Mutualisms and Population Regulation: Mechanism Matters

For both applied and theoretical ecological science, the mutualism between ants and their hemipteran partners is iconic. In this well-studied interaction, ants are assumed to provide hemipterans protection from natural enemies in exchange for nutritive honeydew. Despite decades of research and the potential importance in pest control, the precise mechanism producing this mutualism remains contested. By analyzing maximum likelihood parameter estimates of a hemipteran population model, we show that the mechanism of the mutualism is direct, via improved hemipteran growth rates, as opposed to the frequently assumed indirect mechanism, via harassment of the specialist parasites and predators of the hemipterans. Broadly, this study demonstrates that the management of mutualism-based ecosystem services requires a mechanistic understanding of mutualistic interactions. A consequence of this finding is the counter intuitive demonstration that preserving ant participation in the ant-hemipteran mutualism may be the best way of insuring pest control.


Introduction
Providing food for the world is fundamentally an ecological issue given that agricultural ecosystems follow the same ecological laws as all other ecosystems [1]. One ecological process that is frequently acknowledged as central to ecosystem organization is the formation of mutualisms. Mutualisms are ubiquitous in both natural and managed ecosystems and are especially integral to global agriculture, since crop species are often dependent on mutualistic interactions such as pollination [2], nitrogen fixation [3], and mycorrhizal associations [4]. A less advertised, but extremely important mutualism in agroecosystems is between ants and their hemipteran partners [5,6], usually seen as detrimental to crops since hemipterans frequently reach pest status and threaten crop production [7,8].
While hemipterans may appear to benefit from ant presence, the mechanism and degree of mutualistic interaction is widely contested [9]. In laboratory settings without predators, the direct benefit of ant presence has been demonstrated to be mediated by the removal of sooty mold which negatively affects hemipteran growth rates [10,11]. In the presence of predators, however, it has long been argued that the indirect release from predation pressure is the primary driver of the ant-hemipteran mutualism [12,13,14,15]. Assuming the latter, it is frequently presumed that hemipteran population control in agroecosystems necessitates elimination of ants.
If an alternate direct mechanism for the mutualism were operative, however, such a proposal could be counter-productive. Assume, as is usually the case, that ants and their mutualistic partners have an aggregated distribution and therefore are concentrated in patches within the agroecosystem space [16]. If, as is frequently assumed, ants provide the hemipterans protection from their specialist predators (i.e. indirect mutualism), then specialist predator populations will not be able to build up sufficiently large densities to operate as effective biological controls in the general agroecosystem space (Fig. 1A). However, if the mechanism of the mutualism is direct, that is, not through interference with predation but rather through improved hemipteran growth and reproduction, then the patches with the mutualism could generate larger densities of specialist predators, effectively becoming a spatial reservoir for the predator (Fig. 1B). In other words, an indirect mechanism of an ant-hemipteran mutualism would interfere with natural control of the pest (Fig. 1A) while a direct mechanism could, in some cases, enhance it (Fig. 1B). Thus, with regard to the ecosystem service of pest control, the mechanism of the mutualism matters.
Despite the ubiquity of ant-hemipteran mutualisms, the specific mechanism by which ant-tended patches support both specialist predator and hemipteran populations has rarely been investigated in the field. In this study, we examine the mechanism of mutualistic interaction between the green coffee scale, Coccus viridis, and the hemipteran-tending ant, Azteca instabilis, in a coffee growing region in Chiapas, Mexico. In this study system, we know that the main predator of C. viridis is a myrmecophylous beetle, Azya orbigera (Coccinellidae) which specializes on C. viridis [17]. Like many other coccinellid beetles [18,19,20], the larvae of A. orbigera are poor dispersers but are immune to ant attacks and thus experience inadvertent protection from their natural enemies in the presence of ants [21,22]; in contrast the adult beetles are strong fliers and have the capacity to locate their prey even in small densities [23], but are vigorously attacked by the ants. Thus, adult beetles must confront physical attack by ants so as to deposit eggs in the vicinity of hemipteran insects, a task accomplished through complex behavioral mechanisms not related to the present communication and reported elsewhere [24]. In this study, we examine the role of the main specialist predator, adult A. orbigera beetles, in mediating hemipteran population dynamics; we additionally provide greater detail of this particular system to emphasize the key aspects relevant to the current study (Fig. 1C).
In the described research, we examine the population dynamics of ant-tended and non-tended hemipterans and investigate the potential of ant-tended patches to serve as specialist predator reservoirs for the remaining portions of the agroecosystem. Utilizing this model system, we posit that the mutualism mechanism itself critically underlies local and landscape-level pest population dynamics.

Methods
The study was conducted in an organic coffee farm managed as a commercial polyculture [25], with coffee bushes growing under a canopy of species-rich shade trees (Finca Irlanda, located at 15u 119 N, 92u 209 W). The ant-hemipteran mutualism forms a clustered distribution, with A. instabilis occupying 3% of nest-sites (trees) [16], and hemipteran populations found throughout the landscape, but in greatest population density in the presence of ants [26]. Given the uniform management practices throughout the plot (in terms of light cover and soil fertility) and the lack of significant effects of abiotic factors (tree species, size, and canopy Figure 1. Diagrammatic representation of the two hypotheses describing the mutualism mechanism and details of the study system. (A) Indirect mutualism situation where the mutualism patches do not generate spatial reservoirs of the predator, and (B) direct mutualism situation, where the mutualism patches generate spatial reservoirs. Solid black lines indicate strong effects, and black dotted lines indicate weak effects. An arrow indicates a positive effect on the target organism, a circle indicates a negative effect on the target organism, and text size indicates relative population density. (C) The details of the study system, where bold lines refer to the generalized interactions and non-bold lines to the details of the particular study system. The direct effect (i.e., not trait-mediated) is that the ant either consumes sufficient honeydew to prevent sooty mold or removes the sooty mold directly. The indirect effect (i.e., trait-mediated) is the behavioral effect the ants have on the ability of the predator adults to lay eggs in the general area where the hemipteran pest is located and the ability of the predator adults to consume the hemipterans. The positive effect the ants have by protecting the larval predators against their parasitoids does not negate the negative effects the ant has on the adult predator through direct interference with its ability to eat and oviposit. The question as to whether the system is a direct mutualism (B) or an indirect mutualism (A) depends on the relative values of all the interactions in the system details (C). doi:10.1371/journal.pone.0043510.g001 cover) on ant nest distribution, the emergent clustered pattern has been attributed to biological local interactions [16]. We conducted two field studies to examine 1) the population dynamics of the beetle A. orbigera, the specialist predator of the hemipteran, and 2) the mechanism of the ant-hemipteran mutualism.
The first field study investigated the population persistence of the specialist predator in a 45 ha plot, in sites with and without ants. The location of the 45 ha plot was randomly selected and the exact location of each ant nest in the plot was marked and recorded [26]. Beetles were sampled systematically by superimposing a 50650 m grid onto the map of the plot. In each quadrant without any ant nests, beetles were sampled on coffee bushes around the centermost shade tree; in quadrants with ant nests, beetles were sampled around the shade tree with an ant nest that was closest to the center of the quadrant. In each site, we searched for adult and larvae beetles on every coffee bush within a 3 mradius of the selected tree for 30 mins. We conducted these samplings four times over two years (rainy season 2006, dry season 2007, rainy season 2007 and dry season 2008) in 55 sites tended by A. instabilis and 60 sites not tended by ants. After recording adult and larval beetle populations across the four field seasons, we used a Mantel permutation test [27] to examine if beetle persistence (i.e. the proportion of times a site was occupied by at least one beetle) was correlated with ant-nest density (i.e. # of nests within a 20 m radius of the sampled tree).
In the second field study, we took advantage of the sessile nature of C. viridis by investigating hemipteran population dynamics using digital photography (Fig. 2). Photographs of the hemipteran populations were taken over two years (weekly from June to July of 2006, February to August of 2007, and July to August of 2008), and from these photos, we examined hemipteran population dynamics across three independent population cycles in ten independent randomly chosen study sites (separated by more than 500 m), six of which were closely associated with a nest of the arboreal nesting A. instabilis and four of which were in an area known to be at least 200 m removed from any such nest (N = 30 populations, N = 8-30 weekly time steps per population) [28]. _ENREF_29Using the photographs, we counted the number of intact and predated crawlers, nymphs, and adults each week on each leaf, in sites with and without ants. The latter was possible by comparing consecutive pictures and counting newly and previously predated hemipterans; the process was further enabled by the fact that the main predator, A. orbigera, leaves telltale feeding marks on predated hemipterans [28]. By directly measuring predation on the scales, we document differential predation rates by predators across ant-tended and non-tended sites.
Using the counts of individuals and predation events, we built a simple Leftkovitch stage-based model [29,30] to describe C. viridis population dynamics. The model has six parameters: crawler-tonymph and nymph-to-adult transition probabilities; crawler, nymph, and adult predation parameters; and mean adult  fecundity. Unlike traditional Leftkovitch models our model has density dependence, with the predation probability dependent on the population number of a leaf. The predation probability for a given life stage on a given leaf for a given predation parameter is equal to: P predation~0 :001e aN 1z0:001e aN where N is the population of scales of that life stage on that leaf and a is the predation parameter. For a given set of parameters we calculated the likelihood of observing our ant-tended and nontended C. viridis time series, and we searched for two sets of parameters that maximized these likelihoods using the fminsearch function in Matlab [31]. For each parameter we created likelihood profiles to see how sensitive the likelihood was to a change in that parameter [31]. To create a likelihood profile, a given parameter is fixed at one of a set of values and then the likelihood is maximized by letting the other parameters vary, again using fminsearch. The result is a curve showing the likelihood of the model predicting the data given a fixed value of one of the parameters. This was repeated for each of the six parameters, and for ant-tended and non-tended populations. As is convention, we plot the negative log-likelihood, so a smaller value is more likely. We then used the likelihood ratio test to determine which parameters were significantly different between ant-tended and non-ant-tended sites [32].

Results
Based on the first field study, our results indicate that both adult beetle persistence (P = 0.002) and larvae beetle persistence (P = 0.004) were highly correlated with ant nest density (Mantel test , N = 115). Neither adult nor larvae beetle persistence was spatially auto-correlated (Moran's I, N = 115, P.0.411 for both). Results from the second field study ( Table 1) indicate that hemipteran nymph and adult predation rates were not significantly different between ant-tended and non-tended sites. Instead, the hemipteran transition probability from crawler to nymph was significantly higher in the presence of ants (Fig. 3A) and hemipteran adult fecundity was significantly higher in the presence of ants (Fig. 3B) (Table 1).

Discussion
In this field-based study, we show 1) that the mutualism patches are indeed spatial reservoirs for the specialist predator, A. orbigera, and 2) that predation rates on the hemipterans are equally high in ant-tended and non-tended patches. Our results indicate that the specialist predator is able to prey equally well inside and outside ant-tended patches. Accordingly, independent studies show that the protective behavior of the ant is effectively cancelled when the ant is under attack by a phorid parasitoid, and A. orbigera is then able to oviposit near and prey upon ant-tended scale colonies [23,33]. However, due to the direct path of the ant-hemipteran mutualism, C. viridis populations grow faster inside the ant-tended areas, creating spatial reservoirs for A. orbigera, which then potentially exert effective biological control over the rest of the farm. Thus we propose that the direct nature of the mutualistic mechanism ( Fig. 1) simultaneously benefits and limits hemipteran population growth, depending on the spatial scale of the interaction.
Effective control of hemipteran populations in this system is a consequence of the self-organized heterogeneity of the anthemipteran mutualism [17,21,22]. While studies conducted across taxa have suggested that mutualisms may be exploited by specialist predators [34,35], the relative cost or benefit of ant-protection to hemipteran populations has not been well-established [9]. Our results indicate that while specialist predator persistence was significantly higher in ant-tended patches, hemipteran predation rates were not significantly different between the patches. Antpresence has no effect on predation rates likely because the specialist predator larvae is immune to ant attacks and also experiences inadvertent protection from its natural enemies in the presence of ants [21,22]. Thus, in this system, we show that if ants are providing hemipterans protection from predation, this protection is only enough to balance increased predation pressure in ant-tended sites.
Most importantly, our results clearly demonstrate that ants provide a benefit to hemipterans, but that the mutualism occurs via enhanced C. viridis crawler to nymph transition probabilities and increased adult fecundity, not protection from predation. Other than protection from predation, the two other benefits of ant-tending hypothesized in past studies, are the 1) increased hygienic conditions for hemipterans via the removal of hemipteran honeydew [11], and 2) better site-selection in ant-tended populations, leading to increased nutrient uptake [36]. The majority of past studies suggest that hemipterans may actually incur a reduction in fecundity in the presence of ants [37], potentially due to a forced increase in honeydew production [38] and/or reduced assimilation of N by hemipterans [9,39,40]. In contrast, our study demonstrates that the primary benefit anttended hemipterans experience is increased fecundity and growth rates, a finding that has been documented only in cage experiments in the absence of natural enemies [36]. While the exact ant-mediated activity leading to the growth-based mutualism is not known, we hypothesize that it may be due to a combination of improved hygienic conditions and feeding site selection mediated by ants.
Given the strong dependence of natural and agricultural systems on mutualisms, pest control decisions made in the light of incorrect assumptions about the mechanism of the mutualism could lead to predator population decline and loss of this critical ecosystem function. In this study, we provide evidence that the ant-hemipteran mutualism is driven by the direct enhancement of hemipteran growth rates and fecundity. Based on the data, we posit that this occurs via the creation of spatial reservoirs for specialist hemipteran predators. Therefore, in agroecosystems where ants tend hemipterans, the recommendation to eliminate ants for pest control may be counterproductive. If, as in this study, the direct mutualism mechanism supports predators that control hemipteran pests at the general level of the farm, then the mutualism itself must be thought of as part of the biological control function and be preserved.