What limits predation rates by the specialist seed-feeder Penthobruchus germaini on an invasive shrub?


*Correspondence author. E-mail: rieks.vanklinken@csiro.au


  • 1Specialist seed-feeders are widely used in weed biological control, but seed predation rates are frequently insufficient to cause the required impacts. Understanding the underlying reasons is prerequisite to predicting efficacy.
  • 2We conducted continental-scale surveys of an introduced, multi-voltine seed-feeder [Bruchidae: Penthobruchus germaini (Pic.)] on an invasive legume (Caesalpinaceae: Parkinsonia aculeata L.). We tested three hypotheses as to what limits seed predation; namely, seed escape through egg aggregation, mortality of immature beetle stages, and failure to track temporal fluctuations in resource availability. We also tested how these factors interacted with the environment and each other.
  • 3Mean seed predation was relatively low (2–30%), despite mean egg densities of between 0·55 and 3·2 eggs per seed. Eggs were slightly aggregated (negative binomial, k = 1·87). Unexplained egg mortality (6–44%), egg parasitism (10–70%) and larval/pupal mortality (62%) were high, but egg parasitism was the only mortality factor that was density-dependent and that varied across climatic regions. Egg densities responded poorly to rapid within-season increases in seed availability.
  • 4All examined factors dampened seed predation rates. However, we developed a deterministic mathematical model which showed that seed predation would still have been relatively low (5–56%) at the observed egg densities, even without direct effects of immature beetle mortality. Also, to achieve a benchmark seed predation rate of 80%, egg densities would need to be over 8·5 eggs per seed with no parasitism, and an unrealistically high 27·8 at 70% parasitism. Available data suggest that seed predation by specialist seed-feeders will often be constrained by one or more of the factors we identified.
  • 5Synthesis and applications. Selecting effective biological control agents is an important challenge for weed biocontrol. Our results suggest that many seed-feeders will not regulate plant populations, but that predictions of their efficacy can be greatly improved by quantifying the way eggs are distributed across seeds and the various mortality factors that affect immature beetle stages. The ability of seed-feeders to track resource fluctuations may also be predictable. For example, multi-voltine insects appear better at tracking between-season and between-site variation in resource availability than sharp changes in within-season availability.


There is an ongoing debate as to what role the release from co-evolved natural enemies plays in allowing plant species to invade new habitats (Liu & Stiling 2006). Biological control proceeds under the assumption that the addition of specialized herbivores will result in a decline in the distribution and abundance of the invasive target plant (Myers & Bazely 2003). However, because there is a risk to non-target organisms with every agent released, choosing effective agents is of great importance (Sheppard et al. 2003; van Klinken & Raghu 2006). Therefore, predicting the effects of herbivores on plant populations has both theoretical and applied implications. In this study, we examine how factors that affect seed predation rates can be used to predict levels of seed predation in new environments.

Specialist seed-feeders can affect plant distribution and abundance (Crawley 1992; Louda & Potvin 1995; Ehrlen 1996; Maron & Gardner 2000) and influence community structure (Louda 1982; Rose, Louda & Rees 2005), and are frequently used as biological control agents (Julien & Griffiths 1998; Dennill et al. 1999; Myers & Bazely 2003). Some seed-feeders have had substantial impacts as biological control agents (Julien & Griffiths 1998), and they have even been advocated as companion species with new agro-forestry species to negate possible invasive tendencies (Hughes 1995; Zimmermann & Neser 1999). However, evidence from field studies and population modelling suggest that the required seed predation levels are typically high, often above 80% (Rees & Paynter 1997; Myers & Risling 2000; Parker 2000; Sheppard et al. 2002), and that this level of damage is rarely achieved by seed-feeding biological control agents (Julien & Griffiths 1998; Raghu, Wiltshire & Dhileepan 2005; van Klinken 2005) and seed-feeding insects more generally (Crawley 1992). Population regulation by seed-feeders may therefore be the exception rather than the rule.

Improving our ability to predict which seed-feeders, if any, will be effective biological control agents requires identification of the underlying factors that limit seed predation rates. This includes how the insects respond to environmental conditions, which may differ between the native and introduced invasive plant distributions (Raghu & van Klinken 2006). Various factors that limit seed predation rates have been identified in the literature, but they are rarely tested. In addition, their interaction with each other and the ways in which they are influenced by the environment have not been considered (Traveset 1991; Ehrlen 1996). In this study, we focus specifically on three factors that we believe may be of particular importance to seed predation rates: (i) spatial patterns in egg distribution across the seed population, (ii) mortality of immature stages of the seed-feeders, and (iii) the ability of seed-feeders to track within-season resource fluctuations.

The way seed-feeders distribute their eggs across seeds differs between species of seed-feeder. Eggs can be laid individually or in clusters, the latter typically occurs on seeds that can support the development of more than one seed-feeder (Janzen 1980). In addition, eggs or egg batches are typically aggregated across the seed population, although this is rarely quantified (Siemens & Johnson 1992). Both the clustering and aggregation of eggs is likely to influence seed predation rates.

Several mortality factors have been identified for specialist seed-feeders. Mortality factors include: eggs falling off the seed, failure to hatch, failure of larvae to penetrate the seed, larval competition within the seed, unexplained mortalities within the seed (Traveset 1991), and natural enemies. Natural enemies include egg parasitoids (van Klinken 2005), larval parasitoids (Janzen 1975; de Steven 1981; Wang & Kok 1986; van Klinken 2005) and pathogens (Janzen 1975). These mortality factors are expected to have a direct effect on seed predation rates and a regulating effect on seed-feeder populations (van Huis, Schutte & Sagnia 1998). However, specialized parasites and pathogens would be expected to be less diverse in the seed-feeder's introduced range (Torchin et al. 2003).

The ability of herbivores to track temporal fluctuation in resources (temporal tracking inertia, Solbrech & Sillen-Tullberg 1986) has been identified as an important constraint to seed predation. It is likely to be particularly important for specialist seed feeders whose phenology needs to be closely synchronised with the seed availability patterns of their host. Tracking within-season fluctuations in resource availability has received limited attention but can be an important determinant of seed predation rates (Westermann et al. 2003; van Klinken 2005; Raghu et al. 2005), and will be influenced by the stage (e.g. immature or mature seeds) and location (e.g. pre- or post-dispersal) of seeds that they oviposit on, and on insect and host phenology. Environment can also be important through its influence on when and for how long seeds remain available to seed-feeders.

In this study, we present the results of an extensive, continental-scale evaluation of a specialist seed-feeding bruchid beetle [Penthobruchus germaini (Pic.)] released in Australia as a biological control agent against an exotic leguminous shrub (Parkinsonia aculeata L.). Most bruchid species are monophagous or oligophagous seed-feeders (Jermy & Szentesi 2003), and they are therefore favoured as biological control agents (Julien & Griffiths 1998). Studies on P. germaini in one climatic region suggest the seed-feeder can be relatively abundant, but that seed predation rates are low (van Klinken 2005). We therefore use this system to test three main hypotheses for what limits seed predation levels: (i) seeds escaping predation as a result of egg aggregation; (ii) the direct effect of a range of mortality factors acting on immature beetle stages, including egg parasitism; and (iii) failure of seed-feeder populations to track fluctuations in seed availability, especially within seasons. In addition, we develop a mathematical model to specifically test the relative importance of observed levels of egg aggregation and mortality factors in determining seed predation rates. Study populations were located in contrasting climatic regions and environments. We expect each factor to contribute, but for their effect to vary according to the diverse climates and habitats in which the seed-feeder and its host co-occur across Australia (Russell & Louda 2005). We use our results, and those from related studies, to identify ways in which the efficacy of seed-feeders can be better predicted prior to their release.

Materials and methods

study system

Parkinsonia aculeata (Caesalpinaceae) is a perennial shrub or tree native to the Americas (Hawkins et al. 2007). Most seed pods are produced in early to late summer, depending on the region. They are indehiscent, typically 80 mm long and 8 mm wide and each contain from 1 to 10 seeds. Seeds are approximately 9 mm long and 4 mm wide, and have hard-seeded dormancy. Pods are shed through abscission, and seeds are subsequently released if pods decay or are damaged. Natural dispersal is probably primarily by water (van Klinken 2005). The shrub is invasive across climatically diverse regions in northern Australia (Parsons & Cuthbertson 1992).

Penthobruchus germaini (Coleoptera: Bruchidae) is native to Argentina. It was released in large numbers as a biological control agent in Australia between 1995 and 1999 after extensive host-specificity testing showed that it would only attack P. aculeata in Australia (G. Donnelly, unpublished data). It was widespread and abundant by early 1999 in three of the four study regions; namely, Central Queensland (Lockett, Gray & Donnelly 1999), the Victoria River District and the Barkly Tablelands (Lukitsch & Wilson 1999). No releases have been recorded in Central Australia, yet it was relatively abundant there when first surveyed in October 2001 (see Results).

In Australia, oviposition has only been reported on mature (or very nearly mature) pods and free, mature seeds (van Klinken 2005). Eggs are glued individually onto the surface of the pod or free seed and covered with a fine membrane. The beetle is multi-voltine, with the life cycle completed in 35–45 days at a constant 30°C (Briano, Cordo & DeLoach 2002). Eggs hatch within 8–9 days and first instar larvae tunnel down through the pod and then into the seed. Pupation occurs within the seed and adults emerge by cutting a hole through the seed coat and pod wall. Although more than one larva can enter a seed, only one adult ever emerges, resulting in mortality of the seed. Lifetime fecundity averages 348 eggs if females are provided with a honey and pollen food source (Briano et al. 2002).

Penthobruchus germaini eggs are commonly parasitized in Australia by a wasp (van Klinken 2005), identified as Uscana sp. (Trichogramatidae). It could not be identified further without revision of the group. Uscana is a genus of egg parasitoids that predominantly attack bruchid eggs. Either one or two individuals emerge per P. germaini egg. Where known, trichogramatid adults live for up to 8 days, females oviposit preferentially into eggs less than 3 days old, and development inside the host egg takes approximately 7–11 days (Kapila & Agarwal 1995).

Study sites were selected to represent the diverse climatic conditions and habitats wherein Parkinsonia aculeata (hereafter parkinsonia) and P. germaini co-occur in Australia. Four climate zones were identified and sites for regular sampling were placed between 0·3 and 38 km apart within a region in each zone (Table 1; Table S1, Supplementary material). Where parkinsonia distributions allowed, replicated sites were placed across the range of invaded habitats, as defined by timing and duration of inundation. Two additional regions were sampled in two of the climate zones to provide additional replication (Table 1). These were only sampled on one or two occasions, each coinciding with peak pod abundance, and data were only used in the sensitivity analysis.

Table 1.  Sampling strategy across the six surveyed regions. Further details on the regions are provided in Table S1, Supplementary material
RegionsHabitatsDistance between sitesSurveys
  • Habitat was defined by inundation regimes: uplands were never inundated; riparian sites were located on river banks and were subject to pulse inundation; wetlands were seasonally inundated for long durations (months); lowlands were periodically inundated for long durations. However, only wetland and Central Queensland lowland sites were inundated during the course of the study.

  • The range of distances is given between sites located in different habitats and sites serving as replicates (Reps) within a particular habitat type.

  • §

    Data published in van Klinken (2005).

  • All trees top-killed by frost in mid-2002, resulting in no pod production in 2002–2003.

Semi-humid wet-dry tropics
 Victoria River DistrictUplands (4 sites)Reps: 0·4 to 6 km apart2000–2001 (8 surveys, August–May)§
  2001–2002 (4 surveys, September–May)
  2002–2003 (6 surveys, October–Apr)
 East KimberleyWetlands (2 sites)Habitats: 0·4–6 km apart28 December 2000, 10 December 2002
Uplands (1 site)Reps:1·2–9·3 km apart 
Riparian (1 site)  
Semi-arid wet-dry tropics
 Barkly TablelandsUplands (1–2 sites)Habitats: 1–38 km apart2001–2002 (9 surveys, June–March)
Floodout (2–3 sites)Reps: 14–38 km apart2002–2003 (7 surveys, August–May)
 Central QueenslandUplands (2 sites)Habitats: 0·3–0·5 km apart2001–2002 (4 surveys, October–April)
Lowlands (2 sites)Reps: 15 km apart2002–2003 (9 surveys, September–July)
 Central Coastal QueenslandFloodout (3 sites)Reps: 0·6–1·2 km apart1 May 2004
 Central AustraliaRiparian (3 sites)Reps: 0·6–1·8 km apart2001–2002 (6 surveys, December–July, Aug 2003)

data collection

Sites were visited at 4- to 6-week intervals between September 2000 and July 2003, depending on the region (Table 1). Data on pod fall, oviposition and seed predation by P. germaini and egg parasitism were collected on each visit following the method of van Klinken (2005).

Seed phenology

At each site, gauze litter traps (diameter 58 cm) were placed under each of 10 healthy, mature trees (five in the Victoria River District in 2000–2001), above the flood-line, approximately midway between the main stem and the edge of the canopy. Trees were selected with limited or no canopy overlap. Litter traps were emptied on each visit, and the number of pods and seeds therein recorded to give an estimate of seed rain through time for each pod generation (pods produced within a single season). Pods that had remained on the tree from the prior season, and were therefore from a previous pod generation, were easily differentiated from new pods by their higher levels of mould and decay. For the purpose of data analysis (see below), the day of maximum pod fall was estimated from litter trap data from each region as 25 November in the Victoria River District, 18 December in Barkly Tablelands, 31 December in Central Queensland, and 7 February in Central Australia.

Egg density, egg fate and seed mortality

Beetle egg density, and the condition of eggs and seeds, at the time of sampling were determined on each visit by collecting pods from each site and freezing them within 48 h. A minimum of 40 pods from each pod generation at each site were collected into paper bags both from the tree itself and the ground under the trees. A maximum of 10 pods were collected from any one tree.

Pods and seeds were examined with a microscope. Pods have swellings along their length, each housing individual seeds. For each seed-swelling on each pod the number of P. germaini eggs was recorded. P. germaini eggs were classified as ‘intact’ or ‘not intact’ (eggs that had fallen off but were recognizable by glue remains, had collapsed, were mouldy or had been damaged by unknown causes). Intact eggs were classified as unhatched, hatched or parasitized.

The seed-swelling was then dissected and the seed within categorised with the naked eye as viable, intact but unviable (seed coat no longer hard and shiny and the germplasm no longer healthy), or consumed. If consumed, then the consumer was identified as one of three possibilities: (i) P. germaini or its larval parasitoid, on the basis of the size and shape of the emergence hole; (ii) ‘lepidopteran’ on the basis of feeding damage, frass and webbing; and (iii) ‘unidentified’ if a clear diagnosis was not possible.

data analysis

Statistical analyses were performed in s-plus 6·2 (November 2003, Insightful Corp., Seattle, WA, USA) unless otherwise stated.


The number of eggs per seed was modelled by a negative binomial distribution, after inspection of the relationship between mean and variance, with mean dependent on site, time of collection and pod location and with a constant dispersion parameter:

Etot = NegBin(log(svij + lk),θ)(eqn 1)

where Etot = number of eggs per seed, svij = effect of site i at survey j, lk = effect of pod location k (tree or ground) and θ = dispersion parameter for the negative binomial. The negative binomial was parameterized so that var(E) = µ + µ2/θ (McCullagh & Nelder 1989).

Egg parasitism

A generalized additive model (Wood 2006) was fitted in r version 2·50 (using the MGVC package) (23 April 2007; r Foundation for Statistical Computing, Vienna, Austria) to the proportion of eggs that were parasitized to identify explanatory variables for egg parasitism rates, including beetle egg density and the time of year in relation to seed availability. The model used was:

image(eqn 2)

where p = the expected proportion of eggs parasitized with the number of eggs parasitized distributed as an over-dispersed binomial (modelled by a scale parameter that captures the tendency of a wasp to parasitize groups of eggs), rq = effect of qth combination of region and pod generation (combined because there was insufficient data to separate effects of region and pod generation), si = effect of site i, lk = effect of pod location k (tree or ground), and f and g are non-linear smoothers estimated as penalised regression splines to model the effect of number of eggs per seed (E) and time (in days) from maximum pod fall (t). g was estimated for each combination of region and pod generation (q). Only a single smoother was fitted for egg density (E). Fitting individual smoothers for each region–year combination did improve the model slightly, but created significant computational difficulties and led to identifiability problems. If individual region–year smoothers were fitted to egg density it was no longer possible to clearly distinguish the effects of egg density. Parameter values were obtained by maximizing the quasi-likelihoods. The degree of smoothing was chosen by generalized cross-validation.

Only seeds collected between 100 days before to 300 days after the estimated time of maximum pod fall were used in the analysis. Outside this range, there was too little data to give meaningful confidence intervals.

Egg hatch rates and egg hatch to seed predation rates

The probability of an intact beetle egg hatching and of seeds with a single hatched egg being consumed for each region and pod generation was estimated using generalized linear mixed models with site as a random effect. Egg hatch rates were estimated using data from pods on the ground collected during the first survey after the time of greatest pod fall, by which time most eggs would be expected to have hatched (as most ovipositions occur on the tree, see Results). Seed predation rates were estimated using pods on the ground collected at least 40 days after the period of maximum pod drop so as to allow the hatched larvae sufficient time to consume the seeds. Effects of larval competition could not be determined as there were insufficient seeds with more than one hatched egg, and they were therefore excluded from the analysis.

The relative importance of predation-limiting factors

The sensitivity of seed predation rates to oviposition patterns and egg and within-seed mortality (including egg parasitism) was modelled, using parameter values and ranges that captured the variation observed in the data analysis (see Results). The probability that seeds would have at least one egg (equation 3, below), one hatched egg (equation 4, below) or be consumed (equation 5, below) was modelled as a function of eggs per seed. Survey data were used to derive parameters for egg dispersion (equation 1), mean egg densities (0–20 eggs per seed), the probabilities of egg hatch for unparasitized eggs (90%) and of larvae consuming the seed (38%), and to produce a realistic range of egg parasitism rates (0–70%).

The proportion of seeds with eggs (pl) assumed both random oviposition (Poisson distribution) and non-random oviposition (negative binomial distribution). Under the negative binomial, the probability of a seed having at least one egg was estimated as:

image(eqn 3)

where θ is the dispersion parameter from equation 1 and µ is the mean number of eggs per seed (0 to 20).

The probability of there being at least one hatched egg on a seed (ph) assumed negative binomial egg counts and a 90% chance of unparasitized eggs hatching (η) and was estimated as:

image(eqn 4)

The probability of a seed being consumed (pi) assumed negative binomial egg counts, a 90% chance of egg hatch, 38% probability of larvae consuming the seed (a), and a range of egg parasitism rates (0–70%) (ν) and was estimated as:

image(eqn 5)

Equation 5 was subsequently used to estimate what seed predation rates would have been in each region and year if there had been no direct effects of mortality to immature beetle stages.


egg distribution across seed populations

There was an average of 1·6 beetle eggs per seed, with 57% of seeds having at least one egg, 1·5% of seeds having more than 10 eggs and 0·2% of seeds having more than 20 eggs. The distribution of eggs across seeds was similar across regions and years, and was best described with a negative binomial with a dispersion factor (θ) of 1·87 (SE 0·04) (Table S2, Supplementary material). Variances were similar to means at low egg densities, and there was therefore no evidence of the beetle laying multiple eggs on a single seed visit.

ability to track fluctuating resources

Pods contained between one and seven seeds (average 1·6). Pod fall was relatively synchronous within each region, but timing varied with region by up to a few months (Fig. 1). There was considerable variation in seed densities between regions and between years within regions (Fig. 1).

Figure 1.

Cumulative mean seed drop and mean egg density on seeds on the tree (closed symbol) and ground (open symbol) given for each pod generation, in the Victoria River District (a, b; data for three-pod generations), Central Queensland (c, d; two-pod generations), the Barkly Tablelands (e, f; two-pod generations), and Central Australia (g, h; one-pod generation).

Egg density differed between regions (0·55–3·2 eggs/seed) (Table 2). Variation in egg density between sites was too high, and replication too low, to test for habitat effects in the Barkly Tablelands (uplands vs floodout) and Central Queensland (uplands vs. lowlands), but no consistent patterns in the data were apparent (data not shown). Data on between-season variation in egg density were limited to three regions (Table 2, Fig. 1). The greatest between-season changes were observed in the Victoria River District, where both the total annual seed production and mean annual egg densities dropped substantially over the 3 years.

Table 2.  Mean egg density (eggs per mature seed), total number of eggs and mean egg parasitism rates (proportion of total eggs that were parasitized; pooled across sites and surveys) on pods collected from the tree and off the ground. Adjusted mean egg parasitism rates were estimated by excluding site effects (s), and setting time of season (t) at 40 days after peak pod fall and beetle egg density (E) at 1·6 (equation 2)
Region SeasonPods on treesPods on ground
Egg densityn (eggs)Egg parasitism meanEgg densityn (eggs)Egg parasitism mean
Victoria River District
 2001–021·00118667·0%73·9%0·20 12670·6%69·2%
 2002–030·55 47453·0%43·6%0·42243163·3%38·1%
Barkly Tablelands
 2002–031·98219814·6% 9·4%1·78250520·0% 7·6%
Central Queensland
Central Australia
 2001–021·571573 9·6%16·7%1·27149912·1%13·7%

The within-season pattern of egg density generally tracked those for pod drop, with egg densities decreasing one to two surveys prior to pod drop, before increasing again through the following few surveys (Fig. 1). Central Australia and 2002–2003 in the Barkly Tablelands were exceptions, with little change in egg density after the main pod drop.

Most ovipositions occurred on pods on the tree as egg densities on the trees were similar to, or higher than, those on the ground (Fig. 1). Also, changes in egg density on ground pods often lagged behind those on tree pods by approximately one survey, suggesting that egg densities on ground pods was being determined by the oviposition dynamics on tree pods.

egg to adult mortality

Egg parasitism rates

Mean egg parasitism was similar on pods collected from the tree and the ground (Table 2). However, parasitism levels varied considerably with region, and was lowest in Central Australia and highest in the Victoria River District and Central Queensland. There were also considerable differences between seasons, particularly in the Barkly Tablelands. Raw means were, however, confounded by a range of factors included in the multiple regression (equation 2), all of which were highly significant (Table 3; Spearman's rank correlation between predicted and observed proportions = 0·46). Egg parasitism rates, once adjusted for egg density and time of survey effects, were qualitatively similar to raw means (Table 2), and were generally highest in the Victoria River District and lowest in Central Australia.

Table 3.  Analysis of deviance for proportion of parasitism (equation 2)
VariableDegrees of freedomDevianceProbability
Region and pod generation (r)    4  280< 0·0001
Site (s)   18 2072< 0·0001
Location (l)    1   51< 0·0001
f (egg density)    7·5 1240< 0·0001
g (time from maximum pod fall)   34·4 1099< 0·0001

There was a strong positive relationship between parasitism rates and beetle egg density, and it was linear at beetle egg densities typically encountered in the field (Fig. S1, Supplementary material; Table S2). For example, the estimated proportion of eggs parasitized (as estimated from equation 2) at a typical site increased from 28% at a mean egg density of one egg/seed to 57% at a mean egg density of five eggs/seed (Fig. S1). Egg parasitism rates were also related to time of podding season but patterns differed between regions and pod generations (Fig. S2; Table S3, Supplementary material). However, variation might be an artefact of relatively low sampling frequencies.

Egg hatch rates for unparasitized eggs

Over 90% of eggs were intact (Table S4, Supplementary material). Egg hatch rates for intact eggs that had not been parasitized averaged over 85%, but varied considerably with region (Table S4), and also between years and surveys (data not shown). Most notable were relatively low egg hatch rates in the Victoria River District, although it was not statistically different when comparing samples taken soon after peak pod drop, when most eggs would have had an opportunity to hatch (Table S4).

Seed mortality

Only Penthobruchus germaini and unidentified Lepidoptera (probably Mesophleps palpigera (Walsingham), van Klinken 2005) were recorded from seeds. P. germaini caused the greatest seed predation, although it was never higher than 50·2% in any one survey and pod location (Table 4). Larval parasitoids of P. germaini were present and, where identified, were Dinarmus simus (Girault) (Pteromalidae). However, larval parasitism rates were not recorded as emergence holes could not be consistently differentiated from those of P. germaini. A substantial proportion of seeds were unviable for reasons other than predation, including imbibition and subsequent decay within the pod (Table 4).

Table 4.  Seed fates (mean), pooled for survey and pod location. The maximum predation rate by Penthobruchus germaini for a particular survey and pod location is given in brackets
Region Seasonn (seeds)ViableUnviable
P. germainiLepidopteraUnknown
Victoria River District
 2000–2001495369·0%19·9% 5·1% (9·6%, n = 418)5·8%0·2%
 2001–2002182790·4% 6·9% 1·5% (3·3%, n = 36)1·0%0·2%
 2002–2003187486·4% 7·5% 5·3% (22%, n = 82)0·1%0·1%
Barkly Tablelands
 2001–2002487176·0% 6·4%17·5% (50·2%, n = 231)0·2%0·0%
 2002–2003251680·0% 9·8%10·9% (27·5%, n = 273)0·0%0·1%
Central Queensland
 2001–2002125977·2% 9·5% 8·6% (25·4%, n = 126)0·7%0·0%
 2002–2003418776·1% 8·8%15·0% (37·3%, n = 330)0·1%0·1%
Central Australia
 2001–2002218251·8%18·2%29·9% (50·2%, n = 251)0·0%0·1%

Egg hatch to seed predation rates

The estimated probability of a seed with a single hatched egg resulting in adult emergence (either P. germaini or its larval parasitoids) was 38% (95% confidence interval of 30–48%). There were no statistically significant differences between regions (F = 1·45; d.f. = 2,586; P = 0·236).

relative importance of predation-limiting factors

The relationship between average egg density and seed predation was explored in the absence of parasitism and with parasitism using the empirical relationships determined above (Fig. 2). The bruchid egg density required for 80% of seeds to have at least one egg was higher (2·5 eggs) as a result of their negative binomial distribution than if they had been randomly distributed (1·6 eggs) (Fig. 2). A 90% probability that eggs would be intact and hatch, which was the maximum observed in any region (Table S3), increased the required egg density slightly to 2·8 eggs per seed. We were not able to determine the relationship between multiple-egg hatch events on a single seed and the probability of seed predation. However, even assuming that there was no intra-specific competition within the seed, an egg density of 8·5 would be required to attain seed predation rates of greater than 80% (Fig. 2).

Figure 2.

The modelled relationship between beetle egg density and the proportion of seeds that was predated. Eggs were assumed to have a negative binomial distribution (θ = 1·873), unless otherwise stated. Egg hatch rates are for unparasitized eggs. Egg parasitism rates, where modelled, are indicated on the graph (0–70%). Predation estimates following egg hatch conservatively assumed no intra-specific competition.

The probability of seed predation at a given egg density was highly sensitive to egg parasitism (Fig. 2) at the levels encountered in the field (Table 2), even when using conservative egg hatch rates (Table S4) and assuming no intra-specific competition. To achieve a seed predation rate of 80%, egg densities would need to be over 11·9 at 30% parasitism and 27·8 at 70% parasitism (Fig. 2).

the effect of mortalities of immature beetle stages on seed predation

Predictions of predation rates based on empirical data were similar to observed values in late season surveys when most adults were expected to have emerged (Table 5). Seed predation was predicted to have been between 1·9- and 10-fold higher if there had been no egg to adult mortality. However, even in the absence of any direct effects of egg to adult mortality, our model predicts that seed predation rates in all regions and years would have been less than 60% at the field-recorded egg densities, and less than 36% once mortalities other than egg parasitism were included (Table 5). Egg parasitism caused the greatest proportional reduction in seed predation in the semi-humid wet-dry tropics, but had little direct effect in arid Central Australia.

Table 5.  Empirical data from seeds collected on the first survey conducted after 40 days since peak pod fall, and the resulting predictions of seed predation rates with and without mortality of immature beetle stages
Region SeasonEmpirical dataSeeds consumed (predicted)
n (seeds)Egg densityEgg parasitismEgg hatch (unparasitized)Seeds consumedAssume all mortality factorsAssume no egg parasitismAssume no mortality
  • Predictions assumed a negative binomial egg distribution and 38% chance of hatched larvae resulting in seed predation (as for Fig. 3, see text) and empirical data on egg density, egg parasitism and egg hatch (data shown).

Semi-humid wet-dry tropics
 Victoria River District
 2001–014400·5169·8%100% 4·3% 4·4%11·1%20·1%
 2001–022560·1765·1% 94·7% 2·0% 2·0% 3·8% 7·3%
 East Kimberley
 28-Dec-003060·2252·9% 55·2% 2·6% 1·8% 3·5%10·6%
 10-Dec-023950·0983·8%100% 0·3% 0·5% 2·5% 5·0%
Semi-arid wet-dry tropics
 Barkly Tablelands
 2001–023701·8738·2% 79·4%17·6%21·0%28·8%53·4%
 Central Queensland
 2001–023611·8850·3% 61·8% 5·3%14·4%23·4%49·1%
 2002–031731·2632·1% 93·6%22·0%19·0%24·7%43·9%
 Central Coastal Queensland
 1-May-04 331·7145·2% 95·1%17·7%20·5%30·3%51·6%
 Central Australia
 2001–025141·4410·5% 94·9%28·4%25·5%27·3%47·9%


Seed predation rates by Penthobruchus germaini varied considerably across Australia and between years, but even the highest predation rates were similar to other seed mortality factors within the pod. Mean predation rates were less than 30% in all regions and years, despite egg densities averaging almost two eggs per seed in some regions. Modelling of the demographic effects of the observed seed predation rates was beyond the scope of this study, but effects would be expected to be limited, based on results from other studies (see Introduction). Relatively low predation has also been observed for all other bruchid species that have been released and evaluated as biological control agents (Julien & Griffiths 1998; Impson, Moran & Hoffmann 1999; Radford, Nicholas & Brown 2001; Paynter 2005; Raghu et al. 2005). The only exception we could find was the uni-voltine bruchid accidentally released on scotch broom (Cytisus scoparius (L.) Link) in north-west USA, which destroyed over 80% of the seeds (Redmon, Forrest & Markin 2000).

We draw three main conclusions regarding factors limiting seed predation rates: (i) the ability of seed-feeders to regulate host populations can be constrained by even modest aggregation of eggs and commonly recorded levels of insect mortality on or in the seed; (ii) seed-feeders, especially multi-voltine species, commonly fail to track within-season fluctuations in resource; and (iii) with the possible exception of egg parasitism, the effects of these factors on seed predation rates are likely to be predictable prior to the release of seed-feeders into new environments. These conclusions have important implications for the use of specialist seed-feeders to regulate invasive plants.

the effect of oviposition pattern and beetle mortality on seed predation

Egg distributions within seed populations were consistently aggregated across northern Australia, and across surveys and years. The observed egg distribution suggested that eggs were laid singly and the number of visits by ovipositing females per seed was random (because the mean was similar to the variance at low egg densities), but that the seeds differed in attractiveness to the egg-layer, and/or seeds differed in their exposure times to the egg-layer. Egg aggregation was relatively slight compared with some other species, but it still resulted in a considerable increase in the egg densities required to achieve high egg coverage compared to if eggs had been randomly distributed. The importance of aggregated distributions and clumping on discrete hosts has received considerable attention as a potential mechanism for avoiding inter-specific competition (Rohlfs & Hoffmeister 2003), but in the case of seed-feeders, and from the perspective of the host, its effect is reduced seed predation.

Egg hatch rates for unparasitized eggs did not differ statistically between regions, but was lowest when egg parasitism was the highest, suggesting that some egg mortality may be the result of failed parasitism. Field egg hatch rates compare well with those recorded in a laboratory study in Argentina where adults were fed on an ideal diet of honey and pollen (c. 92·1%, n = 4173; Briano et al. 2002). Larval and/or pupal mortalities (c. 62%) were surprisingly constant across the surveyed regions and years. Furthermore, substantial unexplained egg, larval and pupal mortalities appears to be common among seed-feeders. Failure of eggs to hatch is commonly around 20% (de Steven 1981; Siemens & Johnson 1992; Sagnia 1994), although it can be as high as 50% (Traveset 1991). This does not include eggs that fall off the seed, which can be high (Traveset 1991), but is probably insignificant for P. germaini which glues its eggs firmly to the seed. Unexplained larval or pupal mortalities are frequently between 30 and 60% (Janzen 1977; de Steven 1981; Traveset 1991; Sagnia 1994). These widely reported low egg hatch rates and high unexplained larval mortalities suggest that egg densities for many seed-feeding species would need to be over eight per seed (assuming an egg distribution as for P. germaini) to result in predation of at least 80% of seeds. The required egg density would increase considerably with increasingly aggregated egg distributions, and for seed-feeders that require more than one individual to consume a seed. The unexplained mortality of immature beetle stages can therefore seriously limit the proportion of seeds that many seed-feeder populations will consume.

Egg parasitism was the only mortality factor detected in our study that was dependent on egg density. A very similar result was found for the egg parasitoid Uscana semifumipennis Girault on the seed-feeding bruchid Stator limbatus (Horn) (Siemens & Johnson 1992). Strong density dependence of egg parasitoids will act to dampen any increase in larval emergence, and hence seed predation, resulting from increases in egg density, even at the relatively low egg densities observed in our study. Indeed, in our study system, it means that unrealistic mean egg densities are required to achieve 80% seed predation. High egg parasitism will also reduce the ability of P. germaini populations to track seed populations (van Huis et al. 1998). Egg parasitoids are rarely considered in seed predator studies. However, high parasitism rates have been recorded for other species (Traveset 1991; Siemens & Johnson 1992; Sagnia 1994; Coetzer & Hoffmann 1997), if not universally so (Wang & Kok 1986).

ability to track fluctuating resources

Within-season variation in egg density was strongly coupled to fluctuations in seed availability, highlighting the importance of understanding within-season resource dynamics, and the ability of seed predators to track those resources, when determining seed-feeder impacts (Solbrech & Sillen-Tullberg 1986; van Klinken 2005; Raghu et al. 2005). Seed dynamics on trees is particularly important as P. germaini beetles showed a strong oviposition preference for pods on the tree (Fig. 1; van Klinken 2005). Maturation of parkinsonia pods was relatively synchronous at any one site, occurred approximately 6 weeks prior to pod drop (van Klinken 2005, unpublished data) and was consistently correlated with a drop in egg density, suggesting predator satiation. Egg densities therefore failed to peak at times when most seeds were available to seed-feeders, thereby limiting total annual seed loss to predation (van Klinken 2005). A similar outcome has been observed where the host of multi-voltine seed-feeding species are consumed by vertebrate herbivores soon after maturation (Impson et al. 1999; Baes, de Viana & Saravia 2001; Radford et al. 2001). Multi-voltine seed-feeders therefore appear poorly equipped to effectively track sharp within-season fluctuations, which prevents high seed predation of the annual seed crop. However, multi-voltine seed-feeders may be better at tracking between-year and between-site variation in the timing of resource fluctuations; in our study, variation in the seasonal timing of pod maturation with climate zone had no apparent effect on the ability of P. germaini to track fluctuations in seed availability. In contrast, synchronisation between uni-voltine insects and their host allows them to exploit sharp within-season peaks in resource availability, but they are often less able to deal with between-year and between-site variations (Redmon et al. 2000; Russell & Louda 2005). Indeed, all the examples that we could find of seed-feeding biological control agents that regulated host populations were uni-voltine (e.g. Hoffmann & Moran 1998; Dennill et al. 1999; Redmon et al. 2000).


To be an effective biological control agent, a seed predator needs to reach sufficiently high egg densities at the right time of the year so as to predate the greatest proportion of the annual seed crop (van Klinken 2005). Our modelling showed that the egg density that is required to achieve a target seed predation rate is determined by the multiplicative effect of the way in which eggs are distributed across seeds, and a wide range of mortality factors of immature beetle life stages. In our study, egg parasitism and unexplained larval/pupal mortality were the most important factors, but this will vary with the study system. However, even without high immature beetle mortality, an inability to track within-season fluctuations in seed availability meant seed predation would still have been quite low at a time when most seeds were available (5–56%). With the exception of egg parasitism, each of the factors we considered that limited seed predation rates acted independently of environmental conditions and their effects in the target environment should therefore be predictable. If this is commonplace, then it should simplify the selection of effective seed-feeding biological control agents.

The inability of seed predators to track fluctuations in seed resources through time was the key factor limiting seed predation in our study system. Strict oviposition preferences, such as for seeds on trees, will have the effect of increasing within-season resource fluctuations. Our results suggest that multi-voltine insects will become less effective the greater the fluctuations in within-season resource are, and that the converse might apply to uni-voltine seed-feeders. The relative efficacy of multi-voltine and uni-voltine seed-feeders may therefore be predictable by considering oviposition preference for seed age and position, and the spatial and temporal fluctuations in resource availability within and between seasons.

The ability of seed predator populations to track fluctuating resources will be dampened by the population effects of both the way eggs are distributed across seeds, and mortality of immature life stages. Although we did not consider their population effects directly, the impact of egg parasitism is likely to be particularly important as it was the only density-dependent mortality factor. Although parasitism is often lower in the introduced range because of a lack of specialist natural enemies (Torchin et al. 2003), this is clearly not always the case. Also, predicting parasitism rates is more difficult than simply predicting parasitoid diversity (van Klinken & Burwell 2005). Nonetheless, predictions should improve as our knowledge of the parasitoid fauna, and their abundance, in the target distribution of potential biological control agents improves. For example, any new seed-feeders being considered as biological control agents for invasive plants in arid and semi-arid Australia should possess mechanisms to avoid known egg parasitoid risks.

Overall, given the multiple inter-relating factors limiting seed predation rates, it is not surprising that seed-feeders frequently do not reach the densities required to regulate host populations. This suggests caution when prioritizing seed-feeders for biological control, and when advocating their use as companion species in agro-forestry. It does, however, suggest that detailed pre-release studies which quantify factors that can limit herbivory will be useful for identifying the most efficacious species for biological control, especially where the interactions between herbivore, environment and plant damage are well understood (Raghu & van Klinken 2006).


We thank: Jim Begley, Bert Lukitsch, John Peart, John Gavin (NT DBIRD), Mike Pattison, John McKenzie (QNRME), Noel Wilson (DAWA), Allan Tompson (CALM), for assistance in the field; Tracee Withers and Celine Clech-Goods for assistance in the laboratory; Chris Burwell (Queensland Museum) for identifying wasps; Theo Evans, Nancy Schellhorn, Shon Schooler and Graham Hastwell for comments on an earlier draft; and the Natural Heritage Trust for funding.