reared on more than one host plant) (Table 1). Hyphantria cunea in MD also showed significant differences in pupal mass based on host, family, sex, and host ⫻ sex and family ⫻ sex interactions (Table 1, Fig. 3a). In MD, three additional indices of performance were measured: survival, development time and feeding efficiency. Of 960 larvae, 82.4% (791) pupated and 72.4% (695) emerged as adults, not including larvae that failed to establish. Two families failed to establish at all on several plant species due
Results Natural patterns of host plant use At no site was tree use by Hyphantria cunea independent of tree identity, (likelihood ratios: CT, χ2 ⫽ 287.52, DF ⫽ 56, p ⬍ 0.0001; MDC&O, χ2 ⫽ 175.55, DF ⫽ 26, p ⬍ 0.0001; MDPWNR, χ2 ⫽ 121.51, DF ⫽ 27, p ⬍ 0.0001), and all tree communities surveyed included species that were abundant, 1512
Figure 2. Least square means of the mean pupal mass of H. cunea families (least square means corrected for sex) reared upon eight preferred host plants in CT. Numbers at the base of bars indicate the number of family means used to calculate host means. Error bars indicate standard error. Hosts sharing letters are not statistically distinct according to Tukey’s HSD test.
Table 1. Results of ANOVAs for performance measures (MD and CT) and enemy-free space (CT) based on host plant, sex, family and interactions. Values in bold type indicate statistical significance. Source of variation Maryland Pupal mass Host plant Family Sex Host plant ⫻ Sex Family ⫻ Sex Survival Host plant Family Development time Host plant Family Sex Host plant ⫻ Family Connecticut Pupal mass Host plant Sex Parasitism frequency Host plant
DF (num, den)
F-ratio
p
7,588 5,5 1,5 7,588 5,588
33.65 8.72 260.10 5.89 2.52
⬍0.0001 0.0165 ⬍0.0001 ⬍0.0001 0.0288
7,35 5,35
2.76 0.45
0.0216 0.81
7,35 5,36 1,570 35,570
28.14 10.04 22.02 3.89
⬍0.0001 ⬍0.0001 ⬍0.0001 ⬍0.0001
7,44 2,44
4.19 13.31
0.0013 ⬍0.0001
6,33
3.44
0.0094
to unknown factors. To avoid biasing the data, we excluded from analyses all individuals from these families, leaving six families with balanced representation among hosts. Survivorship (% pupated) ranged from 75.2% on Liquidambar to 95.0% on Diospyros, and variation due to host plant was statistically significant (χ2 ⫽ 20.49, DF ⫽ 7, p ⬍ 0.0001). When family was not significant in the two-way ANOVA, we dropped it from the model. In this reduced model, the percent survival on the highest quality host (Diospyros) was significantly higher than on the lowest quality host (Liquidambar; F7,40 ⫽ 2.96, p ⫽ 0.01) (Fig. 3b). Development time varied based on host, family, sex and the host ⫻ family interaction. On average development time was 4.5% longer in females than males, and ranged from 35 days on Diospyros to 50 days on Carya (Table 1, Fig. 3c). We also found that feeding efficiency (slope of pupal mass vs. frass production) varied among host plant species, ranging from 0.11 in Cercis to 0.24 in Diospyros (Fig. 4). Fall webworm populations in CT differed significantly among host plants in parasitism frequency; on average, mortality from parasitism of H. cunea feeding on Alnus was more than six times higher than parasitism on Populus (Table 1, Fig. 5). Since H. cunea were collected as eggs in the MD study, enabling more thorough analysis of performance, analysis of parasitism was not possible (i.e. larvae were not exposed to parasitoids in the field). Although ecologically significant host-associated variation in both performance and parasitism frequencies was observed, neither appears to account for the observed patterns of host use in the fall webworm. In MD, the four measures of host plant quality were not significant predictors of host use (for individual sites and combined data, all R2 ⬍ 0.33, and p ⬎ 0.32), and neither host-associated performance nor EFS explained observed patterns of host use in CT (pupal mass: R2 ⫽ 0.10, p ⫽ 0.27; parasitism frequency: R2 ⫽ –0.06, p ⫽ 0.86).
Figure 3. Least square means of three measures of H. cunea larval performance in MD when reared upon eight preferred host plants: pupal mass (A), survival (B, and development time (C). LSmeans were derived from the models detailed in Table 1, except for in B, in which only host is used as a main effect. Numbers at the base of bars indicate the number of individuals used to calculate means, except in B, where they indicate the number of families. Error bars indicate standard error. Hosts sharing letters are not statistically distinct according to Tukey’s HSD tests.
1513
Figure 4. Feeding efficiency on eight host plants, indicated by the linear relationships between fecal mass produced and body mass gained (ANCOVA, host plant ⫻ fecal mass: F7,534 ⫽ 2.90, p ⫽ 0.0056). Symbols represent individual caterpillars. Slopes of lines represent feeding efficiencies for different host plants, the identities of which are indicated by italicized letters.
Tradeoffs In MD, no correlation was found between any measure of host quality and abundance at either site (for individual sites and combined data, all ρ ⬍ 0.47, and p ⬎ 0.35). In CT, where both host quality and the degree to which hosts offered EFS was evaluated, no trade-off was identified between host quality and EFS (r ⫽ 0.02, p ⫽ 0.94). However, host abundance was negatively correlated with performance in CT fall webworms (Fig. 7).
Figure 5. Least square means of the average percent mortality from parasitoids experienced by H. cunea on seven preferred hosts in CT (ANOVA: F6,33 ⫽ 3.44, p ⫽ 0.01). LSmeans were derived from the models detailed in Table 1. Numbers at the base of bars indicate the number of families used to calculate means. Error bars indicate standard error. Hosts sharing letters are not statistically distinct.
1514
Figure 6. The relationship between host plant abundance and use by H. cunea in CT and at two sites in MD. Each point represents a different host plant species, and the solid lines represent observed relationships (CT: R2 ⫽ 0.54, p ⬍ 0.0001; MDC&O: R2 ⫽ 0.90, p ⬍ 0.0001; MDPNWR: R2 ⫽ 0.96, p ⬍ 0.0001). Broken lines represent the expected relationship if host plants are used in proportion to their abundance.
performance of families on different host plant species may stem from gene-by-environment interactions (G ⫻ E effects), reflecting family-based differences in performance hierarchies. If host abundance is the only reliable determinant of host choice in H. cunea, populations would likely be under selection to optimize fitness on the most abundant host. Although in both CT and MD sites there are numerically dominant host species, we do not find that those species confer the highest fitness, either with regard to bi-trophic performance or EFS. The maintenance of genetic variation in the ability to exploit alternative hosts may be explained by gene flow from adjacent localities where different sets of hosts predominate in abundance. A high degree of gene flow seems particularly likely given the large and continuous distribution of H. cunea, and its tendency to inhabit edge-habitats including the forest edges adjacent to major roads. Differences in EFS among host plants Figure 7. The relationship between host plant abundance and H. cunea performance (pupal mass) on various host plants in CT (r ⫽ –0.61, p ⫽ 0.02). Each point represents a host plant species.
Discussion Abundance drives host use Host tree abundance was the only strong predictor of host use identified among the potential drivers of use evaluated in this study. Despite minor differences among sites in the methods used to determine local host abundance and differences in the time of year host use by Hyphantria cunea was assessed, local abundance explained a large portion of variation in host use in CT (55% for all sites combined) and for each of the two MD sites (90–96%; Fig. 6). This is a notable observation given that studies of host choice, even among generalists, typically show herbivores to exhibit distinct preferences among suitable hosts (Singer and Stireman 2001). These preferences are typically driven by the fitness consequences of feeding on different plants (Thompson 1988b, Gripenberg et al. 2010). In this study, however, host use was not explained by four measures of bi-trophic performance or EFS. Differences in bi-trophic performance among host plants All performance measures of H. cunea were shown to vary among host plants. In the MD study, in which individuals from different H. cunea families were distributed among hosts for rearing, we identified significant family ⫻ host interactions, indicating that genetically distinct cohorts differ in bi-trophic performance on different hosts. This may be explained in several ways. For example, it is possible that families whose mothers developed on a given plant species would in turn develop best on that plant due to environmentally based maternal effects. However, empirical tests for such maternal conditioning have rarely yielded support (Via 1991, McLean et al. 2009). Alternatively, the differential
We identified significant differences in mortality from parasitoids on different host plants in spite of small sample sizes in CT. These differences may be attributed either to hostassociated variability in attack rates or to diet-mediated differences in resistance to parasitoids after attack (Lill et al. 2002, Smilanich et al. 2009). The weak, positive correlation between EFS and pupal mass in the CT population indicates that some hosts are of high quality from both bi- and tri-trophic perspectives. For example, webworms feeding on cottonwood (P. deltoides) attain high body mass while experiencing a low frequency of parasitism (Fig. 2, 5). If quality were generally consistent between fitness parameters, we would expect webworms to choose hosts on that basis. However, many hosts used by the fall webworm are high quality with regard to growth and low quality with regard to EFS (e.g. A. rugosa), or vice versa. It is therefore not surprising that quality, as measured by single fitness components, was not a significant driver of host use at either site. It is also possible that host associated enemy-free space varies from year to year. Though our limited sampling precludes a meaningful investigation of the host ⫻ year interaction, the statistical significance of host-associated parasitism reported here from multiple-year data suggests that parasitism on a given host was somewhat consistent during our period of study. Tradeoffs The tradeoff between host quality and host availability seen in CT may offer some explanation for the lack of choosiness observed among fall webworms. If the search for high quality hosts incurs the risk of predation before oviposition, natural selection would favor a strategy of less extensive searching. The risk of bat predation, for example, may contribute to the trade-off between host tree quality and abundance seen in CT, especially since some arctiid moths are known to sense the biosonar of echolocating bats (Miller and Surlykke 2001, Conner et al. 2009), and oviposition patterns in H. cunea may be partially contingent upon bat activity. However, without knowledge of the relative predation risk between sites, differences in tree community composition are sufficient to explain the presence of a tradeoff at one site and its absence at the other: high quality hosts are rare in CT and 1515
not in MD. For example, webworms grow best on mulberry (Morus rubra) and boxelder (Acer negundo) in CT, which are among the rarest of trees at CT survey sites, whereas black cherry (Prunus serotina) and green ash (Fraxinus pennsylvanica) are low quality hosts and quite common there. Alternatively, the tradeoff between host quality and host availability in CT might result from generations of abundancebased use, mediated by the induction of plant secondary metabolites (either within or across years). If the most abundant trees are used repeatedly from year to year by gregarious species like H. cunea, they should be chemically induced to a greater degree on average, which may lead to declines in the quality of their foliage to herbivores. If H. cunea females choose hosts based solely on their abundance, the most abundant trees may possess the highest level of chemical defense, creating the observed negative correlation between host quality and host availability. This hypothesis is difficult to evaluate, given the gaps in our knowledge regarding the inducibility of the host trees in our study system, and how inducibility may vary seasonally. However, the observation that fall webworms are less likely to attack cherry trees that have been infested with the gregarious eastern tent caterpillar (Malacosoma americanum, Lasiocampidae) earlier in the season (Travis 2005), suggests that they do respond to prior chemical induction or some correlate thereof, at least in one host-plant species in one part of their geographic range (see also Williams and Myers 1984). Regardless of mechanism, an adaptive model of dietary generalization wherein selection favors polyphagy in the face of unpredictability or rarity of high quality resources has gained recent support by studies of the orange tip butterfly (Anthocharis cardamines) (Wiklund and Friberg 2009). Here, the unpredictability in quality would be the result of induction of defensive compounds in a resource that is predictable in its availability (trees). Community studies of foliage quality variation within and between seasons are necessary to assess these hypotheses. Geographic variation in host use A positive relationship between bi-trophic and/or tri-trophic performance and host use would have provided evidence for local adaptation of H. cunea to the three sites under study (Thompson 1988a, 2005). That we did not observe such a result does not preclude the possibility of local adaptation to ecological circumstances not considered here, especially since only offspring fitness was measured in this study. In fact, the tradeoff between host quality and abundance in CT suggests that adult moth searching behavior may be locally optimized. Preliminary results of a population genetics study (Stireman et al. unpubl.) suggest that MD and CT populations have differentiated to some degree. However, without further investigation we cannot ascertain whether phenotypic and genetic differences between populations are due to natural selection or genetic drift. Rather than local adaptation, our results suggest that different patterns in host use between localities may be based upon a common rule, applied to two different vegetative communities. To avoid senescence before ovipositing, a reasonable rule might be, ‘only search for X number of hours or days before choosing’. In CT, where high quality hosts are rare, and low quality hosts abundant, such a rule would 1516
lead to the quality/abundance tradeoff. In MD, where high quality hosts are relatively common, no tradeoff would result. This illustrates that the choice criteria of herbivores need not differ among localities to create geographic mosaics in host use. It is also possible that abundance-based host use has contributed to the maintenance of dietary generalization in H. cunea. Under circumstances where high quality hosts are rare, reduced search-times may compel moths to use a more abundant alternative, even if it incurs small reductions in offspring fitness (Williams 1983, Williams et al. 1983). Under circumstances where the most abundant host confers the highest fitness, lack of choosiness may inhibit dietary specialization. Even though natural selection would favor adaptations to improve performance on the abundant host, specialization should not occur in the absence of a preference-allele for that host plant species (Bush 1975, Thompson 1988a, Jaenike 1990, Thompson and Pelmyr 1991). If such an allele were to arise, its proliferation might be hindered by the high proportion of the population lacking the mutation and using the high quality host solely by virtue of its abundance. In this case, specialization might still arise if, for example, the maternal host were more likely to be selected, or if the preference allele enables individuals to select hosts more efficiently. Although the former, a maternal effect known as Hopkins’ host selection principle, has received very limited empirical support (Barron 2001, Janz et al. 2009), there is strong theoretical and empirical evidence that dietary specialization confers the advantages of efficient decision-making and high quality choices for offspring fitness (Bernays 2001, Tosh et al. 2009). Alternatively, abundance-based use in the fall webworm is compatible with the idea of local specialization due to a restricted proportion of utilizable plants found in a given habitat (Fox and Morrow 1981). This mode of specialization may be acting in parts of H. cunea’s range, like areas of the southwestern USA, where only Salicaceae provide suitable forage. Ecological specialization of this kind seems likely for H. cunea, because being contextual rather than genetic, it would be robust to the high degree of gene flow expected for the species based on its habitat preferences (discussed above). Expanding this study to include more localities across the vast geographic range of this species, especially in parts of its range where diet breadth is narrower, would provide valuable insights into how abundance-based host use influences the evolution of diet breadth. Acknowledgements – The authors thank Bob Marquis and John Stireman for their participation in developing this project, Angela Smilanich for valuable comments on the manuscript, and Kevi Mace and Konni Schießl for their work in the field and laboratory.
References Barbosa, P. and Greenblatt, J. 1979. Effects of leaf age and position on larval preferences of the fall webworm, Hyphantria cunea (Lepidoptera: Arctiidae). – Can. Entomol. 111: 381–383.
Barron, A. B. 2001. The life and death of Hopkins’ host selection principle. – J. Insect Behav. 14: 725–737. Berger, E. W. 1906. Notes on the fall webworm (Hyphantria cunea) in Ohio. – Ohio Nat. 6: 453–456. Bernays, E. A. 2001. Neural limitations in phytophagous insects: implications for diet breadth and evolution of host affiliation. – Annu. Rev. Entomol. 46: 703–727. Bernays, E. A. and Graham M. 1988. On the evolution of host specificity in phytophagous arthropods. – Ecology 69: 886–892. Bernays, E. A. and Chapman R. F. 1994. Host plant selection by phytophagous insects. –Chapman and Hall. Bush, G. L. 1975. Sympatric speciation in phytophagous parasitic insects. – In: Price, P. W. (ed.), Evolutionary strategies of parasitic insects and mites. Plenum Press, pp. 187–206. Conner, W. E. et al. 2009. Sound strategies: accoustic aposematism, startle and sonar jamming. – In: Conner, W. E. (ed.), Tiger moths and woolly bears: behavior, ecology and evolution of the Arctiidae. Oxford Univ. Press, pp. 177–192. Denno, R. F. et al. 1990. Role of enemy-free space and plant quality in host-plant selection by willow beetles. – Ecology 71: 124–137. Doak, P. et al. 2006. Fitness consequences of choosy oviposition for a time-limited butterfly. – Ecology 87: 395–408. Egan, S. P. and Ott, J. R. 2007. Host plant quality and local adaptation determine the distribution of a gall-forming herbivore. – Ecology 88: 2869–2879. Feeny, P. 1995. Ecological opportunism and chemical constraints on the host-associations of swallowtail butterflies. – In: Scriber, J. M. et al. (eds), Swallowtail butterflies: their ecological and evolutionary biology. Scientific Publishers, pp. 9–15. Fox, L. R. and Morrow, P. A. 1981. Specialization: species property or local phenomenon? – Science 221: 887–893. Greenblatt, J. A. et al. 1978. Larval feeding preferences and inducibility in the fall webworm, Hyphantria cunea. – Ann. Entomol. Soc. Am. 71: 605–606. Gomi, T. and Takeda, M. 1996. Changes in life-history traits in the fall webworm within half a century of introduction to Japan. – Funct. Ecol. 10: 384–389. Gripenberg, S. et al. 2010. A meta-analysis of preference–performance relationships in phytophagous insects. – Ecol. Lett. 13: 383–393. Jaenike, J. J. 1978. On optimal oviposition behavior in phytophagous insects. – Theor. Popul. Biol. 14: 350–356. Jaenike, J. J. 1990. Host specialization in phytophagous insects. – Annu. Rev. Ecol. Syst. 21: 243–273. Jaenike, J. J. and Selander, R. K. 1980. On the question of host races in the fall webworm, Hyphantria cunea. – Entomol. Exp. Appl. 27: 31–37. Janz, N. et al. 2009. No effect of larval experience on adult host preference in Polygonia c-album (Lepidoptera: Nymphalidae): on the persistence of Hopkins’ host selection principle. – Ecol. Entomol. 34: 50–57. Lill, J. T. et al. 2002. Host plants influence parasitism of forest caterpillars. – Nature 417: 170–173. McLean, A. H. et al. 2009. Effects of the maternal and pre-adult host plant on adult performance and preference in the pea aphid, Acyrthosiphon pisum. – Ecol. Entomol. 34: 330–338. Michaud, J. P. 1990. Conditions for the evolution of polyphagy in herbivorous insects. – Oikos 57: 278–279. Miller, L. A. and Surlykke, A. 2001. How some insects detect and avoid being eaten by bats: tactics and countertactics of prey and predator. – BioScience 51: 570–581. Morris, R. F. 1972a. Predation by insects and spiders inhabiting colonial webs of Hyphantria cunea. – Can. Entomol. 104: 1197–1207. Morris, R. F. 1972b. Predation by wasps, birds, and mammals on Hyphantriacunea. – Can. Entomol. 104: 1581–1591.
Morris, R. F. 1976. Relation of mortality caused by parasites to the population density of Hyphantria cunea. – Can. Entomol. 108: 1291–1294. Murphy, S. M. 2004. Enemy-free space maintains swallowtail butter fly host shift. – Proc. Natl Acad. Sci. USA 101: 18048–18052. Oliver, A.D. 1964. A behavioral study of two races of the fall webworm, Hyphantria cunea (Lepidoptera: Arctiidae), in Louisiana. – Ann. Entomol. Soc. Am. 57: 192–194. Schluter, D. 2009. Evidence for ecological speciation and its alternative. – Science 323: 737–741. Singer, M. S. 2001. Determinants of polyphagy by a woolly bear caterpillar: a test of the physiological efficiency hypothesis. – Oikos 93: 194–204. Singer, M. S. 2008. Evolutionary ecology of polyphagy. – In: Tilmon, K. J. (ed.), Specialization, speciation and radiation. Univ. California Press, pp. 29–42. Singer, M. S. and Stireman, J. O. III 2001. How foraging tactics determine host-plant use by a polyphagous caterpillar. – Oecologia 129: 98–105. Singer, M. C. et al. 1992. Distinguishing between ‘preference’ and ‘motivation’ in food choice: an example from insect oviposition. – Anim. Behav. 44: 463–471. Singer, M. S. et al. 2004. Roles of food quality and enemy-free space in host use by a generalist insect herbivore. – Ecology 85: 2747–2753. Smilanich, A. M. et al. 2009. The insect immune response and other putative defenses as effective predictors of parasitism. – Ecology 90: 1434–1440. Sokal, R. R. and Rohlf, F. J. 1995. Biometry: the principals and practice of statistics in biological research. – W. H. Freeman. Strong, D. R. 1988. Insect host range. – Ecology 69: 885–915. Strong, D. R. et al. 1984. Insects on plants: community patterns and mechanisms. – Harvard Univ. Press. Thompson, J. N. 1988a. Variation in preference and specificity in monophagous and oligophagous swallowtail butterflies. – Evolution 42: 1223–1234. Thompson, J. N. 1988b. Evolutionary ecology of the relationship between oviposition preference and performance of offspring in phytophagous insects. – Entomol. Exp. Appl. 47: 3–14. Thompson, J. N. 1993. Preference hierarchies and the origin of geographic specialization in host use in swallowtail butterflies. – Evolution 47: 1585–1594. Thompson, J. N. 1998. The evolution of diet breadth: monophagy and polyphagy in swallowtail butterflies. – J. Evol. Biol. 11: 563–578. Thompson, J. N. 2005. The geographic mosaic of coevolution. – Univ. Chicago Press. Thompson, J. N. and Pellmyr, O. 1991. Evolution of oviposition behavior and host preference in Lepidoptera. – Annu. Rev. Entomol. 36: 65–89. Tosh, C. R. et al. 2009. Theoretical predictions strongly support decision accuracy as a major driver of ecological specialization. – Proc. Natl Acad. Sci. USA 106: 5698–5702. Travis, H. J. 2005. The effect of eastern tent caterpillar (Malacasoma americanum) infestation on fall webworm (Hyphantria cunea) selection of black cherry (Prunus serotina) as a host tree. – Am. Midl. Nat. 153: 270–275. Uva, R. H. et al. 1997. Weeds of the Northeast. – Cornell Univ. Press. Via, S. 1984. The quantitative genetics of polyphagy in an insect herbivore. I. Genotype-environment interaction in larval performance on different host plant species. – Evolution 38: 881–895. Via, S. 1991. Specialized host plant performance of pea aphid clones is not altered by experience. – Ecology 72: 1420–1427.
1517
Wagner, D. L. 2005. Caterpillars of eastern North America: a guide to identification and natural history. – Princeton Univ. Press. Warren, L. O. and Tadic, M. 1970. The fall webworm, Hyphantria cunea (Drury). – Arkansas AES Bull. 759: 106. Wiklund, C. 1981. Generalist vs specialist oviposition behavior in Papilio machaon (Lepidoptera) and functional aspects of the hierarchy of oviposition preferences. – Oikos 36: 163–170. Wiklund, C. and Fribert, M. 2009. The evolutionary ecology of generalization: among year variation in host plant use and offspring survival in a butterfly. – Ecology 90: 3406–3417.
1518
Williams, K. S. 1983. The coevolution of Euphydryas chalcedona butterflies and their larval host plants. III. Oviposition behavior and host plant quality. – Oecologia 56: 336–340. Williams, K. S. and Myers, J. H. 1984. Previous herbivore attack of red alder may improve food quality for fall webworm larvae. – Oecologia 63: 166–170. Williams, K. S. et al. 1983. The coevolution of Euphydryas chalcedona butterflies and their larval host plants. II. Maternal and host plant effects on larval growth, development and food-use efficiency. – Oecologia 56: 330–335. Zar, J. H. 2010. Biostatistical analysis, 5th ed. – Pearson Prentice-Hall.
