East meets west: adaptive evolution of an insect introduced for biological control


Correspondence author. E-mail: craig.phillips@agresearch.co.nz


  • 1A possible explanation for low success rates when introducing natural enemies to new regions for biological control of insect pests is that they fail to adapt to their new conditions. Therefore it has been widely recommended that biological control practitioners increase the probability of local adaptation by maximizing the genetic variation released. An alternative recommendation is to use climate matching to identify native populations that may already possess traits suited to the new region. However, support for these recommendations is weak through lack of empirical evidence that local adaptation is important to biological control.
  • 2This study examined how genetic drift and selection influenced the population frequencies of two asexually reproducing, genetically differentiated parasitoid biotypes that were introduced to New Zealand from South America for biological control. Other than by mutation, the biotypes were genetically fixed due to the absence of recombination both within and between biotypes. This meant that adaptive evolution could occur only if selection acted on any traits that varied between the biotypes introduced from South America.
  • 3The two parasitoid biotypes were released simultaneously at 14 sites and their frequencies were monitored for up to 10 years. Changes in biotype frequency were consistent with strong directional selection favouring one of the South American biotypes, thus generating established parasitoid populations that were better adapted to New Zealand conditions than those that had originally been released. This local adaptation of the control agent contributed to greater mortality of the pest.
  • 4Synthesis and applications. This study provides the first clear demonstration of the importance of releasing natural enemy genetic variation in new regions to foster adaptive evolution and improve success rates in classical biological control. However, the benefit to biological control of maximizing the genetic variation released needs to be balanced against possible risks to non-target species. The results do not support the concept of choosing sampling sites for putative biological control agents based solely on climatic similarities between the source location and the intended region of introduction.


Most introductions of species to new regions, either intentional or inadvertent, fail to establish self-sustaining populations (Williamson & Fitter 1996). Even in classical biological control of insects, where natural enemies are intentionally introduced to help suppress insect pests, only ≈30% establish (Hall & Ehler 1979; Williamson & Fitter 1996) and ≈20% provide some degree of control (Hall, Ehler & Bisabri-Ershadi 1980; Hokkanen 1985). Such low success rates have prompted pleas for greater adoption of experimental approaches to help biological control become a more predictive science (Roush 1990; Kareiva 1996; Ehler 1997; Roderick & Navajas 2003).

Establishment probabilities appear to increase with the number of founding individuals (Hopper & Roush 1993; Grevstad 1999; Berggren 2001; Memmott et al. 2005), possibly because larger populations should suffer less from Allee effects, have lower chances of extinction due to demographic and environmental stochasticity, and possess greater potential to adapt to local conditions. Adaptive potential should increase with founder population size, as large populations are expected to harbour greater genetic variation, have higher probabilities of producing beneficial mutations, and be less susceptible both to inbreeding depression and to the randomising effects of genetic drift on natural selection (Reznick & Ghalambor 2001; Reed et al. 2003; Kawecki & Ebert 2004).

It has been recommended that biological control practitioners increase the probability of local adaptation by maximizing the genetic diversity obtained from the native range, minimizing inbreeding and selection for laboratory-fit populations during prerelease culturing, and releasing as much genetic variation as possible into the new region (Simmonds 1963; Force 1967; Remington 1968; Lucas 1969; Mackauer 1972; Delpuech, Carton & Roush 1993; Reed et al. 2003). Another related recommendation has been to use climate matching to identify native populations that may already possess adaptations suited to the new region (McDonald 1976), thereby reducing any need for adaptive evolution following introduction. Although anecdotal observations have supported these recommendations, robust empirical evidence for the importance of intraspecific genetic variation and local adaptation to biological control is still lacking (Mackauer 1972; Roush 1990; Hopper, Roush & Powell 1993; Kareiva 1996; Ehler 1997; Hufbauer 2002; Louda et al. 2003; Roderick & Navajas 2003; Hufbauer & Roderick 2005). Moreover, the strategy of maximizing the genetic diversity released conflicts with concerns that released populations could adapt to attack new, non-target hosts (Simberloff & Stiling 1996; Louda et al. 2003). Clearly, any benefits and risks associated with genetic variation in introduced populations cannot be elucidated while uncertainty remains as to whether evolution even plays a role in biological control.

This study capitalized on a biological control introduction of two asexually reproducing, genetically differentiated biotypes of a South American parasitoid wasp to compare the genotype frequencies that were released throughout New Zealand with those that became established subsequently. The absence of recombination both within and between biotypes greatly simplified the task of measuring evolution and identifying the role of natural selection because, aside from mutations, the biotypes remained genetically fixed following their release. This meant the only traits that were exposed to selection were those that had varied between the biotypes when they were originally introduced from South America. The null hypothesis investigated was that differences between biotype frequencies in the released populations and those subsequently observed at each site would be the random result of genetic drift. Selection and local adaptation would be evidenced by non-random changes in biotype frequency that could not be attributed to genetic drift.

A morphometric analysis of wasps collected at five New Zealand sites from 1992 to 1994 found that frequencies of one biotype exceeded expected levels at all sites (Phillips, Baird & Goldson 1997). The present study provides a more comprehensive and powerful test of the null hypothesis by: (i) monitoring biotype frequencies at 14 sites for up to 10 years; (ii) correcting a partly erroneous interpretation of the morphometric variation; (iii) assigning specimens to biotype more accurately; (iv) comparing the biotype frequencies observed at each site with those originally released at that site, rather than comparing with a single, average frequency over all sites; and (v) applying more rigour to the data analysis, including an assessment of the effects of released population size.


the study system

The South American weevil Listronotus bonariensis (Kuschel) (Coleoptera: Curculionidae) was first recorded in New Zealand in 1927 (Marshall 1937) and is now ubiquitous. Population densities of 50–400 L. bonariensis adults m−2 typically occur in pastures (McNeill et al. 2002; Barker & Addison 2006) that cover over 10 million ha (≈40%) of the country (MAF 1997). New Zealand L. bonariensis populations are genetically homogeneous and exhibit less variation than South American populations, possibly due to a bottleneck during colonization (Williams et al. 1994).

In 1991, the South American parasitoid wasp Microctonus hyperodae Loan (Hymenoptera: Braconidae) was introduced for biological control of L. bonariensis (McNeill et al. 2002). It lays one egg per adult L. bonariensis (its only known host), the larva develops within the living, active weevil then emerges to pupate, while the host dies due to the parasitism (Loan & Lloyd 1974). Microctonus hyperodae reproduces by apomictic parthenogenesis (Loan & Lloyd 1974; Iline & Phillips 2004). No males of M. hyperodae have been observed in South America (Loan & Lloyd 1974), nor have any been seen during >10 years of mass rearing and field monitoring in New Zealand (McNeill et al. 2002). Microctonus hyperodae females were obtained from eight South American sites (Fig. 1) and their descendants were cultured separately to facilitate the experiment described here.

Figure 1.

Maps showing sites where Microctonus hyperodae was collected in southern South America, and New Zealand sites where it was subsequently released, then sampled for this study.

The M. hyperodae genetic variation obtained from South America has been assessed using morphometric (Phillips & Baird 1996), protein electrophoresis (Iline & Phillips 2004), and PCR-based (Winder et al. 2005) methods involving assays of 81, 69 and 82%, respectively, of the 141 asexual female lineages released in New Zealand. Morphometric variation occurred in the number of antennal segments, the size of the radial cell, stigma and basitarsus 3, and the shape of the petiole (Phillips & Baird 1996). Protein variation (Iline & Phillips 2004) occurred at loci for malate dehydrogenase (MDH), dihydrolipoamide dehydrogenase (DDH2), and a calcium-binding protein (CBP9), while nucleotide sequence variation occurred in microsatellite and mitochondrial 16S gene regions (Winder et al. 2005).

Based on morphometric data, Phillips & Baird (1996) suggested that the eight South American geographical populations comprised three biotypes with one originating from west of the Andes (Chile), another from east of the Andes (Argentina, Uruguay and Brazil) and a third, intermediate between the first two biotypes, from within the Argentinean Andes at San Carlos de Bariloche (Fig. 1). Data from the protein (Iline & Phillips 2004) and genomic (Winder et al. 2005) studies corroborated the existence of the eastern and western biotypes, but showed that the San Carlos de Bariloche population was a mixture of the eastern and western biotypes (3 E : 5 W), rather than a third biotype. Moreover, the protein and genomic results indicated that one of seven female lineages assayed from Concepcion, Chile belonged to the eastern biotype (Iline & Phillips 2004; Winder et al. 2005). The degree of variation between the M. hyperodae biotypes was consistent with intra- rather than interspecific variation, and was much less than the variation found within either of the two major, host-associated clades of the congeneric parasitoid Microctonus aethiopoides Loan (Vink et al. 2003).

Following an initial release in New Zealand at Lincoln (Fig. 1), M. hyperodae dispersed at only 2 km year−1, so a mass rearing and distribution programme was conducted to assist its spread (McNeill et al. 2002). By 1998, M. hyperodae had become established at 112 release sites, where rates of L. bonariensis parasitism often exceeded 60% within 3 years of release (McNeill et al. 2002; Barker & Addison 2006).

data collection

Microctonus hyperodae was obtained from 14 New Zealand release sites (Fig. 1) for 1–10 years following each release (Table 1). The released populations comprised between 1981 and 18 146 individuals (Table 1) and contained eastern biotype frequencies that ranged from 0·63 to 0·79 (Table 1). Study site locations (Fig. 1), the low dispersal rate of M. hyperodae (McNeill et al. 2002), and periodic surveys for M. hyperodae between sites (data not presented) indicated that there was no significant dispersal of M. hyperodae between sites during this study.

Table 1.  For each of 14 New Zealand sites, the number of individuals released, percentage of eastern M. hyperodae released and total number of specimens classified by sampling year
SiteNumber released% ENumber of Microctonus hyperodae classified by year
  • *

    Parasitoid releases were made in 1991 except at sites where R indicates an alternative year of release.

Wellsford117036511 2310528 2       
Ruakura123326510 17 574136       
Cambridge1139079   56        
Reporoa1225665  24 644862       
Meeane 492167   R26   32   
Bridge Pa 492167   R30   23   
San Angelina 198169      R 36  42 
Watiro 198169      R 23    
Te Muna 198169      R  3  53 
Wharekauhau 198169      R  2  1317
Hororata1814665 1  9          
Lincoln1814665 3129 23693074553846616  
Ophir 844068   2 162641 5      
Sutton 526363    7  7 6      

Adults of the host, L. bonariensis, were collected by suction sampling, then maintained in cages at ≈20 °C, 14 : 10 light : dark. Any immature M. hyperodae that emerged were reared to the adult stage (Goldson et al. 1992), then stored either in 70% alcohol or at –40 °C prior to analysis. At Lincoln (Fig. 1), additional parasitoids were collected as adults by suction sampling (Phillips, Proffitt & Goldson 1998). Parasitoids were not obtained from all sites in all years due to the combined effects of limited resources, the large geographical scale of the experiment, sampling sometimes coinciding with seasonal periods when insects were in their below-ground or endophytic life stages, and difficulties in obtaining insects during wet weather.

data analysis

Assessment of M. hyperodae genetic variation

Of the 1853 M. hyperodae collected, 1324 (71%) were classified by discriminant function analysis of morphometric variables (described below), and 524 by MDH genotype (Iline & Phillips 2004). The morphometric approach was used because: (i) it was accurate and inexpensive (see Results); (ii) it was compatible with concurrent studies of M. hyperodae egg loads (Phillips et al. 1998); (iii) the older specimens were no longer suitable for study by enzyme electrophoresis or PCR-based methods; and (iv) the morphometric data of Phillips & Baird (1996); Phillips et al. (1997) were available for further analysis. Variation in MDH genotype (Iline & Phillips 2004) was also used to assign M. hyperodae to biotype when fresh or frozen specimens were available and not required for other experiments.

improvement of morphometric method

The following methodological developments were based on morphometric (Phillips & Baird 1996), protein (Iline & Phillips 2004) and genomic (Winder et al. 2005) data that had been obtained from the same M. hyperodae single-female lineages, all of known South American origin.

Phillips & Baird (1996) mistakenly inferred from their morphometric data that M. hyperodae from San Carlos de Bariloche represented a third biotype, and therefore computed discriminant functions for classifying specimens of unknown origins as either eastern, western or San Carlos de Bariloche. Here, this error was eliminated by using results from molecular analyses (Iline & Phillips 2004; Winder et al. 2005) to assign single-female lineages from San Carlos de Bariloche in the morphometric data set as either eastern or western (Phillips & Baird 1996). Similarly, the lineage from Concepcion in Chile that had exhibited eastern molecular markers was assigned as eastern in the morphometric data. These corrected data were used to calculate new discriminant functions for classifying M. hyperodae of unknown origins to biotype (Phillips & Baird 1996). One function was calculated using the set of 11 variables employed in the original morphometric method (Phillips & Baird 1996), while others were calculated using different sets of fewer, arbitrarily selected morphometric variables to test if similar accuracy could be achieved with fewer measurements. The performance of each set of morphometric variables in classifying M. hyperodae to biotype was assessed by cross-validation, which involved randomly allocating morphometric data for different specimens to one of two subsets. A discriminant function computed from one data subset was then used to classify specimens in the other subset to biotype, and the accuracy of these classifications was recorded. Each set of variables was subjected to 100 cross-validation test cycles involving >14 000 test classifications per set.

test of congruence between allozyme and morphometric variation

To seek additional, direct support for congruence between the morphometric and molecular methods for classifying parasitoids to biotype, M. hyperodae individuals of unknown South American origin were classified as either eastern or western using both morphometrics and MDH allozyme genotype. The 215 parasitoids tested were obtained from five of the 14 New Zealand study sites (Fig. 1) during 1999 and 2002. The antennae, a forewing and a hind leg were removed from each specimen for morphometric analysis before the remainder was used to determine its MDH genotype.

tests for sampling biases

Microctonus hyperodae were collected from Lincoln (Fig. 1) both as eggs or larvae within their hosts, and as free-living adults. To check for sampling bias, the within-year biotype frequencies observed in each life stage were compared using a generalized linear model (GLM) for modelling binomial proportions employing a logit link in genstat ver. 7·1.

The dates when parasitoids were collected varied between sites and years, and this could have created bias if there had been phenological variation between biotypes. However, such phenological variation could not have occurred during winter (June–September), when M. hyperodae exists only as eggs and larvae. To test for bias, the biotype frequencies observed in summer were compared with those observed in winter using the GLM described above.

tests of null hypothesis

The null hypothesis was tested three ways. First, the observed and expected biotype frequencies were compared using a GLM for modelling binomial proportions with a logit link (McCullagh & Nelder 1983). An offset was used to fit the expected frequencies (those released at each site) as the base model, and the size of the released population was added to the model as a covariate. Second, counts of each biotype within year and site were analysed by calculating χ2 statistics to test the goodness of fit between the observed and expected values. Third, a computer simulation model was developed to generate frequency distributions for the responses of biotype frequencies to genetic drift and sampling variation in each of the 14 released populations. Genetic drift was simulated by using the biotype frequencies and population sizes released at each field site (Table 1) to initiate corresponding modelled populations, and the simulations continued for the number of years after release that the last field collection was made at each site (Table 1). Most populations were assigned three generations per year, with the exceptions of Wellsford, Ruakura and Cambridge, which were assigned four; and Ophir and Sutton, which were assigned two (Fig. 1). These were the potential maxima for each site calculated by a M. hyperodae phenological model (Barlow et al. 1994), although they were slightly higher than was subsequently observed in the field at Ruakura (Barker & Addison 2006) and Lincoln (Goldson et al. 1998). When used in tandem with a modelled assumption of zero annual population growth (described below), they provided conservatively high estimates of drift. In accordance with an estimate of 90% mortality of immature M. hyperodae each spring at Lincoln (Phillips et al. 1998; C.B.P., unpublished data), released individuals had a 10% random chance of survival, as did first generation individuals in each subsequent year. Reproductive rates were randomly assigned to individuals from a Poisson distribution with a mean that was redefined each generation to constrain overall annual population growth to zero. For example, for the four sites where 1981 insects were released (Table 1), the populations were randomly reduced to 198 in each of the five years of the simulation, and they never exceeded 1981 individuals. Although, in reality, the released M. hyperodae populations increased rapidly at all sites (Goldson et al. 1998; McNeill et al. 2002; Barker & Addison 2006), the assumption of zero growth was more conservative as drift has a greater effect in small populations. Ten thousand simulations per site were completed, and the resulting annual biotype frequency distributions arising from genetic drift at each site were recorded.

The effects of sampling variation in field collections of M. hyperodae were combined in the model with those of genetic drift by convolution. During each of the 10 000 drift simulations per site, an annual sample of individuals was randomly taken from the modelled population using sample sizes identical to those obtained from the field (Table 1), and biotype frequencies in these samples were measured. The distributions of annual biotype frequencies arising from the 10 000 simulations of both genetic drift and sampling variation were recorded for each site, thus enabling 95 and 99% confidence intervals to be calculated for each site and year.

The probability that the observed frequencies of eastern M. hyperodae arose from random causes was assessed by comparing the observed values with their corresponding 95 and 99% CIs for genetic drift plus sampling variation.


congruence between allozyme and morphometric variation

Classifications based on both morphometrics and MDH genotype gave congruent results for 199 (93%) of the 215 specimens tested. Assuming MDH gave the correct result in the remaining 16 cases, eastern and western M. hyperodae were equally likely (P ≈ 0·07) to be misclassified by morphometric analysis. Due to this absence of bias, the overall results from MDH and morphometric classifications were similar: MDH gave 180 E : 35 W, while morphometrics gave 178 E : 37 W.

tests for sampling biases

During 1997, 1999 and 2000, M. hyperodae were collected from Lincoln both as eggs or larvae within their hosts, and as free-living adults. Within years, there were no significant differences in biotype frequencies between life stages (P =0·671), or within sites and years between summer and winter (P = 0·312). Due to this lack of bias, data for samples collected on different dates from the same site in the same year, or as different life stages within the same year, were pooled.

improvement of morphometric method

Using the 11 variables originally employed by Phillips & Baird (1996), 99% of eastern (n = 264) and 97% of western (n = 106) parasitoids were correctly classified during 100 cross-validation test cycles. However, similarly high accuracy (98% of 17 000 classifications correct) was obtained when only data for total number of antennal segments, length of basitarsus 3, width of stigma and length of radial cell were used. These four variables were therefore used here to make all morphometric classifications. The data required to calculate the discriminant function are shown in Table 2, and the function took the form:

Table 2.  Morphological character means and canonical coefficients for the discriminant function to separate eastern and western Microctonus hyperodae
ParameterWestern (n = 106)Eastern (n = 264)Overall meanRaw coefficient
Sum of segments in two antennae 41·1 38·7 39·40·8400
Length of basitarsus 3236·9252·8248·1–0·0386
Width of stigma139·5153·1149·9–0·0551
Length of radial cell519·7483·4493·80·0225
Z = Σc1(X1 – Y1) + ... + c4(X4 – Y4)

where Z is the discriminatory score, c the raw coefficient for a variable, X a sample value for the variable, and Y the overall M. hyperodae variable mean. For eastern M. hyperodae Z < 0, and for western M. hyperodae Z > 0.

results of computer simulations

Results from the simulations are shown in Fig. 2. As expected, greater genetic drift occurred in smaller populations. For example, the 95% CI for San Angelina, where 1981 parasitoids were released (Table 1), reached 0·33 after just 4 years (Fig. 2B), while at Lincoln, where 18 146 parasitoids were released (Table 1), the 95% CI reached only 0·18 after 10 years (Fig. 2A), even though three generations per year were modelled at both sites. The model also generated drift rates that, as expected, were positively correlated with the number of generations per year. For example, three generations occurred at Reporoa and four at Ruakura, and after 5 years the 95% CI was 0·16 for Reporoa and 0·17 for Ruakura (Fig. 2A).

Figure 2.

Frequency of eastern Microctonus hyperodae in annual samples collected from 14 New Zealand release sites. Dashed horizontal lines, 95 percentile of 10 000 genetic drift simulations; solid horizontal lines, 95 percentile of 10 000 simulations of sampling error + drift. Annual biotype frequencies that significantly differ from those expected under the no-selection hypothesis: *, P < 0·05, **, P < 0·01. (a) Sites where M. hyperodae was released in 1991; (b) sites where M. hyperodae was released after 1991.

As expected, the contribution of sampling variation to random variation in biotype frequencies was negatively correlated with sample size (Fig. 2). For example, only three parasitoids were sampled from Lincoln in 1992 (Table 1) and the 95% CI for drift plus sampling variation equalled 1, with sampling variation contributing 94% of the error (Fig. 2A). In contrast, 384 parasitoids were sampled at Lincoln in 1999 and the 95% CI for drift plus sampling variation was 0·19, with sampling variation contributing only 14% of the total variation (Fig. 2a).

tests of null hypothesis

The GLM indicated that significantly higher than expected frequencies of eastern M. hyperodae (P < 0·001) occurred at all sites except Lincoln (Fig. 2a) and Meeane (Fig. 2b). Western biotype frequency was positively correlated with released population size (P < 0·001), but this was solely due to Lincoln, where one of the two largest releases occurred (Table 1) and where the western biotype was most successful (Fig. 2a). When Lincoln data were omitted, the relationship became non-significant (P = 0·718).

The χ2 tests and computer simulations gave very similar results, so details are given only from the simulations. These showed that, contrary to expectation, 24 of 47 observed values lay outside their corresponding 99% CIs, and a further four lay outside their 95% CIs (Fig. 2). These significant departures from random variation occurred in at least 1 year at all 14 study sites (Fig. 2). Significantly higher proportions of eastern M. hyperodae occurred at all 14 sites (27 of 47 samples; Fig. 2), while only one sample collected at Lincoln in 1998 contained significantly more western M. hyperodae than expected (P < 0·01; Fig. 2a). Except for Lincoln and Meeane, all sites exhibited broadly similar temporal patterns of change in biotype frequency where eastern M. hyperodae rapidly became dominant, reaching frequencies of 0·9 or more, often only 1–2 years after release (Fig. 2). At Lincoln (Fig. 2a), frequencies of eastern M. hyperodae were significantly higher than expected from 1993 to 1994 (P < 0·01), returned to expected levels during 1995–97, declined below these levels in 1998 (P < 0·01), then returned to expected levels during 1999–2001. At Meeane, frequencies of eastern M. hyperodae increased above expected levels 1 year after release (P < 0·05), but declined to expected levels 5 years after release (Fig. 2b).


This study evaluated the recommendation that biological control practitioners foster adaptive evolution in introduced natural enemies to increase success rates in biological control. Implicit in this evaluation were three main questions: (i) do natural enemies evolve after being introduced to new regions? (ii) is this evolution adaptive? and (iii) does adaptive evolution contribute to biological control success? The data clearly answered the first two questions because biotype frequencies changed over time at all sites, and these changes were frequently too large to be explained solely by drift, which indicated that selection had played a significant role. Moreover, the selection was strongly directional, favouring eastern M. hyperodae at all but one of 14 study sites.

It is also evident that this adaptive evolution contributed to biological control success because it increased L. bonariensis mortality above what would have occurred if the biotype frequencies had changed only due to drift. This conclusion is based on the observed greater reproductive success of eastern M. hyperodae in New Zealand combined with careful consideration of the parasitoid's biology, as follows. Theoretically, the selective advantage of eastern M. hyperodae could have arisen from superiority during interference competition (Collier & Hunter 2001) with western parasitoids. For example, eastern larvae may have been superior during combat with western larvae in superparasitized hosts (only one M. hyperodae larva ever completes its development in a host). In this situation, increased frequencies of eastern M. hyperodae would not lead to greater L. bonariensis mortality because eastern larvae would merely replace western larvae in the same host individuals. However, M. hyperodae females strongly avoid superparasitism (McNeill, Goldson & Baird 1996), so there is minimal potential for such interbiotype competition. Also, M. hyperodae adults do not host-feed, and the only way they kill L. bonariensis is through parasitism (Phillips 2002). This strict dependence of M. hyperodae population growth on L. bonariensis mortality meant that the greater reproductive success of eastern M. hyperodae in New Zealand was directly linked with increased L. bonariensis mortality, irrespective of the underlying reasons for the eastern biotype's advantage (e.g. higher attack rates or greater survival of immature stages).

These results provide the first robust, empirical support for the recommendation to maximize the natural enemy genetic diversity that is obtained from the native range and released in a new region to increase the likelihood of adaptive evolution and successful biological control (Simmonds 1963; Force 1967; Remington 1968; Lucas 1969; Mackauer 1972; Delpuech et al. 1993; Reed et al. 2003). The more limited potential for recombination to produce genetic variation in asexual compared with sexual species suggests that releasing multiple, genetically variable lineages should be relatively more important for asexual species. Nevertheless, genetic variation between sexual lineages should still facilitate local adaptation in released populations, so the results are broadly applicable. To exploit intraspecific variation more effectively, better knowledge is needed of the causes and spatial patterns of genetic variation exhibited by natural enemies in their native ranges, and of the genetically variable traits that are important to biological control outcomes.

The probable origin of New Zealand's L. bonariensis is eastern South America, near Colonia (Fig. 1; Williams et al. 1994), which is consistent with the possibility that eastern M. hyperodae is better adapted to New Zealand's L. bonariensis (Phillips et al. 1997). However, the pattern of genetic variation between South American geographical populations of L. bonariensis (Williams et al. 1994) does not match the east–west structure seen in M. hyperodae (Iline & Phillips 2004; Winder et al. 2005), and provides little support for a correspondence between M. hyperodae biotype and L. bonariensis genotype. Moreover, the relatively high frequency of the western biotype at Lincoln during 1995–2001 (Fig. 2a), in the presence of only minor variation between New Zealand sites in L. bonariensis randomly amplified polymorphic DNA (Williams et al. 1994), and in the absence of variation between New Zealand sites in L. bonariensis allozyme frequencies and mitochondrial cytochrome oxidase 1 sequences (C.B.P., I.I.I. and C.J. Vink, unpublished data), strongly suggests that L. bonariensis genotype is not a key factor in the selective advantage of eastern M. hyperodae in New Zealand.

Currently, the best insight into this selective advantage has arisen from field sampling at Lincoln. Although laboratory measurements showed no significant development rate variation between South American geographical populations of M. hyperodae (Barlow et al. 1994), field sampling has revealed that there is interbiotype phenological variation: while both biotypes overwinter as first instar larvae, the eastern biotype reaches the adult stage in early summer, ≈1 month before the western biotype (C.B.P., unpublished data). Moreover, there is a significant positive correlation between the frequency of western M. hyperodae at Lincoln and annually accumulated degree days (C.B.P., unpublished data). These observations suggest that the generations of western M. hyperodae could lag behind those of the eastern biotype throughout summer, and reproduction by the last summer cohort of western adults might be more seriously truncated by cool autumn conditions than that of eastern adults. The basis for the interbiotype phenological variation observed at Lincoln, and the extent to which the above hypothesis explains the biotype frequencies observed at other sites, are currently being examined.

The conclusion that some biological control agents do undergo postintroduction adaptive evolution has implications for minimizing impacts against non-target species (Simberloff & Stiling 1996; Barratt et al. 1997; Louda et al. 2003), because it may support the view that ecological risks from new introductions are largely unpredictable. In this model system, however, M. hyperodae generally responded to selection pressures within a year of release, indicating that some hypotheses regarding evolutionary changes in risks from natural enemy introductions could realistically be made prior to release and tested shortly afterwards, leading to improved predictive capabilities.

The demonstration by M. hyperodae that natural enemy intraspecific genetic variation exists for traits that are important to the outcome of biological control introductions indicates there may be potential to screen biotypic variation in putative biological control agents prior to release with the explicit objective of minimizing ecological risks. For example, although intensive postrelease monitoring has revealed no impacts of M. hyperodae on non-target species (Barratt et al. 1997), the observed biotypic variation in the timing of the early summer emergence of M. hyperodae adults could, in theory, have been exploited prior to the species’ introduction to New Zealand to minimize any temporal overlap between parasitoid adults and the potential non-target host species identified during prerelease testing (Goldson et al. 1992). The need to regulate new organism introductions at below the level of species is becoming increasingly recognized by government authorities (Gerard et al. 2006).

This study illustrates some limitations in using climate matching to identify the most suitable geographical sources of natural enemies for biological control introductions. The implicit assumption that populations from different climates will exhibit variations in adaptations related to climate has not been strongly supported by analyses of M. hyperodae intraspecific genetic variation. For example, no genetic variation was detected among the eastern M. hyperodae collected from five climatically disparate locations in Argentina (excluding San Carlos De Bariloche), Uruguay and Brazil (Fig. 1), while the Andes mountains apparently had a more direct influence on M. hyperodae genetic variation than did climatic variation between sampling sites. It may be that the apparent absence of variation in M. hyperodae over large portions of its South American range is partly due to its asexual reproduction, but the same point still applies to the arrhenotokous, congeneric parasitoid M. aethiopoides. Vink et al. (2003) studied specimens from 14 countries, and found little correlation between M. aethiopoides genetic variation and sampling location, while considerable variation was associated with host species. Indeed, nearly all of the M. aethiopoides genetic variation documented by Vink et al. (2003) could be obtained by rearing M. aethiopoides from the different host species present in one field in southern France (C.B.P., C.J. Vink, A. Blanchet & K.A. Hoelmer, unpublished data). The studies of genetic variation in M. hyperodae and M. aethiopoides therefore suggest that biological control practitioners would be ill advised to design sampling regimes for putative biological control agents based solely on climatic similarities between the source location and the intended region of introduction.


We are grateful to G. Barker, P. Addison (AgResearch, Ruakura), M. Slay (AgResearch, Poukawa), C. Ferguson, A. Evans and B. Barratt (AgResearch, Invermay) for contributing specimens; to H. Townsend and R. Cane (AgResearch, Lincoln) for technical assistance; and to M. Barron (Landcare Research, Lincoln), N Richards (AgResearch, Lincoln) and three anonymous referees for reviewing earlier versions of this manuscript. This research was funded by the FRST ‘Ecosystems Bioprotection’ programme, contract LINX0304.