Pollination is a key ecosystem service which most often has been studied in isolation although effects of pollination on seed set might depend on, and interact with, other services important for crop production. We tested three competing hypotheses on how insect pollination and pest control might jointly affect seed set: independent, compensatory or synergistic effects. For this, we performed a cage experiment with two levels of insect pollination and simulated pest control in red clover (Trifolium pratense L.) grown for seed. There was a synergistic interaction between the two services: the gain in seed set obtained when simultaneously increasing pollination and pest control outweighed the sum of seed set gains obtained when increasing each service separately. This study shows that interactions can alter the benefits obtained from service-providing organisms, and this needs to be considered to properly manage multiple ecosystem services.
Animal-mediated pollination is a key ecosystem service that substantially contributes to global food supply and human nutrition [1,2]. Global declines of pollinators owing to habitat loss, agricultural land-use changes and other pressures  have resulted in declines in the level and stability of seed set in insect-pollinated crops [4,5]. In a vast majority of existing studies, crop pollination has been studied as a benefit obtained from a single context-independent interaction between a plant and its pollinators.
Ecosystem services can be linked to each other through a common environmental driver, or through direct interactions between the services themselves . Bos et al.  challenged the view of context-independent crop pollination and argued that the net result of a change in the pollinator community on seed or fruit yield is likely to depend on and interact with other ecosystem services, such as regulation of pest pressures. Direct interactions between ecosystem services have important practical implications for how to combine management of multiple services in cropping systems and in the landscape (figure 1), but they remain poorly studied .
The interactions between pollination and pest control for yield may be informed by plant compensation theory, which predicts that plants in high-resource environments have a higher ability to compensate for herbivory (i.e. absence of pest control) compared with plants in low-resource environments . We are not aware of any studies explicitly testing how pollen limitation affects plant compensation to seed predators, but it is possible that plants have a higher flexibility to reallocate resources to undamaged ovules following pest attack from seed predators when a higher proportion of these ovules have been pollinated (corresponding to a high-resource environment). This may be a pathway under which pollination can compensate for a lack of control of seed predating pests (figure 1b).
There are, however, some mechanisms which speak against a compensatory interaction between pollination and control of seed predating insects. Plants should be less capable of compensating for herbivore attack late in the season . Herbivore attack immediately before and during flowering, and on plant reproductive parts, might be severe and irremediable for insect-pollinated plants. Floral traits can be altered such that pest-infested flowers or grazed plants are less preferred by pollinators [10,11]. Moreover, herbivores that continue to consume plant parts after pollination, such as flower-feeding grazers or seed predators, might switch their foraging to selectively consume well-pollinated and seed-rich flowers, which are highly rewarding . Hence, lack of control of pests attacking plant reproductive parts may cancel positive effects of pollination, resulting in a synergistic interaction between the two services (figure 1c).
Interactions between pollination and herbivory, and their consequences for plant reproduction, has so far largely been investigated from a plant evolutionary and demographic perspective with a focus on herbivory of leaves, shoots or flowers [13,14]. These studies found multiple potential pathways for interactive effects. Despite this, interactions between insect pollinators and herbivores remain largely uncharted in crop plants (but see [15,16]). To our knowledge, there is no study that tested how the control of pests that attack seeds or fruits interact with pollination services for yield in a crop plant. We conducted an experiment with the aim to separate between the competing hypotheses on combined effects of insect pollination and pest control on yield (figure 1). For this, we combined two levels of insect pollination with two levels of seed predation (that simulated two levels of pest control) in a cage experiment in red clover (Trifolium pratense L.) grown for seed.
2. Material and methods
(a) Study system
Red clover is an important nitrogen-fixing crop used for fodder and as green manure [17,18], and the seeds are produced for legume seed mixtures. Red clover is obligately outcrossed by insect pollinators, mostly bumblebees (Bombus spp.) and honeybees (Apis mellifera) . A red clover plant typically flowers for several weeks, and an inflorescence (flower-head) normally have 50–200 flowers which open continuously over 6–10 days . Successful pollination normally leads to development of a single seed per flower. Weevils of the genus Apion (Coleoptera: Brentidae), specifically the clover seed weevils A. trifolii L. and A. apricans Hbst., are the major pest insects in European red clover seed production . Adult weevils oviposit in clover inflorescence buds and the larva develops inside a single clover inflorescence, where it consumes 6–10 ovules and developing seeds . Damage by Apion weevils to flowers can be quantified by inspecting mature inflorescences. The larva chews a hole in the calyx covering each ovule when it passes from one flower to the next (O. Lundin 2008, personal observation). Parasitic wasps within the order Hymenoptera attack the larvae which can result in high parasitation rates with potential to control the abundance of red clover seed weevils [20,22].
(b) Pest control treatment
We simulated pest control by placing experimental plants on locations with either a low or high abundance of the pest (figure 2a), which ensured that pest abundances were kept at realistic levels in the treatments. The abundance of Apion weevils was assessed in six red clover seed fields (pest control sites) in southern Sweden in early June 2010, with three to five pan traps (white plastic soup plates, 45 mm deep, inner ø 160 mm) per field. The choice of field locations were based on earlier observations of the distribution of Apion weevils in the area (O. Lundin 2008, unpublished data). Locations were minimum 7 km and maximum 70 km apart. The pan trap assessments verified that the fields provided good and replicated representation of either low or high pest control, as mean number of pest insects caught was low on three fields (0.1, 0.1 and 0.1 Apion weevils per trap and day), and high on the three other fields (2.8, 5.2 and 5.2 Apion weevils per trap and day). The contrasts in pest control were not created by pesticide use, as insecticides were not applied until late in the period for the pest control treatment. Furthermore, insecticides were applied on all three fields with low pest control, but only on one field with high pest control.
Sixty red clover plants of the tetraploid variety SW Sara, from a red clover field (the experimental field) located in the southernmost part of Sweden (55°56′ N, 12°50′ E), were transplanted into 10 l pots early in June 2010. After being potted, all plants were exposed to the same watering regime throughout the experiment. When the plants were about to form buds, about 1.5–2 weeks later, 10 plants were moved to each of the six pest control sites. Three weeks later, the potted plants on all pest control sites had produced numerous buds that had been exposed to oviposition by Apion weevils. At this stage, inflorescences with five or more open flowers were cut off from the plants in order to avoid uncontrolled pollination of the plants. All buds (mean 15 per plant, s.d. 6) which had four or fewer open flowers were marked on each plant by placing a piece of plastic straw on the stalk, just under the bud. The plants were then relocated back to the experimental field for the pollination treatment. The number of inflorescences cut off from plants prior to the pollination treatment (mean 18 per plant, s.d. 11), did not differ between plants that received contrasting pest control or pollination treatments (data not shown).
(c) Pollination treatment
To create two levels of pollination, plants from each pest control site were pollinated by bumblebees for a contrasting amount of time in cages on the experimental field in a fully crossed design (figure 2). We allowed some pollinator to access flowers also at the lower level of pollination because red clover will produce virtually no seeds without insect pollinators . Each level of pollination had five replicates (pollination sets) that each contained six plants, one from each of the pest control sites (figure 2b). Pollination cages were 1.8 × 1.8 × 1.8 m, had wooden frames, and nets of polyester mesh (approx. 1 × 1 mm mesh size) to prevent bees from entering or exiting caged areas. We used two Bombus terrestris L. bumblebee colonies (NATUPOL, Koppert Biological Systems, The Netherlands), placed in one cage each for the pollination treatments. Bombus terrestris is a potentially less efficient pollinator of red clover compared with more long-tongued bumblebee species, but nevertheless it is the most abundant species in the field  and we used it because it is the only species that is commercially available in colonies in the study region. Approximately, 1 m2 of the crop (with a plant density of approx. 10–40 plants per m²) was left inside pollination cages. This made it possible for bees in the cages to visit other than the experimental plants thereby promoting cross-pollination under field-like conditions. Additional cages were used to store the experimental plants without access for pollinators. The experimental field had low levels of pest abundance (previously assessed to 0.1 Apion weevils per trap and day), which minimized the risk of affecting the pest control treatment.
Each time before pollination of a set of plants started, newly emerged unmarked inflorescences with open flowers were first counted and cut off from the plants, leaving only those inflorescences which had received the pest control treatment. At the start of each pollination round, bumblebees were let out from their nest until 10 workers were present in the cage. A valve ensured that foragers could return freely to the colony without any more bees exiting during pollination treatments. To determine ‘low’ and ‘high’ pollination levels, we approximated from pilot observations that a red clover inflorescence had 90 flowers, and that a bumblebee visited 20 flowers per minute when actively foraging in the cage. In the pilot observations, we observed that individual bees typically spent a few hours in the cage before returning to the nest. We only observed legitimate flower visits and no nectar robbing behaviour. We counted the total number of inflorescences in all pollination sets that were assigned to receive a ‘low’ pollination and then allowed 1.5 bumblebees × 1 minute × number of inflorescences of total foraging on the group of experimental plants. Thus, if the bees visited only virgin flowers and had 100 per cent pollination success per visit, we expected approximately one-third of the flowers to be pollinated in this treatment. Foraging time was equally divided between two separate pollination rounds, one with each of the colonies, with 4 days in between. Total numbers of bumblebees foraging on the group of experimental plants were counted each minute to ensure pollination according to the protocol in the ‘low’ pollination sets. A ‘low’ pollination set needed in total about 1 h in the cages to be pollinated according to the protocol. In ‘high’ pollination sets, a starting density of 10 bumblebees were left unsupervised for at least 3 h in four different pollination rounds, two rounds with each of the colonies and with 1–2 days in between rounds. We restricted pollination in the ‘high’ pollination treatments, to avoid unrealistically high bee visitation rates. All plants were pollinated with multiple rounds, with some days in between treatments to ensure that both early and late opening flowers on inflorescences received pollination.
(d) Seed production measures
The plants were left in pollinator exclusion cages for about a month for ripening of marked inflorescences. All marked inflorescences were then cut from the plants and put individually in paper bags. Additional unmarked inflorescences which had emerged after the pollination treatment were counted on all plants. In the laboratory, the calyx of all individual flowers on 10 harvested inflorescences from each plant were manually inspected; on seven plants where only five to nine inflorescences (average 8.7 inflorescences) were harvested, all inflorescences available were inspected. The proportion of calyces with a whole seed, the proportion of weevil damaged ovules (indicated by calyx having a bite hole and not containing a whole seed), and the proportion of empty undamaged calyces was determined for each inflorescence inspected.
(e) Statistical analyses
Data (see the electronic supplementary material) were analysed using linear mixed models with identity links and Gaussian errors (SAS proc Mixed) in SAS v. 9.1 for Windows (SAS, Cary, NC, USA). REML was used as estimation method, the denominator degrees of freedom were calculated with the Satterthwaite method, and the nobound option allowed (non-significant) negative within subject variances to be estimated . We analysed the fixed effect of pollination (‘low’/‘high’), pest control (‘low’/‘high’) and their interaction on the mean (arcsine square root transformed) proportion of damaged ovules and seed set per plant. The transformations of response variables, which were done to achieve approximately normal distribution of model residuals, might qualitatively affect ordinal (non-crossing) interactions (figure 1b,c), which are dependent on measurement scale . To account for this, we verified that any differences in slopes for ordinal interactions on the arcsine square root scale were conserved or augmented on the linear scale, which seems more appropriate to inform management and policy. To account for non-independence in the study design, pest control site, pollination set and pest control treatment nested within pollination set (figure 2) were included as random factors in analyses. When significant interactions were found, simple main effects of a factor (pollination or pest control) were analysed separately for each level of the other factor by splitting the dataset. To be able to test for a simple main effect of pest control under ‘low’ pollination, the random factor pollination set first had to be excluded in order for the model to converge. We also tested and verified that our primary yield measure, which was seed set measured on the flower-head level, could be translated to whole-plant effects (see the electronic supplementary material).
(a) Proportion damaged ovules
Proportion of damaged ovules per inflorescence increased when pest control was low (F1,3.8 = 24.73, p = 0.0089), whereas there was no main effect of pollination on the proportion of damaged ovules per inflorescence (F1,8.0 = 0.07, p = 0.79). There was however a significant interaction between pollination and pest control (F1,8.0 = 5.70, p = 0.044; figure 3a). Tests for simple effects showed that high pest control decreased the proportion of damaged ovules both at low (F1,3.6 = 12.35, p = 0.029) and high (F1,2.5 = 51.09, p = 0.010) pollination, but with a larger effect size in the latter case. Pollination was significantly associated with a lower proportion of damaged ovules at high pest control (F1,8.0 = 7.17, p = 0.028), but not at low pest control (F1,8.0 = 2.04, p = 0.19).
(b) Seed set
Both pollination (F1,8.0 = 215.19, p < 0.0001) and pest control (F1,2.4 = 137.55, p = 0.0039) increased seed set. There was, however, also a significant interaction between the two (F1,8.0 = 15.14, p = 0.0046). The gain in seed set owing to higher pollination was larger at high pest control (figure 3b). Tests for simple effects showed that the effect of augmented pollination on seed set was positive both at low (F1,8.0 = 33.22, p = 0.0004) and at high (F1,8.0 = 261.52, p < 0.0001) pest control, but with a larger effect size in the latter case. The effect of pest control tended to be positive at low pollination (F1,1 = 82.83, p = 0.070) and was positive at high pollination (F1,1.9 = 75.33, p = 0.016).
We found that the gain in seed set obtained in red clover when simultaneously increasing pollination and pest control outweighed the sum of seed set gains obtained when increasing each service separately, supporting the hypothesis of a synergistic effect of pollination and pest control on yields (figure 1c).
The effect on red clover seed set by individually manipulating the supply of the two ecosystem services followed expectations. We observed a considerable increase in seed set following augmentation of insect pollination, which is in line with earlier reports on the importance of insect pollinators for seed set in red clover [19,25]. Maximum average seed set under high levels of pollination and pest control was still under 60 per cent, which may be due to a limited proportion of fertilized ovules setting seed in tetraploid red clover, or partial inability of B. terrestris to successfully pollinate tetraploid red clover flowers [26,27]. The level of pest attack was also important for seed set, and plants exposed to low levels of seed weevil abundance produced a higher seed set compared with plants exposed to high weevil abundances. This is line with a recent study showing that Apion spp. weevils strongly can limit red clover seed yields in the study area . The contrasts in pest attack might have been caused by natural enemy parasitoids attacking weevil larvae [20,22] or bottom-up factors, such as weather or habitat quality. Future experimental studies of service interactions between pollination and pest control that directly manipulate the service-providing communities might reveal potential additional mechanisms for interaction which are not directly mediated by pest densities, such as behavioural interactions between pollinators and natural enemies to pests.
Importantly, we found that the ecosystem services collectively affected red clover seed set in an interactive manner. The seed set benefit obtained from a constant pollination treatment was larger at high pest control. Similar results have been found in studies with simultaneous manipulations of pollinators and herbivores attacking flowers, fruits or seeds of wild plants [12,28,29]. In these studies, positive interactive effects on plant fitness were attributed to target herbivores cuing in on well-pollinated flowers, or herbivores decreasing plant attractiveness to pollinators. Decreased pollinator visitation caused by the pests may have played a part in the results we obtained and this deserves further observational study. Moreover, if decreased pollinator visitation is found important, it has to be confirmed that this result holds for differences in pest load between entire fields in larger-scale studies.
Seed predators can, however, also shape yields interactively with pollinators via other pathways. It is possible that the interactive effect on seed set was caused by a proportional rather than constant reduction in seed set by the pests, such that more seeds were lost by a given amount of seed predators when pollination was high. Furthermore, increasing pollination had contrasting effects on ovule damage depending on the pest control regime, and this also points to that pest damage was dependent on the level of pollination. Perhaps increased pollination was able to relax seed predators from resource limitation under low pest control only and thereby contributing to increased pest damage, as opposed to under high pest control when resource limitation is less likely to occur. The interpretation of this result is, however, complicated by the fact that it was not possible to assess if ovules that had been damaged by the pests had been fertilized or not. In conclusion, more detailed studies on seed predator performance and consumption of fertilized versus unfertilized ovules under contrasting levels of pollination would be needed in order to further understand the mechanisms by which seed predating herbivores may alter pollination benefits.
Together these mechanisms suggest several pathways rendering the interactive effect of pollination and pest control on yield, where lack of pest control cancels pollination benefits. From a more general perspective, pest outbreaks following low pest control should strongly limit the potential of pollination to affect yield if they cause complete, or close to complete, destruction of plant reproductive parts. In other cases, the interactive outcome might be dependent on the relative timing of herbivory and pollination, and the type of damage that is done by the herbivore. For example, there is some support for that pollination can compensate for leaf herbivory  while pollination and herbivory to shooting stalks might have independent and non-interactive effects on plant performance . The guild of herbivorous pests controlled by natural enemies might be an important predictor of how pollination and pest control interactively determine yield, and this deserves further study in more insect-pollinated crops and with several types of herbivore guilds.
Our results have implications for estimating the values of interacting and complementary ecosystem services, in this case provided by pollinators and natural enemies to pests. Correct estimation of ecosystem service values is important for economically efficient conservation of the organisms providing the service. The context-dependent nature of pollination benefits shown in this study puts in question current benefit transfer functions, where the benefits of an ecosystem service from one area are extrapolated to another seemingly similar area . Such benefit transfer is increasingly practiced for spatial mapping of pollination, pest control and other ecosystem services , but may give misleading results if interactions between ecosystem services are not taken into account.
The exploration of correlations in space and time between regulating ecosystem services, such as pollination and pest control, emerge as a research priority. Decreasing level and stability on yield in insect-pollinated crops has so far solely been attributed to pollinator declines, without considering how pest control services have changed in tandem [4,5,33]. If pollination and pest control services tend to covary positively, for example, owing to similar responses to land-use change , pressures put on both these services may collectively affect yield level and stability in a way that can only be understood by quantifying their collective effect on yield. Our results also suggest that it is crucial to identify the regulating ecosystem service in shortest relative supply, as management for other services in such cases might only have minor effects on yield. Finally, coordinated management for multiple ecosystem services can have positive synergistic effects, which can outweigh the summed benefits of managing ecosystem services spatially or temporally separate.
Crop plants depend on several other regulating (e.g. pathogen control) and supporting (e.g. soil nutrient supply) ecosystem services from both below- and above-ground organisms, which may have interactive effects on plant performance [35,36]. These interactions need to be better understood to provide adequate guidelines for the combined management of multiple ecosystem services in agricultural landscapes.
We thank Lorenzo Marini, Camilla Winqvist and three anonymous reviewers for constructive comments and the farmers for allowing us to work on their land. The work was financed by grants from the Swedish Research Council for Environment, Agricultural Sciences and Spatial Planning (FORMAS) to R.B. and H.G.S.
- Received September 21, 2012.
- Accepted November 30, 2012.
- © 2012 The Author(s) Published by the Royal Society. All rights reserved.