Ecological Interactions Affecting the Efficacy of Aphidius colemani in Greenhouse Crops

Aphidius colemani Viereck (Hymenoptera: Braconidae) is a solitary endoparasitoid used for biological control of many economically important pest aphids. Given its widespread use, a vast array of literature on this natural enemy exists. Though often highly effective for aphid suppression, the literature reveals that A. colemani efficacy within greenhouse production systems can be reduced by many stressors, both biotic (plants, aphid hosts, other natural enemies) and abiotic (climate and lighting). For example, effects from 3rd and 4th trophic levels (fungal-based control products, hyperparasitoids) can suddenly decimate A. colemani populations. But, the most chronic negative effects (reduced parasitoid foraging efficiency, fitness) seem to be from stressors at the first trophic level. Negative effects from the 1st trophic level are difficult to mediate since growers are usually constrained to particular plant varieties due to market demands. Major research gaps identified by our review include determining how plants, aphid hosts, and A. colemani interact to affect the net aphid population, and how production conditions such as temperature, humidity and lighting affect both the population growth rate of A. colemani and its target pest. Decades of research have made A. colemani an essential part of biological control programs in greenhouse crops. Future gains in A. colemani efficacy and aphid biological control will require an interdisciplinary, systems approach that considers plant production and climate effects at all trophic levels.

population as an alternate food source for the wasp when pest aphid levels are low [16]. This provides prophylactic aphid control by providing a constant source of wasps [15]. As an aphid specialist, A. colemani is also compatible with biological control programs for other greenhouse pests [17].
Despite the low cost, ease of use, and comparatively high efficacy of this parasitoid, no biological control agent is foolproof. Failures of A. colemani have been reported many times in the literature (e.g., [15,[18][19][20]), and are an unfortunate reality for growers (Acheampong et al. [21]; S.E. Jandricic, personal observation [22]). For example, in the large floriculture production area of Ontario, Canada, 69% of growers currently use biological control as the main means of pest management. However, for many of these growers, incomplete aphid control with natural enemies is the primary reason their pest management programs still require pesticide use (J. Aalbers, personal communication [23]). So, what precipitates failures of a natural enemy like A. colemani that is generally successful?
According to ecological theory, for prey populations to be suppressed, the following model must be true: r < aP* where r = prey growth rate, a = attack rate per-predator per-unit-prey, and P* = predator abundance at equilibrium [24]. Thus, any factor which reduces parasitoid abundance (e.g., affects development time, survival, longevity, reproduction), attack rate (e.g., affects searching ability, flight capacity, host preference, prey defenses), or increases prey growth rate (e.g., affects aphid development time, survival, fecundity, defenses), could ultimately allow pest populations to grow. Greenhouses are managed, relatively closed environments compared to natural or other cropping systems. Yet, many ecological factors are still present that could negatively or positively affect the life history, fitness, or behavior of A. colemani.
Our objective was to identify the biotic and abiotic factors that may affect A. colemani efficacy by acting on "a" or "P*". Our goal is not to review all studies relating to A. colemani. Instead, we focus on factors at each trophic level that could affect efficacy of A. colemani ( Figure 1). Trophic levels include the crop plant, the host aphid, intra-and inter-guild interactions with other natural enemies, and hyperparasitoids. Abiotic factors such as temperature, humidity, air flow, and toxic chemicals in the environment can also inhibit the functioning of A. colemani in isolation or in concert with biotic effects. By identifying and synthesizing the complexities and interactions inherent in each trophic level and in the greenhouse environment, we hope to identify and prioritize the research topics which will improve the stability and reliability of A. colemani for aphid control in greenhouse crops (Table 1).  [25] Increased aphid handling time [24] Non-defensive morphological traits (e.g., PGR effects on plant architecture) Negative Host-finding is more difficult [26] May negatively affects mummy abundance, percent emergence, female parasitoid size, and sex ratio [27] Volatile organic compounds (e.g., plant species cues alone) Variable Could affect host plant choice [28][29][30][31][32] Volatile organic compounds (e.g., natal-host effects) Positive A. colemani may prefer host complex on which it is reared [29,30] Resource provisioning (e.g., Flower nectar)

Positive
Can increase fecundity, percent emergence, female sex ratio, and longevity [33] Negative Could benefit pests and hyperparasitoids NA Indirect Good plant quality Positive Can increase fecundity, percent emergence, female sex ratio, and longevity [33] Herbivore resistance traits (e.g., toxic allelochemicals) Negative May negatively affects life history traits [32] Fertilizers Positive Increase percent emergence, mummy weight, male longevity and adult size [34] Negative Percent emergence may be reduced on some species, compared to other. [35] Variance in female development time [36] Variance in number of mummies [36] Aphid hosts Direct

Aphid species Variable
Offspring survival, female ratio, and size were lower on R. padi than M. persicae, A. gossypii, and S. graminum [37] Using a poor quality aphid as an alternate host on a banker  [37] May not perform well in multi-pest environment NA Clones Variable Parasitism levels vary with clone (red clone > light green > dark green) [40] Negative Less effective against insecticide-resistance clones [41] [52,53] Can also infect parasitized aphids and reduce mummy formation and adult emergence [54] A. colemani does not detect infected hosts, so wastes eggs/energy [55] Neutral Verticillium lecanii is safe for A. colemani in mummy form (5 days post-parasitism) [54] Hyperparasitoids (e.g., Alloxysta victrix and Dendrocerus aphidum) Negative Parasitize A. colemani [56,57] In the summer, when hyperparasitoid population is high, aphid control can fail [14] Can affect parasitoid population on banker plants [20] Indirect

Direct Plant Effects
Morphological characteristics of plants, such as trichomes, spines, and waxy layers can deter herbivore colonization and feeding [75]. Such morphological defenses can also reduce biological control by reducing natural enemy colonization and foraging efficiency [25,[75][76][77][78][79]. Desneux and Ramirez-Romero [25] compared A. colemani efficiency at attacking Myzus persicae (Hemiptera: Aphididae) on Brassica napus with and without epicuticular wax. Aphidius colemani spent 20% more time grooming and had 15% more failed stings on waxy plants than on the non-waxy B. napus [25]. Similarly, glandular trichomes caused A. colemani to spend more time searching, leading to nearly ten times fewer parasitized aphids on potato species with glandular trichomes than on the species without [80]. Most of the aphid control was actually due to the effects of trichomes, with only 5.6% attributed to parasitism [80]. And, parasitoid mortality was nearly three times higher when trichomes were present than when absent [80]. Determining the release rate that provides effective control on plants with more challenging morphological characters, would allow growers to minimize their costs, while maximizing aphid control.
Other morphological traits such as leaf size, leaf texture, and number of branches, which have not necessarily evolved as plant defenses, can also affect biological control. While these traits have not been studied for A. colemani, these architectural characteristics have been shown to impede foraging efficiency of other parasitoids and natural enemies [76,[81][82][83][84]. Growers also actively alter plant architecture with horticultural techniques, including pruning [85] and by applying plant growth regulators (PGRs)-organic compounds used to modify plant growth and/or development [27]. The effect of PGRs on A. colemani has been studied, and Paclobutrazol, commonly used to create more compact, bushier plants, was shown to reduce A. colemani foraging efficiency by providing aphids with more concealed feeding locations than the sparser, untreated plants [26]. PGRs can also reduce A. colemani mummy abundance, percent emergence, female parasitoid size, and sex ratio [27]. In fact, no adults emerged from mummies on plants treated with ancymidol, a PGR used to reduce internode elongation. The mechanisms of how many PGRs affect parasitoid fitness are unclear but the direct and indirect effects of these compounds should be considered when designing or evaluating a biological control program. Whether plant architectural differences are grower-induced or due to natural variation, it is important for growers to understand the effects these changes may have on biological control outcomes. Plant architecture variation could explain potential failures in biological control between seasons and years and between plant species/cultivars. Morphology is not the only plant attribute that can directly affect A. colemani. Plants also release volatile organic compounds, which are used by natural enemies to track prey [86]. Aphidius colemani searching efficiency is highly dependent on these volatiles, and can vary with the aphid-plant species combination [29][30][31][32]56]. The host-plant species' chemical cues, alone, can also influence host selection [29,32]. For example, when given the choice between the odor of uninfested rape (Brassica napus) leaves and uninfested Chinese cabbage (Brassica rapa ssp. chinensis) leaves, A. colemani showed a preference for the rape, regardless of the plant on which they had been reared [29]. Thus, A. colemani foraging efficiency is very likely to vary across greenhouse crops due to differences in plant chemical cues and defensive volatile profiles.
Plant derived cues can also affect parasitoid host choice via natal-host effects. Using a Y-tube olfactometer, Storeck et al. [29] found that A. colemani showed a preference for the host-plant complex on which it was reared. Douloumpaka and van Emden [30] found that A. colemani emerging from aphids on plants preferred the odor of that plant species, whereas those emerging from aphids reared on artificial diet showed no preference. Additionally, parasitoids can be imprinted with the odor of plants that are near the one on which they were reared [87]. Aphidius rhopalosiphi emerging from aphids on wheat grown near tomato were imprinted with the tomato volatiles, and subsequently preferred the odor of wheat grown near tomatoes [87]. Given the observations by Douloumpaka and Van Emden [30], similar behavior can be expected for A. colemani. Though foraging experience can override this conditioned preference [29], it may still be helpful to position A. colemani banker plants near the target crop to ensure attraction to the target area.
Adult parasitoids often consume flower nectar for carbohydrates and other nutrients [47]. Such resource provisioning by plants can benefit parasitoid life history traits [33,88,89] and parasitism [88,90]. Aphidius colemani feeding on nectar from the shrub Photinia x fraseri Dress (Rosaceae) had higher fecundity, percent emergence, female sex ratio, and longevity when compared to a blank control [33]. However, nectar sources may also work against biological control programs by benefiting pests [91][92][93][94] and hyperparasitoids [89]. Although longevity of Aphidius ervi (Haliday) (Hymenoptera: Braconidae) increased in the presence of flowering buckwheat, Fagopyrum esculentum (Moench), longevity of its hyperparasitoid Dendrocerus aphidum (Rondani) (Hymenoptera: Megaspilidae) also increased, causing it to live 2.5 to 3 times longer than A. ervi [89]. Thus, resource provisioning could indirectly increase aphid abundance. Pest aphid populations can also directly benefit from resource provisioning. When oats were interplanted with fava beans, R. padi population densities nearly doubled compared to oat monocrops [91]. In this case, parasitism was inversely density dependent, and parasitoids were not able to keep up with the growing aphid population. But, in other cases, an increase in pest abundance can be accompanied by an increase in parasitism (See [93]). Consequently, the outcome of aphid biological control programs using A. colemani is likely to vary in effectiveness over time, as plants flower. Effectiveness is also likely to vary between greenhouses containing flowering plants and those with vegetative plants only. Greater research in this area is needed to determine the greenhouse crop type (flowering vs. non-flowering crops, mixtures) in which A. colemani is likely to function optimally.

Indirect Plant Effects
Plants can indirectly affect parasitoids via their quality ( Figure 1). Plants of low quality, due to low nutritional value or high plant defenses, can reduce the size, quality, or fecundity of aphid hosts. Thus, fewer or lower quality aphid hosts may be present for parasitoid reproduction [75,95]. The reverse is also true: High quality plants and resources can improve A. colemani life history traits, making them potentially better biological control agents [33]. However, it is often difficult to separate effects of host plant quality from the effects of aphid species or clone quality on parasitoid fitness. Thus, summation of indirect plant effects on A. colemani can be complicated. With this in mind, in this section, we isolated host-plant effects by highlighting papers that manipulated plant species or cultivar, while keeping the aphid species constant.
Production of toxic allelochemicals is one way plants reduce the survival and reproduction of herbivore pests [96,97]. In agricultural crops, such chemical resistance traits are often increased through plant breeding [98] and their effects on herbivores are relatively well understood [95]. Comparatively few studies have assessed the effects of herbivore resistant plants on A. colemani or other natural enemies. Kalule and Wright [32] compared the effects of three common cabbage cultivars with varying levels of resistance to the aphids Brevicoryne brassicae L. and M. persicae on A. colemani life history traits. Aphidius colemani females emerging from aphids reared on highly resistant cabbage cultivars had reduced adult longevity, though no other fitness characteristics were affected [32]. Other possible negative effects of plant allelochemicals on A. colemani, such as reduced clutch size, longevity and parasitism rates, have to be surmised from other non-aphid parasitoid species [99,100]. For example, longevity of Mexican bean beetle parasitoid larvae [Pediobius foveolatus (Hymenoptera: Eulophidae)] was negatively affected by herbivore resistance in soybeans, and was lowest on the cultivars most resistant to Mexican bean beetle hosts [Epilachna varivestis (Coleoptera: Coccinellidae)] [99]. If parasitoids are less likely to parasitize hosts on toxic plants, and have reduced abundance via decreased longevity and clutch size, they are unlikely to be effective biological control agents. This may not be a cause for concern in ornamental crop production, where varieties are not usually bred for herbivore resistance, but this is important in greenhouse vegetable crops such as tomato, pepper, and lettuce.
Fertilizers can improve plant quality for herbivores and subsequently affect the hosts' suitability for parasitoids [101]. Again, although the effects of fertilizers on herbivores have been well studied [102], little is known about their effects on parasitoids. Aqueel et al. [34] found that parasitism, percent emergence, mummy weight, male adult longevity, and A. colemani size were increased by adding nitrogen fertilizer. Other fitness indicators, such as sex ratio, were unaffected. Although this study showed positive effects of nitrogen on A. colemani, nitrogen can also increase pest hyperparasitoid population growth [103][104][105], which may lead to negative effects on biological control programs. The type of fertilizer used can also affect parasitism, as was shown by decreased parasitism of the aphid Brevicoryne brassicae on cabbage plants fertilized with organic chicken manure compared to those using synthetic fertilizer [106]. Additionally, fertilizers affect plant defensive compounds, including glucosinolates [107,108], which have the potential to affect herbivores and their natural enemies [109]. Fertilizers should therefore only be used to the extent that they improve plant growth, yield, or aesthetics, as excess fertilizer may cause more harm than good [110,111].
Plant symbionts such as rhizobacteria can affect interactions between herbivores and their natural enemies [112], by increasing crop vigor and potentially increasing plant tolerance to pests ( [113] but see [114]). Initial studies on parasitoids showed that rhizobacteria can alter the volatile composition of plants by interfering with the jasmonic-acid pathway [115]. Thus, rhizobacteria-colonized plants infested with M. persicae were less attractive to the parasitoids due to a breakdown in defensive chemical signaling [115]. Zytynska et al. [116] demonstrated a significant variation in size of A. rhopalosiphi when reared on aphids feeding on barley plants infected with rhizobacteria. However, the variation in parasitoid size was affected by both plant genotype and the aphid host, demonstrating the complexity of interactions that plant symbionts can have with the 3rd trophic level. A study assessing the effect of rhizobacteria on parasitism by A. colemani using a single aphid and plant species could inform biological control program recommendations.
Endophytes are symbiotic fungi which can protect plants from herbivores by producing toxic alkaloids in exchange for nutrition from the plant [117]. The effects of endophytes on herbivores have been studied in many systems, but their effects on natural enemies are relatively unknown [112]. No study, to date, has investigated the effects of endophytes on A. colemani. However, Härri et al. [112,118] investigated the effects of endophytes on A. ervi reared on a grain aphid. They found that plants infected with the endophyte Neotyphodium lolii negatively affected the reproductive ability of the F1 generation of A. ervi, and fewer mummies were produced [112]. Understanding endophyte effects on aphid and parasitoid population dynamics may help explain failures of A. colemani in particular crops and will improve biological control recommendations. For example, for crops with high endophyte populations, release rates and natural enemy composition may need to be modified to reduce aphids below threshold abundance.
The above examples detail cases where plants indirectly affect A. colemani through a known mechanism. However, there are also many cases where the mechanisms responsible are unknown or not yet identified. Jandricic et al. [35] observed a lower percent emergence of A. colemani reared on R. padi on oats (Avena sativa L.) compared to those reared on barley (Hordeum vulgare L.), rye (Secale cereal L.) or wheat (Triticum aestivum L.). The number of parasitoids available for biological control was therefore affected by plant species, though the reasons for this are unclear. Rearing A. colemani on different cultivars/varieties can also affect parasitoid fitness [36]. Plant genotype can cause a variance of almost 10% in female A. colemani development time, and 14% in number of mummies produced [36]. Studies like these provide information that can help growers select plants to optimize A. colemani production, and should be continued in the future. Such studies also draw attention to the unintended effects of plant breeding programs on the efficacy of biological control. While identifying the mechanisms by which the plants may be affecting the parasitoid is important, we understand that in many cases, the mechanisms may be difficult to tease apart, due to the indirect interactions between plants and natural enemies (see [98]).

Host Aphid Effects on Pest Suppression by A. colemani
The larvae of koinobiont parasitoids are intimately associated with their host [119] (Figure 1). Therefore, distribution, abundance, and performance of A. colemani in the greenhouse will depend on the quality of their host aphid. Parasitism is generally restricted to a comparatively (to predators) narrow range of species which the parasitoid has the behavioral and morphological adaptations to locate and successfully attack. Within this host range, characteristics such as instar, body size, color morph, and colony density affect pre-and post-parasitism success, as well as prey preference [120] ( Table 1). Impacts of these factors on the success of A. colemani as an aphid biological control agent are detailed below.

Host Aphid Effects on Parasitoid Development and Fitness
Although A. colemani can reproduce in over 41 aphid host species [2], it is primarily used as a biological control agent for M. persicae and A. gosypii, two of the four ubiquitous pest species in greenhouses. Aphidius colemani life history traits can differ by aphid host [37], which may affect its population size and ability to suppress pests. A well-studied example of how aphid host affects wasp fitness is with Rhopalosiphum padi compared to other aphids. Despite its common use in banker plant systems for A. colemani, several studies have shown that R. padi is actually a relatively poor host for this parasitoid [5,37,38]. Parasitoid offspring survival, proportion of female offspring, and size were significantly lower on R. padi than M. persicae, A. gossypii, and S. graminum [37]. Similarly, Bilu et al. [5] suggested that R. padi was the least suitable host for A. colemani compared to A. gossypii and M. persicae, as determined by offspring body size. Reduced female offspring size and offspring survival was also found by Prado and Frank [38] when using R. padi compared to M. persicae as host. Smaller parasitoids can carry fewer eggs [121], may be less efficient at host searching [122], and will fly shorter distances than larger parasitoids [123]. Taken together, these effects make it seem like R. padi is an extremely poor choice for banker plant systems in greenhouses, however, in this case, relatively lower fitness on the alternate host is a benefit. This is because A. colemani females tend to leave R. padi on the banker plants to forage for better hosts (e.g., pest aphids) on the crop plants [38]. While a number of recent studies have investigated the use of banker plants at suppressing a single pest species (e.g., van Driesche et al. [15]; Prado and Frank [38]), using banker plant systems to suppress multiple pest aphids in a greenhouse is yet to be investigated. This could be especially important given that studies with other aphid biological control agents have suggested that prey preferences can lead to differential control in the greenhouse [124].
Aphidius colemani fitness can also be affected by aphid clonal lines within the same species-an important consideration given that aphids reproduce parthenogenetically in greenhouses. A few key studies have suggested that different secondary endosymbionts in aphids can confer resistance to parasitoid attacks by causing larval mortality of developing parasitoids [125,126]. However, to our knowledge, only one study has investigated the effects of aphid endosymbionts on A. colemani parasitism. Clones of M. persicae and Aphis fabae infected with the endosymbiont Regiella insecticola were strongly resistant to A. colemani, as indicated by lower number of mummies [39]. When presented with aphids with and without such endosymbionts, the closely related species A. ervi was equally attracted to both, suggesting that parasitoids may be wasting eggs and energy parasitizing resistant aphids [125]. Larval mortality, and wasted energy and eggs can reduce parasitoid abundance and efficacy. Although a logical alternative to parasitoids in this scenario would be predators, consumption of aphid secondary symbionts can reduce predator survival as well [127,128]. Thus, if growers notice a lack of efficacy of biological control agents for an emerging aphid population, the presence of resistance-conferring endosymbionts may be to blame, and pesticides may have to be used (but see Section 5). Currently, studies are investigating the use of antibiotics to eliminate aphid endosymbionts as a means of improving aphid control [129].

Host Effects on Aphid Acceptance and Suppression by A. colemani
Biological control outcomes for aphids can be significantly affected by prey preference of the natural enemy involved (e.g., Bergeson and Messina [130]). Although A. colemani parasitizes many aphid species, it seems to have an apparent genetic bias in attack performance, leading it to attack A. gossypii more readily than M. persicae [1]. This bias was strong enough to cause A. colemani which had been reared on M. persicae for multiple generations to switch hosts when given the option [1]. Similar observations made by Sampaio et al. [131] and Bueno et al. [64], and anecdotal observations made by biological control specialists in Canada confirm that A. colemani seems to better control populations of A. gossypii than M. persicae (G. Murphy, personal communication [132]). There is no simple explanation for this apparent preference, however. No fitness benefits for A. colemani emerging from A. gossypii have been noted (but see [133]-potential benefit for male mating ability), and in fact, female size for A. colemani is smaller when emerging from A. gossypii compared to M. persicae [37]. As Benelli et al. [3] suggest, however, preference may not be precisely matched with host quality variables. Further studies are needed to clarify both preference and performance of this parasitoid on greenhouse aphid pests, especially if it is being marketed to growers as an effective solution for both M. persicae and A. gossypii. Additionally, studies should determine how A. colemani reacts in the presence of suitable (e.g., crops infested with both M. persicae and A. gossypii) and/or unsuitable aphid host species, as studies with other aphid biological control agents in greenhouse crops have suggested that such multi-pest environments can lead to differential control [134,135].
Parasitoids use aphid physical, chemical, and behavioral attributes to select a host [120,136]. Thus, variation between aphid clonal lines within a same species can influence aphid acceptance and suppression by A. colemani [40,120,137]. For example, A. colemani performance varied when presented with three M. persicae clones (light green, dark green and red) collected from greenhouse pepper plants [40]. Aphidius colemani parasitized more red clones than light green and dark green clones, with dark green clones least parasitized. When assessing the long-term (4-week) effects of these interactions on aphid suppression, Gillespie et al. [40] found that parasitism reached 100% in cages with the red and light green clones, but only about 50% of the dark green clones. The authors suggest that the differences in aphid suppression are related to the aphids' life history traits. Dark green clones (which lacked expression of a particular esterase involved in insecticide resistance) had a higher reproductive rate, and thus may have been least vulnerable to suppression by A. colemani [40]. Different clones of M. persicae have also been associated with different levels of insecticide resistance [137,138], and A. colemani is less effective against certain insecticide-resistant clones [138]. This reduced control could be due to differential survival and development rates in the insecticide-resistant aphids [138]. Though this may be worrisome for growers, insecticide-resistant clones may have fitness tradeoffs including reduced reproductive rates, and reduced response to alarm pheromones, which could lead to selection against these clones in the absence of insecticides [138]. Furthermore, combining aphid predators, such as coccinellid species, with kaolin (an insect repellent) have shown promising results for the control of insecticide-resistant M. persicae [139]. Altogether, this provides an argument for relying on biological control as the primary means of aphid control. Introducing pesticides can interact with both aphids and their natural enemies in unexpected ways, and complicate biological control-based integrated pest management (IPM) programs for other pests in the same system.
Differences between aphid instars can also affect A. colemani efficacy [43,140]. In terms of pest control, the theoretical ideal would be for A. colemani to prefer 1st and 2nd instars, as this would kill pest aphids before they reproduced. If later instars (3rd or 4th) are attacked, both M. persicae and A. gossypii are able to reach adulthood and produce a limited number of offspring before becoming mummies [41]. Similarly, Lin and Ives [43] noted that although they had reduced fecundity, later-instar Aphis glycines were able to reproduce after reaching adulthood for up to three days following parasitism. Overall, results from studies on the actual instar preference of A. colemani are widely variable. This suggests that A. colemani preference for specific instars varies with aphid species and/or host plants. For instance, Perdikis et al. [41] showed that A. colemani preferred to parasitize 1st and 2nd instars of A. gossypii and M. persicae over the older, larger hosts. Martinou and Wright [42], however, observed a preference for intermediate instars (2nd and 3rd) of M. persicae reared on pepper (Capsicum annuum L.), with the preference shifting to 1st-3rd instars when M. persicae was reared on aubergine. Such differences in instar preference may be due to differences in instar size between species [121], parasitoid size, host defensive behaviors, and immune responses [141].
Older and larger aphids can better defend themselves from parasitoids [142]. These defenses, which include kicking, dropping, shaking their body, and running away [142] can increase parasitoid handling time and risk of injury [44,[143][144][145]. As parasitoids alter their host-selection behavior in relation to their own body size, large and small parasitoids of the same species prefer larger and smaller aphids, respectively [144,146]. The inverse relationship between parasitoid size and handling time means that smaller aphid hosts are more often parasitized by smaller Aphidius individuals (because they have not yet developed effective defenses; [144]), while larger A. colemani have a wider range of accepted instars [44,45]. A strict preference for specific instars could affect the potential of the parasitoid to exploit aphid populations that differ in size structure. Thus, A. colemani ability to parasitize a wide range of host instars may be beneficial in biological control in greenhouses, where multiple crops and aphids combinations co-occur.

Host Density Effects on Aphid Suppression by A. colemani
Aphid density can influence parasitoid searching time and number of patch visits [4,46]. Stadler and Volkl [46] found that density of the aphid Pentalonia nigronervosa Coq. (Hemiptera: Aphididae) was positively correlated with the amount of time A. colemani spent foraging, which, in turn, was positively correlated with number of ovipositions [46]. Aphidius colemani showed similar behavior in a different study with A. gossypii, suggesting this is independent of aphid species [4]. In this study, A. colemani also arrived earlier and searched longer on heavily infested than lightly infested leaves, but did not always discover hosts on low-density leaves [4]. This led the authors to suggest that A. colemani exhibits a type III functional response on low-density patches (linear increase in parasitism with increasing host density, until a maximum is reached), while exhibiting a type II functional response (decreasing parasitism with increasing host density) at higher aphid densities. However, the converse was found by Byeon et al. [147]. Differences in these findings could be explained by different leaf sizes or aphid species used. A more consistent comparison of A. colemani functional response on different aphid species on the same sized leaf could provide insight into how well A. colemani will react to differentially infested plants. More importantly, it may provide a clearer picture of the efficacy of A. colemani at low densities. Currently, A. colemani is often recommended for aphid control at low densities simply because it is cheap [51], rather than because of its effectiveness.
Honeydew production by aphids could also explain some of the differences in A. colemani response to varying aphid densities, as Aphidius wasps are known to use honeydew as contact kairomones to locate host aphids [148]. The presence of honeydew can increase the amount of time Aphidius nigripes (Hymenoptera: Braconidae) dedicates to searching for aphids, and host-searching is focused closest to the honeydew [148]. As honeydew concentration increases with aphid densities, it is likely that A. colemani will focus initial efforts on highly infested plants, rather than plants with few aphids. Honeydew, like flower nectar, can also provide non-prey food for parasitoids [47]. Host species can affect honeydew quality for A. colemani, which can affect its longevity [47]. Aphidius colemani lived nearly 9 days longer when feeding on honeydew produced by M. persicae than it did when feeding on honeydew produced by B. brassicae on the same plant [47]. Aside from investigating the life history benefits of honeydew on parasitoids, few studies have actually assessed the impact of honeydew on biological control [47]. Further research is needed to determine if, and how honeydew affects the outcome of biological control.

Considerations at the 3rd and 4th Trophic Levels: Effects of Competition, Intraguild Predation, Hyperparasitism and Multiparasitism
Interactions between A. colemani and other organisms at the 3rd trophic level have the potential to affect aphid control outcomes (Figure 1; Table 1). One consideration is competition with other parasitoid species for hosts. Parasitoid abundance can be reduced through direct competition for prey, or multiparasitism (multiple parasitoid species laying eggs in the same host). In the latter case, one larva usually outcompetes the other (e.g., [149,150]) as in the case of A. colemani and Lysiphlebus testaceipes (Hymenoptera: Aphididae) (also commercially available), where L. testaceipes usually "wins" [48]. Sampaio et al. [48] suggest this may result in the displacement of A. colemani if the two are used together for A. gossypii control at low aphid densities. However, displacement of A. colemani by another parasitoid species has yet to be shown in a greenhouse study. Overall, displacement is unlikely to be a concern for growers using high, weekly releases of various parasitoid species. However, it may become an issue if banker plants are used as the main source of wasps, as there have been anecdotal reports of A. ervi taking over banker plants intended for the open rearing of A. colemani [124]. This would likely go unnoticed by the grower and outbreaks of M. persicae and A.gossypii may follow.
As concerns have recently been raised that A. colemani shipments might actually be a mix of closely-related species, analyses of species competition within these populations is an important avenue for future research. Tomanović et al. [151] used morphological and genetic analysis to reveal that species previously synonymized with A. colemani are likely separate species. Thus, parasitoids sold as "A. colemani" may actually be a complex of A. colemani, A. platensis, and A. transcaspicus (though work by Frewin et al. [152] does not support this). If some commercial suppliers have contamination with multiple parasitoid species, there is likely to be subtle physiological or behavioral differences between these species, as well as host range differences [151]. Thus, aphid control by commercial populations of this natural enemy may be affected by the ratio of one species to another. Further, confirmation of the identity and characteristics of commercial populations of A. colemani is needed to limit any unpredictability in control that may result from mixed-species populations.
Aphidius colemani also interacts with predators in the greenhouse. These can be predators which are purposefully released for aphid control (e.g., lacewings, ladybird beetles, predatory bugs like Dicyphus spp.) or naturally occurring (e.g., syrphid flies). Many predators readily consume aphids that have already been parasitized by Aphidiine wasps (e.g., [50,[153][154][155]) obviously reducing parasitoid abundance. Though A. colemani does not appear to avoid predator-infested plants like some parasitoid species [49,86], predator cues may somewhat reduce parasitism rates by Aphidius species [50,156]. Despite these effects, few field-crop studies have demonstrated disruption of biological control when a generalist predator and aphid parasitoid co-occur (see [157,158]). Far mo re agricultural studies have revealed positive effects of natural enemy diversity on aphid control (e.g., [154,[159][160][161][162][163]). This is thought to be due largely to both additive and synergistic effects from a diversity of natural enemies [51,154]. This positive trend appears to hold up the greenhouse. Snyder et al. [164] and Bilu and Coll [50] both show that the use of a coccinellid predator with an aphid parasitoid improved aphid control over time. Messelink et al. [51] demonstrate improved aphid control with A. colemani and the addition of the generalist predatory bug Orius majusculus in sweet pepper. The combination of predators with A. colemani may be especially useful when multiple aphid species are present in the greenhouse (see Section 3.1), when other soft-bodied greenhouse pests co-occur (see [165], or on complex plant species [26,27]).
The combination of A. colemani with entomopathogenic fungi for aphid biological control can be problematic. The most popular commercially-available fungal products, based on strains of Beauveria bassiana (Balsamo) Vuillemin, can infect and kill adult A. colemani at high rates (>55% of the population) in the lab [52] and the greenhouse [53]. This fungus can also infect already parasitized aphids, reducing mummy formation and adult parasitoid emergence by up to 83% in closely related wasp species [166]. Similar results have been seen with other fungal species, such as Verticillium lecani (Zimm.), where A. colemani emergence was reduced by 90%-100% [54]. Intra-guild interactions between parasitoids and fungi are asymmetrical, and the "winner" is determined by timing of oviposition/infection [167]. Generally, the fungi will outcompete parasitoids unless wasp oviposition takes place at least 4 days before infection [54,166,168]-meaning sprays of entomopathogens are generally only compatible with A. colemani when it is in the mummy form [54]. Further, Aphidius species cannot readily detect entomopathogenic fungi in an aphid host, only rejecting an infected aphid if it is sporulating [55].
Currently-available fungal products, with the exception of Verticillium species (not yet available in North America as of 2015) generally have low pathogenicity against the top greenhouse pest aphids (see [169]). Given the side-effects on parasitoids, the use of current fungal-based products should be avoided in IPM programs for aphid control. However, the reality in the greenhouse environment is that fungal products will regularly be present, as they are highly effective for other greenhouse pests such as whiteflies (Aleyrodidae) and thrips (Thysanoptera) [170,171]. As much as possible, growers need to consider their entire biological control program, even when applying reduced-risk pesticides. Thus, timing of the application of entomopathogenic fungi should be made to ensure most of the A. colemani population is in mummy form [172] or re-releases of this parasitoid post-spray will be a likely necessity.
Hyperparasites are parasites that develop on or in another parasite, killing it in the process. A variety of hymenopterous hyperparasites-known as hyperparasitoids-attack Aphidius species and can considerably reduce their numbers. Studies have shown significant non-consumptive effects of hyperparasitoids on parasitic wasps, including deterrence of Aphidius foraging in patches with hyperparasitoid volatiles (e.g., Holler et al. [173] and Petersen et al. [174]). The most common hyperparasitoid species of A. colemani include Alloxysta victrix (Westwood) (Hymenoptera: Cynipidae), the most common hyperparasitoid of A. colemani in greenhouse pepper crops in England [56], Dendrocerus aphidum (Hymenoptera: Megaspilidae), the most abundant species in pepper greenhouses in the Netherlands [57], and Dendrocerus carpenteri, the dominant hyperparasitoid in pepper greenhouses in British Columbia, Canada [21]. A plethora of other hyperparasitoid species are also reported from around the world [14,57,175,176]. The effects of hyperparasitoids on biological control in greenhouses are not well studied, but it is thought that some hyperparasitoid species may be more successful in the greenhouse than the field due to temperature differences [21]. Additionally, general trends in hyperparasitoid density based on growing seasons are suggested in the literature. van Steenis [14] found that hyperparasitism did not interrupt aphid control in the spring in the Netherlands, when hyperparasitoid density was relatively low. However, in the summer, the ratio of hyperparasitoids to A. colemani was much higher, and aphid control failed. A similar study from Japan showed a high rate of hyperparasitism on A. colemani banker plants in late spring, across 4 years of study in commercial greenhouses [20]. In some years, hyperparasitism rates reached almost 100% on banker plants by June. Acheampong et al. [21] found that hyperparasitism rates generally peaked from June-August in British Columbia, Canada. Together, these studies raise concerns over the efficacy of releases of A. colemani in summer months especially, where high rates of aphid population increase, temperatures unfavorable to A. colemani (i.e., ≥30 °C), and higher numbers of hyperparasitoids may act together to derail aphid control [14,63,177]. However, some operations (especially organic growers) may observe hyperparsitoids disrupting aphid biocontrol as early as the spring if the hyperparasitoids are able to overwinter in the greenhouse [57].
The reliability of A. colemani banker plants as a biological control strategy is also affected by hyperparasitoids [20]. Hyperparasitism may be largely responsible for the 30% failure rate of banker plants to control aphids, as reported by growers in Japan [20]. It is likely that long-term use of the same banker plants by Nagasaka et al. [20] (who only re-seeded every 3-4 months) contributed to high levels of hyperparasitoids over time due to their longevity (ca. 4-6 months for adults of some hyperparasitoid species; [178]). For best performance, we recommend the replacement of banker plants every 3-4 weeks, to ensure healthy populations of banker-plant aphids [35], reduce the incidence of mildew like molds [35,179], limit space taken up by banker plants (T.J. McClure and S.D. Frank, unpublished data [180]) and remove reservoirs of hyperparasitoids. However, replacement rates may need to be even higher in summer months. Some IPM consultants in Ontario, Canada have found that the only way to prevent aphid problems from June to August is to replace banker plants every 2.5 weeks (M. Short, personal communication [181]), or even remove them entirely during this time-period and rely on pesticide sprays for aphid control. Further studies are needed to determine: (i) the timing of hyperparasitoid infestations in different regions; (ii) how different maintenance schemes for banker plants can potentially mediate risk from these organisms; (iii) if there are ways to provide parasitoids with refuges from hyperparasitoids within the greenhouse; and (iv) if the addition of another biological control agent during the summer can stabilize aphid control by A. colemani in the face of hyperparasitism.

Abiotic Considerations
Many abiotic factors are present in greenhouses that can act alone or interact with biotic factors to affect biological control programs ( Figure 1; Table 1). Pesticides are one of the most important abiotic factors in agricultural systems. Non-target effects of pesticides on A. colemani have been well studied in terms of direct contact activity (e.g., [58][59][60][61][62]), and there is a clear need to evaluate novel chemicals as they come on the market. Residual toxicity of pesticides (i.e., the period after application when they still pose a threat) on A. colemani is also recognized as an important factor (e.g., [59,68,69]), with most growers understanding that chemicals with shorter residual times are generally safer for natural enemies (S.E. Jandricic, personal observation [22]). However, A. colemani is more susceptible to insecticide residues on the foliage than many other natural enemies used in greenhouse biological control [68], and can continue to be exposed to systemic insecticides through aphid hosts, honeydew and nectaries [182]. Thus, greenhouse IPM programs for aphids would benefit from research to specify safe release intervals following application of different pesticides for A. colemani.
Open-access databases summarizing information of contact effects on A. colemani and persistence of chemicals can be found at http://www.biobest.be/neveneffecten/3/none/ and http://side-effects.koppert.nl/. However, these databases should not be used to find pesticides to regularly spray in conjunction with Aphidius. Rather, they should serve as a guide when selecting pesticides to use in "hot spots" only (smaller areas of aphid outbreaks), in order to provide A. colemani with refugia from pesticide effects. This is important, because even pesticides classified as "harmless" by compatibility databases have the potential to cause up to 25% mortality of A. colemani populations. "Slightly harmful" pesticides can cause up to 50% mortality. Though these categories concur with IOBC guidelines for pesticide compatibility with natural enemies [183], it is likely that these mortality levels represent a significant loss of protection against aphids, especially since natural enemy releases at effective, continuous rates are key to aphid biological control [10]. Similarly, pesticides can have important "indirect" effects (i.e., sub-lethal or latent effects [71]) on natural enemies (see [184]), but these have not yet been clearly factored into open-access databases. Indirect effects of pesticides on A. colemani and other Aphidius spp. include decreased attraction to aphids on treated plants and re-invasion of insecticide treated areas, reduced foraging, fecundity, oviposition, increased development time and a strongly male-biased sex ratio [59,62,70,[185][186][187][188][189]. Private databases (e.g., ipm-impact.com) contain some of this information, but their subscription costs are likely prohibitive to growers and smaller IPM consulting companies.
Aphidius colemani populations are heavily influenced by temperature, since this directly affects wasp development time (time from egg to adult emergence). Though reports of optimal development temperatures for A. colemani vary in the literature [17,[63][64][65], development is roughly fastest between 22 °C and 28 °C. Variations in populations are likely due to host plant effects (as demonstrated by Zemek et al. [190], physiological differences in different host aphids at different temperatures, biotype effects, or even variation in wasp source population [191]. At extremely high temperatures (30 or 31 °C), larvae of A. colemani generally cease development [63,64], meaning that the efficacy of this natural enemy will be greatly reduced in summer. Even short periods of high heat can negatively affect populations of Aphidius spp. One hour spent at 36 °C decreased A. avenae populations by 50% and reduced fitness traits of the survivors [192]. Populations of A. gossypii and M. persicae can still increase at 30 °C-33 °C [67,73], making aphid biological control difficult at high temperatures.
Aphidius colemani can develop at temperatures as low as 10 °C and will still attack aphids at 10-15 °C [63,65]. But, A. colemani handling rates decrease linearly with temperature, meaning that 40%-50% fewer aphids are handled at 10-15 than at 20 °C [193]. Given that intrinsic rates of increase of aphids are also much lower at low temperatures [67], successful control of aphids in cooler months is still possible, however, and has been successfully demonstrated by Kim et al. [194] in sweet pepper.
Such studies of development times and thresholds do not allow us to predict natural enemy population growth rates at different temperatures, however. This is because a lone metric such as "fast development" can have fitness trade-offs, such as reduction in parasitoid size [66], which also often corresponds to a decrease in female adult longevity and fecundity [66]. Temperature-based population models, e.g., calculations of intrinsic rate of increase (rm), are an improvement, since they take into account survival, attack rate, fecundity, and sex ratio along with development time to estimate population growth rate [195]. However, few studies of rm have been conducted for A. colemani, and only at a few temperatures (see [17,190,196]). Yet, even these models are limited, as greenhouses do not function at a steady state. Fluctuating daily temperatures and other environmental conditions (see below) are complicating factors in insect development and fecundity (e.g., [67]). Further, insect behavior can be affected by changing environmental conditions, e.g., by shifting activity to parts of the day that are more suitable or increasing patch residence times [197]. Thus, to have any predictive power, more comprehensive models conducted under actual greenhouse conditions are needed. These should include life history studies on the parasitoid wasp, as well as the prey in the presence of the wasp (as in [6]). This should ideally be done on several economically important crop plants, possibly with different architectures (see Section 2.1), grown at different times of year. The results of such studies could potentially be used to predict optimal seasonal usages across plant types.
An unknown environmental effect on A. colemani efficacy is the use of "dynamic climate regimes". This novel method of plant production takes advantage of the adaptability of plants by providing high temperatures during the day, and low temperatures at night (rather than using constant pre-sets) to save energy while still providing optimal long-term average temperatures for plant productivity [198,199]. Recent studies show that this environment management style can reduce heating costs by 10% without negative effects on production [199]. However, this method can also increase A. gossypii populations compared to traditional temperature regimes, due to higher short-term mean temperatures favoring aphid development [72]. How A. colemani responds in dynamic versus constant climate greenhouses remains to be seen, but there are several possible negative effects of this strategy. For example, maximum daytime temperatures may exceed the upper development threshold for A. colemani, or reduce the ratio of oviposition activity to other activities, leading to instability in parasitoid-host population dynamics.
Humidity also has important effects on natural enemies in the greenhouse. For example, it can strongly affect Aphidoletes aphidimyza (Diptera: Cecidomyiidae) fecundity [200], as well as hatching and predation rates of predatory mites [201][202][203]. Flight and dispersal of Orius are highly contingent on temperature and humidity combinations [204,205]. Knowing this, it is surprising that only two papers in the literature have directly addressed humidity effects on an Aphidius species. Yan and Chen [206] showed that humidity levels for optimal eclosion and adult longevity for Aphidius gifuensis were between 75%-85% RH. At humidity levels above and below this range, eclosion and longevity fell. However, Fink and Volkl [207] did not see differences in foraging abilities (residence times, time allocation, or oviposition success) for Aphidius rosae when exposed to low humidity in the field (ca. 40% RH). Given the paucity of studies and the disparity between them, this is certainly an important avenue for further investigation. Growers are able to regulate RH to some degree, and currently do so to optimize transpiration or avoid condensation on plants to minimize plant diseases. Control of humidity to optimize A. colemani performance may turn out to be an important mediator of aphid biological control. Similarly, effects of "precipitation" in the greenhouse on A. colemani, e.g., from misting systems, are currently overlooked. However, simulated light rain reduced foraging of A. rosae by >80%, and increase cleaning time by >15% post-rain [207]. Precipitation may also serve to removing searching cues (honeydew) for A. colemani.
Recently, new lighting techniques have been a research focus in the greenhouse industry. Offering lower energy costs or improved plant growth, manipulations such as light emitting diodes (LED), photoselective screens (e.g., UV absorbing), and changes in photoperiod may soon become the norm (see review [208]). Given the relative novelty of these techniques, our understanding of how they may affect both pest and natural enemy biology and behavior is far from complete. Both positive and negative effects on various pests and natural enemies have already been noted (see review [209]). For example, several processes that reduce UV-levels are likely improve aphid control, as they have been shown to reduce aphid attraction, dispersal, reproduction and virus transmission [210][211][212]. However, only one study to date has confirmed that reduction of UV light did not interfere with the performance of A. colemani [74]. Other lighting techniques and regimes (e.g., changes in day length, light intensity, light quality, etc.) also need to be tested to avoid potential disruption of current aphid biological control programs. It is naïve to think that changes in lighting technologies would require no adjustment in pest management, especially since they can alter plant nutritional quality, physical or chemical defenses, and/or volatile emissions or profiles [213,214], which can directly or indirectly affect biological control (see Section 2).
Lastly, wind speed created by cooling fans in greenhouses is also a factor to be considered when using A. colemani. Wind speeds of just 2 m/s (4.5 mph) can reduce oviposition and increase resting behavior of the parasitoid Aphidius rosae [207], and winds of >0.5 m/s (0.1 mph) can interfere with male mating flights of Aphidius ervi and Aphidius nigripes [215,216]. A well designed greenhouse has fans that produce wind speeds of 0.9-1.3 m/s (2-3 mph) [217], but many greenhouses operations may be even higher than this in actuality. For the aphid biological control agent A. aphidimyza, commercial suppliers often recommend that the fans be turned off for a period of time after release, as wind can interfere with their settling and oviposition behavior [124]. Thus, the potential for improved host seeking and oviposition with a period of no wind should perhaps be investigated for A. colemani.

Conclusions
For over four decades, A. colemani has been used for biological control of aphids. This parasitoid is among the most cost effective, and successful biological control agents in greenhouses and is widely used worldwide for controlling M. persicae and A. gossypii. We identified ecological interactions at each trophic level that can affect A. colemani efficacy by affecting either abundance or parasitism rate. Effects present at the 3rd and 4th trophic levels (specifically, the presence of fungal-based insecticides and hyperparasitoids), can clearly have strong, negative effects on A. colemani populations. Though these effects can lead to parasitoid population crashes, and potential loss of aphid control, their effects are acute and short term. Proper timing of entomopathogenic fungi sprays and careful management of banker plants could lessen their effects. On the other hand, low quality plants, due to poor nutrition or strong plant defenses, can consistently reduce foraging efficiency and fitness of A. colemani. Such effects at the 1st trophic level seem to be the greatest threat to the efficacy of A. colemani for aphid control, as they are more chronic, and are more difficult to manage than those posed at other trophic levels. This is because plant traits are often not within a grower's control (frequently dictated by market demands and limited by current breeding programs). Further, current focus is on optimizing growing conditions for the plant in the most economical way; not necessarily on the wider effects of these production conditions on other trophic levles.
We identified many interactions at each trophic level for which more research is needed. At the first trophic level, more research on how trichome type and density affect A. colemani foraging will help predict biological control success in different crops. Further investigation of the net effects of grower practices (e.g., fertilizers, plant growth regulators) on aphid abundance is also necessary, given that such practices can affect plant quality for aphids and parasitoids, and ultimately affect the outcome of biological control programs. Of course, it would be impossible to conduct experiments on each plant species and variety available under different fertilizer regimens. Instead, we need enough research on these interactions to predict general consequences, and make informed recommendations that are applicable to a wide range of crops.
The combined effects of greenhouse temperature, humidity, lighting, and day length on A. colemani, and on biological control in general, are also poorly known compared to our knowledge of their effects on plant production [213]. Horticulturalists would never make plant culture recommendations without knowing optimal growing conditions for the plant. Yet, holistic, optimal environmental conditions for effective biological control by a specific beneficial organism are usually not known, and thus not included in recommendations by companies or extension professionals. Basic research on temperatures at which biological control agents die or become inactive is often available, and, in some cases, even the optimum range for parasitism rate is known. However, this information has limited usefulness in a vacuum. Future research needs to more clearly recognize that temperature and other environmental factors also change pest population growth rates in concert with parasitoid life history and behavior. Again, we need to have enough research to predict the net effect of increasing or decreasing temperature (or other conditions) on pest abundance in the presence of a natural enemy, not just on isolated factors such as parasitism rates. Further, research surrounding abiotic factors on aphids and their natural enemies should include both research on the greater greenhouse environment, as well as how this relates to the microclimate within the plant canopy where arthropods organisms generally function [213].
Genetic analysis of commercial and wild A. colemani is an important avenue for research. This will be important to reduce inbreeding, and to develop "strains" that are easy to rear but without negative trade-offs such as small size. Commercial insectaries may want to consider capitalizing on traits of wild populations, as potentially useful variations in life history characteristics have been identified in parasitoid populations outside of greenhouses (see [191]). Further, confirming the identity and characteristics of commercial populations of A. colemani should be an industry priority. Should A. colemani shipments prove to be a mix of cryptic species, then any research predictions based on the biology or behavior of a single population would be hopelessly muddled.
Since the 1970s researchers and growers have learned much about A. colemani. Our review demonstrates the abundance and complexity of interactions that could affect A. colemani efficacy. Many of these interactions are driven by plant selection, grower inputs, and the abiotic environment which directly and indirectly affect plant quality for pests and parasitoids. The research needed to advance our understanding and recommendations will require an interdisciplinary approach rather than entomologists working in isolation. For example, working with plant breeders could result in plant varieties that strike a balance between consumer-valued traits, pest resistance traits, and traits that are favorable to biological control. Likewise, working with horticulturalists could help guide plant growth regulator and fertilizer recommendations toward those that reduce positive effects on pests or negative effects on parasitoids. Collaboration between researchers, extension personnel, the pest control industry, and growers will be essential to advance A. colemani efficacy and support the growing use of biological control in greenhouses.
thank Brian Spencer (Applied Bio-nomics Ltd.) and Gerben Messelink (Wageningen University) for their input on various questions. This work was supported by grants from the American Floral Endowment and Fred C. Gloeckner Foundation to SDF, a USDA SARE grant to SGP, and by the North Carolina State University, Department of Entomology.

Author Contributions
Sara G. Prado, Sarah E. Jandricic and Steven D. Frank conceived of and wrote the manuscript.