Size-dependent predation risk in tree-feeding insects with different colouration strategies: a field experiment

Authors


Correspondence author. E-mail: triinu.remmel@ut.ee

Summary

1. Body size is positively correlated with fecundity in various animals, but the factors that counterbalance the resulting selection pressure towards large size are difficult to establish. Positively size-dependent predation risk has been proposed as a selective factor potentially capable of balancing the fecundity advantage of large size.

2. To construct optimality models of insect body size, realistic estimates of size-dependent predation rates are necessary. Moreover, prey traits such as colouration should be considered, as they may substantially alter the relationship between body size and mortality risk.

3. To quantify mortality patterns, we conducted field experiments in which we exposed cryptic and conspicuous artificial larvae of different sizes to bird predators, and recorded the incidence of bird attacks.

4. The average daily mortality rate was estimated to vary between 4% and 10%. In both cryptic and conspicuous larvae, predation risk increased with prey size, but the increase tended to be steeper in the conspicuous group. No main effect of colour type was found. All the quantitative relationships were reasonably consistent across replicates.

5. Our results suggest that the size dependence of mortality risk in insect prey is primarily determined by the probability of being detected by a predator rather than by a size-dependent warning effect associated with conspicuous colouration. Our results therefore imply that warningly coloured insects do not necessarily benefit more than the cryptic species from large body size, as has been previously suggested.

Introduction

Body size is a central parameter in animal life histories as it inevitably interacts with most fitness-related traits (Stearns 1992; Blanckenhorn 2000; Roff 2002). As a general rule, large body size in females is thought to be favoured by fecundity selection; indeed, body size has been shown to be a major determinant of female fecundity in many insect species (Honěk 1993; Tammaru, Kaitaniemi & Ruohomäki 1996). While selection for larger body size is often easy to detect and may be fairly consistent across species, counterbalancing factors tend to be less obvious (Blanckenhorn 2000). Therefore, it is a challenge for evolutionary ecology to find the mechanisms preventing insects from evolving towards increasingly larger sizes (Endler 1986; Tammaru, Esperk & Castellanos 2002; Blanckenhorn 2005).

While constraint-based explanations may be sufficient to explain the absolute upper limit of insect body sizes (e.g. Chown 2001), the high level of between-species variability in this trait suggests that adaptive explanations for body sizes perhaps deserve greater attention. One such adaptive explanation is that the short development time associated with small body size permits faster generation turnover (Kozłowski 1993; Roff 2002). However, as the number of generations per season is typically fixed in temperate insects, this argument is unlikely to have widespread validity.

Alternatively, the high level of predation risk typically associated with juvenile stages has been proposed to be a crucial selective factor (Berger, Walters & Gotthard 2006; Berger et al. 2006; Mänd, Tammaru & Mappes 2007) that could have a similar effect on optimal size in various species. Indeed, it is widely recognized that high predation pressure on juvenile stages selects for earlier maturation and, consequently, smaller body size (e.g. Stearns 1992; Stoks 2001; Beketov & Liess 2007). Nevertheless, to build case-specific models of optimal body size in insects, reliable quantitative estimates of predation risk are needed; such estimates are not, however, readily available.

A constant mortality rate that is independent of body size may still be insufficient to counterbalance the fecundity advantage of being larger. This may be particularly true when weight gain per unit time is high, as is typical in moth larvae (Tammaru 1998). Under such conditions, an increase in predation risk with increasing prey size could have a much stronger effect on the size optimum of the prey.

Bird predation is the most likely factor that could cause positively size-dependent mortality in insects. Insectivorous birds are major predators of tree-feeding insect larvae (e.g. Cornell & Hawkins 1995; Grushecky et al. 1998) and, importantly, they are visual hunters. Various arthropods represent another important predator guild that exploits folivorous insect larvae. However, as the threat posed by arthropods is often found to weaken as prey size increases (Roger, Coderre & Boivin 2000), they are less likely to cause positively size-dependent predation risk in their prey, and are therefore of less relevance in the current context.

The probability of an insect larva being eaten depends both on its detectability and its acceptability to predators, both of which may be related to larval body size (Mänd et al. 2007). Moreover, the degree to which these aspects of predation risk are dependent on size may radically differ for insects with different defence strategies. Aposematic insects use conspicuous warning colours to signal their unpalatability to potential predators (Poulton 1890), and large body size can enhance the warning effect (Nilsson & Forsman 2003), making them even less acceptable prey objects. Indeed, in a previous study we showed that wild birds prefer larger cryptic but smaller conspicuous prey items (Mänd et al. 2007). This is consistent with other work that has reported similar trends in birds’ preferences for different sized prey insects (Gamberale & Tullberg 1996, 1998; Forsman & Merilaita 1999). In addition to large body size, large pattern elements (Forsman & Merilaita 1999; T. Remmel & T. Tammaru, unpublished data) and large groups of individuals (Mappes & Alatalo 1997; Gamberale-Stille 2000; Riipi et al. 2001) have been shown to increase the repellent effect of conspicuous warning colouration in prey. Thus, it seems reasonable to predict that aposematic insects should gain more from large body size than cryptic insects do, and should therefore evolve larger body sizes on average (Forsman & Merilaita 1999; Nilsson & Forsman 2003). However, empirical evidence for the predicted difference in average body sizes is still lacking (Nilsson & Forsman 2003).

The size dependence of detectability also appears to differ between cryptic and conspicuous insects. An aviary experiment with artificial prey demonstrated that the detectability of conspicuous baits for great tits Parus major Linnaeus was considerably more steeply dependent on size than that of cryptic baits (Mänd et al. 2007). Such a relationship has the potential to counteract the advantage of lower acceptability associated with larger conspicuous species, assuming that warning colouration does not guarantee full protection to larvae (Exnerova et al. 2003; Endler & Mappes 2004). In field conditions, the opposing effects of acceptability and detectability may potentially result in size-dependent mortality patterns of differing direction and strength for conspicuous prey, depending on which of the two aspects is more relevant in natural situations. Specifically, if acceptability is the more important size-dependent determinant of predation risk, we would expect to find a negatively size-dependent mortality rate in conspicuous but not cryptic larvae. Conversely, if detectability is more important, we would expect to observe positively size-dependent mortality rates in both larval colour types.

Here we report the results of experiments that aimed to demonstrate the net outcome of these two determinants of insect mortality – acceptability and detectability – on the size-dependent survival of cryptic and conspicuous larvae in a natural setting. For this purpose, we exposed artificial insect larvae of different sizes and both colour types to bird predation in temperate forest habitats and recorded the occurrence of bird attacks on the larvae. The experiments were designed to provide realistic, quantitative estimates of size-dependent predation risk in tree-feeding insect larvae, which are essential for constructing empirically based life-history models. Predation experiments were repeated several times between May and August in order to detect possible seasonal changes in the relative predation risk of differently sized and coloured larvae. The results concerning the temporal dynamics of predation risk per se have been published in a separate paper (Remmel, Tammaru & Mägi 2009).

Materials and methods

The two experiments reported in this study (performed at Järvselja in 2003 and near Tartu and Kilingi-Nõmme or K-N in 2005–6) aimed to quantify and compare the size dependence of bird predation risk experienced by cryptic and conspicuous tree-feeding insect larvae. For this purpose, differently sized and coloured artificial prey items, designed to imitate Lepidopteran larvae, were exposed to predators in the field, and bird-inflicted damage was recorded after 2 or 4 days. The two experiments shared the same general design but differed in the timing of trials relative to the seasonal cycle, the material used to prepare artificial prey items and some minor details of experimental set-up (see next).

In 2003, we used: (i) edible pastry cylinders, consisting of lard and flour (Church, Jowers & Allen 1997) and (ii) similarly shaped inedible plasticine cylinders to imitate insect larvae (Fig. 1a). In 2005–6, we used commercially available live blowfly pupae (c. 1 cm in length) arranged in rows to form ‘larvae’ of 1–5 cm in length (Fig. 1b,c). The reason for using both edible and inedible materials in 2003 was to test for the possible bias in the results caused by birds learning. The use of inedible larvae was abandoned in 2005–6, as edibility had no effect on the ‘mortality’ of larvae (see Results).

Figure 1.

 Examples of artificial larvae used in the experiments: (a) a schematic drawing of the size and colour classes used in all experiments; (b) an intact larva from the 2005–6 experiment; (c) a larva damaged by bird(s) from the 2005–6 experiment (arrows indicate the bird-inflicted damage).

Artificial larvae of four different sizes were used in all trials: 1, 2·3, 3·6 and 5 cm in length (for more details, see Remmel et al. 2009). Larvae were painted either green (cryptic), or black and yellow (conspicuous). However, different conspicuous colour patterns were used in different trials to enable broader generalizations to be drawn from the results. In 2003, the conspicuous pattern consisted of four yellow dots on a black background; in 2005–6, two different patterns were used: coarse (5-mm wide) or fine (1-mm wide) yellow stripes on a black background. The coarse pattern was expected to be more conspicuous than the fine pattern. In 2003, pastry larvae were coloured with non-toxic finger-paints, while plasticine larvae were made by combining pieces of differently coloured non-toxic plasticine. The water-based colours applied to the puparia of blowflies in 2005–6 were similarly non-toxic, and, in any case, birds did not consume the puparia but only the soft inner tissue, and so did not ingest the paint. With no difference in the materials used, the cryptic and conspicuous larvae were assumed to be equally palatable (except the plasticine baits in 2003). The colours of the cryptic larvae were tested for ultraviolet reflection to confirm their non-conspicuous appearance.

In August 2003, predation trials were conducted at 16 sites (transects along forest edges) in a southern boreal forest habitat around Järvselja, Estonia (58°16′ N 27°19′ E). In 2005–6, the experiments took place in similar forested habitats in Estonia, in two locations c. 130 km apart: near the town of K-N (58º09′ N, 24°58′ E) in 2005 and 2006, and near the city of Tartu (58°23′ N, 26°43′ E) in 2006.

In 2003, five larvae (either pastry or plasticine) of each colour × size combination (i.e. 40 larvae in total) were glued, singly, to the narrow (< 5 mm in diameter) apical twigs of deciduous trees along the forest edge at each site. The distance between adjacent larvae was > 5 m, such that the artificial larvae were arranged along a linear transect of c. 200 m in length. In 2005–6, the trials were performed at 12 sites (> 1 km apart) in each area (K-N and Tartu). In each temporal replicate (i.e. date of the experiment), two larvae of each colour × size combination (i.e. 16 larvae in total) were placed at each site, with adjacent larvae > 5 m apart, forming linear transects of about 80100 m. Such replicates (consisting of 16 larvae exposed at different sites and different periods during the summer) will hereafter be referred to as site × time replicates. In all trials, the exact position of each larva was chosen before its type (colour and size) was randomly allocated. To avoid any damage by predatory insects (particularly ants), the basal sections of twigs bearing larvae were surrounded with a 2-cm wide belt of insect glue (Tanglefoot®, The Tanglefoot Company, Grand Rapids, Missouri, USA). In total, 640 larvae were exposed in 2003, and 4256 larvae (in 168 site × time replicates) in 2005–6.

The experiment was repeated five times between May and August in K-N 2005, eight times in K-N 2006 and nine times in Tartu 2006 (see Remmel et al. 2009 for exact dates). At any given site, successive trials were carried out in slightly different locations, > 100 m from one another, to avoid potential effects associated with learning by the birds inhabiting a particular location. In 2006, exact overlap in the timing of trials in K-N and Tartu was avoided: generally, they were performed on an alternating weekly basis. This was performed to minimize the confounding effects of weather on predation rates; by avoiding similarities in short-term weather conditions, it was considered justified to include the two areas as independent replicates in the analyses.

Larvae were checked for beak marks or partial consumption by birds after 2 (2003) or 4 (2005–6) days. Beak marks are clearly recognizable in soft materials like pastry or plasticine, and bird predation was also clearly identifiable from the remains of blowfly pupae: the puparia were typically torn apart with only the soft pupa missing (Fig. 1c). The appearance of the attacked ‘larvae’ in the field matched well with that of the blowfly pupae consumed by great tits in cage trials. Moreover, predatory insects were effectively excluded in all but a couple of cases in which larvae were attacked by ants (the appearance of ant-damaged pupae is clearly different from those attacked by birds). Cases where larvae were missing or where damage could not be reliably attributed to bird attacks (2·5% in 2003 and 1·4% in 2005–6) were excluded from statistical analyses.

Generalized Linear Models (sas proc glimmix; SAS Institute Inc. 2008) were used to estimate the effect of the factors colour (cryptic or conspicuous) and site of exposure (a random variable) on the binary response variable (attacked or not), with size included as a covariate. In 2003, material (plasticine or pastry) was included as an additional fixed factor, while in 2005–6, time (the date of experiment) was included as an additional covariate. Size and time were standardized by subtracting the mean size or time, respectively, such that the main effects of the categorical variables in the resulting heterogeneous slopes model could be interpreted as applicable to an average-sized larva in mid-summer (Littell, Stroup & Freund 2002). By default, in heterogeneous slopes models, the main effects of factors are interpretable for a situation where the continuous variables are equal to zero. Hence, without the described transformation, the main effect of, e.g. colouration would have been applicable to a larva of size zero, which is not biologically meaningful.

In 2005–6, the dependence of mortality rate on larval size and colour was analysed separately for the three combinations: K-N 2005, K-N 2006 and Tartu 2006. Additionally, the effect of pattern crudeness in the conspicuous group was analysed using a separate Generalized Linear Model.

Results

Experiment of 2003: differential prey survival

In 2003, a mean of 18% (median 20%) of exposed larvae were attacked by birds (i.e. had recognizable beak marks) during the 2 days of exposure. Larger larvae were attacked more frequently than smaller ones (Table 1 and Fig. 2). When the two colour groups were analysed separately, this trend was significant in conspicuous, but not in cryptic prey items (conspicuous: F(1, 298) = 4·74, = 0·03; cryptic: F(1, 318) = 0·75, = 0·39). However, there was no significant size × colour interaction in the full model (Table 1). The average daily mortality rate was calculated as: inline image, where m is the mean mortality recorded during the 2-day period. Daily mortality was 10·2% (SD = 14·3% across the 16 sites) in the smallest and 12·4% (SD = 14·9%) in the largest cryptic larvae, but 5·4% (SD = 6·9%) in the smallest and 12·7% (SD = 11·6%) in the largest conspicuous larvae. In both colouration groups, there was little difference in the survival of the two largest size classes (Fig. 2). Pooled over all size classes, the cryptic larvae were attacked about 9% more frequently than the conspicuous ones; however, this difference was not statistically significant (Table 1).

Table 1.   Type III analysis of repeated measures Generalized Linear Model investigating the effect of larval colour and size on the probability of bird attack; Järvselja 2003. Omitting the non-significant effects from the model did not qualitatively change the results
EffectNumerator d.f.Denominator d.f.F*P
  1. *The F-values were obtained by the glimmix procedure in sas.

  2. d.f., degree of freedom.

Colour16151·370·24
Size16154·830·028
Material110·660·500·49
Size × colour16151·060·30
Figure 2.

 Per cent of bird-damaged cryptic and conspicuous larvae in different size classes during the exposure trial, in 2003 (Järvselja, mean ± SE over the 16 study sites) and 2005–6 (Kilingi-Nõmme and Tartu, mean ± SE over the 168 site × time replicates).

There was a substantial difference between study sites with respect to predation rates: daily larval mortality in different sites ranged from 0% to 24% [mean = 10·3% ± 6·5% (SD)]. The material used to prepare the artificial larvae – pastry or plasticine – did not have an effect on overall survival, neither did it interact with colour (F(1, 614) = 3·56, = 0·60) or size (F(1, 614) = 1·19, = 0·28). This indicates that the results were not affected by birds learning to avoid or accept certain larvae; the two materials were therefore pooled.

Experiments of 2005–2006: seasonal dynamics in differential prey survival

The mean level of bird predation on exposed larvae was 14% (median 6·7%) in K-N 2005, 12% (median 6·3%) in K-N 2006 and 18% (median 12·5%) in Tartu 2006. Daily mortality, calculated as: inline image, where m is the mean mortality recorded during the 4-day period, was on average c. 5%, but ranged from 0% to 34% among the different site × time replicates. In total, 1·4% of the larvae were either not found at inspection or damaged by causes other than bird predation. Scoring the missing larvae as attacked by birds did not substantially alter the results.

The mortality of both cryptic and aposematic larvae was positively dependent on size (Table 2 and Fig. 2). The main effect of colour on prey survival was not significant, although conspicuous larvae were attacked about 5·7% more often than the cryptic ones. The significant size × colour interaction term in the models for K-N 2005 and 2006 indicated that the slope of size dependence of mortality was steeper in the conspicuous group; however, this interaction was not significant in Tartu 2006 (Table 2). The mean daily mortality rate in the cryptic group was 2·6% (SD = 2·3 over the site × time replicates) in the smallest larvae and 4·9% (SD = 2·3) in the largest larvae; in the conspicuous group, it was 1·9% (SD=1·9) in the smallest and 6·1% (SD = 3·2) in the largest larvae. Again, the difference between the two largest size classes was low in the larvae of both colour types (Fig. 2).

Table 2.   Type III analysis of repeated measures Generalized Linear Model investigating the effect of larval colour and size on the probability of bird attack; Kilingi-Nõmme (K-N) 2005, 2006 and Tartu 2006. Omitting non-significant effects from the model did not qualitatively change the results
Effectd.f.F*P
  1. *The F-values were obtained by the glimmix procedure in sas.

  2. d.f., degree of freedom.

K-N 2005
 Colour10·110·74
 Size122·42<0·0001
 Time10·620·43
 Size × colour14·470·035
 Error973  
K-N 2006
 Colour10·050·83
 Size122·43<0·0001
 Time124·65<0·0001
 Size × colour13·960·047
 Error1492  
Tartu 2006
 Colour10·500·50
 Size121·47<0·0001
 Time145·74<0·0001
 Size × colour10·050·82
 Error1712  

The daily survival of exposed larvae varied considerably in relation to the study site (from 0% to 34%) and the date of exposure (discussed elsewhere: Remmel et al. 2009).Nevertheless, no significant time × size, time × colour or time × size × colour interactions were apparent: all size and colour classes followed approximately the same temporal dynamics (Fig. 3).

Figure 3.

 Temporal dynamics of predation risk in cryptic (open circles) and conspicuous (closed circles) larvae (mean ± SE over all larvae), Kilingi-Nõmme and Tartu 2005–6 pooled.

We detected no main effect of pattern crudeness on survival (F(1, 1230) = 0·38, = 0·54) or pattern × size interaction in conspicuous larvae (F(1, 1230) < 0·01, = 0·97). This suggests that the results of size-dependent mortality can be generalized to prey animals with different conspicuous colour patterns.

Discussion

Predation risk was positively dependent on size for both cryptic and conspicuous artificial larvae. Expressed in terms of a daily mortality rate, the largest cryptic larvae were attacked about 1·5-fold more often than the smallest ones, and the largest conspicuous larvae about 3-fold as frequently as the smallest ones. The results were reasonably consistent across the replicates in our experiment, and the patterns of mortality risk were unaffected by the type of material used to make artificial larvae or the details of the warning colouration. We therefore believe that these patterns of size-dependent mortality can be used as realistic estimates in optimality models predicting the adaptive value of adult body size in tree-feeding temperate insects. The absence of an interaction between calendar date and body size in determining mortality risk is also a noteworthy result in a general life-history context, even if the seasonal dynamics in mortality risk themselves cannot be ignored (Remmel et al. 2009).

The small differences in mortality risk that were apparent between the two largest size classes in both of our replicates (2003 and 2005–6) is another result of qualitative importance and may indicate that the size dependence of predation risk reaches a plateau at larger sizes. Such a pattern would undoubtedly reduce the potential of bird predation to form the selective force behind the cost of large size in insects: the possibility that mortality risk would substantially increase when the actual maximum size is surpassed appears not to be supported by our data. However, as the apparent deceleration of the size-dependent mortality curve was present in conspicuous larvae in Järvselja and Tartu, but not in K-N, and because there were only four size classes, it is currently premature to draw firm conclusions about the shapes of these curves. Nevertheless, we feel that this question may deserve further attention.

Surprisingly, the largest warningly coloured larvae in our experiments were attacked slightly more frequently than equally sized cryptic larvae (Fig. 2). Although the repellence of the colour combination used in this experiment has been proven in aviary trials (birds are much more cautions in attacking such objects: Mänd et al. 2007), it is clear from this experiment that the warning colouration did not prevent artificial larvae from being attacked. In terms of the two aspects determining mortality risk (see Introduction and Mänd et al. 2007), our results suggest that, in field conditions, detectability is far more important than acceptability in determining the size dependence of predation rate in tree-feeding insect larvae. If the opposite had been true, we would have expected to observe a negative relationship between mortality risk and body size in warningly coloured larvae, as their acceptability has been found to be negatively related to size (Mänd et al. 2007).

Moreover, in both years in K-N, the relationship between size and mortality rate was more strongly positive in the conspicuous than the cryptic prey (Fig. 2 and Table 2), although the slopes of size dependence appeared similar in the other two locations. Nevertheless, it can be concluded that in a natural setting, the positive size dependence of detectability was not balanced by a positive size-dependent increase in the warning signal strength. The steep slope may have reflected the tendency of small conspicuously coloured larvae to be harder for birds to find than the equally sized green ones (Fig. 2). This difference is likely to be attributable to the disruptive effect (Cott 1940; Merilaita 1998; Stevens & Cuthill 2006) created by the contrasting colour patches in the warning colouration, and is consistent with our previous results (Sandre, Tammaru & Mänd 2007). Alternatively or additionally, the low mortality of small conspicuous larvae may have resulted from optimal foraging decisions by birds (Stearns 1992; Rutten et al. 2006), whereby the smallest conspicuous larvae may not have been regarded as sufficiently rewarding to risk intoxication.

Furthermore, our finding implies that warningly coloured insects may not actually benefit more from growing large than cryptic species, as has previously been suggested (Forsman & Merilaita 1999; Nilsson & Forsman 2003). While the warning effect of conspicuous colouration may indeed be too weak to pay off for the smallest individuals (Mänd et al. 2007), it may equally be counteracted by increased detectability among the largest individuals. Indeed, there is an increasing body of evidence to suggest that various bird species readily attack warningly coloured prey (Exnerova et al. 2003; Endler & Mappes 2004; Sandre et al. 2007), and that even the strongest warning signals cannot confer protection from all predators. Thus, it appears that warning colouration should primarily be expressed in medium-sized insects, and there is indeed some empirical evidence to support this (Sandre et al. 2007).

Highly size-dependent predation risk in conspicuous species may also help to explain why so many aposematic insects appear to display sub-maximally conspicuous warning patterns (Endler & Mappes 2004). Given that warning colouration is unlikely to confer complete protection from predators, insect larvae may benefit from expressing relatively small warning signals, which could help to conceal them from foraging predators, but which would deliver an aposematic signal once they were detected (Tullberg, Merilaita & Wiklund 2005). Our results indicating a tendency for a steeper slope of size dependence in conspicuous larvae suggest that expressing sub-maximally intense signals may also be useful to flatten the size-dependent rise in detection risk.

There were no significant differences in overall predation risk between cryptic and conspicuous larvae, although the latter are probably easier to find, at least among the larger size classes (Mänd et al. 2007). The most likely explanation for this is that the acceptability of warningly coloured larvae to bird predators was reduced to the extent that it counterbalanced the cost associated with their higher detectability. Birds’ avoidance of warningly coloured prey can be either learned or innate, and the latter may be mediated by different psychological mechanisms, such as dietary conservatism (Marples, Roper & Harper 1998; Marples & Kelly 1999) or an inborn aversion to certain colours (Exnerova et al. 2007) or to any novel-looking prey (neophobia, Barnett 1958; Marples & Roper 1996). In the context of this study, we suggest that the mechanism behind the low overall predation rate of conspicuous larvae is more likely to be an aversion to the classical black-yellow warning signal than neophobia. This is because previous aviary studies have shown that great tits (major predators of Lepidopteran larvae throughout the Palearctic boreal zone) readily attack and consume different novel food items (e.g. Exnerova et al. 2007; Rowland et al. 2007). Moreover, neophobic reactions are often overcome in a matter of minutes (Marples & Kelly 1999; Sandre et al. 2007). However, the comparison of overall mortality in cryptic and conspicuous larvae must be made with caution, because both types of artificial larvae were immobile and equally well exposed in our experiments; in contrast, in natural conditions, warningly coloured larvae are likely to be more active and exposed during the daytime than cryptic larvae (e.g. Bernays & Singer 2002).

We also found that the seasonal dynamics of predation risk were similar for larvae of both colour types. This result differs from the findings of Ojala (2006), whose work in a geographically close area showed that predation rates on conspicuous models of Parasemia caterpillars were highest, compared with the cryptic models, just after the inexperienced fledglings of most insectivorous birds started foraging. The discrepancy between our study and that of Ojala may have resulted from various differences in the experimental settings. In particular, our experiments were conducted in forest habitats, whereas those of Ojala took place in grassland. These habitats are likely to contain different bird communities. While it is known that avoidance of conspicuous prey is innate and does not need to be learned in some, but not all, of the main bird predator species of tree-feeding insects (Exnerova et al. 2007), it is not clear whether the same is true in grassland bird communities.

While birds are likely to represent the dominant size-selective predators for tree-feeding insect larvae in temperate areas (Cornell & Hawkins 1995), other natural enemies exist, mainly in the form of predatory arthropods and parasitoids, which could also potentially cause size-dependent mortality. Although arthropods generally pose a higher risk for smaller larvae (e.g. Roger et al. 2000; Cogni, Freitas & Amaral Filho 2002), it would still be worth investigating whether they can substantially affect the optimal sizes in their prey. Similarly, while some parasitoids have no preference for host size (e.g. Teder & Tammaru 2001, 2003), others may prefer either smaller (e.g. Li et al. 2006) or larger host individuals (e.g. Teder, Tammaru & Pedmanson 1999; Lin & Ives 2003), and may therefore create selection pressure favouring either smaller or larger body size. The methods used in the present study were not designed to study the effect of predators other than birds: artificial prey is not suitable for investigating arthropod predation, which is strongly dependent on the prey animals’ ability to fend off the attacks (Chiravathanapong & Pitre 1980; Roger et al. 2000). Therefore, other complimentary methods should be used to derive the complete picture.

In conclusion, we found consistent trends of positively size-dependent predation risk in both cryptically and conspicuously coloured larvae. The respective estimates can be used to model mortality in quantitative studies on various aspects of insect life history. Our results suggest that the effect of detectability is more important than that of acceptability in determining the size-dependent mortality risk of larvae. At least within the size range of prey investigated here, predation risk seems to be more strongly dependent on size in conspicuous larvae. Bird predation is thus a more likely selective factor against large body size in conspicuously rather than cryptically coloured insects.

Acknowledgements

The authors thank Oksana Gluško who participated in the experimental work. Mike Boots, John Davison, Andrew Higginson, Tom Sherratt and Elin Sild made valuable comments on the manuscript. The study was supported by Estonian Science Foundation grant 7522, targeted financing project SF0180122s08 and by the European Union through the European Regional Development Fund (Center of Excellence FIBIR).

Ancillary