Synergistic interactions of ecosystem services: florivorous pest control boosts crop yield increase through insect pollination

Abstract

Insect pollination and pest control are pivotal functions sustaining global food production. However, they have mostly been studied in isolation and how they interactively shape crop yield remains largely unexplored. Using controlled field experiments, we found strong synergistic effects of insect pollination and simulated pest control on yield quantity and quality. Their joint effect increased yield by 23%, with synergistic effects contributing 10%, while their single contributions were 7% and 6%, respectively. The potential economic benefit for a farmer from the synergistic effects (12%) was 1.8 times greater than their individual contributions (7% each). We show that the principal underlying mechanism was a pronounced pest-induced reduction in flower lifetime, resulting in a strong reduction in the number of pollinator visits a flower receives during its lifetime. Our findings highlight the importance of non-additive interactions among ecosystem services (ES) when valuating, mapping or predicting them and reveal fundamental implications for ecosystem management and policy aimed at maximizing ES for sustainable agriculture.

1. Introduction

Ecosystem services (hereafter ES) encompass a large set of goods and functions provided by ecosystems, vital for human well-being [1]. While the global demand for reliable provisioning of ES is increasing, many of these services are declining due to anthropogenic-driven ecosystem changes [2]. Considerable effort has been made to quantify, map and identify the drivers and consequences of individual ES in agriculture [3], and it is increasingly recognized that ES rarely act in isolation, but interact with each other in complex ways [4,5]. Alterations in shared drivers can indirectly result in synergies or trade-offs between multiple ES and a range of potentially profound positive or negative interactive effects [4]. Consequently, there have been repeated calls for research aimed towards a better understanding of the relationships among multiple ES, and their underlying mechanisms, as a prerequisite for effective ecosystem management to sustainably maximize ES [4,6]. Maximizing multiple ES in agro-ecosystems is crucial to meet the challenge of long-term food security through sustainable crop production, without jeopardizing biodiversity and ecosystem health [7]. To achieve this goal, we need to understand if and how the management of one service has positive, negative or negligible effects on other services.

Among the multiple provisioning and supporting ES which contribute to yield in agro-ecosystems, animal-mediated crop pollination (hereafter pollination) represents a key service with an estimated economic value for global crop production of €153 billion per year [8]. At the same time approximately one-third of the potential global crop yield is lost to pests [9], highlighting the enormous potential and urgent need for pest control. Pollination directly increases and stabilizes the yield of ca 70% of the world's most important crops [10,11]. Pest control directly reduces the negative impact of pests on crop plants [1]. While the importance of pollination and pest control for crop production are well recognized individually [10,12], their interactive effects on crop yield and farmer's economic gain remain largely unexplored. The few studies that have recently started to explore the combined effects of pollinaton and pest control suggest that such interactions may indeed exist [13,14], but unfortunately we still lack (i) a robust quantification of the effect size of such interacive effects compared with their single effects, which would enable the contribution of such interactive effects on the final crop yield and its economic importance to be assessed; (ii) a mechanistic understanding of ecological drivers governing interactive effects among pollination and pest control. However, a quantitative knowledge of the economic importance of the potential interactive effects of pollination and pest control on crop yield, and a better understanding of the mechanisms driving such effects, is fundamental in order to reliably valuate and predict interactive effects, and those of their potentially shared drivers, such as land-use or climate change [15]. This is a vital prerequisite for improved management of agro-ecosystems and ecological intensification (sensu [3]).

In this study, we experimentally examine the single and combined effects of pollination and simulated pest control shaping yield quantity and quality in winter oilseed rape Brassica napus L. (hereafter OSR). OSR is among the most important food, fodder and biofuel crops worldwide, and its economic importance has continuously increased in the last few years. Yield losses of OSR due to herbivory by the pollen beetle Meligethes aeneus Fab. could be severe without pest control [16]. Although OSR is considered to be mainly wind pollinated, recent studies indicate that the contribution of animal-mediated pollination to OSR yield can be considerable [17,18]. Thus, interactive effects of pest control and animal pollination in OSR production may be of great economic importance, yet remain to be explored experimentally.

The potential non-additive effects of pollination and pest control, which shape crop yield, may occur via a multitude of pathways. Here, we test three hypothetical mechanisms (M1–M3). Such effects could arise where alterations in the attractiveness of crop plants to pollinators—through changes in floral traits or direct repellence of pollinators by florivorous pests—reduce flower visitation and thereby pollination services [19] (M1); through compensatory responses of crops to herbivory by pests, such as compensatory growth [20] or even over-compensation [21] resulting in overall higher yields (M2). A further potentially important, but to our knowledge unexplored, synergistic effect of pollination and control of florivorous pests may act via florivorous pest-induced changes in flower lifetime (M3). In OSR, for example, flower lifetime is shorter if pollen is removed from the stamens [22]. If florivorous pests remove pollen from the anthers, thereby reducing the lifetime of flowers, the probability of a flower being visited and the average number of visits a flower receives by pollinators during its lifetime may be reduced.

The main objectives of this study are, therefore, (i) to quantify the relative importance of animal pollination and pest control, and in particular their interactive effects on OSR yield quantity and quality, and the potential economic value of these effects and (ii) to test the mechanistic pathways (M1–M3) driving potential interactive effects among pest control and pollination. We show that the potential economic benefit for a farmer from synergistic effects is 1.8 times greater than the individual contributions of pollination and pest control and provide a mechanistic explanation for this striking finding. Our findings reveal the importance of taking non-additive pollination–pest control interactions into account when predicting and managing multiple ES for sustainable food production.

2. Material and methods

(a). Study system

OSR production in Europe suffers from a series of herbivorous pests. The most severe yield losses in Europe (up to 100% without pest control) are due to herbivory by the pollen beetle, M. aeneus Fab. (Coleoptera: Nitidulidae) [16]. The adult beetles feed on the pollen of open and closed flowers, the latter leading to flower development abortion and consequently reduced fruit set. OSR is considered to be mainly wind pollinated, because currently grown cultivars with restored fertility are self-fertile, but recent evidence suggests that insect pollination can significantly increase seed set in commonly grown varieties [17]. Bumblebees, in particular, the buff-tailed bumblebee Bombus terrestris L. (Hymenoptera: Apidae, hereafter ‘bumblebees') together with honeybees and a series of solitary bee and hoverfly species are the most important pollinators of OSR in Europe [23].

(b). Experimental design

The experiment was conducted in spring 2014 at Agroscope-Reckenholz (lat 47.430868°, long 8.518491°, 442 m.a.s.l.) in Zurich, Switzerland. An area of 0.9 ha was sown in early September 2013 with the winter OSR variety ‘Visby’ (Rapool-Ring GMBH Isernhagen, Germany), a commonly grown OSR variety in Central Europe. The entire field was managed according to standard practices of conventional OSR production until the establishment of the experimental treatments. Immediately after the fertilizer application in early spring, 24 cages (4 × 2 × 2 m) were assembled—before colonization of the field by pollen beetles. Cages were spaced 4 m apart from one another to avoid reciprocal shading. They were covered with a fine mesh fabric (HD-polyethylene, 0.74 × 1.12 mm; Howitec, The Netherlands) which excludes pollinators, pollen beetles and their natural enemies, including small hymenopteran parasitoids and ground-dwelling arthropods, but should not affect wind pollination of the caged plants [24]. Cages were randomly assigned to one of four treatments (n = 6) in a fully crossed design with two pollination levels (pollination versus no pollination) crossed with two pest control levels (weak versus strong pest control). Cages were arranged within six spatial blocks on the experimental field such that each block contained one cage of each of the four treatments.

(c). Simulated pest control treatment

Pest (pollen beetle) control was simulated by experimentally establishing two different pollen beetle densities in the cages, with 12 cages per simulated pest control treatment. This experimental pest control thus reflected pest control irrespective of the identity of the pest control agent, and ensured that the level of pest control could be precisely and uniformly established at field-realistic levels for the two treatments [13]. To simulate a strong level of pest control we added a total of approximately 1600 adult pollen beetles to each of 12 randomly selected cages, resulting in approximately nine beetles per caged OSR plant (approx. three beetles per main raceme, e.g. [25]). This corresponds to a situation with successful pest control, reducing pest levels below the threshold where no significant yield loss is expected [26], even when assuming a negative linear relationship among pollen beetle density to OSR yield [25]. In each of the remaining 12 cages approximately four times more pollen beetles (approx. 6800 adults, resulting in approx. 36 beetles per plant) were introduced. These densities, subsequently monitored to test for treatment performance, corresponded to average natural densities of adult pollen beetles in OSR fields in the study region during colonization by pollen beetles (L. Sutter 2014, unpublished). Adult pollen beetles were collected by sweep netting surrounding OSR fields and were introduced into the cages at the time of natural pollen beetle colonization of OSR fields in the study region.

(d). Insect pollination treatment

Shortly after the onset of flowering, on 14 April 2014, half of the randomly selected cages of both pest control treatments were each equipped with a colony of B. terrestris consisting of a queen and approximately 9–12 workers (‘mini hive’, Biobest, Westerlo, Belgium). Bumblebee hives were mounted 15 cm above the ground and protected against rain with a plastic roof. To achieve natural levels of OSR flower visitation under field conditions, the number of bumblebee workers and the amount of time they were allowed to visit OSR flowers in cages was controlled by adjusting the time the hive's separate inlet and outlet hole was open based on the following formula:

where V_{rate flower field} is the observed flower visitation rate by pollinators in the study region under field conditions (L. Sutter 2014, unpublished field study), L is the average flower lifetime (h) observed under field conditions [22], F_{tot cage} is the estimated total number of flowers per cage (estimated at the beginning of the experiment, this study) and V_{single worker field} is the average number of OSR flower visits per time observed for a single B. terrestris worker under field conditions [27]. The estimated value of ‘bumblebee hours’ obtained by this formula tells us how many bumblebee workers are allowed to forage for how long in a cage to achieve visitation rates of caged OSR flowers that are in the range of natural visitation rates of OSR flowers under field conditions in the study region. During the remaining time period bumblebees fed on a sugar solution provided inside the hive.

(e). Pollinator visits and flower phenology

Flowering onset, flower lifetime and flower visitation by pollinators were assessed for both pest control treatments. At each of three observation rounds during the flowering period of OSR the number of flowers visited and the time spent on a single flower (visit duration) of each of two bumblebee workers was recorded during 2 min (200 visits recorded on average per cage). The time of day of observations was randomized across cages and observation rounds and pest control treatments. Flower abundance (i.e. number of open flowers in a 50 × 50 cm wooden frame, averaged over four counts in different places, calculated for the cage area) was estimated on each observation date, which was used to estimate average pollinator visitation rate per flower of each cage.

To detect potential differences in flowering onset, the total number of open flowers of 10 randomly selected plants of each cage was counted at the beginning of the flowering period, before pollinators were introduced into the cages. To test for possible differences in flower lifetime among treatments, all open (but not yet senescent; hereafter ‘open’) flowers of the main shoot inflorescence of 10 randomly selected plants of each cage were counted and marked (using fine wire tagging the upper and lower limit of the range of open flowers). Because senescence (sensu [22]) could not be observed directly, the number of senescent flowers and the number of flowers still open from the previous marking were counted 72 h later. To calculate the relationship between the percentage of senescent flowers and flower lifetime, a calibration was necessary. To this end an additional experiment was conducted, in which 10 independent inflorescences of potted OSR plants with 10 freshly opened flowers were each exposed to different densities of adult pollen beetles (0, 0.1, 0.5, 1, 5 pollen beetles per OSR flower; n = 10 for each pollen beetle level). Flower lifetime and the proportion of senescent flowers were recorded. As flower lifetime (L) and the proportion of senescent flowers after 72 h (P_tot) are linearly proportional (L = 0.2824 × P_tot + 9.72; R² = 0.97), the slope of this relationship allows the calculation of flower lifetime from the percentages of senescent flowers measured in the cages. This experiment was also used to demonstrate and parametrize the negative relationship between pollen beetle density and flower lifetime. Reducing flower lifetime can consequently reduce the number of pollinator visits per flower lifetime (M3). These results are shown in the electronic supplementary material, figure S1. The number of pollinator visits a single flower receives during its lifetime (V_tot) was then estimated for each cage as:

where L is the estimated average flower lifetime per cage (h), V_rate is the total number of flower visits per cage during 1 h (visits × h⁻¹) and F_tot is the estimated total number of flowers per cage.

(f). Yield measurements

Yield and other plant parameters were determined for 10 randomly selected OSR plants of each cage, collected on the 17 July 2014, when fruiting was complete (seeds dry and fully developed), but before ripe fruits started to split and disperse seeds. For each harvested plant the total number of shoots was measured. Additionally, the total number of fruits containing seeds (fruit set) of each main and fifth side shoot was measured. For each of these shoots, seed set (i.e. the mean number of seeds per fruit), mean seed mass (i.e. the mean weight of 1000 seeds) and total seed mass per fruit (i.e. mean seed mass × seed set) were measured for 10 randomly selected fruits per shoot. Shortly after harvesting these 10 plants, each entire (previously caged) 2 × 4 m plot was threshed with a threshing machine (Wintersteiger Classic Plot Combine). In addition to the standard quantitative measure of agronomic yield (seed mass of threshed plants per cage (t ha⁻¹); hereafter ‘yield’), we analysed the oil content of the pooled seeds of the threshed plants per cage (g kg⁻¹) with near infrared spectroscopy (NIRS; Foss NIRSystem, Inc. Silverspring, MD, USA, calibration according to ISO standard 12099 [28]) as a measure of yield quality. Farmer's potential economic gain (€ ha⁻¹) based on the actual market price for rape-seed oil in 2014 in Switzerland [29] was calculated as seed mass (t ha⁻¹) × oil content (g kg⁻¹) × market price (€ ha⁻¹).

(g). Statistical analysis

The response variables ‘yield’, ‘oil content’, ‘farmer's potential economic gain’, ‘total number of shoots per plant’ and ‘flowering onset’ were analysed with linear mixed effect models (LMM) using the R-package lme4 [30] with treatments ‘pollination’ and ‘pest control’ and their interaction as fixed and ‘block’ as random effects. ‘Fruit set’, ‘seed set’, ‘mean seed mass’ and ‘total seed mass per fruit’ were analysed by means of LMM with the same model structure described above and the additional random factors ‘shoot’ nested in ‘plant’. ‘Number of visits per flower lifetime’, ‘visitation rate’ and ‘visit duration’ were modelled only for cages with pollinators using LMM with ‘pest control’ treatment as fixed and ‘block’ as random effect. Residual variances of all models were homoscedastic and normally distributed except those of ‘visitation rate’, which were log-transformed to meet LMM assumptions. The p-values for fixed effects were calculated based on residual degrees of freedom estimated with the Kenward–Roger approximation [31].

As parameter estimation in linear mixed effect modelling is at the frontier of statistical research, we cross-checked the robustness of the model predictions by also estimating all the parameters from these models in a Bayesian framework. Figures show means of posterior distributions from 10 000 samples drawn from three MCMC in JAGS [32] ± the respective standard deviations. Priors were set vague as flat normal distribution with standard deviation of 1 000 000. All statistical analyses were performed in R v. 3.1.1 [33].

3. Results

(a). Synergistic pollination–pest control effects on yield, oil content and farmer's potential economic gain

In the absence of pollinators, OSR yield (total seed mass), oil content and farmer's potential economic gain were increased by 6%, 1%, and 7%, respectively, at strong compared with weak pest control levels (figure 1; table 1). Furthermore, pollination by bumblebees significantly increased OSR yield by 7% on average and farmer's potential economic gain by 7% at weak pest control. Although no effect of pollination on oil content was detected at weak pest control, pollination resulted in a 1.1% increase under strong pest control conditions (figure 1; table 1). Importantly, the positive effect of pollination was significantly stronger at stronger pest control (figure 1). This synergistic effect (positive interaction) of pollination and pest control accounted for a pronounced increase in yield (11%) and farmer's potential economic gain (12%), and a slight but significant increase in oil content (3%; figure 1; table 1).

Figure 1.

Mean of posterior distribution ± s.d. of (a) OSR yield, (b) oil content and (c) farmer's potential economic gain with ‘insect pollination’ (bumblebee pollinators present (solid line) or absent (dashed line)) under weak versus strong pest control (PC) (n = 6).

Table 1.

Summary of the results of linear mixed effect models testing the effects of the fixed factors ‘insect pollination’ (bumblebee pollinators present or absent), ‘pest control’ (weak versus strong pest control (PC)) and their interactive effect on investigated response variables. The response variables ‘flowering onset’, ‘visits per flower lifetime’, ‘visitation rate’ and ‘visit duration’ were only assessed for the ‘pest control’ treatment. Denominator degrees of freedom (DDf), F-values and corresponding p-values (in bold if < 0.05) from linear mixed effect models, based on Kenward–Roger approximations, are shown (see Material and methods section for detailed description of explanatory variables, response variables and statistical models).

	DDf	F-value	p-value
yield (total seed mass ha−1)
pollination	10.60	66.69	<0.001
pest control	12.58	52.54	<0.001
pollination × pest control	17.01	13.94	0.002
oil content
pollination	12.08	4.33	0.059
pest control	14.99	11.67	0.004
pollination × pest control	14.99	3.96	0.065
farmer's potential economic gain
pollination	11.74	56.31	<0.001
pest control	14.40	49.84	<0.001
pollination × pest control	15.55	16.24	<0.001
number of fruits
pollination	12.08	0.97	0.344
pest control	14.99	20.22	<0.001
pollination × pest control	14.99	0.06	0.816
seed set
pollination	11.94	124.72	<0.001
pest control	14.75	18.58	<0.001
pollination × pest control	15.23	16.83	<0.001
mean seed mass
pollination	11.28	8.34	0.014
pest control	13.65	6.50	0.023
pollination × pest control	16.18	1.77	0.201
total seed mass per fruit
pollination	12.08	93.42	<0.001
pest control	14.99	6.42	0.023
pollination × pest control	14.99	10.46	0.006
total number of shoots per plant
pollination	219.69	1.46	0.229
pest control	184.76	0.81	0.370
pollination × pest control	23.01	0.22	0.645
flowering onset
pest control	16.01	0.98	0.337
mean number of visits per lifetime
pest control	10.00	7.84	0.008
visitation rate
pest control	10.00	0.07	0.792
visit duration
pest control	10.00	0.14	0.712

The reduction in yield due to lower pest control was caused by an overall reduction in the number of fruits per plant (fruit set), irrespective of the level of pollination (figure 2a; table 1). Yield increase due to pollination, on the other hand, was driven by an increase in the number of seeds per fruit (seed set) (figure 2a; table 1), resulting in a higher total seed mass per fruit, despite a slightly reduced mean seed mass under the pollination treatment (figure 2b; table 1). The positive effects of pollination on seed set and consequently total seed mass per fruit were significantly stronger at strong pest control, indicating that increased seed set per fruit, together with the higher number of fruits, was the principal driver of the synergistic effects of pollination and pest control on OSR yield.

Figure 2.

Mean of posterior distribution ± s.d. of (a) seed set per fruit (triangles) and number of fruits per shoot (fruit set; circles) and (b) mean seed mass per seed (mean mass of 10 seeds for display, triangles) and total seed mass per fruit (seed set × mean seed mass per seed, circles) of OSR as a function of insect pollination (bumblebee pollinators present (solid line) or absent (dashed line)) under weak versus strong pest control (PC) (n = 6).

(b). Mechanisms driving synergistic pollination–pest control effects

To detect potential changes in flower visitation behaviour of the pollinators as a response to different levels of pest control (M1), flower visitation rate and visit duration were analysed. However, there was no significant difference in the flower visitation rate or the duration of visits between pest control treatments (figure 3; table 1). To detect potential compensatory growth mechanisms of plants exposed to different levels of pest control treatments (M2), the numbers of side shoots and flowers per plant were analysed. However, there was no indication of over-compensatory growth as the number of fruits decreased with weak pest control and the numbers of shoots did not differ between treatments (figure 2a; table 1). Moreover, flower onset did not differ between pest control treatments (table 1). However, the estimated number of visits an individual flower received during its lifetime (M3) was reduced by 41% under weak pest control (figure 3; table 1).

Figure 3.

Mean of posterior distribution ± s.d. of the average number of OSR flowers visited by bumblebee pollinators per second (average visitation rate; dashed line) and the predicted number of pollinator visits per flower lifetime (solid line) under weak versus strong pest control (PC) (n = 6).

4. Discussion

We found strong synergistic effects between pollination and pest control on the quantity and quality of OSR yield. These positive interactive effects contributed 1.6 and 2.3 times more to quantitative yield gains (total seed mass) than their individual effects, respectively. We found significant synergistic effects of pollination and pest control not only on seed set and total seed mass, but also on the oil content of seeds. Although the increase in oil content due to this interaction was rather small (15 g, equivalent to 2.2%), the gain in harvested oil is highly economically relevant, in particular, when considering the vast areas planted with OSR in Europe and worldwide. Insect pollination has also been found to affect oil content and nutritional quality in other oil crops, e.g. almonds [34], but the underlying mechanisms of this remain poorly understood. For plants, investing in grain fitness by increasing its fat content is a possible way to strengthen offspring fitness, in particular, if pollination occurs through outcrossing [35]. Interestingly, such effects on oil content were only detected at low pest levels (figure 1b), possibly because a plant's ability to allocate resources is otherwise exhausted by the need to compensate for pest-induced damage. Owing to the combined increase of yield quantity and quality, the economic gain for a farmer resulting from the synergistic effect of the two ES (€ 311 ha⁻¹) was 1.7 and 1.8 times that of the individual benefits from pollination (€ 118 ha⁻¹) and pest control (€ 110 ha⁻¹), respectively.

(a). Pest control and pollination driving yield

Pest control and pollination affected OSR yield through distinct pathways: pest control resulted in an increased yield through an increased fruit set (12% reduced flower abortion at strong pest control), while pollination did not affect fruit set. By contrast, pest control had no effect on the number of seeds per fruit when pollinators were absent (figure 2a), whereas pollination increased seed set, with significantly more pronounced increases under strong pest control. This increase in the number of seeds per plant, due to a higher number of seeds per fruit and an increased number of fruits, was the major driver of overall quantitative yield gains. These findings corroborate recent evidence that insect pollination can significantly enhance seed set and yield in commonly grown OSR varieties [17,18,36,37]. Moreover, and most importantly, they demonstrate that these yield gains strongly depend on the level of pest control. Indeed, pollination increased average seed set from 12 to 16 seeds per fruit under weak pest control, but up to 22 seeds per fruit under strong pest control. Our analysis reveals that although mean seed mass was slightly reduced where more seeds were produced per fruit, a pattern in line with previous studies in OSR [38], this decrease was by far outweighed by the marked increase in seed number, such that the total seed mass per fruit was still significantly higher when pollinated by insects (figure 2b).

(b). Pest-induced reduction in flower lifetime as a key driver of synergistic pollination–pest control effects

Research on the reproduction of wild plants proposes a multitude of potential pathways for synergistic processes between pollination and pest control. For example, herbivory, and in particular florivory, may modify flower traits such as flower display or floral resource quality [39]. Alternatively, florivores may directly repel pollinators. Both of these processes can lead to altered plant attractiveness to pollinators [40,41] and consequently to reduced flower visitation and pollination [19]. Although bumblebees were confined to cages in our study and were thus only exposed to a reduced set of possible flowers to visit, there were many flowers free of pollen beetles available, which could have preferentially been visited by bumblebees. Selective flower visitation would have forced bumblebees to spend more time searching for pollen beetle-free flowers and hence would have resulted in reduced visitation rates or altered flower visit duration. We could, however, not detect any sign of altered flower visitation behaviour across pest control treatments, indicating that this potential mechanism (M1) did not play a significant role in explaining the pronounced synergistic effects found in our study.

Another possible pathway driving synergistic pollination–pest control interactions involves compensatory responses of plants to florivory (M2) [20]. If over-compensation had contributed to the observed synergistic pollination–pest control interactions, either the number of shoots or the number of fruits produced per shoot should have increased with pest levels or plant damage levels, resulting in overall higher yields. However, as the number of shoots remained unaffected, and the number of fruits decreased with decreasing pest control, over-compensation should therefore not have played a major role in contributing to the observed interactive effects in our experiment either.

Furthermore, it is conceivable that the amount of pollen available for pollination could be reduced by florivores or pollen thieves to such an extent that pollination success becomes compromised [42]. Although we cannot exclude the possibility that this pathway contributed to the strong pollination–pest control interactions found in our study, the fact that OSR flowers produce large amounts of pollen [43] and many flowers remained uninfested by pollen beetles, including in the cages with high pollen beetle densities (L. Sutter 2014, personal observation), may suggest that the pollen pool available for pollination was probably sufficient and this interaction pathway, therefore, probably did not play a major role in our study.

Here, we propose an alternative and—to our knowledge—novel mechanism as the principal driver of the strong synergistic interactions of pest control and pollination: florivory-induced reduction in flower lifetime (M3). Acceleration of flower senescence in OSR occurs via the removal of pollen from the stamens, rather than pollen deposition on stigmas [22]. Our findings provide a strong indication that pollen beetles trigger such accelerated flower senescence through their removal of pollen from stamens. Pollen beetles reduced flower lifetime by an average of 50% at high compared with low densities. This shortening in flower lifetime, demonstrated in a complementary experiment specifically designed to test this hypothesis (see electronic supplementary material, figure S1 for detailed results), reduces the estimated average number of pollinator visits a flower receives during its lifetime from 2.0 to 1.2 visits. This decrease in total pollinator visitation was associated with a decline in seed set of 26%. At an average number of pollinator visits of 1.2 at low pest control, a large proportion of flowers are likely to remain unvisited, probably contributing to the observed reduction in seed set. Lower total pollen deposition, lower proportions of outcross pollen and disadvantages due to weaker pollen competition [44,45] may have thereby reduced the seed set. This should be most pronounced when pollinator densities are limited in real agro-ecosystems; a recent study indeed indicates that enhancing pollinator densities can increase OSR yield, at least in the studied region [37].The aim of this study was to experimentally test a set of possible mechanisms that act on a local scale (M1–M3). However, future work should also address other potential pathways of interactions on a larger scale (field or landscape), including direct interactions between pollinating and pest control-providing organisms, which may reveal additional pathways for interactive pollination–pest control effects that have not been studied here. While controlled experiments allow for rigorous hypothesis testing, a potential drawback is the limited applicability of findings to real-word systems. In the present experimental study, however, we believe this potential limitation is minimized by (i) using two different, naturally occurring levels of pest control, (ii) calibrating pollinator visitation rates based on our own and published field data of natural visitation rates and by (iii) measuring yield parameters according to standard agronomic practice. Hence, yield and other crop plant parameters, as well as crop damage and pollinator visitation rates are all in the range reported in other field studies (e.g. [14]). It is important to measure agronomic metrics of yield because damage or effects on seed set do not necessarily translate into crop yield [46].

5. Conclusion and implications

Our study clearly shows that insect pollination and pest control can interact in highly non-additive ways with profound consequences on crop yield and economic value. To improve predictions of the contribution of pest control and pollination to crop yield, current models (e.g. [47]) should be refined by integrating these interactions. Our results could provide a basis for such improved predictions of OSR yield. It remains an important challenge for future ES research to obtain such data for other important crops in a range of agro-ecosystems. Without taking non-additive interactions among multiple ES into account, estimations of ES and their use in single and multiple ES models [48], spatial ES value mapping [49] or benefit transfer functions [50] are not reliable and can even be misleading. Our findings also have profound implications for ecosystem management [51]. Although the drivers of pest control and pollination in agro-ecosystems have been studied well in isolation, there is evidence that shared drivers, such as land-use change, can jointly affect multiple ES [15]. Our findings highlight that the effectiveness of measures aimed at mitigating pollinator losses, to enhance crop pollination services, may fail to deliver economic yield benefits if pest control services are not concomitantly addressed. By contrast, integrated management of multiple ES could be a promising and cost-effective approach towards ecological intensification (sensu [3]) by taking full advantage of synergies among multiple ES. Yet, to effectively and sustainably manage agro-ecosystems for multiple ES, more research aimed at a better understanding of the interactions among ES is vital.

Data accessibility

All data associated with this manuscript are available at the Dryad Digital Repository: http://dx.doi.org/10.5061/dryad.gm11d.

Authors' contributions

L.S. and M.A. designed the study, L.S. and M.A. performed the research, L.S. analysed the data and L.S. and M.A. wrote the manuscript.

Competing interests

We have no competing interests.

Funding

This project has received funding from the European Union's Seventh Framework Programme for research, technological development and demonstration under grant agreement no. 311879.