Introduction
The spatial distribution of herbivorous insects is in most instances non-random. A large body of literature has demonstrated, for example, that insect herbivores exhibit oviposition preferences, niche partitioning, and use of enemy-free space (e.g. Dethier, Reference Dethier1959; Strong et al., Reference Strong, Lawton and Southwood1984; Casey, Reference Casey, Stamp and Casey1993; Bernays & Chapman, Reference Bernays and Chapman1994; Price, Reference Price1997). Apart from such factors, species that differ in life-history strategy, defence characteristics, host plant specificity and microclimate preferences may be expected to have different distribution patterns (e.g. Strong et al., Reference Strong, Lawton and Southwood1984; Wallner, Reference Wallner1987; Holmes & Schultz, Reference Holmes and Schultz1988; Stork et al., Reference Stork, Hammond, Russell and Hadwen2001; Kessler & Baldwin, Reference Kessler and Baldwin2002; Ribeiro et al., Reference Ribeiro, Codeco and Fernandes2003). Even within a species, different life stages are subject to different key factors influencing survival and the selection imposed is likely to result in a range of behaviours and microhabitat preferences (Price, Reference Price1997), and consequently, differences in distribution.
Generally, at the between-host plant scale, insect herbivore distributions may be influenced by host plant density (Williams et al., Reference Williams, Jones and Hartley2001), distance from the edge of a site (McGeoch & Gaston, Reference McGeoch and Gaston2000), habitat structure (Ellingson & Andersen, Reference Ellingson and Andersen2002), direct or plant-mediated interactions between herbivores (Riihimäki et al., Reference Riihimäki, Kaitaniemi and Ruohomäki2003), avoidance of conspecifics (Stamp, Reference Stamp1980), spatial escape from natural enemies (Bernays, Reference Bernays1997; Williams et al., Reference Williams, Jones and Hartley2001), and species dispersal characteristics (McGeoch & Price, Reference McGeoch and Price2004). In addition, host plant selection may be based on host plant size or quality characteristics (Floater, Reference Floater1997; Hodkinson et al., Reference Hodkinson, Bird, Hill and Baxter2001), as well as previous levels of herbivory (Gilbert et al., Reference Gilbert, Vouland and Grégoire2001). Within plants, spatial distribution may be affected by heterogeneity in plant quality and defence (Orians & Jones, Reference Orians and Jones2001; Kessler & Baldwin, Reference Kessler and Baldwin2002), niche partitioning (Dubbert et al., Reference Dubbert, Tscharntke and Vidal1998), within- and between-species interactions (Cappuccino, Reference Cappuccino1988; Cappuccino et al., Reference Cappuccino, Damman, Dubuc, Cappuccino and Price1995; Faeth & Hammon, Reference Faeth and Hammon1997), larval behaviour (Anstey et al., Reference Anstey, Quiring and Ostaff2002), avoidance of natural enemies (Stamp & Wilkens, Reference Stamp, Wilkens, Stamp and Casey1993; Wermelinger, Reference Wermelinger2002), or environmental thermal regimes (Stamp & Bowers, Reference Stamp and Bowers1990; Klok & Chown, Reference Klok and Chown1998, Reference Klok and Chown1999). Therefore, identifying the specific factors responsible for the fine-scale abundance and distribution of insects is fundamental to explaining the patterns observed, understanding species population dynamics and, consequently, the habitat requirements necessary for their conservation and sustainable use (Ranius, Reference Ranius2001).
The spatial distributions of sessile life stages are often easily determined and thus useful for examining the mechanisms responsible for observed distribution patterns of species (e.g. Heads & Lawton, Reference Heads and Lawton1983; Hails & Crawley, Reference Hails and Crawley1992; Brewer & Gaston, Reference Brewer and Gaston2002; Veldtman & McGeoch, Reference Veldtman and McGeoch2004). In addition, the condition of individuals in the pupal stage is often largely a summary of the fate of previous or future life stages, e.g. larval performance, final instar parasitism and adult potential fecundity (Wickman & Karlsson, Reference Wickman and Karlsson1989; Veldtman et al., Reference Veldtman, McGeoch and Scholtz2004). The pupal cocoons of two wild silk moth species native to southern Africa, Gonometa postica Walker and Gonometa rufobrunnea Aurivillius (Lepidoptera: Lasiocampidae) are economically valuable (Veldtman et al., Reference Veldtman, McGeoch and Scholtz2002). Cocoons can be degummed to produce high quality silk, which rivals the silk produced from Bombyx mori (Linnaeus) (Lepidoptera: Bombycidae). Currently, the pupal stage is the target of harvesting practices that are totally dependent on the availability of pupae from natural populations (Veldtman et al., Reference Veldtman, McGeoch and Scholtz2002). These pupae almost exclusively occur on the branches and stems of woody plant species (Hartland-Rowe, Reference Hartland-Rowe1992). Because of the harvesting demand, and poor knowledge of the species biology, there is substantial interest in understanding factors influencing the distribution of pupae among and within trees for both Gonometa species. Apart from this applied value in predicting where individuals of these species occur, they provide an ideal study system to identify which factors determine the fine-scale abundance and distribution of the pupal stage of an insect herbivore. Furthermore, pupal information on Gonometa species will contribute to the development of an appropriate conservation strategy for these economically important species (McGeoch, Reference McGeoch2002). Consequently, this study investigates if between and within-tree pupal distributions in Gonometa postica and G. rufobrunnea are non-random, and if so, if there are relationships between pupation site use and tree characteristics.
Materials and methods
Study area
Gonometa postica and G. rufobrunnea populations were examined at six and five sites respectively within the known (historic and recent records) eruptive range of these species (described fully in Veldtman et al., Reference Veldtman, McGeoch and Scholtz2002). The dominant woody host species utilized by G. postica (at three sites each) was Acacia erioloba Meyer and Acacia tortillis Hayne (both Mimosaceae), while G. rufobrunnea only utilizes Colophospermum mopane Kirk ex Benth. (Caesalpiniaceae).
Sampling was standardized by delimiting an approximately rectangular area incorporating 100 trees at each site, to compensate for possible tree-density differences between host-plants and localities. An initial minimum of 40 first-generation cocoons per site was a prerequisite for selection, with at least three sites per host plant selected.
Life history
The females of both Gonometa species have limited flying ability and are short-lived (4–7 days). Within the study area, when diapause is broken in early spring (September to October), emerging moths mate and lay eggs to form the first generation. This generation develops for approximately two months before final instar larvae start to pupate (November to December). A varying proportion of these pupae undergo rapid development and emerge to give rise to the second generation in mid-summer (December to January), with pupation occurring in early autumn (March to April). The remaining first generation pupae and surviving second-generation pupae enter diapause, emerging only the following spring (Hartland-Rowe, Reference Hartland-Rowe1992; R. Veldtman et al., unpublished). The cocoons of G. rufobrunnea are cryptically coloured (red) while those of G. postica are not (white) (Veldtman et al., Reference Veldtman, McGeoch and Scholtz2002).
Cocoon sampling
Surveying of plots commenced in winter (June to July, 2000) and was repeated in mid summer (January, 2001). This sampling procedure was repeated the following year, all sites being surveyed four times by the end of January 2002. Newly formed pupae counted in the first, second, third and final survey are referred to from here on as generation one, two, three and four, respectively.
For each of the 100 trees per plot, the species, maximum height, number of branches and geographic spatial position were recorded. Tree species used for pupation were divided into three functional types namely, larval host plant (H); non-host plant (N); non-host plant with thorns (NT), as the use of each represents a different pupation strategy. Remaining on the host plant to pupate can guarantee that the correct host is oviposited on (Bernays & Chapman, Reference Bernays and Chapman1994). On the other hand, using non-host plants can disrupt the search image of natural enemies (Guildford, Reference Guildford and Crawley1992). Tree height was measured to the nearest 0.25 m and divided into three size categories: small (<1.75 m), medium (1.75–3.00 m) and large (>3.00 m). In addition, to standardize for three-dimensional size differences between trees, the number of branches per tree was estimated. At each site the smallest sampled tree (0.75 m tall) of the dominant woody host species present was taken to represent one branch, all other trees in the site were then expressed relative to this unit. The position of each tree within a site was measured at the main trunk of the tree with a hand-held Global Positioning System Receiver (GPS: Garmin Etrex, Garmin, International Inc., Kansas; ∼3 m accuracy during measurement, see Veldtman, Reference Veldtman2004 for further details).
Every tree was carefully searched and all pupae of the present generation (cocoons covered by setae) were counted (the time spent searching for pupae was proportional to number of branches per tree). For each pupa, its sex (see Veldtman et al., Reference Veldtman, McGeoch and Scholtz2002), cocoon size, height in the tree (to the nearest 5 cm), distance from the main tree trunk (to the nearest 10 cm), branch position and aspect were recorded. Branch position was divided into seven categories: edge (E, within 15 cm from terminal branch end); edge middle (EM, 15–30 cm from terminal branch end); edge stem (ES, terminal branch directly from main trunk); middle branch edge (ME, start of terminal branch 60 cm from edge); middle (M, middle branch); middle stem (MS, start of main branch); and stem (S) on tree trunk (fig. 1). Aspect was determined with a compass, dividing measured directions into four sectors, each centred on a cardinal compass direction, i.e. north, east, south and west. At the start of the study, the number of pupae per aspect was not recorded directly in the first generation, but the number of first generation cocoons found in the second survey was counted instead. Consequently, the site sample sizes for which data on aspect use were available could be lower than for other variables, if some pupae became detached and were not resampled in the second survey.
Fig. 1. Within-tree, branch position categories of Gonometa species pupae: edge (E, within 15 cm from terminal branch end); edge middle (EM, 15–30 cm from terminal branch end); edge stem (ES, terminal branch directly from main trunk); middle branch edge (ME, start of terminal branch 60 cm from edge); middle (M, middle branch); middle stem (MS, start of main branch); and stem (S) on tree trunk.
Data analysis
Alpha level corrections for multiple testing were performed using the step-up false discovery rate (FDR) correction procedure, which has been shown to be the least over-corrective of current alpha-level correction methods (García, Reference García2004).
Between-tree scale
At a between-tree scale, the objective was to determine if variation in pupal abundance could be explained by tree characteristics such as tree functional type, tree size, or by across-tree aggregation patterns. To determine if tree functional types (H, N, NT) or larval host plant size classes (small, medium, and large) had a greater or lower proportion of the pupae than expected from their recorded frequencies, Chi-square goodness of fit analyses were performed (Zar, Reference Zar1984). Expected frequencies were calculated as the product of the proportion of trees of a category with the sites' total pupal abundance (expected pupal frequencies ≥5). For both groupings three categories were generally available for comparison. In cases where some groups did not have sufficient pupae to allow analysis (see Zar, Reference Zar1984 for bias in Chi-square values), a two-category comparison was made.
Second, spatial analysis by distance indices (SADIE) methodology (Perry, Reference Perry1995) was used to quantify the degree of departure from spatial randomness for the spatially-referenced (X,Y) recorded branch and pupal count data. Spatial non-randomness is based on the distance to regularity, which is the minimum cumulative distance to achieve a regular distribution of counts (Perry & Dixon, Reference Perry and Dixon2002). The index of aggregation (I a) describes overall aggregation and values approximately ⩽1.5 indicate significant aggregation (Perry, Reference Perry1995; Perry et al., Reference Perry, Winder, Holland and Alston1999).
The degree of clustering in number of pupae and branches was also quantified, using the index of clustering, v, that provides information on the degree of clustering for each spatially referenced point based on the magnitude of the count and its occurrence in relation to neighbouring counts (patches – counts greater than the sample mean, v i and gaps – counts smaller than the sample mean, v j; see Perry et al., Reference Perry, Winder, Holland and Alston1999; Perry & Dixon, Reference Perry and Dixon2002). For each site–generation combination, I a, mean v i and mean v j were calculated if pupae were found on more than 20% of the trees. At densities lower than this (e.g. mean count per tree <0.2), it is not possible to quantify overall aggregation and spatial clustering (Korie et al., Reference Korie, Perry, Mugglestone, Clark, Thomas and Mohamad Roff2000; Winder et al., Reference Winder, Alexander, Holland, Woolley and Perry2001).
Thereafter spatial matching between the spatial clustering in pupal abundance and number of branches was determined with spatial association statistics (see Winder et al., Reference Winder, Alexander, Holland, Woolley and Perry2001; Perry & Dixon, Reference Perry and Dixon2002 for full description of method). All spatial non-randomness and association analyses were done using SADIEShell (v. 1.22 software, Kelvin F. Conrad and IACR-Rothamsted 2001).
Finally, to determine the amount of variability in pupal abundance explained by spatial and environmental variables (tree variables), trend surface analysis and stepwise model building approaches were applied (Legendre & Legendre, Reference Legendre and Legendre1998). Trend surface analysis was first used to determine the best fit set of spatial variables (significant terms from the third order polynomial of GPS recorded latitude and longitude of each tree) that significantly contributed to explaining variation in pupal abundance (Legendre & Legendre, Reference Legendre and Legendre1998). Thereafter, a stepwise model-building procedure (generalized linear model, Poisson distribution, log link function) was used to determine the additional variation explained by tree variables (number of branches, tree height and tree functional type) after spatial non-independence was accounted for. To prevent the sequence of additive model-building influencing which variables are included in the final model (Abraham et al., Reference Abraham, Chipman and Vijayan1999; Randic, Reference Randic2001), best subset analyses of only tree variables were done to rank them in order of the magnitude of variation explained. The tree variables were then sequentially added to the spatial model according to rank, until the percentage of deviance explained was not increased significantly, or all tree variables were included (see Legendre & Legendre, Reference Legendre and Legendre1998).
Within-tree scale
At the within-tree scale, the objective was to quantify patterns in pupal abundance, and to determine how much of the within-tree distribution in pupal abundance is explained by pupal and tree variables. These included branch position, aspect, standardized cocoon height, cocoon height and distance from the tree trunk. First, the significance of differences in the numbers of pupae between different branch positions or aspects was determined by Chi-square goodness of fit (Zar, Reference Zar1984). This was done for each site-generation combination separately, as well as for each Gonometa species in total. Expected frequencies were calculated as the expected proportion of pupae per category multiplied with a sites' total pupal abundance. For branch position, given the physical space constraints in the number of possible pupation sites in tree shape, all positions further than 30 cm from the tree's outer edge were lumped into one category, assuming that E, EM and all other categories combined would have equal frequencies of pupae by chance. For both branch position and aspect, the influence of sex was also taken into account (expecting equal numbers, see Veldtman et al., Reference Veldtman, McGeoch and Scholtz2002) with Chi-square analysis of two-way contingency tables (Zar, Reference Zar1984).
Second, the height frequency distribution of pupae for each primary host plant species was described after controlling for tree height differences between trees. To determine how pupae across sites were distributed in terms of relative tree height, the height recorded for each cocoon was divided by the height of the tree on which it was found. Thus, if pupae are found near the crown of trees, the standardized cocoon height value should be close to one. Distributions were determined for both species, and for G. postica populations on different dominant host-plant species separately. The hypothetical crown volume and distribution of each dominant host-plant species (i.e. Acacia erioloba, A. tortillis and Colophospermum mopane) were estimated from descriptions and drawings from Palgrave (Reference Palgrave1977), as well as from observations in the field.
Finally, potential factors responsible for within-tree pupal distribution patterns of G. postica and G. rufobrunnea were identified by determining how much of the variation in cocoon height and distance of the cocoon from the tree trunk could be explained by cocoon position attributes or tree characteristics. Functional type and height of tree, as well as branch position of the cocoon and sex were used as explanatory variables for cocoon height. Only tree functional type, tree height, and cocoon sex were used as explanatory variables for distance to trunk because branch position was logically correlated with distance to trunk. For the analysis of both continuous dependent variables, a generalized linear model assuming a normal distribution (log link function) was used (McCullagh & Nelder, Reference McCullagh and Nelder1989).
Results
Sites differed in the absolute and mean (±SE) number of branches, tree height, between sites, and spatial randomness, therefore offering a range of conditions to investigate pupal abundance patterns (appendix 1). In all but a few cases, counts of the number of branches per tree were randomly distributed within sites (appendix 1).
Between-tree variability
Significant patterns of non-randomness (both over- and under-utilization) were observed, after accounting for differences in the number of trees per site for each tree functional type (fig. 2a). For G. postica, the host plant was usually significantly over-utilized (ratio of observed to expected number of pupae greater than one) and only under-utilized (ratio smaller than one) in one case. In contrast, the host plant of G. rufobrunnea was under-utilized (fig. 2a). Both non-host tree functional types were significantly under-utilized by G. postica in most cases (only two cases of over-utilization). In contrast, either non-hosts with or without thorns were always significantly over-utilized by G. rufobrunnea (fig. 2a). Thus, G. postica pupated mostly on its primary host plant, while G. rufobrunnea tended to pupate on non-host plants in general. More G. rufobrunnea females were found on non-host plants and more males on the primary host plant relative to the opposite sex, and both sexes were significantly larger if occurring on non-host plant species (females: 40.02±0.15 (n=353) vs. 41.34±0.18 mm (n=218), t-value=−5.49, P-value <0.001; and males: 32.46±0.09 (n=719) vs. 34.09±0.17 mm (n=195), t-value=−8.17, P-value <0.001). Gonometa postica showed similar trends, but both sex ratio and female cocoon size were only significantly greater in non-hosts species where A. tortillis was the primary food plant (45.26±0.14 (n=356) vs. 46.13±0.34 mm (n=55), t-value=−2.22, P-value=0.027).
Fig. 2. Ratio of observed to expected number of Gonometa postica and G. rufobrunnea pupae for each site–generation combination (as in table 1) accounting for (a) tree functional type, (b) height differences between trees, and (c) branch position within trees. If bars are above one it indicates that a category is over-utilized (more individuals than expected), while when below one, under-utilization is indicated. The dotted line indicates when the observed and expected frequencies were equal. H, primary host plant; N, non-host plant without thorns; NT, non-host plant with thorns. Small (S), <1.75 m; medium (M), 1.75–3.00 m; and large (L), >3.00 m. (E), (EM) and (rest) denote edge, near edge, and all other branch positions. Sample sizes for b and c are indicated (a and c were similar). *, **, and *** indicate significance between categories at P<0.05, 0.01, and 0.001. ‘∫’ and ‘∫∫’ indicate bias in chi square values when 20% of frequencies are below five, or if any frequency is below one respectively. ns, not significantly different; na, not applicable.
Categorizing tree height of only host plant trees, marked differences in utilization were found between height classes, even after standardizing for frequency differences. In all cases large trees were over-utilized while small trees were consistently under-utilized. Where medium sized trees formed the largest category (Kopong), this size class was over-utilized (fig. 2b). Thus the largest of trees available within the site were over-utilized, independent of the actual size of the plant.
Among all trees, pupal abundances of both species were generally spatially random (table 1). Furthermore, in two out of three cases where aggregation was detected, other generations sampled at the same site were spatially random (table 1). However, despite spatial randomness in pupal abundance at the site scale, local clustering indices identified certain trees as contributing significantly to the formation of patches of pupal abundance. Thus, pupae were aggregating on specific trees. Spatial association between number of pupae and number of branches was significant in almost all cases for G. postica, while few significant cases were found for G. rufobrunnea (table 1). Local spatial association values were usually significant for only a few single trees. Thus the number of pupae per tree was independent of tree spatial position within the site.
Table 1. Spatial clustering of Gonometa species pupae and association between number of pupae and number of branches of a sample tree.
Significant positive association (5% level, two tailed test) was determined using SADIE. I a, v i, v j and X are the overall index of aggregation, mean clustering values of patches and gaps and overall association value respectively. *, **, and *** indicate significance at P<0.05, 0.01, and 0.001. Underlined values were non-significant after column wide step-up false discovery rate (FDR) correction at the 0.05 level.
The total percentage deviance in pupal abundance explained for G. postica and G. rufobrunnea ranged between 15–69% and 19–75% (table 2). For both species the spatial component contributed little to explaining pupal abundance in most cases, explaining more than 20% of the deviance in only two out of 26 cases. In contrast, generally more than 30% of the deviance was explained by the pure environmental component (spatial non-independence taken into account) (table 2). Number of branches followed by tree height was the most important variable explaining the pupal abundance of G. postica between trees. For G. rufobrunnea this pattern was not as general, with tree functional type and height adding greater percentages of explained deviance in several data sets. For both species, number of branches and/or tree height was positively related to pupal abundance in all cases (table 2).
Table 2. Forward stepwise regression of pupal abundance used to determine the percentage of deviance explained (DE) by spatial and environmental (sample tree) variables.
The total %DE by the spatial component (pure spatial and spatially structured environmental; see Legendre & Legendre Reference Legendre and Legendre1998), as well as the increase %DE by sequentially added significant tree variables (additively the pure environmental component) is shown. The order of adding significant tree variables and their respective coefficients is also shown. NBR, number of branches; HT, tree height; FTT, tree functional type (H, primary host; N, non-host; NT, non-host with thorns).
Number of pupae for each site–generation combination is as specified in table 1.
There was, however, a major difference between the species in the relationship between the tree functional type and pupal abundance. For G. postica, pupal abundance was significantly higher on its primary host plant than non-host plants in both Acacia veld types, whereas G. rufobrunnea pupal abundance was significantly lower on its host plant (table 2). Even though tree functional type added significantly to the percentage of explained deviance in 10 cases for G. postica, in half of these the regression coefficients were non-significant. In contrast, in four out of five cases tree functional type coefficients were significant for G. rufobrunnea (table 2). Thus, tree size seems to largely explain between-tree variation in pupal abundance for G. postica, while tree functional type was also important for G. rufobrunnea.
Within-tree variability
For each site–generation combination, the difference between expected and observed numbers of pupae per branch position was significant in most cases, with the edges of terminal branches and/or near edges of branches usually being over-utilized by pupae, while the pooled remaining branch positions were under-utilized (fig. 2c).
There were also significant differences between males and females in the frequencies of branch position occupied. For both species, males usually significantly over-utilized the edges of terminal branches, and in a few cases near edges of branches, while females mostly over-utilized the grouped remaining branch positions. Sex differences were significant for G. postica in nine cases (50%) (Vryburg2: generation (gen) 1, 2, and 4; Hotazel: gen 2, 4; Gabane: gen 1, 2; Kumukwane: gen 1, 4) and for G. rufobrunnea in three cases (38%) (Shashe1, Shashe3 and Dumela1: gen 1). The same utilization patterns for G. postica and G. rufobrunnea were evident when the total number of male and female cocoons per branch position was compared across the entire study. The percentage female cocoons in the ‘rest’ category was greater than that for males for both G. postica and G. rufobrunnea (fig. 3a).
Fig. 3. Percentage of total pupal population categorized by (a) branch position and (b) aspect for Gonometa postica and G. rufobrunnea. E, EM, ES, M (including ME and MS), and S denote edge, near edge, stem edge, middle of branch and main stem respectively (see fig. 1).
The difference between expected and observed numbers of pupae between aspects was significant in most cases for G. postica (81%), but not G. rufobrunnea (25%). Where such differences were significant, northern and/or eastern aspects were over-utilized, while southern and/or western aspects were under-utilized (results not shown). Nonetheless, the same pattern was evident for both G. postica and G. rufobrunnea when the total number of male and female cocoons per aspect was considered across the entire study (fig. 3b). There were, however, no significant differ-ences in the frequencies of males and females with respect to aspect (results not shown).
The distribution of pupae in terms of standardized cocoon height showed marked differences between- and within-species (G. postica). For G. postica at sites with Acacia erioloba, cocoon height was normally distributed, with most cocoons just above mid-tree height (fig. 4a). At sites with Acacia tortillis cocoon height was also normally distributed, but in this case most cocoons were found just below mid-tree height (fig. 4b). In contrast, G. rufobrunnea had a left skewed distribution with most individuals at the two-thirds tree height mark (fig. 4c). However, in all cases most pupae were found below the height where the greatest available canopy volume of the primary host plant was expected to occur (fig. 4a–c).
Fig. 4. Frequency distribution of standardized cocoon height of Gonometa postica pupae on Acacia erioloba (a) and on A. tortillis (b), as well as G. rufobrunnea on Colophospermum mopane (c). Shaded area next to distribution indicates hypothetical available pupation site volume. Dashed line indicates mid tree height.
In all cases the relationship between cocoon height and tree height was significantly positive (table 3). An analysis of cocoon height revealed that branch position, tree functional type and tree height, but not pupal sex, always contributed significantly to the percentage of deviance explained for Gonometa species (table 3). Cocoons with branch position category E, EM or ME were consistently found higher in the tree, while S category cocoons were found significantly lower. With respect to tree functional type, in all three regressions the cocoons on primary host trees were significantly higher than they were on non-hosts (table 3). For cocoons of G. postica on A. tortillis and G. rufobrunnea, cocoons on undefended non-host plants were significantly lower. This indicates that even when tree height is accounted for, tree functional type may still influence pupation height.
Table 3. Generalized linear regression of the height and distance from the tree trunk where pupation occurred for Gonometa postica (for both host plants) and G. rufobrunnea.
The fit and percentage deviance explained (d.e.) by the total model as well as the significance of independent variables is shown. Branch position: E, EM, ES, ME, M, MS, and S; denote edge, near edge, stem edge, edge of branch, middle of branch, start of branch, and main stem respectively. Sex: female (F) and male (M); Tree functional type: primary host (A.e., Acacia erioloba; A.t., A. tortillis; C.m., Colophospermum mopane), non-host no thorns (N) and non-host with thorns (NT).
Finally, for both Gonometa species cocoon distance to the tree trunk always had a significant positive relationship with tree height (table 3). Gonometa postica cocoons were significantly further from the tree trunk if on a primary host plant, while G. rufobrunnea had a tendency to be closer if on a non-host without thorns, although tree functional type did not significantly explain this distance. Gonometa postica on A. erioloba and G. rufobrunnea were significantly closer to the trunk if cocoons were female, while for G. postica on A. tortillis, sex had no significant effect (table 3).
Discussion
By documenting between-tree and within-tree pupal distribution in Gonometa this study takes the first step to generating testable hypotheses to explain these patterns. The fine-scale pupal distributions of G. postica and G. rufobrunnea was markedly non-random (in a non-spatial context) at both scales considered and was significantly explained by tree characteristics. This suggests that there may be a selective advantage to the choice of oviposition and pupation sites in these species. However, at the between-plant scale different factors potentially determine the distributions of the two Gonometa species, while within plants similar factors may result in common pupal distributions.
Between-tree pupal patterns
At a between-tree scale, most G. postica pupae were found on large primary host trees, while G. rufobrunnea used large primary host trees as well as non-host trees (one-third of all pupae) irrespective of their size. Also, tree size explained more of the variation in G. postica pupal abundance, and had a stronger positive spatial relationship with abundance (i.e. areas with large numbers of branches had high pupal abundance) than G. rufobrunnea. Nonetheless, for both species pupal abundance patterns were not explained by the spatial position of trees, but rather specific properties of the tree (i.e. size and tree functional type). This suggests that trees used as pupation sites are individually selected by adults or larvae (evidence given below) irrespective of their position relative to other trees (see also Rodeghiero & Battisti, Reference Rodeghiero and Battisti2000). For example, if an unsuitable tree occurs immediately next to a highly suitable tree, pupae will only be found on the latter, and never, or rarely on the former. The strong trend in G. rufobrunnea towards more females, and larger pupae in general, on non-host plants is a curious result. It is possible that large larvae are more likely to disperse, or have greater dispersal distances, from the host plant before pupation (see also Gutierrez & Menendez, Reference Gutierrez and Menendez1997; Etienne & Olff, Reference Etienne and Olff2004; Ness et al., Reference Ness, Bronstein, Andersen and Holland2004). As a result the pupae found on non-host plants will be larger and have a greater probability of being female (the larger sex in Gonometa species). Therefore, at the between-plant scale the two Gonometa species differed in the extent to which non-larval-host plants were used for pupation, as well as the importance of tree size in explaining pupal abundance.
Although several mechanisms can be used to explain why bigger trees have more pupae, two reasons suggest that oviposition behaviour of Gonometa adults is responsible for this pattern. First, host plant apparency is well known to affect the oviposition patterns of Lepidoptera (Courtney, Reference Courtney1982). For example, the oviposition pattern of Imbrasia belina (Westwood) (Saturniidae) (a species ecologically similar to G. rufobrunnea) is related to the apparency of the host plant, quantified as tree size and the proximity of neighbouring host plants (Wiggins, Reference Wiggins1997). During oviposition site selection, location of host plants is partly visual in most butterflies, and if the host plant is conspicuous oviposition is usually limited to host plants (Wiklund, Reference Wiklund1984). The primary hosts of both Gonometa species were highly apparent, generally the largest trees at the site, and most abundant. Large trees may thus be more apparent to ovipositing females and consequently receive more egg batches (Courtney, Reference Courtney1982; Batzer et al., Reference Batzer, Martin, Mattson and Miller1995; Wiggins, Reference Wiggins1997).
Second, large host plants have a greater probability of sustaining higher numbers of final instars and larvae are thus more likely to remain and pupate on these plants (Batzer et al., Reference Batzer, Martin, Mattson and Miller1995). Alternatively, larvae may die of starvation if the eggs they emerged from are located on small hosts (which are quickly defoliated, see Floater, Reference Floater2001; Rhainds et al., Reference Rhainds, Gries, Ho and Chew2002) or co-occuring non-host plants (Dethier, Reference Dethier1959; Steinbauer et al., Reference Steinbauer, McQuillan and Young2001; Hódar et al., Reference Hódar, Zamora and Castro2002). The first instar larvae of Lepidoptera species that commonly oviposit on non-host plants (generally species that overwinter as eggs or small larvae) use silk threads to ‘select’ host plants (Bernays & Chapman, Reference Bernays and Chapman1994). Consequently larvae will only survive if a suitable host plant is in close proximity (Leyva et al., Reference Leyva, Clancy and Price2003). The limited early instar dispersal ability of Gonometa suggests that if females oviposit on non-hosts, first instars may at best be able to disperse to suitable hosts directly next to the oviposited plant. Based on the large distances between the primary host plants of especially G. postica, early instar larvae are unlikely to successfully disperse to suitable hosts if oviposition occurs on unsuitable hosts. In general, oviposition on the host plant is typical of southern African Lasiocampidae (Scholtz & Holm, Reference Scholtz and Holm1985). Pupal distributions may thus simply be a result of host plant size.
Conversely, pupation patterns of Gonometa species are unlikely to be the result of secondary larval host plant selection by later instars. Although Lepidoptera larvae are more likely to move to an object that is visually conspicuous (Bernays & Chapman, Reference Bernays and Chapman1994), dispersal success to alternative hosts is usually low (Floater, Reference Floater2001). The low number of pupae relative to available foliage on host plants suggests that defoliation by Gonometa is rare and remaining on the host plant will be less costly than moving to a secondary host plant of the same species (Batzer et al., Reference Batzer, Martin, Mattson and Miller1995). There is thus little evidence to suggest that density dependent dispersal of larvae to secondary host plants occurs (see Rhainds et al., Reference Rhainds, Gries, Ho and Chew2002), and therefore adult oviposition patterns are the likely primary determinant of pupal distributions, at least for G. postica. However, the frequent use of non-host plants by G. rufobrunnea suggests that a secondary mechanism is required to explain the pupal distribution of this species. As an alternative, factors that influence pupal survival may influence the distribution of G. rufobrunnea. Pupal survival may be influenced by both abiotic (e.g. solar radiation) and biotic factors (e.g. natural enemy attack or avoidance) (Nowbahari & Thibout, Reference Nowbahari and Thibout1990; Kukal, Reference Kukal1995; Lyon & Cartar, Reference Lyon and Cartar1996; Hazel et al., Reference Hazel, Ante and Stringfellow1998; Bennett et al., Reference Bennett, Lee, Nauman and Kukal2003). Non-host plants used by cryptic G. rufobrunnea pupae, which are vulnerable to bird predation (Veldtman, Reference Veldtman2004), may serve as a form of enemy-free space. Predators, especially vertebrates, using visual cues may not only select high-density prey patches, but also form search images of prey against certain backgrounds (Guildford, Reference Guildford and Crawley1992). Using non-host plants may thus be a method of escaping bird predation, by disrupting the search image of the predator (Brower, Reference Brower1958; Hazel et al., Reference Hazel, Ante and Stringfellow1998). Apparent G. postica pupae, which are virtually immune to predation (Veldtman, Reference Veldtman2004), were seldom found on non-host plants, thereby supporting this hypothesis. Furthermore, when host plants have high larval densities, pupating on the same host plant will decrease the effectiveness of cocoon crypsis as an anti-predator defence (Brower, Reference Brower1958).
Within-tree pupal patterns
At a within-tree scale Gonometa pupae of both species showed similar patterns of branch position and aspect use, as well as patterns of (non-standardized) cocoon height and distance from the trunk. It is thus unlikely that natural enemy avoidance played a role here. Rather, interspecific similarities in within-tree use suggest a common explanation (e.g. the influence of abiotic factors). Most pupae were found on the edge or near the edge of branches on the eastern and northern sectors of trees, on larger trees, and occurred higher and further away from the main stem. Although there are more pupation sites on terminal branches, within-tree pupation patterns were not simply a matter of space availability, as more exposed branch positions were used than would be expected. Instead, differences in solar radiation may explain these patterns. For pupae in trees there may be a trade-off between maximum rate of development and avoiding hot midday-temperatures that are potentially detrimental to their survival (Denlinger, Reference Denlinger2002).
Branch positions near the trunk will receive the least solar radiation, while terminal branch positions will receive minimum shading (Kotzen, Reference Kotzen2003). Therefore, it is possible that the cooler microclimates near the tree trunk (see Klok, Reference Klok1998) are less favourable for pupal development, compared to those on the edge of branches that are most likely to receive oblique, early morning radiation (see Bryant et al., Reference Bryant, Thomas and Bale2002). Differential aspect use within trees may also be explained by differences in thermal microclimate properties (Stork et al., Reference Stork, Hammond, Russell and Hadwen2001). In the Southern Hemisphere, northern and eastern aspects of trees will receive more solar radiation in the morning than southern and western aspects, while the reverse is the case in the afternoon (see Kotzen, Reference Kotzen2003). Therefore, pupae positioned to receive maximum morning radiation may maximize developmental rates, without being exposed to detrimental afternoon radiation.
The difference in standardized cocoon height between Gonometa species (and between G. postica populations on different host plants) corresponded with differences in the shape of the primary host plants and provides further support for pupae avoiding direct solar radiation (i.e. high maximum temperatures). In all cases the maximum frequency height classes of both Gonometa species corresponded to regions below the maximum canopy volume of their host plants. Thus pupation site availability itself was not a major determinant of the position of pupae within trees, but rather, selection of pupation sites shaded at midday (see Kotzen, Reference Kotzen2003). Therefore, within-trees, branch position, aspect and tree shape may influence pupation site choice by providing favourable microclimate conditions for pupae.
However, sex differences in pupation site use suggest an added unknown mechanism resulting in within-tree pupation patterns. Possibly, using terminal branch edges is advantageous for the rapid, post-eclosion daytime-dispersing males, while more sheltered branch positions allow cover until nightfall for females that have limited powers of dispersal (R. Veldtman, personal observation). Nonetheless, the stronger and more consistent patterns supporting the favourable microclimate hypothesis suggest that differences in received solar radiation is currently the most parsimonious explanation for within-tree pupal distributions.
Conclusions
It has been shown that pupae have distributions that maximize their survival, because selection for pupation sites by larvae largely determines pupal survival probability (Ruszczyk, Reference Ruszczyk1996). However, when pupal survival is not affected by the distribution of the pupae, it appears that a herbivore insect will not modify its original spatial distribution in earlier life stages, and consequently similar patterns may still be visible in later life stages (see also Batzer et al., Reference Batzer, Martin, Mattson and Miller1995). The marked differences between Gonometa species at a between-tree scale, but strong similarities at a within-tree scale, illustrate the scale dependence of factors influencing herbivorous insect distributions (see also Hamid et al., Reference Hamid, Perry, Powell and Rennolls1999). In the case of Gonometa species, the present study study describes the pupal distribution at two scales relevant to its commercial use and conservation. For example, when searching for pupae between trees, non-host plants can be largely ignored for G. postica, but may harbour many G. rufobrunnea pupae. In addition, seeding pupae within trees may be more successful when following observed natural pupation patterns.
Acknowledgements
The authors thank A. Botes for assisting with fieldwork; F. and M. Taylor, (Development Consultancies) for making it possible to collect data in Botswana; farmers of the Northern Cape and North West Provinces for their hospitality and assistance, as well as P. le Roux, C.J. Klok, B.J. Sinclair, J.S. Terblanche and E. Marias for commenting on previous drafts of the manuscript. Liberty Life Trust and the National Research Foundation are thanked for funding the project. This material is based upon work supported by the National Research Foundation under Grant number GUN2053618 and GUN2053665. Any opinions, findings and conclusions or recommendations expressed in this material are those of the authors and therefore the NRF does not accept any liability in regard thereto.
Appendix 1
Vegetation characteristics of sites (consisting of 100 trees each) where Gonometa species were sampled. The frequency of trees according to functional type (primary food plant (H); non-larval host plant (N); non-larval host plant with thorns (NT)) and primary host plants according to tree size (small <1.75 m; medium 1.75–3.00 m; large >3.00 m) is given. * and *** denote significant difference at P<0.05 and 0.001, while ** indicates P>0.90. I a=Index of overall aggregation.
Underlined values lost significance after column wide step-up false discovery rate (FDR) correction at the 0.05 level.
