1 The shield beetle Cassida rubiginosa Müller (Coleoptera, Chrysomelidae) is one of the most prominent herbivores on creeping thistle, Cirsium arvense (L.) Scop. and is considered as a biological control agent against this weed. In order to investigate if augmented densities of this species could be sustained, predation on C. rubiginosa larvae was studied in the field with the aid of continuous video surveillance of individual prey during three seasons at two different study sites.
2 The paper wasp Polistes dominulus Christ (Hymenoptera, Vespidae) was responsible for 99·4% predation on C. rubiginosa larvae at both study sites. Prey densities were always reduced to a few individuals at the sites after a few days only. Thus, P. dominulus may annihilate the effect of augmentative releases of the beetle as a biocontrol agent.
3P. dominulus showed no aggregation in patches of high host density (numerical response). The observed density dependence was attributed to the functional response of the wasp.
4P. dominulus showed a Type III functional response. In mark-and-recapture experiments we found no indication that this type of functional response was the result of single wasps specializing on this particular prey type.
5 Type III functional responses in generalists can result by switching to prey species which become temporarily abundant. Since in a natural setting a mixture of prey species coexists in different abundance and which are differentially accepted by generalist predators, it is imperative to study their functional response in the field.
The functional response of a predator is a key factor in the population dynamics of predator–prey systems. The functional response can determine if a predator is able to regulate the density of its prey (Murdoch & Oaten 1975). For this to be the case the functional response must show density dependence, i.e. the predator must respond to higher prey densities by consuming an increasing proportion of the available prey over a range of prey densities. Such functional response curves are classified as Type III (Holling 1966), and are characterized by their sigmoid shape when plotting the number of prey items eaten by the predator against the number of prey available.
In contrast, field studies on the functional response of smaller predators, in particular insects, in the field are almost absent. The only true field study on the functional response of insect predators is, to our knowledge, the one by Van den Berg et al. (1997), who calculated predation rates by direct field observations of individual coccinellid and staphylinid predators at various densities of aphids. However, direct observations are very time-consuming in order to obtain a reasonable number of predation events, and are probably not feasible for mobile predators preying on non-gregarious prey. Data from the study by Van den Berg et al. (1997), for example, did not allow for a rigorous testing of the type of the functional response due to the low number of data points.
Recently, remote video surveillance has been suggested as the method of choice for non-intrusive observation of predator–prey interactions in the field because many prey individuals can be observed simultaneously (Kidd & Jervis 1996). Video techniques have been used extensively in laboratory studies but so far were rarely applied to study insects in the field (Wratten 1994; Meyhöfer 2001). We used remote video surveillance to study the predator complex of larvae of the shield beetle Cassida rubiginosa Müller (Coleoptera, Chrysomelidae). This species is considered as a potential native biological control agent against creeping thistle, Cirsium arvense (L.) Scop, one of the world’s worst weeds in agriculture (Holm et al. 1977). In order to investigate if augmented densities of C. rubiginosa could be sustained, we also quantified density dependent predation by manipulating prey densities in the field.
EXPERIMENTAL SITE AND SET-UP
Fourth and early fifth instar larvae of C. rubiginosa used in the experiments were reared in the laboratory on cut shoots of creeping thistle. In 1999 and 2000, the experiments were conducted on a natural 30 m2 patch of c. 800 C. arvense shoots in Bern (site 1). This patch was part of a larger, partly uncultivated meadow of about 1000 m2 with a mixture of flowering plants, and was surrounded by orchard trees and hedges. In 2001, the experiment was carried out on an ecological compensation area of 6 × 150 m located within arable farmland in Belp, near Bern, dominated by creeping thistle, teasel (Dipsacus fullonum L.), parsnip (Pastinaka sativa L.), burdock (Arctium lappa L.) and grasses (site 2). The thistle patch was 50 m2 in size and included approximately 300 C. arvense shoots. Thistles were searched for naturally occurring C. rubiginosa larvae which were removed prior to the experiments. Predation on C. rubiginosa was investigated by continuous video surveillance of individual beetle larvae. This was feasible because C. rubiginosa larvae behave rather sedentary, generally remaining on the same leaf throughout the day. In previous experiments we established that C. rubiginosa larvae will not disperse to plants other than thistles, and that mortality from causes other than predation is virtually nonexistent in late C. rubiginosa larval instars (S. Bacher, unpublished). Thus, all larvae missing could be attributed to predation.
Predation registration took place from 2–8 August 1999, from 26 May to 16 August 2000 and from 4 June to 7 August 2001. Four video cameras were installed in autumn 1999, nine cameras in 2000 and 2001.
VIDEO EQUIPMENT AND RECORDING
Video recording equipment in the field consisted of one infrared (for day and night observations) and eight colour cameras (CCD colour camera module, Conrad Electronics), and two sets of each one time-lapse video cassette recorder (VCR, Panasonic AG-6124) and one multiplexer (Sony YS-SX310P). The use of a multiplexer allows that the feed of multiple cameras can be recorded on a single VCR. Both sets were placed in a weatherproof metal box (57 × 37 × 33 cm) in the field adjacent to the thistle patches. The boxes were ventilated by a fan (10 cm diameter) to avoid overheating of the equipment inside the box. The power for the equipment (220 V) was supplied from the nearest building. The sets were equipped with four and five cameras, respectively. Cameras were fixed to poles and adjusted with the aid of a portable video monitor (b/w LCD monitor, Conrad electronics) such that they focused on a single C. arvense leaf with a larva feeding on it. Video recordings were done in 24-h time-lapse mode. Thus, a conventional E-180-min video tape recorded the events of four and five cameras, respectively, for 24 h. The VCR setting allowed for a resolution of one picture every 1·28 s from each camera. Cameras were checked several times each day and re-focused on the larvae if necessary. Missing larvae were replaced. Tapes were searched for predation events in the laboratory on a third VCR-multiplexer set connected to a video monitor (Sony PVM-14 N6E). Recordings from cameras were investigated only if the larva disappeared during the observation period. Predation events were copied to VHS video tape on a conventional VCR (Panasonic NV-FJ610) for analysing and archiving purposes.
DENSITY-DEPENDENT PREDATOR RESPONSE
In 2000 and 2001, prey densities were manipulated during 42 and 30 consecutive days, respectively, starting in the beginning of June with prey densities of 58 and 69 individuals, respectively. Larvae were distributed uniformly over the patch. Each early morning the entire thistle patch was searched for all remaining larvae in order to determine predation rates of the previous day. In addition, once or twice per week the density (number of foraging predators at the site) of the predator species identified by video analysis was determined during two hours in the early afternoon (peak activity); all predators observed foraging at the site were caught by sweep netting and released after the observation period. In 2000, all wasps captured were marked individually with numbered colour plates used for bee queens.
We first analysed if density dependent predation rates were due to a numerical response of predators (i.e. aggregation in patches of high prey density) by linear regression of predator on prey density. Linear regression was also used to investigate the relation between predator density and weather data (air temperature, wind speed, relative humidity, percentage cloud cover; all data taken from 13·00 h) obtained from the Swiss meteorological station in Bern, Liebefeld. If independence of predator and prey density was found density-dependent predation rates were attributed to the functional response of predators. Because our experiments were conducted with prey depletion we used the ‘random predator equation’ (Rogers 1972) as a description of the functional response:
Ne = N0[1 − exp(−a(T − ThNe))](eqn 1)
where Ne is the number of prey eaten by the predator, N0 the initial prey density, a the attack rate, T the time that predator and prey are exposed to each other, and Th the handling time associated with each prey eaten. Equation 1 models a Type II functional response with a constant attack rate a. When the attack rate itself becomes a function of prey density, functional responses of Type III can be modelled (Hassell, Lawton & Beddington 1977). We used the most general useful form in which a is a hyperbolic function of N (Juliano 2001):
a = (d + bN0)/(1 + cN0)(eqn 2)
For Type III responses, the precise form of the model incorporating prey depletion we used was thus:
Functional responses of Type II and III are difficult to discriminate; several studies focused on this problem (e.g. Trexler, McCulloch & Travis 1988; Casas & Hulliger 1994). To distinguish between Type II and Type III responses we first investigated if the attack rate a increased with prey density. The handling time Th (defined as the time needed for a wasp to kill and eat a larva until take-off from the leaf) was determined from the video recordings. For each recording of a predation event the attack rate was calculated by solving eqn 1 for a (Hassell et al. 1977). We set the value of T to 24 h in our experiment. Density dependence was analysed by Spearman’s rank correlation. If a significant correlation was found, eqn 2 was fitted to the data using the non-linear least-squares curve-fitting procedure of Systat version 9. We also used a more sophisticated procedure to distinguish between Type II and Type III responses (Juliano 2001). We performed logistic regression of the proportion of larvae eaten as a function of initial prey density (Trexler et al. 1988). We first fitted a polynomial function
(Juliano 2001) where Ne/N0 is the probability a larva is eaten by a predator with the CATMOD procedure of SAS version 8·1. A criterion for separating Type II and Type III functional responses is to test for significant positive or negative linear coefficients in the expression fit by the method of maximum likelihood to data on proportion eaten vs. N0 (Juliano 2001). We applied non-linear least-squares regression of eqn 3 to our data for parameter estimation using the procedure NLIN in SAS. Non-significant parameters were removed from the model and the model recalculated until all parameters were significant. Although considered a less desirable choice for distinguishing between functional response types (Trexler et al. 1988), b > 0 obtained by non-linear least-squares regression of equation 3 is sufficient to conclude that the functional response is of a Type III form (Juliano 2001).
PREDATOR SPECIES, PREY DENSITY AND PREY MORTALITY
Of the 157 video recordings of predation events during the three seasons, 156 (99·4%) showed the paper wasp Polistes dominulus Christ (Hymenoptera: Vespidae) as predator of C. rubiginosa larvae (site 1: N = 50; site 2: N= 107, including one recording of a larva of Chrysoperla carnea Stephen preying on a C. rubiginosa larva). Predation events took place between 10·39 h and 17·27 h in 1999 and 2000 and between 10·12 h and 19·36 h in 2001, with activity peaks between 11·00 h and 17·00 h (site 1: 92%; site 2: 87·8%; Fig. 1). No predation was observed during the night.
The time-course of C. rubiginosa larval densities is shown in Fig. 2. In 2000, prey density dropped to three or two individuals on the site (after 16 days), remained constant at this level for 4 days, was increased to 30 larvae (after 20 days) and again decreased to a low level and stayed there for some days. The lowest density was one larva on the whole site. In 2001, prey density was increased six times over the experimental period. Between the peaks, prey density dropped to low density levels, with a minimum of one larva on the whole site. Daily number of C. rubiginosa larvae eaten by predators ranged between 0 and 22 individuals in 2000 and between 0 and 51 in 2001 (Fig. 2).
Sweep-netting revealed a density of foraging P. dominulus at the study site of one to 12 wasps in 2000 and one to four wasps in 2001 during the 2-h sampling period. Predator density was independent of prey density (Table 1). Therefore, we assume a constant number of predators foraging at all prey densities during the experiment. Predator density did not vary with daily weather conditions (Table 1). Only one of the marked wasps was recaptured in 2000.
Table 1. Linear regressions of predator density on prey density and weather conditions (N = 6)
Wind force (knots)
Cloud cover (octas)
DENSITY DEPENDENCE OF HANDLING TIME AND ATTACK RATE
Because P. dominulus appeared in the video recordings virtually as the only predator, we attributed the observed density dependent predation to this species only. Since there was no numerical response of the predator during the experiment, the shape of the prey mortality curve is determined by the predator’s functional response alone.
Handling time Th, defined as the time period between landing of a wasp on a leaf with a C. rubiginosa larva and its take-off after the meal, was calculated from all suitable video recordings of predation during the experiment (2000: N = 30; 2001: N = 77). Handling times of P. dominulus lasted between 11 and 871 s in 2000 and 15 and 695 s in 2001, with a median of 54 (2000) and 73 s (2001; Fig. 3). Handling time Th seemed to decrease with increasing prey density in both years, but this relation was only significant in the second year because of the larger number of recorded predation events in 2001 (2000: Th = 138·48–0·34 N, P = 0·864, r2 = 0·001; 2001: Th = 136·12–1·46 N, P = 0·011, r2 = 0·084). However, because the proportion of the variance explained by the regression was small in both years we used the median of the handling time to calculate attack rates. Calculated attack rates a (from the 30 and 77 video recordings, respectively) yielded values between zero and 7·5E−6 (2000) and zero and 2·4E−5 (2001). Attack rates increased with increasing prey densities (Spearman’s rank correlation 2000: rs = 0·789, P < 0·001; 2001: rs = 0·604, P < 0·001; Fig. 4). Curve-fitting yielded a = 1·3E−7 + 3E−7N/1 + 0·03 N (r2 = 0·6) in 2000 and a = 2·6E−6 + 5·5E−7N/1 + 0·021 N (r2 = 0·48) in 2001 (Fig. 4), indicating Type III functional responses for the data of both years.
DETERMINING THE FUNCTIONAL RESPONSE
In both years, none of the parameters except the intercept L0 estimated by logistic regression were significant (Table 2). The linear parameter L1 of the data from 2000 is negative indicating a Type II functional response, whereas L1 of the 2001 data is positive and the quadratic parameter L2 negative, indicating a Type III curve. A lack of finding significant support for a Type III functional response with the method of logistic regression is not sufficient to conclude that the underlying data was not generated by a Type III functional response. Parameter estimates for eqn 3, as performed by NLIN in SAS, revealed that in both years b and Th were significant and positive, while c and d were not significant, and therefore removed from the model (Table 3). This indicates that a was linearly related to N0 (eqn 2, and the functional response of Type III in both years. The choice of T, the total time predator and prey were exposed to each other, did not affect the proportion of variance explained by the regression (2000: r2 = 0·96; 2001: r2 = 0·98), but the size of the estimates for b and Th. To achieve realistic and meaningful parameter estimates we set the actual daily time of exposure of predators and prey in accordance with the activity period of wasps (Fig. 1) to 6 h rather than the total duration of each experiment (24 h). Handling time estimated by the model was 0·17 ± 0·02 h (= 10·2 min) at site 1 in 2000 and 0·07 ± 0·01 h (= 4·2 min) at site 2 in 2001 (Table 3). The fit of the final model is shown in Fig. 5.
Table 2. Maximum likelihood estimates of the logistic regression of proportion eaten on prey density (eqn 4)
Table 3. Least squares estimates of the reduced model (eqn 3, d=c = 0; T= 6 h)
Sum of squares
Video observations showed an unexpected predator spectrum. With one exception, we registered exclusively the paper wasp P. dominulus as predator of C. rubiginosa larvae during three seasons and at two different sites. One reason may be that C. rubiginosa, as do other shield beetle larvae, carries a faecal shield over its abdomen which provides mechanical protection against predation (Eisner, van Tassell & Carrel 1967). However, this protection is not perfect. For example, Eisner & Eisner (2000) showed that while the faecal shield of Haemispherota cynaea (Say) larvae (Cassidinae) was effective against most predators, a single carabid species (Calleida viridipennis Say) was able to circumvent the protection (due mainly to its large body size), feeding on H. cynaea larvae in the field. This seems to be similar in our system. While small generalist predators such as ants, predatory bugs and lacewing larvae were abundant on both sites (personal observation) only the large P. dominulus actually utilized C. rubiginosa larvae as prey.
Until now, Vespidae have never been reported as important predators of shield beetles. Only Olmstead (1991) reported vespid wasps preying on shield-bearing larvae of Deloyala guttata Olivier. A literature survey by Cox (1996) lists Nabidae, Pentatomidae, Asilidae, Coccinellidae and Formicidae as field predators of Cassidinae. All these records, however, are based on anecdotal observations only. On several occasions, we also observed lacewing larvae preying on C. rubiginosa larvae in the field at different sites (personal observation). Several authors studied the defence of shield beetles against generalist predators in the laboratory (e.g. Eisner et al. 1967; Olmstead & Denno 1993; Müller & Hilker 1999; Eisner & Aneshansley 2000; Eisner & Eisner 2000). However, in all these studies the authors failed to establish the importance of their predators in the field. Our study is the first detailed analysis of predator–cassidinae interactions in the field. We believe that it is crucial to assess predation in a field setting in order to correctly interpret the importance of a defence mechanism.
In a previous study we found that predation is a serious mortality factor in late C. rubiginosa instars, while the early instars suffer mainly from abiotic mortality factors (S. Bacher & L. Kaufmann, unpublished). Under field conditions, the C. rubiginosa instars used in this study require about 5–8 days to reach the pupal stage (Ward & Pienkowski 1978; D. Schenk & S. Bacher, unpublished). After this time, larval densities were almost always reduced to only few individuals on the entire site (Fig. 2). Moreover, in the field C. rubiginosa pupates in the vegetation glued to leaves or stems of herbaceous plants. Therefore, pupae may also be prone to predation by wasps. Thus, the presence of P. dominulus may seriously reduce local densities of C. rubiginosa, and may even lead to the local extinction of beetle populations. In reducing beetle densities it also reduces the impact of C. rubiginosa larvae on its host plant creeping thistle (Bacher & Schwab 2000), and may annihilate the effect of augmentative releases of the beetle as biocontrol agent.
Handling times estimated by the model were longer than only the time needed to kill and eat a prey as revealed by the video analysis. Considering the travel time for the wasp from and to the colony and the time needed to pass the food to other colony members handling times of 10 min at site 1 and 4 min at site 2 as suggested by the model seem realistic. The difference between the handling times at both sites can most probably be attributed to the fact that at site 2 the wasp colonies were located adjacent to the experimental site, whereas at site 1 the location of the wasp colonies could not be discovered in the direct vicinity of the site and was probably farther away.
A likely mechanism how functional responses of Type III could be generated in generalists such as P. dominulus is that predators preferentially hunt for prey that become temporarily abundant. This kind of learning is common in generalist predators (e.g. Murdoch 1969) and will lead almost invariably to a Type III functional response. In the mark-and-recapture experiment we found no indication that the Type III functional response was the result of single wasps specializing on this particular prey type; we rather assume that a large fraction of the wasp population switched to the beetle as prey. Functional responses of Type II were often found in laboratory or field cage studies where only single prey species were offered. However, laboratory studies do not take into account alternative prey as it occurs in the field. Therefore, the prevalence of Type II functional responses found under restricted conditions may overestimate their occurrence and importance in nature. We recommend determination of functional responses of not strictly monophagous predators, most of which are attacking different prey species, in the field.
We thank Britta Tschanz and Ursula Wunder for their field assistance, Thomas Friedli and Tanja Steiner for statistical advice and the two anonymous referees for commenting on a previous version of this manuscript. This study was funded partly by the Swiss National Science Foundation through the National Centre of Competence in Research (NCCR) ‘Plant Survival’.