Florally rich habitats reduce insect pollination and the reproductive success of isolated plants

Abstract Landscape heterogeneity in floral communities has the potential to modify pollinator behavior. Pollinator foraging varies with the diversity, abundance, and spatial configuration of floral resources. However, the implications of this variation for pollen transfer and ultimately the reproductive success of insect pollinated plants remains unclear, especially for species which are rare or isolated in the landscape. We used a landscape‐scale experiment, coupled with microsatellite genotyping, to explore how the floral richness of habitats affected pollinator behavior and pollination effectiveness. Small arrays of the partially self‐compatible plant Californian poppy (Eschscholzia californica) were introduced across a landscape gradient to simulate rare, spatially isolated populations. The effects on pollinator activity, outcrossing, and plant reproduction were measured. In florally rich habitats, we found reduced pollen movement between plants, leading to fewer long‐distance pollination events, lower plant outcrossing, and a higher incidence of pollen limitation. This pattern indicates a potential reduction in per capita pollinator visitation, as suggested by the lower activity densities and richness of pollinators observed within florally rich habitats. In addition, seed production reduced by a factor of 1.8 in plants within florally rich habitats and progeny germination reduced by a factor of 1.2. We show this to be a consequence of self‐fertilization within the partially self‐compatible plant, E. californica. These findings indicate that locally rare plants are at a competitive disadvantage within florally rich habitats because neighboring plant species disrupt conspecific mating by co‐opting pollinators. Ultimately, this Allee effect may play an important role in determining the long‐term persistence of rarer plants in the landscape, both in terms of seed production and viability. Community context therefore requires consideration when designing and implementing conservation management for plants which are comparatively rare in the landscape.


| INTRODUCTION
Changes to the availability and diversity of floral resources through altered land use, including increased landscape fragmentation and simplification, can have considerable impacts on the structure, abundance, and diversity of pollinator communities (Potts et al., 2016;Senapathi et al., 2015;Vanbergen et al., 2013). With an estimated 87.5% of flowering plant species worldwide at least partly reliant upon pollinators for reproductive success and long-term survival, this will have direct implications for plants (Ollerton, Winfree, & Tarrant 2011).
By transferring conspecific pollen between plant individuals, pollinators not only facilitate seed production but have important effects on fitness and population genetic diversity by increasing outcrossing and the exchange of novel alleles (Frankham, 2005;Levin & Kerster, 1974;Mannouris & Byers, 2013).
Plant-pollinator interactions vary with plant population size, density, and habitat context (Essenberg, 2012;Mayer, Van Rossum & Jacquemart 2012). Habitats supporting a high abundance and species richness of flowering plants may either enhance or disrupt the transference of pollen to plants (Blaauw & Isaacs, 2014;Vanbergen et al., 2014). The outcome depends on pollinator visitation patterns, which are determined, in part, by the demography and characteristics of a species' population relative to heterospecific co-flowering plants (Essenberg, 2012). For instance, when at low floral densities, co-flowering heterospecific plants can facilitate pollinator visitation to a plant population by enhancing the overall attractiveness of a floral patch (Rathcke, 1983). At high floral densities, co-flowering heterospecific plants may result in inter-specific competition for pollinators, which can reduce per capita visitation to a plant population, resulting in an insufficient supply of pollen that limits potential seed set (Ghazoul, 2006). Alternatively, although pollinators may prefer foraging on particular plant species (Chittka, Thomson, & Waser, 1999;Gegear & Laverty, 2005;Waser, 1986), such fidelity may be relaxed in communities with high floral diversity, increasing the potential for inter-specific pollen transfer (Fontaine, Collin, & Dajoz, 2008). This has potential negative implications for plant reproduction. The supply of conspecific pollen to a plant can be reduced if it is lost during visitation to heterospecific plants (Wilcock & Neiland, 2002); moreover, the deposition of heterospecific pollen, by clogging the stigma and style of conspecific plants, can inhibit pollination (Holland & Chamberlain, 2007). Both lead to reduced pollination effectiveness and ultimately a reduction in plant seed set.
Pollinators face a metabolic trade-off when foraging for pollen and nectar (Charnov, 1976;Vaudo, Patch, Mortensen, Tooker, & Grozinger, 2016) and optimal foraging theory predicts that they will maximize gain and minimize loss of energy (Charnov, 1976). Thus, pollinators may forage slowly through habitats rich in floral resources, minimizing travel distances between flower visits, and either avoid or promptly traverse florally poor habitats (Lander, Bebber, Choy, Harris, & Boshier, 2011;Pasquet et al., 2008). Moreover, pollinator forging distances have been shown to exhibit an inverse relationship with the proportion of available foraging habitat (Carvell et al., 2012). Pollinator sensitivity to the dispersion of floral resources at different spatial scales is partly influenced by traits, such as body size, that predict their mobility and capacity to forage and disperse pollen (Greenleaf, Williams, Winfree, & Kremen, 2007;Redhead et al., 2016). Given the capacity of pollinators to mediate plant gene flow, changes in foraging behavior or pollinator community composition (e.g., body size distributions) in response to variation in habitat floral resources may profoundly affect plant fitness (Vanbergen et al., 2014;Ward, Dick, Gribel, & Lowe, 2005). This may be particularly important for spatially isolated populations of uncommon plant species because increases in floral diversity might lead to greater inter-specific plant competition for pollinators (Ghazoul, 2006) and reduce the probability of long-distance pollen dispersal (Eckert et al., 2010).
One approach to understanding the interaction between floral community diversity and pollinator-mediated gene flow in locally rare plant populations is to analyze plant mating patterns using highly variable molecular markers (microsatellites). This permits inference, and even direct observation, of patterns of gene movement and mating (Ashley & Dow, 1994), enabling the quantification of relatedness between plants (Ashley & Dow, 1994). The use of such molecular methods has revealed that plant populations often exhibit spatial genetic structure, where relatedness declines with distance between individuals (Loveless & Hamrick, 1984). Increased frequency of mating between close relatives within plant populations can lead to biparental inbreeding, resulting in reduced allelic diversity and greater homozygosity, which has been linked to a reduction in the fitness and long-term survival of plants (Byers & Waller, 1999). Low allelic diversity is particularly detrimental for self-incompatible plants whose reproduction requires allelic variation at a single locus (the 'S-locus'; Byers & Meagher, 1992). Although mutations can cause self-incompatibility systems to break down, resulting in partial self-compatibility, self-fertilization and mating between close relatives in these plants is typically prevented (Richards, 1997). As S-alleles are frequently lost through genetic drift, plant populations could face a reduction in compatible mates with negative implications for plant reproduction (Wagenius, Lonsdorf, & Neuhauser, 2007). Self-incompatibility coupled with spatially structured populations may therefore render some plant species vulnerable to reductions in gene flow due to altered pollinator foraging behavior.
In this study, we investigated how the genetic connectivity and reproductive success of a locally rare and partially self-compatible plant species was affected by habitat floral cover and the activity and richness of pollinator communities. To simulate a species occurring at low population densities, we deployed small arrays of Californian poppy (Eschscholzia californica) into a landscape-scale field experiment where floral cover had been manipulated through agri-environment planting of wildflower patches. In these experimental arrays, we measured pollinator activity, insect-vectored pollen movement using microsatellite genotyping, seed set, and progeny viability. Based on previous observations of altered pollinator behavior in response to floral cover (Heard et al., 2007), we hypothesized that: i. Habitats supporting high floral cover will increase the activity densities and richness of pollinator species in the vicinity of experimental arrays of a partially self-compatible plant (E. californica); ii. The body size distribution of pollinators would be greater in florally rich habitats, reflecting the preference of Bombus spp. to flower species within sown wildflower patches (Carvell, Meek, Pywell, Goulson, & Nowakowski, 2007); iii. Pollen movement between introduced experimental arrays of E. californica would be reduced in florally rich habitats, leading to pollen limitation, lower outcrossing rates, and fewer long-distance pollination events; iv. The reproductive success (seed set and progeny viability) of Eschscholzia californica would be reduced in florally rich habitats, reflecting a higher incidence of self-fertilisation.

| Experimental site and study system
The experiment was conducted on the Hillesden estate in Buckinghamshire, UK (1°00′01′'W, 51°57′16′'N), an intensive arable farm (~1000 ha) situated on heavy clay soils with a relatively flat topography. Since 2005, a number of experimental landscape management "treatments" have been established and managed across the estate within a randomized block design. These treatments, applied to 50-60 ha replicated land parcels, comprise varying proportions (0-8% of land out of production) of a range of wildlife habitat restoration options (including pollen and nectar-rich flower margins and wildflower patches for pollinators) under compliance with the English agri-environment scheme (Pywell et al., 2015). Overall, these wildlife habitats comprised ~4% of the total area ( Figure 1).
To test our hypotheses, we introduced the Californian poppy, Eschscholzia californica Cham., (Papaveraceae) (Seed source: Chiltern seeds Ltd., Wallingford, UK). Although considered naturalized in the UK (Preston, Pearman, & Dines, 2002), E. californica was locally absent, allowing us to unequivocally ascribe paternity in mating events. Eschscholzia californica is a diploid species, with a partially self-compatible mating system, characterized by a low propensity to self-fertilize (Wright, 1979), and thus predominantly requires insects for pollen transfer (Becker, Gleissberg, & Smyth, 2005). It possesses large, open flowers and is visited by a variety of insects from the orders: Diptera, Hymenoptera, and Coleoptera (summarized in; Cook, 1962 separated by >500 m to minimize between block movement of insect pollinators ( Figure 1). At the center of each block, four experimental arrays were placed at 50 m intervals along a 150-m transect laid symmetrically across the boundary between an established wildflower patch (henceforth "florally rich" habitat) and bare, fallow ground (henceforth "florally poor" habitat) ( Figure 1). This ensured the first two arrays on a transect were located within the florally rich habitat, and the second two arrays within the florally poor habitat. The use of agri-environment scheme wildflower patches, sown with a common mix of 25 species, including Trifoilum pratense, Centurea nigra, and Leucanthemum vulgare at a rate of 37 kg/ha , allowed for the standardization of florally rich treatments across the four blocks. To ensure our habitat classification was accurate, prior to the start of the experiment, we  Table S1).

| Pollinator activity and species richness
Pan traps are typically deployed to describe pollinator species richness and activity densities (Westphal et al., 2008). They have also been used to provide a surrogate measure of visitation, allowing for longer periods than standard observation methods (Ricketts et al., 2008). However, this survey method has been recognized to exhibit bias (Roulston, Smith, & Brewster, 2007) because the attractiveness of pan traps depends upon habitat and landscape context (Baum & Wallen, 2011). Pollinators are less likely to encounter traps when floral resources are abundant and more likely to encounter traps when floral resources are scarce, that is, capture rates are proportional to visitation rates per unit flower area (Veddeler, Klein, & Tscharntke, 2006). We exploited this phenomenon to measure the attractiveness and pollinator activity density at our experimentally rare plant populations located within different habitats.
Pan traps comprised three water-filled circular plastic bowls (80 × 200 mm) painted with nontoxic fluorescent paint (1 yellow, 1 blue and 1 white; UV Gear, UK) placed in the center of each array.
Traps were deployed for 24 hr at each of the 16 arrays on the same day, twice weekly over the 16-day study period (totaling four surveys).
Each survey was performed in randomized order, between 0930 and 1700. Emptied traps were left in situ to maintain the same levels of visual attractiveness to foraging insects throughout the experiment.
All insects from the main pollinator groups (Hymenoptera: Apoidea, Diptera: Syrphidae and Lepidoptera) were counted and identified to species level. In addition, given that insect pollinator body mass correlates with foraging range (Greenleaf et al., 2007) and to a lesser extent, pollen deposition (Larsen, Williams, & Kremen, 2005), we measured the intertegular span (the distance between the wing bases) of each insect from the main pollinator groups using digital calipers (given the relationship between intertegular span and body mass (Cane, 1987)). From this we determined the body size distribution of pollinator communities.
To ensure pollinators caught within pan traps could be used as a proxy for visitation, these data were calibrated by direct visitor obser- was recorded and identified to a broad pollinator group as above.
Once at seedling stage, 50 mg of fresh leaf material was removed

| Pollen movement
To detect pollination events, we genotyped approximately ten progeny per plant from each of the 48 field exposed plants (Mean ± SE = 9.52 ± 0.39) using 50 mg of fresh leaf material and following protocols as above. The incidence of self-fertilization in plants from each habitat was calculated manually by individually comparing each successfully amplified progeny against their maternal plant. If, at each of the seven loci, the progeny was a complete match for the maternal genotype, or was homozygote for one of the maternal plants alleles, it was scored as selfed. Alternatively, if any novel alleles were observed in the progeny that were not present in the maternal plant, the progeny was classified as outcrossed. Paternity was determined using Cervus 3.0.7 (Kalinowski, Taper, & Marshall, 2007), where each progeny sample was listed detailing alleles at the seven microsatellite loci, specifying the known maternal sample as well as the potential paternal samples. Here, we analyzed all progeny from within a block against all potential parents within that block. We accounted for selffertilization and selected for the most likely paternal parent based on a derivative of likelihood ratios; the delta score (∆), which is the difference between the likelihood score of the most likely parent and the second most likely parent (Marshall, Slate, Kruuk, & Pemberton, 1998). We only included assignments with a trio ∆ confidence (the likelihood score of a mother-father-offspring match) above 95%, which is classified as high confidence (Marshall et al., 1998). For all paternal assignments, we recorded which habitat, if any, the pollen had crossed together with the distance travelled.

| Plant fitness components: seed production, germination rates, and progeny traits
All open flowers were removed from the 48 genotyped E. californica plants, prior to their placement in premarked locations across the landscape. They remained in the field for 16 days to ensure full anthesis of new flowers (which takes 3-4 days; Becker et al., 2005) and to allow for multiple pollination events. After this period, all fruit were tagged to ensure that only fruit development arising from the period of the field experiment were included in analyses. Plants were then collected and stored under controlled glasshouse conditions (as above) until fruit maturation. Upon maturation, tagged fruit were collected and the number of filled seeds per fruit was counted to quantify seed set per plant.
To determine whether field exposed plants were limited by pollen, we supplemented a flower from each of the 48 plants with outcrossed pollen. This involved methodically wiping four dehiscing anthers from a donor plant onto the receptive stigma of a field exposed plant with dissecting tweezers. Supplemented flowers were then covered with fine muslin to protect against accidental windborne transfer of pollen from the glasshouse air-conditioning system. Once matured, fruit were collected and the number of seeds per fruit was counted to determine maximum seed set. The degree of pollen limitation was expressed as a ratio between the actual seed set (field exposed plants) and the potential seed set (supplemented) in each of the 48 field exposed plants.
To measure the viability of progeny from field exposed plants, 20 seeds from each of the 48 plants were sown into compost and kept under glasshouse conditions (as above). Germination was recorded daily over a 30-day period, and any seeds which had not germinated after 90 days were recorded as nonviable. The germination success was expressed as a ratio between the number of seeds which successfully germinated against the number of seeds which failed to germinate in each of the 48 field exposed plants. Indeed, some species and populations of E. californica can exhibit seed dormancy (Cook, 1962), although this was found to be absent within our experimental plants (personal observation).
To further assess how reproduction by self-fertilization affects the viability and growth traits of a partially self-compatible plant, we performed a glasshouse experiment using 40 artificially crossed plants.
On each plant, we emasculated two flowers and supplemented the first with outcrossed pollen and the second with self-pollen. This involved methodically wiping two dehiscing anthers from a donor plant or the focal plant onto the receptive stigma with dissecting tweezers, before covering it in fine muslin. From each supplemented plant, we sowed a seed from the outcrossed fruit and from the selfed fruit (given that selfed fruits predominantly only produced one seed) into 1L pots.
These were then stored under glasshouse conditions (as above). We recorded the following fitness traits: the germination rate, the duration from germination to reproductive maturity (time of first flower), together with the height (cm) and the number of buds at reproductive maturity (biomass).

| Statistical analysis
Pollinator activity density (a proxy for visitation) and the cumulative counts of pollinator species recorded at each experimental array were modeled using generalized linear mixed models (GLMMs) with a Poisson error distribution. When analyzing the body size distribution of pollinator species caught within pan traps however, a log-normal error distribution was instead used to account for non-integers. Plant fitness components were similarly analyzed using GLMMs with a combination of Poisson (seed production per plant) and binomial (pollen limitation of each plant and the germination success of progeny) error distributions.
Within our models, fixed effects comprised of habitat type (florally rich/florally poor). Experimental block (Figure 1) was fitted as a random effect to account for the spatial structure of our experimental design. For pollinator activity models, additional random effects were included to account for survey date and the pollinator species, when analyzing the activity densities (64 surveys) and body size distribution

| Pollinator activity and species richness
Considering insect taxa generally thought to be the most effective pollinators (i.e., Apoidea, Syrphidae, and to a lesser extent Lepidoptera),

| Pollen movement
As expected for a partially self-compatible species, levels of selfing were low in field exposed plants. However, the proportion of progeny that were produced by self-fertilization was marginally greater from plants within florally rich habitats (Florally rich=15%; Florally poor = 9%; χ² = 3.69, df = 1, p = .055). The incidence of selfing was not, however, correlated with pollinator activity densities (p = .097).

| Plant fitness components: seed production, germination rates, and progeny traits
The number of fruits and seeds produced per plant was highly variable (fruit range = 4-23, seed range = 0-589). However, total seed set in arrays within florally poor habitats was 1.8-fold greater than in those within florally rich habitats (GLMM z = −1.980, df = 613, p = .048; Figure 6). Furthermore, the number of additional seeds produced by pollen supplementation was greater in florally rich habitats (GLMM z = 2.396, df = 38, p = .017; Figure 6), indicating that plants were more pollen limited in florally rich habitats. of seeds germinated when produced by self-fertilization, compared to seeds which were a product of outcrossing (outcrossed seeds = 0.8 (n = 33); selfed seeds = 0.6 (n = 24); χ² = 3.91, df = 1, p = .048, phi = 0.25). However, we found no effect of self-fertilization in E. californica on later stage fitness traits (time to reproductive maturity (first flower) p = .210; height at reproductive maturity GLMM p = .078; biomass at reproductive maturity p = .143). The negative implications of self-fertilization were thus limited to reduced germination.

| Habitat effects on pollinator visitation
Consistent with previous work (Veddeler et al., 2006), we found a negative association between florally rich habitats and the activity density and species richness of pollinators. Elsewhere, the abundance and richness of pollinators has been observed to increase with floral cover (Williams et al., 2015), especially where this cover is limited within the wider landscape (Heard et al., 2007). However, our results suggest that despite the increased aggregation of pollinators in habitats providing abundant, diverse floral resources, pollinator visitation, and fidelity is effectively "diluted," which may result in lower per capita visitation and greater interspecific competition for pollination (Sjodin, 2007;Veddeler et al., 2006). Consequently, when embedded within a diverse community of co-flowering heterospecific plants offering a variety of floral pollen and nectar, rare plant species may be unable to co-opt pollinators (Ghazoul, 2006). In contrast, where coflowering, heterospecific competitors were scarce, our findings suggest that available pollinators would become concentrated, leading to potential increases in per capita visitation rates at the individual plant level (Tscharntke et al., 2012;Veddeler et al., 2006).
A diverse community of pollinators can provide niche complementarity (Pisanty, Afik, Wajnberg, & Mandelik, 2016), often leading to enhanced pollen deposition (Larsen et al., 2005) and seed production (Martins, Gonzalez, & Lechowicz, 2015). Alternatively, a high diversity of pollinators visiting diverse plant assemblages can result in an increase in heterospecific pollen deposition, which can interfere with conspecific pollination by stigma clogging (Holland & Chamberlain, 2007). The extent to which the diversity of pollinator species provides a benefit to plants is determined by the functional diversity and pollination effectiveness of communities (Perfectti, Gomez, & Bosch, 2009). Indeed, pollinator species vary in their specialization, pollen carrying behavior, and daily activity preferences, all of which affect pollination effectiveness (Martins et al., 2015;Rader, Edwards, Westcott, Cunningham, & Howlett, 2011). Furthermore, pollination effectiveness has been associated with body size, where larger pollinator species can travel greater distances (Greenleaf et al., 2007) and deposit a larger amount of pollen per visit (Larsen et al., 2005). In this study, however, we found no difference in the size distribution of pollinators between florally rich and florally poor habitats, indicating that by this measure, there was no difference in the trait structure of pollinator communities between habitats with different floral cover that could alter pollination effectiveness. Instead, pollination effectiveness may be driven by changes to the foraging behavior of pollinator communities.

| Habitat effects on pollen movement
Consistent with previous studies, our findings indicate that pollen movement between local populations was strongly affected by the floral composition of a habitat (Dyer, Chan, Gardiakos, & Meadows, 2012;Lander et al., 2011). Pollen movement between experimental arrays (50 m) was greater when the surrounding and intervening habitat comprised livestock grazed grassland or fallow ground with low richness of floral resources. In addition, we found very few pollination events between arrays separated by habitats of high floral cover or those with heterogeneous intervening habitats (i.e., a mixture of habitats comprising high and low floral cover). These results are consistent with our hypothesis that the foraging behavior of pollinator communities is highly determined by habitat composition.
This higher level of pollen movement between populations in florally poor habitats supports research which shows pollinators to conform to the weighted line foraging principle when encountering heterogeneous landscapes (Lander et al., 2013). This principle assumes that pollinators will occupy optimal foraging habitat until resources are depleted, thus making short, energy efficient, movements between flowers. Conversely, pollinators are expected under this principle to move greater distances within habitats that are nutritionally suboptimal (Lander et al., 2013). By altering the insect-mediated connectivity considerably shorter distances (Rader et al., 2011), remaining in localized resource patches (Pasquet et al., 2008). This results in a distance decay distribution of pollen movement (Matter, Kettle, Ghazoul, Hahn, & Pluess, 2013), suggesting that between block movement (>500 m) in this experiment would be minimal. In spatially genetically structured plant populations, reduced long-distance pollination events, particularly in florally rich habitats, will result in a higher frequency of mating between close relatives. As a consequence, self-incompatible and partially self-compatible plants will suffer from increased biparental inbreeding and a reduction in compatible mates (Turner, Stephens, & Anderson, 1982). This will negatively impact plant seed set and viability (Ward et al., 2005), together with the adaptive potential and consequently the long-term survival of rare plant populations (Etterson, 2004).

| Implications for plant reproductive success
Reductions in the activity densities and richness of pollinator species in florally rich habitats reflect the increased pollen limitation and reduced individual plant reproduction observed within experimental arrays located in florally rich habitats. Pollen limitation has been related to competition for pollinator visitation, with similar results observed in response to an increase in diversity (Vamosi, Steets, & Ashman, 2013) or density (Jakobsson, Lazaro, & Totland, 2009) of co-flowering plants. Low pollen receipt, a cause of pollen limitation, can result either in an increase in self-fertilization (Kalisz, Vogler, & Hanley, 2004), or in the case of self-incompatible or partially self-compatible plants, where it is particularly detrimental, a direct reduction in seed production (Wagenius et al., 2007). Given the limited duration of stigma receptiveness, the ability of a plant to attract pollinators is therefore important for both pollen receipt and seed production (Bernhardt, Mitchell, & Michaels, 2008).
As well as the supply of pollen, the quality of pollen is also critical to plant reproduction and fitness. Pollen quality refers to both the deposition of heterospecific pollen, which can result in physical or chemical inhibition of seed set (Holland & Chamberlain, 2007;Kanchan & Jayachandra, 1980) and to the genetic relatedness of pollen, which can lead to inbreeding depression (Fischer, Hock, & Paschke, 2003

| Implications for the conservation of rare plants
Rarity in plants can be driven by biological or anthropogenic factors and is often characterized by populations comprising low genetic variation together with restrictions in size, local abundance, geographical range, and/or habitat specificity (Espeland & Emam, 2011). In this study, by simulating rare plant populations, we show that restrictions in a plant's population size, over the longer term, could lead to an Allee effect, whereby increases in mating between close relatives, coupled with higher self-fertilization rates further reduces genetic variation and ultimately, increases the risk of local extinction (Etterson, 2004).
We suggest that conservation efforts for plants facing conditions associated with rarity may benefit from focus on enhancing visitation and movement of pollinators between conspecifics. This could be achieved through a combination of: i) increasing the competitive

| CONCLUSION
Our findings show that habitat context mediates plant-pollinator interactions and alters the reproduction of rare plant populations. In florally rich habitats, rare plant populations are at a competitive disadvantage for pollinator visitation when faced with more abundant co-flowering heterospecific plants. Consequently, rare plant populations in these habitats suffer from increased rates of self-fertilization, limited pollen movement, and reduced reproductive success. The implication is that plant populations dependent on insect pollinators may become less connected and more genetically depauperate when located in florally rich habitats, increasing the risk of genetic drift and extinction. Such an effect may hold for not only rare plants but also plants that are widespread but occur at low frequency within the environment.
Although not touched upon here given the small scale of the study, this might be expected to affect plant and pollinator interactions at the habitat level and therefore warrants future study.

ACKNOWLEDGMENTS
We would like to thank Ivan Wright, Anna Oliver, Lindsay Newbold, Annika Perry, Marc Botham, Sarah Hulmes, Lucy Hulmes, and Nadine Mitschunas for valuable assistance and advice during the experiment.
Thanks to the Hillesden estate for allowing unlimited access to the site. We also thank David Gill and anonymous reviewers for helpful comments on the manuscript. This project was funded by the Natural Environment Research Council; NERC DTG NE/L501645/1.

STATEMENT OF AUTHORSHIP
All authors contributed substantially to the design and planning of the experiment, TME conducted the experiment, collected the data, and performed the analysis. TME, SC, and RE performed the paternity analysis, TME wrote the first draft of the manuscript. All authors contributed substantially to revisions and gave final approval for publication.

DATA ACCESSIBILITY
Data supporting results is archived in the NERC Environmental Information Data Centre (EIDC), and the data DOI will be included at the end of the article.