Introduction

Plant–pollinator interactions are central to the survival of many plant species (Ollerton, Winfree & Tarrant 2011), but our understanding of the processes that govern pollinator activity at the community and patch scale is limited, particularly in relation to plant species distributions (Hegland & Boeke 2006). Nevertheless, it is accepted that pollinator choices at the patch scale are associated with two aspects of the plant community (Feinsinger 1987). The first is the flower density of the patch, which is often positively correlated with increase in pollinator visitation rate (e.g. Ghazoul 2005; Hegland & Boeke 2006; Dauber et al. 2010), probably due to an increase in patch attractiveness (e.g. Sih & Baltus 1987) and a minimization of foraging costs (Hegland & Boeke 2006). The second aspect is the patch species composition. The decision of a particular pollinator to visit a flower is a complex outcome of its innate preferences and its current learning experience (Lazaro, Lundgren & Totland 2009; Hegland & Totland 2012). Therefore, the available flower resources and the variation in the flowers’ reward properties in a specific patch will play a major role in the choice process of the pollinators (Clegg & Durbin 2000; Klinkhamer, de Jong & Linnebank 2001).

The outcome of pollinator choices strongly affects the plant reproductive success in a particular patch. One of the well-known examples for a potential positive influence of a patch composition on the reproductive success of the species growing in it is the magnet species effect. Here, a rewarding, highly attractive species increases pollination success of less attractive species growing in its vicinity by locally increasing pollinator abundance. The extent of the positive effect depends on a close spatial association between the species (e.g. Laverty 1992; Johnson et al. 2003; Molina-Montenegro, Badano & Cavieres 2008). On the other hand, several studies have shown that the proximity of highly attractive species creates a competitive disadvantage for the less attractive species. Two mechanisms were shown to create this disadvantage. First, the attention of pollinators may be directed to the more attractive species, causing them to avoid other flowers in the patch (e.g. Yang, Ferrari & Shea 2011). Secondly, if the pollinators can chose between several flowers in a patch, this may increase heterospecific pollen transfer (e.g. Brown, Mitchell & Graham 2002; Cariveau & Norton 2009).

Recently, Hanoteaux, Tielbörger and Seifan (2013) suggested that the spatial arrangement of the species within a patch may be considered as a third relevant aspect for the pollinator's choice process. In their model, the spatial arrangement of two species, which varied in their attractiveness for pollinators, interacted with the species’ relative density in governing the population persistence of the less attractive species. In particular, a less attractive species had higher pollination success at high densities when segregated from the attractive species, while at low densities, its pollination success was higher when the two species were uniformly distributed. Overall, the model suggests that changes in the spatial pattern of plant species may considerably affect the way pollinators perceive their pollination landscape (sensu Hanoteaux, Tielbörger & Seifan 2013). This may modify foraging behaviour and hence plant reproductive success, by both changing the probability of a certain plant patch to be approached by a potential pollinator from a distance and the way in which the pollinator then forages on individual flowers within the patch (Sieber et al. 2011; Yang, Ferrari & Shea 2011).

To date, only limited attempts have been made to experimentally test the effects of the three factors on pollinator choices (but see Caruso 2002; Feldman 2008; Yang, Ferrari & Shea 2011). To fill this knowledge gap, and to study the potential effects of the three factors together, we experimentally manipulated the presence of a highly conspicuous plant species within a species-rich meadow, and tested the response of pollinators to the patch's flower density, flower composition and spatial arrangement. By working within a natural meadow, we were able not only to test the combined effect of the three factors, but also to maintain a more realistic experiment, in which more than two plant species interact with each other and affect the pollinator's foraging decisions.

Pollinator response was considered as a two-step process, in which the pollinators first selected a particular patch in the meadow and then choose among flower species within the patch. Namely, we first studied whether the flower density and the composition of the patch (i.e. the addition of a highly conspicuous species) affected pollinator patch choices. Secondly, we studied whether the visit choices within a patch were affected by the density and spatial arrangement of a highly conspicuous species. We expected (i) the addition of a conspicuous species to increase the number of visits to a plant patch. Within a patch, we predicted that (ii) neighbouring species will be differently affected by the experimental manipulations. Specifically, the presence of a conspicuous species will have a stronger effect on neighbouring plants that are considered attractive by the pollinators (see also Gibson, Richardson & Pauw 2012). Furthermore, we predicted that (iii) an aggregated distribution of the conspicuous species will increase the number of visits to the neighbouring plants relative to a uniform distribution. Moreover, this pollinator response will generate a positive effect on the reproductive success of neighbours (in terms of visits and seed production). (iv) Finally we assumed that different pollinator groups will respond differently to the experimental manipulations.

Material and methods

Results

Overall, 272027 visits from potential pollinators were observed on nine plant species (the above-mentioned focal species plus all ‘others’) in 27 observation rounds.

Choices among plots

Overall, the number of pollinator visits in each of the observation dates was not significantly affected by the location of the plot within the field (Fig. S2). However, the number of pollinators visiting a plot was positively affected by the total flower density (F_1,132 = 277.72; P < 0.001; Fig. 1a) and the density of C. cyanus (F_1,132 = 42.23; P < 0.001; Fig. 1b) in the plot.

Choices within plots

The number of pollinator visits to the manipulated species C. cyanus was positively affected by its density (Table 1; Fig. 2a) regardless of its spatial distribution. In addition, the interaction between density and spatial pattern was significant, showing a faster positive response to density when C. cyanus was aggregated. Although not significant, there was a tendency for more visits and higher seed set in plots in which C. cyanus was aggregated (Table S4 and Fig. S3).

Table 1. Effect of density and spatial pattern on the number of pollinator visits to the studied plant species (per 15 min)

Species	Total	Target	cyanus	Spatial	cyanus × spat
The following effects were tested in the models: total – total inflorescence density per plot; target – intraspecific density of the studied species per plot; cyanus – the manipulated species Centaurea cyanus density per plot; spatial – the spatial arrangement of C. cyanus per plot; cyanus × spat – the interactions between spatial arrangement and C. cyanus density. Numbers represent F values (with 1, 103 d.f. for C. cyanus and 1, 123 d.f. for the other studied species, except spatial pattern, where 2, 103 and 2, 123 d.f. were used, respectively). ↑/↓ – indicate a positive/negative coefficient values (values can be found in Table S3). *0.01 < P-value < 0.05; **0.001 < P-value < 0.01; ***P-value < 0.001.
Manipulated
C. cyanus	53.03↓***		835.80↑***	0.05	61.73***
Congeneric
C. jacea	16.13↑***	297.36↑***	24.59↓***	4.40*	5.85*
Conspicuous
K. arvensis	0.01↓	205.71↑***	97.01↓***	0.89	5.45*
S. pratensis	14.22↑***	54.87↑***	0.58↓	0.03	0.07
R. acris	7.72↑**	0.01↓	20.54↓***	3.41*	4.48*
Less conspicuous
L. corniculatus	1.01↓	7.92↑**	0.88↑	0.09	4.55*
T. pratense	4.47↓*	44.18↑***	7.05↑**	0.26	0.08
A. millefolium	0.32↓	54.02↑***	3.80↓*	1.65	8.38**

Effect on neighbouring species

While the number of visits to each of the neighbouring species had a different response to the total plot density, their response to their intraspecific density and to the density of the manipulated species C. cyanus was similar (Table 1). In all observed neighbouring species, except R. acris, the number of pollinator visits was positively associated with the species’ own density (Fig. 3), and in all cases except for T. pratense and L. corniculatus, the number of visits was negatively associated with C. cyanus density.

The interaction between C. cyanus density and its spatial pattern showed a different effect on the visits to the conspicuous and less conspicuous neighbours (Table 1). The effect on the conspicuous species group (including the congeneric species) was twofold. First, although generally C. cyanus density decreased pollination visits to the neighbouring conspicuous species, the number of pollinators visiting these neighbours was higher at low C. cyanus density than in the control plots, where no C. cyanus was present (Fig. 2b,c; S4). Secondly, the number of pollinator visits to conspicuous neighbours, at least at low densities of C. cyanus, was higher when the manipulated species was regularly distributed relative to when it was aggregated. The response of the less conspicuous neighbouring species was less unified, but in all three cases the response to aggregated C. cyanus was more positive than when it was regularly distributed, at least at low densities (Fig. 2d; S4).

The effects of C. cyanus density and its spatial pattern on the seed production of the neighbouring species were relatively weak and showed only significant associations with intraspecific densities (see Table S4). However, although not significant, the response of each of the neighbour groups to C. cyanus spatial distribution was detected also in the neighbours’ seed production (Fig. S3).

Pollinator group responses

All pollinator groups responded positively (i.e. increased visitation rate) to increased density of all focal plant species (Table 2). An increased density of C. cyanus generally had a negative effect on the number Apis visits to neighbouring species (Table 2; Fig. 4). However, in the cases of C. jacea and K. arvensis, the number of visits in the presence of low densities of C. cyanus was significantly higher than in the control plots (Fig. 4b,d). These two neighbouring species also responded significantly to the spatial arrangement of C. cyanus spatial pattern, though the effect was not consistent.

Table 2. Effect of density and spatial pattern on the number of visits (per 15 min) to the studied plant species of specific pollinator groups

Effect	Manipulated	Congeneric	Conspicuous		Less conspicuous
	C. cyanus	C. jacea	K. arvensis	S. pratensis	L. corniculatus	T. pratense	A. millefolium
The following effects were tested in the models: total – total inflorescence density per plot; target – intraspecific density of the studied species per plot; cyanus – the manipulated species C. cyanus density per plot; spatial – the spatial arrangement of C. cyanus per plot; cyanus × spat – the interactions between spatial arrangement and C. cyanus density. Numbers represent F values (with the relevant d.f.). ↑/↓ – indicate a positive/negative coefficient values (values can be found in Table S5). *0.01 < P-value < 0.05; **0.001 < P-value < 0.01; ***P-value < 0.001. †Negative binomial distribution.
Apis
Total	63.5 (1,103)↓***	25.1 (1,83)↑***	4.6 (1,123)↑*	15.2 (1,83)↓*	6.2 (1,123)↑*
Target		226.4 (1,83)↑***	195.1 (1,123)↑***	54.6 (1,83)↑***	0.1 (1,123)↑
cyanus	867.3 (1,103)↑***	47.0 (1,83)↓***	163.2 (1,123)↓***	2.3 (1,83)↓	24.5 (1,123)↓***
Spatial	0.2 (2,103)	5.5 (1,83)**	3.6 (1,123)*	0.1 (1,83)	1.5 (1,123)
cyanus × spat	77.9 (1,103)***	10.6 (1,83)**	0.1 (1,123)	0.4 (1,83)	0.1 (1,123)
Bombus
Total	50.9 (1,103)↓***	7.2 (1,83)↓**	0.3 (1,123)↓		12.0 (1,123)↓**	1.7 (1,123)↑
Target		24.3 (1,83)↑***	15.2 (1,123)↑**		1.2 (1,123)↑	1.4 (1,123)↑
cyanus	81.3 (1,103)↑***	14.4 (1,83)↑***	0.8 (1,123)↓		9.0 (1,123)↑**	0.1 (1,123)↓
Spatial	5.4 (2,103)*	1.6 (1,83)	4.1 (1,123)*		0.3 (1,123)	0.1 (1,123)
cyanus × spat	6.1 (1,103)*	11.3 (1,83)**	0.4 (1,123)		11.2 (1,123)**	3.2 (1,123)
Flies
Total	0.8 (1,103)↓	2.6 (1,83)↓	39.8 (1,123)↓***				0.9 (1,117)↑
Target		4.3 (1,83)↑*	5.5 (1,123)↑*				33.4 (1,117↑***
cyanus	7.2 (1,103)↑**	6.9 (1,83)↑*	62.4 (1,123)↑***				11.7 (1,117)↓***
Spatial	1.8 (1,103)	1.9 (1,83)	3.8 (1,123)*				0.8 (1,117)*
cyanus × spat	0.5 (1,103)	0.3 (1,83)	5.2 (1,123)*				10.3 (1,117)**
Hoverflies
Total	0.6 (1,39)↑	0.5 (1,30)↑	0.5 (1,46)↓				2.8 (1,50)↑
Target		3.4 (1,30)↑	0.4 (1,46)↑				0.4 (1,50)↑
cyanus	1.1 (1,39)↑	0.2 (1,30)↑	0.1 (1,46)↑				2.6 (1,50)↓
Spatial	0.1 (1,39)	1.4 (1,30)	0.2 (1,46)				0.5 (2,50)
cyanus × spat	1.5 (1,39)	0.1 (1,30)	0.1 (1,46)				0.4 (1,50)
Butterflies
Total	3.7 (1,43)↓	2.0 (1,27)↑	0.7 (1,27)↓
Target		9.1 (1,27)↑**	9.9 (1,27)↑**
cyanus	14.7 (1,43)↑***	0.2 (1,27)↓	0.4↓(1,27)↓
Spatial	0.1 (2,43)	0.4 (2,27)	1.3 (2,27)
cyanus × spat	3.7 (1,43)	0.4 (1,27)	0.1 (1,27)

Unlike the Apis individuals, the number of visits to neighbouring plants made by Bombus individuals was generally positively associated with an increase in the density of C. cyanus (Table 2). In the case of C. cyanus and K. arvensis, Bombus individuals responded to C. cyanus spatial distribution, visiting the focal species more in plots in which C. cyanus was regularly distributed relative to those in which it was aggregated. The response of flies was similar to those of Bombus. Generally, there was an increase in the number of visits to neighbouring species with increased density of C. cyanus. In addition, flies visiting K. arvensis and A. millefolium showed a response to C. cyanus spatial arrangement, in which at least at low C. cyanus densities, aggregated distribution increased the number of fly visits to the two neighbouring species. The hoverfly and butterfly groups did not show any significant trend in their response to the presence of C. cyanus or its spatial arrangement (Table 2).

Discussion

Overall, our study supports the assumption that flower density is a crucial factor in the determination of pollinator choices at the patch level (Goulson 1999; Hegland & Boeke 2006). In addition, we showed that the presence of a highly conspicuous species strongly contributed to the attractiveness of the patch and that this increase in attractiveness had an effect also on within patch choices of pollinators. In particular, we demonstrated that the conspicuous species may function as a magnet species to its neighbours, but that this role may change into strong competition for pollinators when density of the conspicuous species is increasing (see also Munoz & Cavieres 2008). Furthermore, the contribution of the conspicuous species to its neighbours shifted predictably not only with density but also with its spatial arrangement.

The results support our assumption that spatial arrangement of flowers within a patch will influence pollinator choices and that this influence is different according to the conspicuousness of the neighbouring plant. This is apparent when looking at the pollinator choices of neighbouring flowers at low densities of the manipulated species: while the more conspicuous neighbours (i.e. the congeneric and conspicuous neighbour groups) received more visits when the manipulated, highly attractive species was regularly distributed, the less conspicuous neighbours received more visits when the highly attractive species was aggregated within the patch. These results may indicate that the two groups of neighbouring flowers benefited from different aspects of the pollinator foraging behaviour. On the one hand, assuming that the pollinators considered the conspicuous neighbours as relatively attractive (even if less than C. cyanus), when the conspicuous species were interspersed with C. cyanus, they might have offered a sufficiently attractive alternative to persuade some of the pollinators to switch between flower resource. In accordance with this interpretation, only when C. cyanus was aggregated within the plot, did it provide enough closely located flower resources to cause the pollinators to avoid other neighbours. On the other hand, the less conspicuous neighbours probably ‘trapped’ pollinators in the local plant choices when separated from the aggregated C. cyanus, even when more attractive alternatives were available in the patch, as predicted in our recent theoretical study (Hanoteaux, Tielbörger & Seifan 2013). A similar observation was reported in a laboratory experiment by Keasar (2000). She found that less attractive flowers (non-rewarding in her case) were visited more often when separated from the more attractive (i.e. rewarding) ones. This may be because the aggregation of the attractive species reduced its encounter rate in other parts of the plot, which led to a reduction in the pollinators’ ability to compare between the attractive and less attractive species, thus increasing visitation rate to other plant species (see also Internicola et al. 2007).

The effect of the composition and spatial pattern of plant species on pollinator choices may also shed light on the interactions between native and invasive plants. Invasive species are often regarded as highly attractive because of their relatively large flowers and reward (e.g. Bjerknes et al. 2007). Nevertheless, recent experimental studies showed that although visitation rates to plots to which invasive species were introduced usually increased, seed set of native species within those plots was reduced (Totland et al. 2006; Bjerknes et al. 2007). Such outcomes may be interpreted in the light of our experiment, because highly attractive invasive species probably show opposing effects at different scales of the pollinator foraging process: at a larger scale, the invasive species attract more pollinators to their vicinity, thus increase visitation rate to patches in which they are present (e.g. Bjerknes et al. 2007). However, at the patch scale, native neighbours may benefit from the additional pollinators only if the manipulated density of the invasive species is low (Munoz & Cavieres 2008; Morales & Traveset 2009) and when there is a spatial segregation between the plant species (Cariveau & Norton 2009).

Interestingly, the effect of the attractive species on shared pollination services in our study cannot be simply categorized according to the floral traits of the species (e.g. morphology, colour). Numerous classical works connected floral trait syndromes with particular pollinator groups (e.g. Goulson 1999; Johnson & Steiner 2000), and especially predicted that more simple floral structures, in which plant rewards are easily collected, will attract more pollinator types (e.g. Johnson & Steiner 2000; Pellissier et al. 2010). However, only few studies focused on the potential effect of floral traits on indirect plant–plant interactions (Gibson, Richardson & Pauw 2012). These studies generally assumed that plant species with similar floral traits will share similar pollinator groups (Danieli-Silva et al. 2012) and thus will have a higher potential to affect each other's pollination success (Hegland & Totland 2005). Nevertheless, our study indicated that within patches (unlike at larger scales, e.g. Pellissier et al. 2010), most of the neighbouring plants were affected (positively or negatively) by the presence of C. cyanus, regardless of their floral traits. This general result suggests that indirect plant–plant interactions were strongly affected by shifts in pollinator choices even if the plant species have different floral traits (e.g. they belong to different ‘pollination guilds’). A possible explanation may lie in the major role of floral display size and nectar level in determining flower attractiveness in temperate regions (Hegland & Totland 2005). Unlike our prediction, because the inflorescences of all our studied species had similar sizes and all of the species are known to produce relatively high amount of nectar (BiolFlor data bank, Klotz, Kühn & Durka 2002), their attractiveness level to the pollinators may have been more similar than we assumed.

Conclusions about group-specific pollinator behaviour should be made with caution, because observations on the pollinator group level included many confounded effects. The coarse classification we made may have grouped together species whose foraging activities and energetic needs are different. Nevertheless, the analyses revealed some common patterns. The most general is that all pollinator groups responded positively to the visited plant intraspecific density, supporting the general notion that plant density is the central factor affecting pollinator foraging behaviour (e.g. Hegland & Boeke 2006).

Among the analyses of the various pollinator groups, the most striking result was the highly negative effect of increased C. cyanus density on the number of visits of Apis individuals to neighbours. This observation may be explained by the species’ fast learning ability, which is combined with highly constant foraging behaviour (see review in Goulson 1999). Because of this combination, the presence of a highly attractive resource (such as the C. cyanus) might have caused Apis individuals to rapidly switch their preference towards the new resource and avoid other plant species. Nevertheless, the presence of C. cyanus at low densities increased Apis individual visits also to many neighbouring plants, and the behaviour was often sensitive to changes in the spatial pattern of the plants. Interestingly, S. pratensis (Lamiaceae) was the only conspicuous neighbour with enough observations that always responded negatively to the presence of C. cyanus. The strong negative response of this species to the presence of another highly attractive species may be a consequence of the clear morphological differences in the shape of the two species and an indication to the lower tendency of honeybees to switch their constancy during a bout (Chittka, Thomson & Waser 1999). Bombus and fly groups also showed a general sensitivity to the presence of C. cyanus and its spatial pattern, although the direction of their response was sometimes contradictory. This may be explained by the lower fidelity of flies and bumblebees to specific flower species within about (Goulson 1999). Unlike the above-mentioned groups, the hoverfly and butterfly groups did not respond to our manipulations. Although a potential explanation may be that the groups were too heterogeneous for a clear trend, our results fit previous conclusions that these groups are highly generalist and are therefore not expected to show clear flower preferences or to act as effective pollinators (e.g. Branquart & Hemptinne 2000; Danieli-Silva et al. 2012). Hoverflies in particular were suggested to be less mobile and to require less energy than bee species, suggesting that they discriminate less among food resources within a patch while foraging (Hegland & Boeke 2006).

Conclusions

Our study created a unique bridge between theoretical explanations concerning pollinators and their effect on the plant community and shared pollination services. We demonstrated that relative density and spatial distribution of highly conspicuous species interact in dictating pollinator choices not only between plant patches but also in their preferences of visited plants within a patch. Namely, as long as the density of the conspicuous species was relatively low and its distribution regular, it induced visits to other relatively conspicuous species in the patch. However, with increasing density, and when its distribution was more aggregated, the effect shifted towards a strong competition for pollinator visits. Less conspicuous neighbours showed a less uniformed response, but often benefitted more from aggregated distribution of the highly conspicuous species. These results may shed light on the process of pollinator foraging and its potential outcome for the visited plants. In particular, our study supports our theoretical prediction that to predict the outcome of plant–pollinator interactions, the specific composition and community structure, including spatial distribution and species identity, should be taken into considerations in addition to the more traditional focus on species density.

Acknowledgement

We acknowledge funding by the DFG (SE1840/1-1) to MS and KT and from the Landesgraduiertenförderung Baden-Württemberg to SH. We are thankful to Bert Van de moortel and Johannes Kondschak for their help in the field, to Itamar Giladi for help with the spatial analysis and to Tal Seifan for constructive criticism in all stages of the project. This is publication number 832 of the Mitrani department of Desert Ecology.

Data accessibility

Data available from the Dryad Digital Repository (Seifan et al. 2014).