Invader–pollinator paradox: Invasive goldenrods benefit from large size pollinators

Mutualistic interactions between alien plants and native pollinators are needed to enable plant invasions. Although the increasing abundance of invasive plants in a habitat causes a dramatic decline of native pollinators, pollination services received by invaders are often sustained. This invader–pollinator paradox might be attributed to differences in pollination effectiveness and varying vulnerability to invasion among pollinators with different life history traits. In an experimental study, we explored the relationships between pollinator body size, pollination effectiveness and abundance of invasive species.


| INTRODUC TI ON
Globalization has accelerated the introduction and spread of alien species across all continents (Amano et al., 2016). Thus, recognition of basic mechanisms behind successful biological invasions allows a better understanding of ecological processes, which are shaping the state of biodiversity worldwide (Richardson, 2010). Many hypotheses designed to explain invasions are grouped around biotic characteristics as the most important factors affecting the success of alien species (Catford et al., 2009). Biotic characteristics include the properties of invading species and native communities, as well as their interactions (Catford et al., 2009;Tylianakis et al., 2008).
After introduction, alien species can lose and gain biotic interactions, and these novel ecological relationships shape an invasive species' impact on a native community (Mitchell et al., 2006;Morales & Aizen, 2006). New mutual interactions between alien and native species can boost the invasive spread (Aizen et al., 2008;Richardson et al., 2000). Otherwise, the inability to establish biotic relationships between native and invasive species can stop the colonization of a new habitat (Levine et al., 2004).
Animal pollination is crucial to successful plant invasions (Burns et al., 2013), as almost 58% of alien plant species in Central Europe are, at least partially, dependent on insect pollinators (Pyšek et al., 2011). Insect pollination seems especially limiting to self-incompatible alien plants, as plants with self-pollination ability are more likely to establish outside their historical range (Razanajatovo et al., 2016). It has been shown that most invasive plant species are sufficiently pollinated by native or invasive pollinators (Chittka & Schürkens, 2001;Pyšek et al., 2011;Richardson et al., 2000). On the other hand, plant invasions can negatively impact the pollinator communities, for example a high abundance of alien plants dramatically decreases pollinator abundance (Bezemer et al., 2014;Bjerknes et al., 2007;Potts et al., 2010Potts et al., , 2016. Thus, the lack of a shortage of pollination services received by invasive plants is a paradox of the invader-pollination relationship ( Figure 1). This paradox might be attributed to ecological filtering, that is differences in vulnerability to invasion, which structures pollinator populations according to life history traits related to pollination effectiveness (Frund et al., 2013). Mobility traits, such as body size, are linked to how pollinators use resources in a landscape (Moroń et al., 2017), as well as their pollination effectiveness (Abrol, 2012;Ne'eman et al., 2010). Therefore, it may be expected that pollinators with large body sizes are able to avoid some of the negative impacts caused by invasive plants by covering longer distances (Greenleaf et al., 2007) and foraging in uninvaded areas in their home range. Also, because pollinators' body size is a good predictor of crops' fruit set (Garibaldi et al., 2015), larger pollinators might be effective pollinators of invasive plants' flowers, if they gather more pollen and transport it over longer distances (Kerr et al., 2019).
North-American goldenrods Solidago spp. are among the most invasive species in Europe and Asia (Axmacher & Sang, 2013;Weber, 2001). The species, in invaded sites, forms dense monospecific stands frequently covering dozens of hectares (Moroń et al., 2009;Skórka et al., 2010). Goldenrod inflorescences are insect-pollinated and need cross-pollination because of their self-incompatibility (Kabuce, 2006). An individual shoot may produce more than 10,000 seeds, which can be dispersed over long distance by wind, whereas the local population size's increase is mainly the result of clonal growth (Kabuce, 2006;Weber, 2000). Earlier findings showed that, in habitats invaded by alien goldenrods, there is a dramatic abundance-dependent decline of the pollinator community suggesting the emergence of intra-specific competition for pollinators (Fenesi et al., 2015;Groot et al., 2007;Moroń et al., 2009. Thus, invasive goldenrods are excellent species to test whether pollination function in invasive species is dependent on pollinators with particular life history traits, (body size, in this case), in habitats of varying invader abundances during the course of invasion. We performed a field experiment, placing pairs of potted goldenrods at sites which differed in goldenrod cover (0%-100%). Floral visitors of planted goldenrods were noted, as were the seed set and their viability, produced by the plants. We expected that, on sites densely covered by invasive goldenrods, pollination services would be provided by large-bodied pollinators able to withstand the invasion effects and pollinate goldenrods effectively.

| Study area
The study was performed in the grassland landscape located in the valley of the Vistula river near the city of Kraków, Southern Poland ( Figure 2). Using current and historical vegetation data, we mapped the wet meadows (Dubiel, 1995(Dubiel, , 1996Kornaś & Medwecka-Kornaś, F I G U R E 1 Conceptual model of the invader-pollinator paradox. With the increasing invasion (1), the native pollinators decrease (2). However, decrease of native pollinators (2) does not cause shortage in pollination services received by invasive plants ( (Gradziński, 1974;Kornaś & Medwecka-Kornaś, 1974;Langer & Szczepanowicz, 1996). Since the late 1980s, the meadows have been maintained by mowing, grazing or burning at several-year intervals which have resulted in differences in invasive goldenrod cover.
From all the mapped meadows, we selected 25 patches of meadows 1.13 ± 0.18 km (mean ± SD) apart, on average ( Figure 2). We ensured that cover of invasive goldenrods ranged from 0% to 100% (38.36 ± 37.48%). To calculate goldenrod cover within meadows, the study sites were carefully inspected and all patches of invasive goldenrods were mapped with the help of GPS. To control for the confounding effects of potential spatial gradients (size of site, number of native flowering plants and the cover of human settlements, farmland, grassland and woodland in the surrounding landscape), sites were selected in a relatively homogenous landscape.
We used Spearman's rank correlation test to assess correlations between spatial environmental gradients and goldenrod cover.

| Experimental setup
To standardize site-related differences in resources other than pol- We confirmed S. gigantea self-incompatibility, as none of the 10 planted goldenrods with inflorescences isolated by plastic bags to protect against cross-pollination produced seeds ( Figure S2).
Inflorescences covered by mesh to protect against insect pollination produced significantly fewer seeds compared to planted goldenrods with open flowers (Wilcoxon test; seeds per inflorescence: 0.14 vs. 7.2, W = 0, p < .001, Figure S2). We confirmed that invasive goldenrods under experimental and natural conditions produced similar quantities of inflorescences and seeds ( Figure S2). Due to goldenrod self-incompatibility, and in order to protect sources of pollen, especially in sites sparsely covered by goldenrods, we placed two pots with plants belonging to two different goldenrod clones at each site. We ensured that the clones' pollen could successfully pollinate each inflorescence, resulting in seed production which was confirmed by a lack of relationship between the seed set produced by inflorescences of potted goldenrods and the distance to the nearest naturally growing conspecific as shown by the generalized linear mixed models (GLMM) fitted with a negative binomial distribution and the site identity as a random factor (Z = −.94, R 2 = .03, p = .350; Figure S3). Two pots with plants belonging to two different clones were selected for each study site in such a manner as to ensure they would bloom at the same time, and placed there, with a distance of 10 m between the pots (buried in the ground). We also ensured that there was no correlation between the number of inflorescences of potted plants with goldenrod cover (GLMM with a Gaussian distribution and the site identity as a random factor; t = −1.48, R 2 = .04, p = .145; Figure S4). The presence of other flowers in close proximity can affect the number of pollinator visits to goldenrod flowers. Thus, to standardize the closest surroundings of planted goldenrods, we cleared a circle (2 m across) around each pot of all flowering plants.
Potted plants shed blooms in mid-September. We collected steam fragments when seeds were fully developed (by the end of October) and stored these at −4°C till May.

| Surveys
We surveyed pollinating insects visiting potted goldenrod inflorescences. Each observation lasted 15 min and was repeated two to six times for each plant (depending on the duration of the potted plant's flowering period) from mid-August till mid-September, between 09:00 and 17:00, during favourable weather conditions. Altogether, each potted plant was observed for 61 ± 14 min (mean ± SD), which is above the recommended duration of observation needed to obtain the flower visitation rate (Fijen & Kleijn, 2017). Whenever possible, floral visitors were identified to species level. In other cases, pollinators were collected and identified in the laboratory. Before pollinator observations, we noted the number of blooming inflorescences of potted goldenrods. However, we did not find a significant relationship between the number of inflorescences of potted plants and the number of pollinator visits (GLMM with a negative binomial distribution and with the site identity as a random factor; bees: Z = 1.22, R 2 = .02, p = .223; flies: Z = −.13, R 2 = .00, p = .899; Figure S5a) or between the number of flowering native species and the number of pollinator visits (GLMM with a negative binomial distribution and with the site identity as a random factor; bees: Z = −.02, R 2 = .00, p = .983; flies: Z = 1.60, R 2 = .06, p = .110; Figure S5b). The order and time of day at which the sites were checked were random. From each potted plant, 100 seeds were evenly collected from multiple inflorescences, resulting in 5,000 seeds overall and sown on petri dishes in May. We recorded all germinated seedlings (those with a developed radicle or hypocotyl) and stopped observation when two weeks passed without the appearance of any new seedlings.

| The invader-pollinator paradox
Inflorescences of potted goldenrods produced about 30% more seeds at densely covered sites compared to sparsely covered ones.

| Production and viability of goldenrod seeds
The increased presence of larger pollinator species was positively associated with goldenrod seed production, leading to a 35% increase in yield and 20% increase in seed viability (seed set; bees:

| D ISCUSS I ON
Dense stands of invasive plants, similar to mass-flowering crops (Westphal et al., 2003), can have a strong negative effect on insect pollinators (Potts et al., 2016), although most studies show that invasive plant species are not pollinator-or pollen-limited (Pyšek et al., 2011;Richardson et al., 2000;Vilà et al., 2009). However, it F I G U R E 4 Relationship between the seed set per inflorescence (a) as well as seed viability (b) of potted goldenrods and body size of flower visitors. Legend as in Figure 3  Goldenrods need cross-pollination to develop viable seeds (Kabuce, 2006). Taking into consideration that the plant forms often one-clonal patches, effective pollination probably requires the pollen to be carried between inflorescences of different patches.
Small-bodied pollinators cover short distances while collecting food (Greenleaf et al., 2007), at least in comparison with large-bodied ones (Ratnieks, 2000), thus might more frequently transfer goldenrod pollen within a clone. However, this hypothesis needs to be tested by measuring the pollen metrics, for example pollen deposition, to fully understand the effects of life history traits in the pollination of invasive plants.
Although most bees have behavioural and morphological adaptations (Michener, 2000), to carry pollen, their contribution to the pollination of goldenrods is not more effective than that of flies. An explanation of this pattern might be that flies, unlike bees, are not central-place foragers (Brock, 2015), so they are not restricted by nest location in their movement through a landscape. Thus, even if flies are not able to transport as much pollen as bees, pollen may be transferred among distant goldenrod patches, leading to more effective pollination (Rader et al., 2020). However, the existence of a trade-off between pollen load and distance of pollen source for pollination effectiveness needs to be tested.
An alternative explanation could be competition between pollinator species. For example, bees might outcompete flies for goldenrod floral resources. However, we did not find evidence of competition between bees and flies ( Figure S7). Moreover, ecological filtering, as a hypothesis explaining the results, is supported by the significant nestedness of the pollinator community, indicating that species-poor sites constituted subsets of species-rich sites ( Figure S8). Also, because floral food sources are positively related to goldenrod cover and visitation rate remains unchanged, relaxation of competition for resources can be assumed at densely covered sites.
Additionally, it could be possible that the pollination behaviour of bees and flies changes with goldenrod abundance. For example, pollinators might spend more time on an inflorescence at sites densely covered by goldenrods, which would change pollination effectiveness. However, there was no significant correlation between goldenrod cover and the duration of pollinators' visits ( Figure S9).
The invader-pollination paradox can have a non-linear character because inflorescences could receive decreased pollen deposition (reduced pollinator availability) but maintain seed set if there are enough flower visitors to fertilize all ovules (Aizen & Harder, 2007).
Accordingly, we should expect, at most, the maintenance of seed set with the increase of invasive species cover. However, we found an increase of seed set along the cover gradient ( Figure 3c). Moreover, the decrease of species richness at invaded sites was significant, with the number of pollinator species decreasing more than twofold ( Figure 3a). Thus, both size-related quality and size-related quantity of pollinators seem important in pollination ecology of invasive goldenrods (Aizen & Harder, 2007 to be strongly context-dependent and to vary according to the traits of the invaders and the invaded community (Stout & Tiedeken, 2017).
Furthermore, since the impacts of invasive plants are likely to be specific to the plant species and its ecology, our understanding is likely to be limited to globally widespread, problematic plant species.
Thus, more studies of the ecology of invasive plants and flower visitors are required before generalizations about the invader-pollinator paradox can be made.
Because goldenrods are self-incompatible, it may be expected that potted plants received more viable cross-pollen as the cover increased. This in turn might affect seed set and seed viability. To lower this possibility, we potted two plants of different clones to ensure sources of pollen independently from goldenrod cover.
Moreover, we did not find a statistically significant relationship between the seed set produced by inflorescences of potted goldenrods and the distance to the nearest naturally growing conspecific ( Figure S3). Also, the observed pattern of seed set along the cover of naturally growing goldenrods mirrored the pattern found for potted plants ( Figure S6c). Thus, although the level of cross-pollination was not controlled in our study, we assume that the effect of goldenrod cover on the seed set of potted plants was not mimicked by the pollen supply.
Preventing the introduction of invasive species seems to be the most cost-effective solution for biological risk management (Keller et al., 2007). However, in the case of established alien species that dominate in native habitats, their further expansion should be prevented (Zalba & Ziller, 2007). Large-bodied pollinators seem to be key pollinators of invasive goldenrods, especially at sites dominated by the invader. Although goldenrods are capable of effectively outcompeting local native plants, due to, for example, fast clonal growth or a strong allelopathic effect Ridenour & Callaway, 2001), seeds are important for long-distance dispersal (Kabuce, 2006). Taking into consideration that propagule pressure is an important factor determining the success of invasions (Catford et al., 2009), large-bodied pollinators could boost the invasion through sustaining feedback processes (Gaertner et al., 2014), that is densely covered sites are still the source of propagules despite the loss of pollinator diversity. Thus, identifying life history traits of pollination insects and invasive plants, particularly in habitats which are greatly changed by invaders, is important in risk management models concerning the spread of invasions.