Introduction

The semi-aquatic grasshopper Cornops aquaticum (Bruner, 1906) is native to South America and inhabits lowlands from southern Mexico to Central Argentina and Uruguay (Adis et al. 2007). It is host- specific to aquatic plants in the genera Eichhornia and Pontederia (Adis & Junk 2003, Adis & Victoria 2001); however, different populations generally only have access to either E. azurea or E. crassipes (Silveira-Guido & Perkins 1975, Lhano et al. 2005, Ferreira & Vasconcelos-Neto 2001). Between 1996 and 1997, grasshoppers were collected from various South American sites and established in a quarantine laboratory in Pretoria, South Africa (Oberholzer & Hill 2001), for eventual release in South Africa as a biological control agent of water hyacinth, E. crassipes (Oberholzer & Hill 2001). Coevolution of the grasshopper and its host plant (Adis et al. 2004, Brede & Beebee 2005) is presently studied under the HICWA Project (Host-Insect Coevolution on Water Hyacinth), involving 10 countries (Argentina, Brazil, Colombia, France, Germany, Nicaragua, South Africa, Trinidad, United Kingdom and Uruguay), under the auspices of the Tropical Ecology Working Group of the Max Planck Institute for Limnology at Plön, Germany ( www.mpil-ploen.mpg.de).

During our studies, we observed that body size in C. aquaticum appeared to vary with geography and host plant. In addition, we noticed that individuals in our laboratory colony appeared to have become increasingly smaller during 10 y in quarantine. Perhaps simple isolation stimulated this change in body size. If this were true, then we would expect the South African population to be more similar to other isolated populations than to nonisolated ones in South America. In this contribution, morphometric data obtained from 11 wild populations in South America and one 10-y-old laboratory population in South Africa are used to test the following hypotheses:

Materials and methods

We collected 480 C. aquaticum from 11 sites in South America (see Table 1Table 1b.). The South African quarantine population at Pretoria represents a mixed population derived from Brazil, Mexico, Trinidad and Venezuela (Hill & Oberholzer 2000). Populations from an agriculturally disturbed site in Trinidad, from within the city of Belém, and from S. Africa, represent geographically “isolated” populations (cf. Fig. 1). Total length, body length and wing length (as defined by Carbonell 2001) of adults (both sexes) were used as morphometric characters for data analysis (Fig. 2). The length of the hind femur, a valid character to define nymphal stages in C. aquaticum (Franceschini et al. 2005), was omitted, as adult specimens easily lose their hind legs when preserved in ≥70% ethanol.

Table 1a.

Mean size (mm) of C. aquaticum from populations feeding on E. crassipes in South America and South Africa.

Table 1b.

Mean size (mm) of C. aquaticum from populations feeding on E. azurea in South America.

Fig. 1.

Sampling locations and reference of C. aquaticum populations in South America. For location of populations see Table 1a/b. Shaded area indicates range of C. aquaticum in South America (according to Adis et al. 2007).

Fig. 2.

Morphometric measurements on the grasshoppers, following Carbonell (2001). A: total length, I: wing length, B: body length (only undamaged specimens measured).

Statistical analysis

To test Hypotheses 1, 3, and 4, we used univariate mixed-effects models (Crawley 2002a,b) to test statistically if morphology varied with geography and host plant (Manly 2000). We considered ‘grasshopper individual’ nested within ‘sample’ as random effects (Crawley 2002a) to avoid pseudoreplication (grasshopper individual) (Hurlbert 1984, Crawley 2002a), and to allow evaluation of the importance of variation among sampled sites (grasshopper populations) on morphometry (Buckley et al. 2003, Pinheiro & Bates 2000). Length was used as the response variable, while the corresponding morphological character (wing vs body length) was considered one of the explanatory factors, with two levels. Because total length included wing length (Fig. 2), these two measurements were statistically correlated. Therefore, we used only body length and wing length in the analyses, disregarding total length. Initially we adjusted a maximal model, and simplified it subsequently by deleting nonsignificant effects (Crawley 2002a). The fixed effects of the maximal tested model used to test Hypotheses 1 and 3, included gender, geography, morphological character (whether wing or body length) and their interaction terms. We used the interaction terms to test whether morphological characters responded differently to geography and gender. Geography was entered as a categorical variable with three levels: ‘nonisolated’, ‘South Africa’ and ‘other isolated populations'.

To test hypotheses that the isolated S. African laboratory population is similar to other isolated, wild populations, but differs from nonisolated ones, we performed contrast analyses, pooling the geographical levels ‘South Africa’ with ‘other isolated populations’. The use of contrasts instead of a posteriori comparisons allowed us to test explicit hypotheses with biological explanations.

We tested for the effect of random variation among grasshopper populations by deleting the random effect ‘sample’ from the minimal adequate model and noting the change in the log of the maximized restricted likelihood, as per Fox (2002). Significance of fixed effects was tested by deleting each explanatory variable from the model and noting the change in maximum likelihood (Crawley 2002a). Significance tests of fixed effects within the minimal model were shown as F-tests.

We tested the hypothesis that the variance of the isolated South African population is smaller than that of the nonisolated populations by using Fisher's F-test for each measured morphological character. We also tested the hypothesis that there are differences in grasshopper morphometry between host plants, by adjusting mixed-effects models with host plant, morphological character and their interaction terms as fixed effects. Because C. aquaticum grasshopper populations collected on E. azurea were all nonisolated, we compared them exclusively with other nonisolated populations using a subset of all available data (n = 319). All analyses were performed at a 5% significance level and assumed normal error distribution, (Fowler & Cohen 1996); they were performed with the statistical package R (R Development Core Team 2006).

Results

1. The inclusion of the sample site from which the grasshoppers were drawn as a random effect, increased significantly the amount of variation explained by the model (likelihood ratio = 86.58, p < 0.0001). This shows that there was random variation among grasshopper populations associated with a significant covariance within sites, i.e., there were morphological differences among populations (Fig. 3), which must be considered in the mixed-effects model.

Fig. 3.

Wing length (mm) of C. aquaticum among populations. “JK”, “Q” and “xL” represent the isolated populations, “N” and “xM” the populations collected on E. azurea. All other populations were exclusively collected on E. crassipes. For location of samples see Table 1Table 1b..

2. The pooling of the levels ‘South Africa’ with ‘other isolated populations’ did not reduce the variation explained by the fitted statistical model (likelihood ratio = 3.71, p = 0.45). This shows that there was no difference in the response between the geographical groups ‘South Africa’ and ‘other isolated populations’. This allowed these two groups to be merged together as a third (named ‘isolated populations’).

3. The minimal adequate model (Table 2) presented highest order interactions of morphological characters with sex and geography. Wing length and body length did not respond in the same manner (Fig. 4). Females were always larger than males and always presented longer bodies than wings (Table 1Table 1b., Fig. 4).

Fig. 4.

Average C. aquaticum dimensions with 95% confidence interval bars, showing the interactions of morphology, sex and geography. Continuous lines: nonisolated populations; dashed lines: isolated populations. All interactions significant (p < 0.05), as in Table 2.

Table 2.

Analysis of variance of the adjusted minimal mixed-effects model to explain length (mm) of C. aquaticum grasshoppers. Random effects were individuals (n = 480), nested within populations (n = 12). The factor Geography included nonisolated vs isolated populations; the South African population did not differ from the remaining isolated populations.

4. Both sexes were larger in nonisolated than in isolated populations (Fig. 4), but male and female morphology responded differently to geography. Females maintained similar morphological proportions in all populations, whereas males in isolated populations were not only smaller overall, but had smaller wing-to-body ratios (Fig. 4: Micropterous males).

5. The variance of morphological characters in the isolated S. African population did not differ from nonisolated populations (Table 3).

Table 3.

Results of independent Fisher's F-tests, comparing the variances of morphological characters between nonisolated South American and isolated South African grasshoppers.

6. In the host-plant analyses, all interactions (morphological character, sex, and host plant) were significant (Table 4). Females that were collected on E. azurea were larger than those on E. crassipes (Fig. 5). In males, host plant influenced wing length, but not body length.

Fig. 5.

Average ± 95% confidence interval of C. aquaticum grasshopper lengths (mm) from nonisolated populations, comparing two host plants: E. azurea (dashed lines) and E. crassipes (continuous lines). All interactions were significant (Table 4).

Table 4.

Analysis of variance of the adjusted minimal mixed-effects model to evaluate if there were differences in C. aquaticum grasshoppers’ morphology between the host plants E. azurea and E. crassipes.

All four hypotheses are valid: morphology varies geographically, with degree of isolation, and with host plant. South African and South American populations differ in body size, with nonisolated South American grasshoppers being larger than the isolated populations (including the South African laboratory population). In addition, females maintain their morphological proportions irrespective of geographical situation and host plant, with those feeding on E. azurea being larger than those feeding on E. crassipes. In contrast, males alter the proportion of body to wing lengths according to geography and host plant. In isolated populations, male C. aquaticum change their morphology, with a greater proportional reduction in wing length in regard to their body size. Where males feed on E. azurea they present larger wings compared to those on E. crassipes, without altering their body length.

Discussion

In C. aquaticum, morphology varies with geography and host plant. This variation may be determined genetically or environmentally. The smaller overall size of C. aquaticum individuals in isolated populations may have resulted from selection for smaller size or due to a loss of genetic diversity in quantitative traits. A selective pressure for size reduction might be isolation per se, which can reduce the probability of far-dispersing individuals encountering mates, and hence, lower fitness for large individuals which might disperse further. This hypothesis is supported by the fact that island insects are often flightless and small (Diamond 1981, Roff 1994). Denno et al. (2002) tested a similar hypothesis but found no evidence that the low incidence of female macroptery in island populations of the delphicid planthopper Toya venilia was attributable to the effect of isolation.

Various hypotheses compete to explain the apparent reduction in body size of our South African laboratory colony. These include founder effect, inbreeding depression, outbreeding depression promoted by the heterogeneous composition of paternal lineages (Gerloff & Schmidt-Hempel 2005, Schaal & Leverich 2005), and selection for small size. Continuous laboratory breeding may select for small, fast-breeding genotypes that can produce more generations per unit time than larger individuals. Isolation per se may also have influenced body size (Peer & Taborskyi 2005). Populations that are isolated for several generations may have lost genes for larger size because of inbreeding depression (Futuyma 1986, Eldrige et al. 2004) or genetic drift (Futuyma 1986, Escobar et al. 2005).

In contrast, our results might be explained by phenotypic plasticity. In some locusts, morphology is plastic and determined by population density (phase transformation) (Uvarov 1966, Sword & Simpson 2000, Simpson et al. 2001, Babah & Sword 2004). In this study, however, there is no straightforward relationship between population density and isolation. Isolated populations might be spatially constrained and densely packed, as could be the case for the Belém city C. aquaticum population, but could also have plenty of empty surrounding habitat, as in the low-density South African laboratory population, which has been maintained at a population size of ~ 100 individuals.

Our results indicate that geography and host plant influence body size more in males than in females. The smaller flexibility in female morphology may reflect developmental-reproductive constraints. Females may increase their individual fitness through a positive correlation between fecundity and body size (but see Barker et al. 2005). However, if a female has a minimum size requirement for successful production of oocytes, this could result in a lower level of resource being available for development of other morphological characters. Alternative explanations for differences in male-female morphological response might be differential linkage of certain genes influencing morphology, or the need for males to invest in large wings if they are the dispersing sex. Finally, adult size is frequently associated with nymphal developmental conditions, whereby well-fed larvae produce larger adults (Price 1997). If nymphal conditions are involved in the observed responses, this may mean that isolation, geography, and poor quality of host plants (i.e., E. crassipes) might be responsible for less favorable conditions.

Conclusions