Biological control of arthropods: parasitoid-host interactions
In biological control of insects, the best-studied system is a parasitoid-host system represented by a wasp
Aphytis melinus used to control California red scale
Aonidiella aurantii, a pest of multiple species of citrus (Murdoch et al.
2003). Competition between multiple enemy species did not undermine success: The more effective natural enemy
Aphytis melinus competitively displaced a less effective natural enemy
Aphytis lingnanensis Compere, 1955. The basic approach used in investigating this system was to analyze the foraging behavior, determine the consequences of behavior for population dynamics using mathematical models, and finally to test by field experiments whether the models correctly described the underlying processes.
Two (among many) hypotheses were tested with a single field experiment (Murdoch et al.
1996). The first was the refuge hypothesis, which grew out of empirical observations showing that a refuge exists for California red scale at the interior of the tree where scales are impervious to parasitism and suggesting that the parasitoid-host interaction might be stabilized by this spatial refuge for California red scale from parasitism. The refuge accounts for ~ 90% of the total scale population, it is a source of scale recruits to the exterior of the tree, and scale density in the refuge is less variable than at the exterior. Investigators hypothesized that the population at the exterior of the tree might be stabilized by the continuous flux of immigrants from the tree interior to the tree exterior. The refuge concept is a popular idea, widely invoked in population dynamics and biological control (Berryman et al.
2006), but until this study, its contribution to equilibrium levels and stability had not been tested in real biological control systems in a rigorous way.
The second was the metapopulation hypothesis—which holds scale populations may be unstable at local scales of observation (i.e. characterized by wide amplitude oscillations, extinctions, and colonization events), but are stabilized at global scales by sufficient asynchrony in local population fluctuations, and sufficient migration among local populations, to provide for persistence of the ensemble of locally unstable, interacting populations (Kean and Barlow
2000; Levins
1969). Within many agricultural systems, it is widely assumed that insect pests and their natural enemies are forced to persist as a metapopulation, undergoing frequent local extinctions and recolonizing patches following disturbances caused by harvesting or insecticides. However, until this study, the metapopulation hypothesis had not been rigorously tested for biocontrol systems.
The two possibilities—the refuge hypothesis and the metapopulation hypothesis—were rejected by a single, elegant 2 × 2 factorial experimental design to test what stabilizes this parasitoid-host interaction: refuge, metapopulation, or some combination. The experiment compared trees ‘open’ (uncaged) and ‘closed’ (caged) to migration, compounded with ‘refuge’ and ‘refuge removed’ trees. The experiment ran for 17 months (three generations of scale and nine generations of wasp), adequate to measure population dynamics. Stability was operationally defined as the inverse of variability, where variability was defined as the coefficient of variation in scale numbers. Removing the refuge actually decreased variability (increased stability), so the refuge is not stabilizing. Closing off a tree to migration had no detectable effect on variability, so the interaction is not stabilized by metapopulation dynamics. This study provides valuable lessons on how to understand complex systems at all levels of biological organization. Where possible to do so, perturb the system experimentally by selectively removing the candidate regulatory mechanism (or making it inoperative) and comparing the dynamics with that in an unmanipulated control treatment.
Feasible remaining mechanisms for stability were explored using a pulse, density-perturbation experiment that might uncover both density-dependence and the regulating mechanisms causing return to equilibrium. The experiment was coupled with a detailed day-to-day stage-structured model of the system of interactions incorporating candidate regulatory mechanisms, with parameters estimated independent of the experiment (Murdoch et al.
2005). The model distinguished scale stages and their differential use by
Aphytis, and it addressed three remaining mechanisms by which the system might be stabilized: (1) there is a long-lived invulnerable adult scale stage, (2) the wasp
Aphytis develops (about three times) faster than the scale, (3) an attack on older immature scale yields more parasitoid offspring than an attack on younger immature scale (the developmental ‘gain mechanism’)—for example, attack on older scale yields one or more female parasitoid offspring directly, while host-feeding on the youngest scale yields fewer offspring indirectly by providing the adult female with nutrients for egg development. Investigators followed the dynamics of these outbreak populations, together with caged and uncaged control populations, over three to five scale ‘development times’ (time intervals from scale birth to adult). Three separate experiments gave the same result. Control of the outbreak and stability (return to equilibrium density) occurred rapidly over a time period equal to about ~ three scale generations—illustrating the dynamic stability of this system. Little variation in density was observed thereafter. The model reproduced the dynamics of the perturbed populations with remarkable accuracy, and the model result was robust to substantial changes in parameter values and even to structural changes in the model. To generalize these results, it was possible to survey other parasitoid-host systems for control of scale insects to discover whether similar ecological attributes are shared across taxonomically-related biological control systems (Murdoch et al.
2005). Few other cases have been studied in sufficient detail, but 16 cases of biological control of coccids appear to share four main features: (1) the interaction is persistent, and perhaps stable, (2) control is attributed (at least locally) to a single parasitoid, (3) the pest has an invulnerable stage, and (4) the enemy development time is shorter than that of the pest.
The takeaway lessons from this decades-long study are (1) the mechanisms regulating biological control systems cannot simply be deduced from mathematical models, they must also be confirmed by rigorous empiricism including field observations and manipulative experiments, (2) this parasitoid-host interaction is highly stable and is regulated at the level of the individual citrus tree, (3) stabilizing mechanisms are linked to stage structure (developmental responses and invulnerable stages), (4) the stage-structured model developed for this system faithfully describes the return to equilibrium following a pulse, experimental perturbation (Murdoch et al.
2005). It is an open question whether conclusions from this study can be generalized beyond parasitoids interacting with scale hosts, mainly because too few systems have been studied in sufficient detail. However, from what we know so far, none of the other biological control systems targeting arthropod pests that have been examined for strong suppression and stability of interactions (winter moth in Nova Scotia, larch sawfly in Manitoba, olive scale in California, walnut aphid in California, and California red scale in Australia) matches the remarkable, intrinsic stability at a local spatial scale exhibited by the
Aphytis-red scale system in California (Murdoch et al.
1985). Furthermore, theoretical rules of thumb developed from these studies concentrate on pest suppression and ignore stability (Murdoch et al.
2003), suggesting that investigating the causes of stability is an activity of leisure rather than of necessity.
Biological control of weeds: herbivore-plant interactions
In biological control of weeds, a well-studied example is an herbivore-plant system involving multiple herbivore species—the cinnabar moth
Tyria jacobaeae (L.) (Lepidoptera: Erebidae), and the ragwort flea beetle
Longitarsus jacobaeae (Waterhouse) (Coleoptera: Chrysomelidae) interacting with a shared target host plant tansy ragwort
Jacobaea vulgaris Gaertn.) (Asterales: Asteraceae) and non-target host plants related to the target. The caterpillars of the cinnabar moth feed on foliage and flower buds (technically capitula). The adults do not feed. The larvae of the flea beetle feed mainly on roots, and adults feed on foliage. Additional control organism species have been introduced for biological control of tansy ragwort, but they play a minor role in regulating ragwort abundance. The entire community represents an Old Association—the weed, the enemies, and background vegetation of perennial pasture grasses all originated from Europe and have been transported around the world to Australia, New Zealand, and North America. The tansy ragwort system has been studied in the plant and herbivores’ native home in Europe (mainly by investigators in England and The Netherlands) as well as abroad in North America (both USA and Canada), New Zealand, and Australia. Studying invasive species in both native and introduced ranges is a longstanding practice in biological control and recently recommended by ecologists for studying the mechanisms that regulate ecological and evolutionary dynamics of invasive plants (Hierro et al.
2005; Williams et al.
2010). Recently, the tansy ragwort system has been used to study the role of rapid evolution in ecological dynamics (McEvoy et al.
2012b; Rapo et al.
2010; Szucs et al.
2012).
The success of biological control must be measured before it can be explained. Early studies documented variability in the biological control of tansy ragwort on a regional scale, identified its causes, and quantitatively evaluated overall success in surveys of field populations in Western Oregon spanning 12 years and 42 sites (McEvoy et al.
1991). The system of interactions led to strong, stable suppression of ragwort to < 1% of ragwort’s former regional abundance, and the speed of control (6–8 years) did not vary with precipitation, elevation, land use, or timing (year) of natural enemy release. In the vernacular favored by ecologists, the outcome was robust, not highly ‘context dependent.’ A local pulse-perturbation experiment showed that introduced insects, within one ragwort generation, can depress the density, biomass, and reproduction of ragwort to < 1% of populations protected from natural enemies (McEvoy et al.
1991). Stability was not confirmed by modeling. However, the stability of this system has been repeatedly demonstrated empirically by pulse perturbation experiments, a generally accepted method of demonstrating population regulation (Murdoch et al.
2005; Murdoch
1970). Create an upsurge in ragwort abundance and natural enemies readily colonize and return the population to pre-perturbation levels in ragwort populations exposed to natural enemies in open cages. This outcome contrasts with persistence ragwort populations at high levels of abundance when protected from natural enemies by closed cages (James et al.
1992; McEvoy et al.
1991; McEvoy and Rudd
1993; McEvoy et al.
1993). Stabilizing mechanisms, whatever they might be, are not found at local scales (e.g. an agronomic field), but at more global, regional scales of observation. The stabilization of ragwort density is not strictly a deterministic process, but perhaps better described by stochastic boundedness (Murdoch et al.
1985): the ceiling on fluctuations in weed density steadily declined after releasing biological control organisms (McEvoy et al.
1991). Strong suppression of ragwort abundance triggered a successional process leading to replacement of ragwort by a plant community dominated by perennial grasses introduced long ago from Europe (McEvoy et al.
1991), but the community changes also allowed recovery of populations of a North American native herb (the hairy stemmed checker mallow
Sidalcia hirtipes C.L. Hitchc., Malvaceae) of conservation concern (Gruber and Whytemare
1997). An invulnerable stage for ragwort exists, represented by a large, persistent seed bank buried in soil (McEvoy et al.
1991). Factors regulating the number of individuals entering, remaining within, and leaving the seed bank of ragwort populations are fairly well known (McEvoy et al.
1991,
1993; McEvoy and Rudd
1993); the consequences of the seed bank for ragwort population dynamics have been investigated using structured population models, under the simplifying but unsatisfactory assumption of a constant annual rate of recruitment of individuals from the seed bank (Dauer et al.
2012). A recipe for neutralizing a seed bank is to minimize disturbance and maintain a competitive grass cover. However, the exposure of populations to various levels of each regulating factor remains to be quantified in a landscape context. The seed bank did not prevent local extinction of actively-growing stages of ragwort populations over the time period of observation, but over the longer term it might buffer against environmental perturbations and reduce the probability of extinction under conditions recently identified in structured population models parameterized for marsh thistle
Cirsium palustre (L.) Scop. (Asterales, Asteraceae) and its seed bank (Eager et al.
2014).
Theory helps identify problems to be solved and suggests ways to go about investigating them. Our theoretical approach was inspired by the theory of activator-inhibitor systems of reaction–diffusion equations that has been used to describe pattern formation in numerous applications in biology, chemistry, and physics. Reaction–diffusion models first gained a foothold in ecology for modeling spatial spread of organisms and genes (Fisher
1937; Skellam
1951), and subsequently for investigating the roles of spatial heterogeneity and movement in the dynamics of predator–prey interactions (Grunbaum
1998; Kareiva and Odell
1987), and in modeling the spread and biological control of invasive species (Fagan et al.
2002; Hastings et al.
2005; Shigesada and Kawasaki
1997). In our framework, the causes of strong pest suppression and persistence of pest-enemy interactions do not rest solely with the natural enemy. The driving forces in our biological control system were initially assumed, and ultimately confirmed, to be disturbance, colonization, and local interactions (resource limitation, plant competition, and herbivory). Patchy disturbances to vegetation and soil remove biomass, open up space, recycle limiting resources, and set the stage for colonization and occupancy. Colonization plays an organizing role, depending on the ratio of diffusion coefficients of the long-range inhibitor to the short-range activator. Local interactions, including the ‘top-down’ effect of herbivory combine with the ‘bottom-up’ effect of plant competition (resource limitation), set in motion the process of community succession that leads to replacement of tansy ragwort with a background vegetation composed mainly of perennial grasses. We incorporated these assumptions in the Activation-Inhibition Hypothesis developed to describe the workings of biological weed control systems: (1) The Activation Hypothesis is that localized disturbance and buried seed (more generally a source of propagules) combine to create incipient weed outbreaks; (2) The Inhibition Hypothesis is that insect herbivory and interspecific plant competition combine to oppose increase and spread of incipient weed outbreaks; (3) The Stability Hypothesis is that the balance in short-range activation and long-range inhibition leads to a general condition of local instability and stable average spatial concentration of the weed.
Our study was implemented as a very large, long-term factorial experimental design varying the timing and intensity of disturbance in the activation phase; observing the timing of arrival by activator and inhibitor in open field plots in the colonization phase; varying the levels of the inhibitors including herbivory by ragwort flea beetle and the cinnabar moth, and interspecific plant competition. One version of this design yielded a total of 24 treatment combinations of two disturbance times (a pulse perturbation achieved by tilling of soil in fall or spring) × two cinnabar moth levels (plots exposed or protected with cages) × two flea beetle levels (exposed or protected) × three plant competition levels (a press perturbation whereby background vegetation was continually removed, clipped to mimic grazing, or left unaltered), with each of 24 treatment combinations replicated four times for a total of 96 experimental plots (McEvoy et al.
1993). We periodically censused insects and plants by stage for six years or ~ 2–3 ragwort generations (ragwort ranges from biennial to short-lived perennial) and six insect generations (the insects are univoltine), long enough to observe population dynamics of interacting plants and insects for each of the 24 combinations of disturbance-herbivory-plant competition treatments.
The theory of structured population models helped us translate variation in environment factors or drivers (disturbance, plant competition, herbivory) into changes in ragwort vital rates and then project the consequences of changes in vital rates into changes in population growth rates (Caswell
2001). Such models have been widely used for managing populations, whether for harvesting, controlling, or conserving (Caswell
2001; Morris and Doak
2002). Population growth was projected using a linear deterministic, stage-structured population model parameterized independently from field populations representing each experimental treatment combination. This single-species model developed to project ragwort dynamics is incomplete—it does not incorporate density-dependence (which is nonetheless implicit in the pulse-perturbation experiments and in field estimates of parameters) and stochasticity (reflected in the region-wide observational studies), and the ragwort population is not explicitly coupled in the model with each of the interacting populations (insects and other plant species featured in the experiment). Our linear, deterministic model best represents a population growing exponentially following disturbance, and it permits elasticity values to be easily obtained analytically, while elasticity methods for non-linear models are more difficult. This simple model performed remarkably well for projecting the speed of control, the time taken to eliminate an incipient outbreak (Dauer et al.
2012; McEvoy and Coombs
1999), a valuable metric for management. Perturbation analysis (including elasticity, sensitivity, decomposition of effects) provided a framework for ‘targeted life cycle disruption’ as a control strategy, identifying which life cycle transitions are potentially most influential on ragwort’s population growth rate, and which of these transitions are actually the most variable and amenable to management manipulations (Dauer et al.
2012; McEvoy and Coombs
1999). The standard approach in biological weed control is to target plant parts like roots, shoots, leaves, and seeds (a way to kill individuals), but targeting life cycle transitions is now generally recognized as a more reliable way to control populations (Shea et al.
2010). The recommended targeting of life cycle transitions (for this short-lived perennial modeled with a one year time step, the ‘biennial’ transitions include the probability a juvenile alive at time
t will survive and develop to adult at time
t +
1, and the fertility parameter or the number of juvenile offspring surviving at time
t +
1 per adult alive at time
t) proved remarkably robust to variation across 15 combinations of disturbance timing and community configuration, ranging for a single-species ragwort population to a multi-species community with a ragwort population interacting with populations of the cinnabar moth, the ragwort flea beetle, and interspecific plant competitors within the background vegetation (Dauer et al.
2012). Once again, the bugbear of ‘context dependence’ was brought to bay.
Our work with the ragwort system was motivated by a desire to test assumptions and predictions of ecological theories offered as explanations for biological control (McEvoy et al.
1993). We sought answers to three broad questions that have nagged ecologists for decades. First, we asked: do herbivores impose a low, stable pest-enemy equilibrium at a local spatial scale, or does local extinction occur? We found local pest extinction (defined from an economic perspective as eliminating all actively-growing stages outside the seed bank at the scale of an agronomic field) occurs and is compatible with success. We used speed of control (i.e. time to local extinction) as a metric for measuring success, on the grounds that even transient weed populations can cause economic and environmental damage. The speed of control was remarkably insensitive to scale: a 40,000-fold increase in spatial scale of the ragwort infestation (from 0.25 to 10,000 m
2) yields less than a three-fold increase in the time to local extinction of the weed population (from 1–3 to 5–6 year) (McEvoy and Rudd
1993). We conclude with others that transient rather than equilibrium dynamics may be of more practical significance for biological control in systems frequently reset by disturbance (Hastings
2001; Kidd and Amarasekare
2012). Second, we asked: is success more likely from a single ‘best’ herbivore species or from the combined effects of multiple herbivore species? We found that the ragwort flea beetle was by far the more effective regulator of ragwort abundance (Dauer et al.
2012; McEvoy and Coombs
1999); the flea beetle epitomizes the ‘search and destroy’ strategy (Dauer et al.
2012; McEvoy and Coombs
1999), offered as an alternative to creating a low, stable equilibrium on a local spatial scale (Murdoch et al.
1985). We found that the cinnabar moth makes a relatively small but detectable contribution to ragwort suppression (Dauer et al.
2012; McEvoy and Coombs
1999), thereby providing marginal support for the ‘complementary enemies’ model (Murdoch et al.
1985). The ragwort flea beetle attacks perennial rosettes that live up to five years. The rosette stage suffers relatively little mortality from a less successful control organism, the cinnabar moth. The ragwort flea beetle reduces all transitions in the ragwort life cycle graph used to develop the model. The cinnabar moth reduces only of one (from flowering plants to rosettes) of the two transitions (the ‘biennial’ transitions) identified as potentially most influential for ragwort population growth (Dauer et al.
2012; McEvoy and Coombs
1999). Third, we asked: is plant competition necessary to augment the impact of natural enemies on the dynamics of weed populations? When we began our study, it was not clear whether natural enemies and interspecific plant competition act sequentially or simultaneously: that is, natural enemies clearly cause weed suppression, but after that, interspecific plant competition possibly maintains low weed density until disturbance reoccurs. We found that herbivory and competition act simultaneously to inhibit an increase in weed abundance following disturbance, and herbivore colonization of ragwort was not reduced by the weed hiding in the background vegetation. We concluded that strong resource limitation at all target organism densities (due to a combination of intraspecific and interspecific plant competition) is a feature distinguishing biological control of weeds from biological control of arthropods.
This framework of activator-inhibitor systems provides a natural way to (1) incorporate the multiple forces driving biological control (disturbance, colonization, and local interactions including ‘bottom-up effects’ of plant competition and ‘top-down effects’ of herbivory), (2) show how perturbations in these processes give rise to plant invasions, and (3) develop biological control systems to oppose the establishment, increase, and spread of invasive species. It has several advantages over standard practice for explaining and managing biological control of invasive species. First, the activation-inhibition model avoids the “either-or” fallacy of false dichotomies common in the literature on biological invasions. For example, it combines processes often studied singly as contributors to invasions and biological control including disturbance (Hobbs and Huenneke
1992), propagule pressure (Simberloff
2009), perturbation of top-down effects of natural enemies (Keane and Crawley
2002) or
bottom-up effects of plant competition (Callaway and Ridenour
2004). The analysis of interacting causes is fundamentally different from the discrimination of alternative causes. As a second advantage, the activation-inhibition model transcends the oversimplification of causes and cures for biological invasions often attributed to biological control: namely, absence of enemies is the cause of invasions, and addition of enemies alone provides the cure. Even textbooks in biological control now question the ‘enemy release hypothesis’ as a universal explanation for invasiveness (Heimpel and Mills
2017). As a third advantage, the activator-inhibitor framework replaces the phenomenology of ‘context dependence’ with mechanistic explanations based on the action and interaction of disturbance, colonization, and local organism interactions, the forces driving the dynamics of nearly all ecological systems (Levin
1989). Finally, the activator-inhibitor framework incorporates environmental disturbance, movement and spatial heterogeneity. This motivated the gathering of detailed information on movement showing that the relative dispersal ability of the host plant (McEvoy and Cox
1987) is much less than that of the natural enemy species [the cinnabar moth
T. jacobaeae, the ragwort flea beetle
L. jacobaeae, and the ragwort seed head fly
Botanophila seneciella (Meade, 1892)] under field conditions (Harrison et al.
1995; Rudd and McEvoy
1996). Thus, natural enemies can easily overtake and suppress a weed population spreading autonomously. Interestingly, the activator-inhibitor recognizes differences in plants and insects in major modes of colonization, pitting the stronger spatial-averaging ability of the insects (by dispersal) against the stronger temporal-averaging ability of the plants (by dormancy, iteroparity, and perenniality). In their separate ways, the insects and plants rely on life history features (dispersal, dormancy, perenniality, and iteroparity) to average out local unpredictability in the environment.