How should the risk associated with the introduction of biological control agents be estimated?
Earl D. McCoy. Tel: +1 813 974 5219;
fax: +1 813 974 3263; e-mail: email@example.com
- 1Florida has an exceptional burden of invasive species. The history of the classical biological control of invasive arthropod pest species in the region largely is one of inadequate pre-release testing for nontarget effects.
- 2A recent analysis indicated that a substantial risk of nontarget effects may exist in Florida, although the risk appears to be confined to a relatively small group of species within approximately ten families and documented cases of nontarget effects are rare, despite previous risky practices.
- 3Great progress has been made recently in creating an organized framework for dealing with the uncertainty accompanying biological control importations in Florida and elsewhere. We suggest some ways in which balancing the risks and associated costs of releasing a biological control agent against the risks and associated costs of not releasing the agent may be improved.
- 4Ultimately, experts will need to set some level of acceptable risk, and the ‘precautionary principle’ has been advanced to guide this process. As it stands, however, the precautionary principle applied to biological control falls short as a guide because it does not provide a prescription for action.
- 5Florida case histories clearly illustrate both the complexity and urgency related to developing a prescription for action.
Disruption of ecological processes by invading species is a severe and well-documented threat (Mooney et al., 2005). The best response to the threat of invasive species probably is to take various steps to prevent their entry into unwanted places (Lodge et al., 2006). One such step is to adopt a quantitative risk analysis procedure and to apply it to every species proposed for importation (Lodge et al., 2006). This step also has been suggested specifically for the importation of biological control agents, which may pose a threat to nontarget species (Louda et al., 2003; Van Lenteren et al., 2003; Wright et al., 2005; Messing et al., 2006). The potential value of applying quantitative risk analysis procedures to every proposed importation of a biological control agent is clear, and researchers have made important progress in demonstrating how such procedures may be implemented (Wright et al., 2005; Messing et al., 2006). Nevertheless, any kind of meaningful risk assessment will be extremely difficult to accomplish because of the large number of documented risk factors, the paucity of data about these risk factors, and the contextual nature of risk analysis (Simberloff, 2003; Messing & Wright, 2006). Perhaps the most severe difficulty associated with a meaningful risk assessment is the need to determine precisely how much risk is acceptable. Despite the recent interest in the potential risks associated with the importation of biological control agents, we do not consider that the question of how much risk is acceptable has received adequate attention. In the present study, we briefly review the risk analysis process in biological control, suggest some mostly-overlooked considerations that we think are important in such an analysis, and then turn to the difficult issue of acceptable risk. Making the best choices given the uncertainty accompanying biological control should be of primary concern to practitioners and critics alike.
We base our comments largely on the record of importations of biological control agents into Florida. Within the U.S.A., the threat of biological control agents to nontarget species in Florida may be second only to the threat in Hawaii (Messing & Wright, 2006). Despite experiencing one of the highest rates of accrual of adventive species in the U.S.A. (Frank & McCoy, 1992; Frank et al., 1997; Frank & Thomas, 2004), Florida has imported and established only 60 biological control agents to combat the harmful species among them (Frank & McCoy, 2007). The backlog of adventive pests yet uncontrolled by this method is very large either because biological control has not been successful, has not been tried, or does not seem appropriate. Much of the backlog of uncontrolled adventive pests may be directly or indirectly attributable to a shortage of funding. It might be assumed that a shortage of funding would mean that those programmes in which pests pose the greatest economic threat would be undertaken first but the criteria for funding (and thus for control) are much more complicated (Habeck et al., 1993). For example, weeds with little documentation of economic effect are now being emphasized as targets, through several state and federal funding agencies, routinely earmarking funds to support biological control research into invasive weeds. With the exception of programmes targeting tropical soda apple Solanum viarum Dunal, which invades pastures (Frank & McCoy, 2007), programmes targeting weeds in Florida tend to have an environmental, rather than an economic, motivation. Examples include Chinese privet Ligustrum sinense (Lour.), which invades disturbed sites, and wetland nightshade Solanum tampicense Dunal, which invades wooded and streamside sites. Standards for pre-release testing of biological control agents against weeds are very stringent in Florida, possibly because of concern that they might attack crop plants and, more recently, native plants; typically, funding is available for pre-release testing for nontarget effects.
Even though biological control of pest insects began in Florida in the 1890s, in contrast to biological control of weeds, which did not begin in Florida until the 1960s, funding support for biological control research for pest insects lags well behind that for weeds (Messing et al., 2006). Funding has remained relatively low and brief, and has originated from granting agencies with no special category for biological control programmes. This funding structure has resulted in time and resources being spent almost exclusively on obtaining potential biological control agents, culturing them if possible, releasing them, and performing a brief evaluation of the effect on the target pest. Although practitioners may have had funds for post-release evaluation of biological control agents against pest insects, these evaluations largely omitted pre-release nontarget testing and relied upon assumed knowledge from field evidence or from the literature. Financial pressures on biological control programmes ultimately limit the duration of the programmes and can spawn risky practices, such as the ‘lottery (shotgun) approach’, in which biological control agents are released in rapid serial fashion against the same host (McEvoy & Coombs, 2000; Michaud, 2002). From a programme administrator's perspective, to release only one biological control agent when several promising candidates have been detected would appear presumptuous in the face of uncertainty about any one agent's ability to control the pest single-handedly. Similarly, to leave an extended gap between successive releases to reduce risk of nontarget effects would appear superfluous when the available evidence indicates a lack of discernible effect on nontarget species.
Because most biological control agents introduced into Florida have targeted insects (Frank & McCoy, 2007) and few of these have been screened for nontarget effects, concern over the risk to nontargets of releases of biological control agents in Florida is warranted. Frank and McCoy (2007) systematically evaluated the risk to nontargets of the introduction of biological control agents into Florida. Several lines of evidence that they developed concerning host ranges of released natural enemies indicated that substantial risk of nontarget effects may exist in Florida. They showed that 42% of the established biological control agents in Florida have native species in their potential host ranges, that releases of broadly generalist agents and agents targeting native species have occurred in Florida, and that the lottery approach to biological control has been practiced in Florida. On the other hand, they showed that the risk of nontarget effects in Florida seems to be confined to a relatively small group of species within approximately ten families and that documented cases of nontarget effects in Florida are rare, despite previous risky practices. The results of a simple modelling exercise suggested that fewer than ten releases of introduced biological control agents in Florida are likely to have produced population changes in nontarget species, and that fewer than four of the ten are likely to have produced substantial population changes. These conclusions, although based on the best available evidence, are far from certain, which leads naturally to the question of how much risk society is willing to accept in the face of uncertainty about the complicated (unintended, indirect, reticulate) effects that biological control agents could have on nontargets, community structure, and food web linkages (Henneman & Memmott, 2001; Pearson & Callaway, 2003; Hoddle, 2004; Messing et al., 2006)?
Risk analysis and assessment
Risk analysis and assessment for the importation of biological control agents should consist of identifying and weighting potential risks of action and inaction and then balancing those risks (Wapshere, 1989; Greathead, 1995; Messing, 2000). Empirical and theoretical observations have indicated that risk of action (of introduced agents to nontargets) may be engendered by a wide variety of practices. Six of these practices are consequences of inadequate taxonomy (sensu lato) or inadequate laboratory specificity testing:
- 1Employing screening trials that inadequately represent situations encountered in the field (e.g. a ‘species’ of control agent that is composed of biotypes, or even sibling species, with different host preferences) (Askew, 1971; Nechols, 2000; Stouthamer, 2006);
- 2Employing screening trials testing potential host species suitability based on taxonomic relatedness when control agents use other indicators of suitability (e.g. disjunct oligophagy) (Askew, 1971; Henneman & Memmott, 2001; Schaffner, 2001; Kuhlmann et al., 2006);
- 3Falsely concluding that a host species not accepted in screening trials also would not be attacked in the field (Schaffner, 2001; Van Lenteren et al., 2006);
- 4Falsely concluding that a host species accepted in screening trials also would be attacked in the field (Wapshere, 1989; Samways, 1997; Orr et al., 2000; Neuenschwander & Markham, 2001; Schaffner, 2001; Van Lenteren et al., 2006);
- 5Accidentally introducing contaminating or misidentified agents (Frank & McCoy, 1993; Lynch & Thomas, 2000; McEvoy & Coombs, 2000; Goettel & Inglis, 2006);
- 6Failing to recognize that host ranges of agents can evolve subsequent to introduction (Hopper, 2001; Hopper et al., 2006).
Another seven of these practices are consequences of inadequate knowledge, or inadequate appreciation of basic ecological relationships:
- 7Lack of ecological insight causing important relationships to go unnoticed (Moeed et al., 2006), relationships that could help in designing better release procedures (e.g. the significance of the typically nonlinear relationship between abundance of a pest and its ecological effect) (Lockwood, 2000); the adequacy of simple food chain models for modeling biological control (Memmott, 2000; Pearson & Callaway, 2003); the importance of niche overlap in predicting nontarget effects (Obrycki et al., 2000);
- 8Failing to detect potentially important nontarget effects because they do not produce obvious demographic responses (Hawkins et al., 1993; Lynch & Thomas, 2000; Kuhlmann et al., 2006; Messing et al., 2006);
- 9Releasing an agent that undermines another, more effective, agent, leading to increased pest density (McEvoy & Coombs, 2000; Barratt et al., 2006; Goettel & Inglis, 2006; Van Lenteren et al., 2006);
- 10Promoting the replacement of one pest by another, harder to control, pest (McEvoy & Coombs, 2000);
- 11Failing to prevent agents introduced elsewhere from immigrating to new places (McEvoy & Coombs, 2000; Mills et al., 2006; Moeed et al., 2006; Van Lenteren & Loomans, 2006);
- 12Releasing specialist agents in places where generalist resident predators/parasitoids can exploit these agents (Holt & Hochberg, 2001; Michaud, 2002);
- 13Diverting scarce resources from more profitable alternatives for managing pests (McEvoy & Coombs, 2000; Michaud, 2002; Babendreier et al., 2006; Stouthamer, 2006).
The final nine practices appear to be manifestations of what has been termed the ‘revenge effect', in which the rush to solve local and acute pest problems leads to diffuse and chronic problems that are even harder to solve (McEvoy & Coombs, 2000). The first two of these practices are now normally avoided:
The other seven practices are consequences of the pressures put on field practitioners to ‘do something’:
- 16Releasing agents onto islands (Funasaki et al., 1988; Lynch & Thomas, 2000; Henneman & Memmott, 2001); note that this practice is seen as advantageous in some locations (Moeed et al., 2006);
- 17Releasing agents that are only moderately effective on the target (Holt & Hochberg, 2001);
- 18Targeting species with high reproductive capacities (Holt & Hochberg, 2001);
- 19Releasing agents with high attack rates (Holt & Hochberg, 2001);
- 20Releasing highly mobile agents (Holt & Hochberg, 2001; Mills et al., 2006);
- 21Releasing agents without full knowledge of past successes and failures (Michaud, 2002);
- 22Using the lottery (shotgun) approach (McEvoy & Coombs, 2000; Michaud, 2002).
Although the list of risk factors accompanying importations of biological control agents already appears large, it most certainly is not complete, at least in its details. The list is likely to continue to grow (Hoddle, 2004; Messing et al., 2006) because the development of a coherent mathematical framework for analysing ecological effects of invasions is very recent (Ruesink et al., 1995; McEvoy & Coombs, 2000). It would appear that the more one looks, the more risk factors accompanying importations one finds, although, many times, several factors may be closely inter-related. On the other hand, no such detailed list of risk factors accompanying inaction (i.e. failure to introduce agents) appears to have been created. It would appear that most authors simply suppose that the risk of inaction is so obvious that details are unnecessary: the potential target species may run amok, ecologically or economically, in its new location, because the native species cannot resist its invasion (Michaud, 2002). We suggest, however, that the list of risk factors accompanying inaction should be just as detailed as the one accompanying action to make the balancing of risks of action and of inaction, as well as the balancing of associated costs (Shrader-Frechette & McCoy, 1992), reasonable processes. For example, in Florida, we should detail the risks of nontarget effects from the biological control agents of fire ants (Samways, 1997) and the risks that fire ants themselves pose to rare native species, such as the gopher tortoise (Allen et al., 2004), and to native ants (Porter, 2000). The probabilities associated with the sets of risks could then be set from two kinds of intense research: research aimed at determining the likelihood of an imported biological control agent causing nontarget effects and research aimed at determining the likelihood of native species being unable to control the target (Paini et al., 2008). Otherwise, we may stray into trying to balance the benefits of action against the costs of action (Babendreier, 2007), which we do not consider to be very productive in improving the decision-making process.
To illustrate the underappreciated risks that can emerge from construction of a detailed list of the risks of inaction, consider the efforts to control the yellow sugarcane aphid Sipha flava (Forbes) in Florida. This species apparently recruited from native grasses onto introduced sugarcane (Frank & McCoy, 2007), which is not necessarily an unusual event (Strong et al., 1977; McCoy & Rey, 1983; Shapiro, 2002; Graves & Shapiro, 2003); and it is the lone instance of a native species being targeted by an established biological control agent. Other examples of native Florida species recruiting onto introduced crops include the two species of weevils in the genus Pachnaeus that are minor pests of citrus (Woodruff, 1981; Frank & McCoy, 2007). That native targets can recruit from native hosts onto adventive hosts suggests that the opposite pathway (i.e. adventive targets recruiting from adventive hosts onto native hosts) could be a largely unrecognized problem. The listed hosts of targets (Frank & McCoy, 1993) are numerous, and often include multiple species (e.g. ‘ornamentals’, ‘fruits’, ‘grasses’), indicating the broadly generalist feeding habits of many of the targets. The potential movements of adventive pests from introduced to native hosts may be of concern for introduced hosts such as sugarcane, and various field crops, cucurbits and trees, although such movements have not been examined in detail. The effects of adventive pests that are recorded from hosts such as dogwood, holly, magnolia, and cycads, on native plants growing outside of cultivation and on native herbivores of these plants also are little known, as are the potential consequences of enhancement of populations of native insects by the additional food resources supplied via cultivation of host plants. Thus, the risk of ‘nontarget’ effects of adventive pests and their host plants may be an important consideration.
We have suggested a few risk factors potentially accompanying the failure to release a biological control agent (inaction), although the complete list of risk factors, such as that accompanying the release of a biological control agent (action), is likely to grow. The considerable sizes of the two lists of risks emphasize the pressing need for the development of a set of realistic probabilities to accompany these lists. Even if the risk factors only can be ranked, such a ranking could be extremely valuable (Wan & Harris, 1997; Messing, 2000). Many considerations make the development of a set of realistic probabilities difficult; however; we detail three of them here. The first consideration is the necessity of detailed confirmation of many of the basic suppositions accompanying the analysis of the risk of biological control. For example, it needs to be confirmed that successful introductions tend to have more major nontarget effects; that more major nontarget effects tend to be direct effects; that parasitoids raised from nontarget hosts tend to be ‘accidental introductions' (hitchhikers, stowaways, and the like), rather than deliberate ones; and that the detrimental role of classical biological control is minor compared with ‘accidental introductions' (Lynch & Thomas, 2000).
The second consideration is holding introductions to a consistent and reasonable set of standards for safety. Calculations of risk probabilities need to keep the currency (ecological versus economic) constant, or at least not confuse the kinds of currency (Lynch & Thomas, 2000; Nechols, 2000; Obrycki et al., 2000; Michaud, 2002). In Florida, for example, the releases of two species, Encarsia smithi (Silvestri) and Cirrospilus ingenuus Gahan, potentially were detrimental to the control of the pests against which they were targeted because these agents function as facultative hyperparasitoids of other introduced agents. Neither species appears to have compromised the control programme to which it was targeted, and neither appears to have been collected in Florida in recent years. It might be considered problematic to count this example as a nontarget effect, when the effect was economic, and yet the supposed reason for concern is the ecological effect of introduced agents on native species. Calculations of risk also need to take into account the nature of both the target (i.e. native; adventive and harmful; adventive, possibly even introduced, and beneficial) and the agent (attacks only the target everywhere; attacks only the target in the location of interest, for lack of alternative hosts; attacks more than just the target but all nontarget hosts are adventive pests; attacks more than just the target but some hosts are native species that are broadly considered to be economically-important pests; attacks more than just the target but some hosts are native inoffensive species). Such distinctions are not always clearly made; for an example, see table 2 in Lynch and Thomas (2000), in which the nontarget species affected by Cryptolaemus montrouzieri Mulsant is not native. If a highly cautious approach were taken to the introduction of agents, it might be acceptable to release agents that attack only the target everywhere or, questionably, attack only the target in the location of interest, and no others. Yet, for example, populations of a native pest species may be high only because the species attacks an introduced plant (Pachnaeus spp. and Sipha flava may be Florida examples; Frank & McCoy, 2007). In such cases, does the risk of releasing an agent outweigh the risk of using some other control method?
The third consideration is allowing introductions a reasonable chance of demonstrating that they meet the standards for safety. Post-release monitoring needs to be conducted not only to find nontarget effects, but also to determine whether potential risk factors actually turn out to be important. For example, a determination is required of whether the conspicuous lottery approach employed in Florida (Frank & McCoy, 2007) actually has led to increased numbers of nontarget effects. The potential role of negative evidence in decisions to release biological control agents should be considered. For example, the virtually complete lack of examples of nontarget effects to date in Florida despite the apparent risks that sometimes have accompanied introductions of agents (Frank & McCoy, 2007) most likely should influence the calculation of the risk probabilities accompanying future releases. If a highly cautious approach were taken to the introduction of agents, it might not be acceptable to allow the release of an agent that, for example, targets a species with native congeners, even if those congeners were shown not to be suitable hosts in host range tests. Similarly, the recent improvement in biological control practices (Frank & McCoy, 2007) should influence the calculation of probabilities of nontarget effects accompanying future releases. Chronological differences in importations and releases may reflect both changing attitudes about releasing generalist species (Hawkins & Marino, 1997; Henneman & Memmott, 2001) and stricter control of releases. For example, there are now at least two levels of review of applications for the release of biological control agents targeting animals in Florida: the USDA-APHIS (under Federal law) and the Florida Department of Agriculture (under Florida law). University of Florida/Institute of Food and Agricultural Sciences employees additionally must first submit applications to a pre-review committee, and USDA employees must file an Environmental Assessment. For agents targeting plants, the USDA-APHIS review also includes a Technical Advisory Group, made up of representatives from several federal agencies, and all applicants must file an Environmental Assessment (Scoles et al., 2005).
Once probabilities have been ascribed to all of the potential risks, both of action and of inaction, the even more difficult process of balancing risks can be undertaken. The balance should be based on carefully testing the ability of the chosen species to survive in the new environment, its potential to grow in unintended places, the potential effects of its growth in unintended places, its potential for accomplishing the task for which it was introduced, and the potential consequences of not introducing it (Frank & McCoy, 1995; Ruesink et al., 1995; McEvoy & Coombs, 2000; Messing, 2000; Michaud, 2002; Goulson, 2003; Osborne & Cuda, 2003). We have suggested that the correct balancing is risk of action versus risk of inaction but this is not the only option. Other options would be to balance risk of action versus benefits of action or to balance risk of action versus risk of alternative action (Osborne & Cuda, 2003). However, we do not view either of these other options favourably. Comparing risks to benefits may be doomed from the outset. First, the risks and benefits of importation of a biological control agent often cannot be compared on an equal footing (Bigler & Kölliker-Ott, 2006). Second, the potential benefits of importation may be judged to be so immense a priori (Van Driesche & Bellows, 1996; Wright et al., 2005) that any risks may pale by comparison. Third, benefits typically can be identified only in relation to replacement of the risks associated with the current control method (Bigler & Kölliker-Ott, 2006; Moeed et al., 2006). Comparing risks of one control method against risks of another method makes the exercise relative rather than absolute, meaning that both methods could be too risky, although this would never be known. The principal argument against these other options, we suggest, is simply that they divert attention away from the important comparison of the likely ecological outcome of a pest problem with and without the intervention of an imported biological control agent.
Some time ago, Samways (1997) asked in relation to classical biological control ‘… what risks are we prepared to accept?’ He never fully answered the question, although it is a very important one. Exactly how much difference between the risk of action and the risk of inaction would we need to see to decide in favour of the importation of a biological control agent? Such importations never can be completely without risk, and decisions about acceptable risk are not easy to make (Lynch & Thomas, 2000; Nechols, 2000; Obrycki et al., 2000; Hoddle, 2004). Such decisions require value judgments to be made by the persons conducting and regulating importation (Bigler & Kölliker-Ott, 2006; Moeed et al., 2006). Several ways of dealing with the uncertainty inherent in such decisions have been proposed. One way is to simplify the matter by worrying only about nontarget organisms that are rare, beneficial, or beautiful, or that act as keystone or flagship species (Hoddle, 2004). We are concerned that acting in accordance with this suggestion will force biological control into the same philosophical dilemma as it has conservation, and we suggest that we stay away from trying to decide which species are members of the ‘desirable native and exotic fauna’. Another proposed way of dealing with uncertainty inherent in decisions about acceptable risk is to adhere to a ‘precautionary principle’ (O’Riordan & Cameron, 1994; McEvoy & Coombs, 2000). Adopting the perspective that it is actually the damage a pest causes that is the ‘target’ of classical biological control, rather than the pest itself (Lockwood, 2000), makes the potential value of the precautionary principle clear. In the face of our profound ignorance concerning the temporally and spatially subtle consequences of ecological relationships (Lockwood, 2000; McEvoy & Coombs, 2000; Osborne & Cuda, 2003); the self-perpetuating, self-dispersing, and irreversible nature of biological control (Barratt et al., 2000; Lockwood, 2000); and at least the potential for biological control agents to cause declines and extinctions of nontarget species (Samways, 1997; Schellhorn et al., 2002), it would appear that a particularly cautious approach to the introduction of biological control agents is warranted. The precautionary principle for biological control has four parts: (i) recognizing that the release of biological control agents can harm nontarget organisms; (ii) recognizing that actual harm to nontarget organisms of sufficient magnitude and severity has occurred to merit new principles for the release of agents; (iii) placing the burden of proof for showing that the release of new agents is necessary, safe, and effective on those proposing the releases; and (iv) ensuring that the procedure surrounding the release of agents is open, informed, and democratic, and involves an examination of all alternatives, including no release (McEvoy & Coombs, 2000).
The precautionary principle is an excellent starting point for addressing the question of acceptable risk. For the most part, the principle is a straightforward, reasonable response to previous incautious behaviours, such as failing to devote proper attention to potential nontarget effects (Neuenschwander & Markham, 2001; Schaffner, 2001; Waage, 2001); failing to respond to evidence that additional testing was needed (Gassman & Louda, 2001); failing to consider possible alternatives to introductions of biological control agents, such as the use of plant competition and native biological control agents (Sheldon & Creed, 1995; McEvoy & Coombs, 2000; Nechols, 2000; Michaud, 2002); releasing agents without adequate justification or necessity (Lynch & Thomas, 2000; Nechols, 2000; Michaud, 2002); and inadequately measuring effectiveness of biological control programmes because pre-release data were missing or no follow-up surveys were undertaken (McEvoy & Coombs, 2000). The principle also is a reasonable response to the general lack of input from the public (Greathead, 1995). Biological control, similar to conservation (Shrader-Frechette & McCoy, 1999), is an activity carried out typically in the public interest, with public funds, and with risk to the public welfare; and, therefore, assumes the responsibility of being openly democratic.
We suggest, however, that the precautionary principle is problematic because it falls short of providing a prescription for action (Shrader-Frechette & McCoy, 1999). Simply requiring those proposing releases to demonstrate that a particular release is simultaneously necessary, safe, and effective without clarifying what evidence, and how much of it, would be needed to satisfy these requirements does not appear to be reasonable. We have observed much the same problem in the conservation arena, where, for example, a party is required to show that a restored wetland is ‘functional’ or that a particular action does not cause ‘significant harm’, when the meanings of these terms are not even clear to the regulatory agency. The lack of a sound prescription for action can lead to inaction, which, in the case of biological control, may have serious consequences, including stimulating illegal importation (Samways, 1997); encouraging continued ecological damage by target species (Lynch & Thomas, 2000; Messing, 2000; Holt & Hochberg, 2001; Neuenschwander & Markham, 2001); and dissuading cooperation by creating an air of negativity, overemphasizing past errors, and failing to appreciate improvements that already have occurred [Stiling and Simberloff (2000, p. 33) appeared to acknowledge this potential problem when they said that they had ‘no intention to suggest that the sky is falling’.].
Loose adherence to the maxim that specialist natural enemies make the most effective biological control agents, and the fortunate general failure of imported generalist natural enemies to establish, largely reflect Florida's classical biological control record. The record shows that many risks were taken and that it is probable there have been a few major nontarget effects and more minor ones (albeit confined to families of insects that the general public regards as pests), although that current evidence of any nontarget effects in the field is scant (Frank & McCoy, 2007). Pre-release testing of biological control agents against target and carefully selected nontarget organisms is the cornerstone of both an effective and a safe classical biological control programme, and should be standard practice. Risk to nontargets could be minimized further by biological control practitioners, carefully balancing the risks and associated costs of action and inaction, and the adoption of a precautionary principle, with a set of ecological and economic considerations going beyond pre-release testing. Details of how a precautionary principle should be implemented are complicated and in need of substantial further elaboration. Although such elaboration is a difficult undertaking, we must rise to the task quickly. The imminent invasion of Florida by the lime swallowtail Papilio demoleus L. (Heppner, 2006) provides a clear illustration of this point. The species has been spreading rapidly among the larger Caribbean islands (Guerrero et al., 2004; Homziak & Homziak, 2006), and genetic analysis has confirmed that the lineage now in the Caribbean feeds on citrus (Eastwood et al., 2006). Once the species arrives in Florida, it will present a potentially catastrophic economic threat to the citrus industry and an ecological threat to native congeners, some of which also feed on citrus and one of which is endangered. Yet, the options for control of P. demoleus will be limited. First of all, unlike other citrus pests, swallowtail butterflies are part of the charismatic microfauna, making it difficult to attempt any sort of control without an adverse reaction from the public. Classical biological control could threaten nontarget congeners. Biological control by generalist native predators may, at best, take years to accomplish, or, at worst, prove ineffective. Chemical control, on the other hand, could disrupt biological control of other citrus pests. Despite all of the interest in the direct and indirect risks of importation of biological control agents, the proper course of action to take is not at all clear at this point.
We thank J. P. Cuda, R. H. Messing, and P. A. Stiling for reading and providing comments on the manuscript.