Are native and non‐native pollinator friendly plants equally valuable for native wild bee communities?

Abstract Bees rely on floral pollen and nectar for food. Therefore, pollinator friendly plantings are often used to enrich habitats in bee conservation efforts. As part of these plantings, non‐native plants may provide valuable floral resources, but their effects on native bee communities have not been assessed in direct comparison with native pollinator friendly plantings. In this study, we performed a common garden experiment by seeding mixes of 20 native and 20 non‐native pollinator friendly plant species at separate neighboring plots at three sites in Maryland, USA, and recorded flower visitors for 2 years. A total of 3,744 bees (120 species) were collected. Bee abundance and species richness were either similar across plant types (midseason and for abundance also late season) or lower at native than at non‐native plots (early season and for richness also late season). The overall bee community composition differed significantly between native and non‐native plots, with 11 and 23 bee species being found exclusively at one plot type or the other, respectively. Additionally, some species were more abundant at native plant plots, while others were more abundant at non‐natives. Native plants hosted more specialized plant–bee visitation networks than non‐native plants. Three species out of the five most abundant bee species were more specialized when foraging on native plants than on non‐native plants. Overall, visitation networks were more specialized in the early season than in late seasons. Our findings suggest that non‐native plants can benefit native pollinators, but may alter foraging patterns, bee community assemblage, and bee–plant network structures.

To provide food resources for bees, many seed companies, botanical associations, and other interest groups have compiled and promoted "pollinator friendly" plant lists and/or seed mixes. The nectar and/or pollen from these mixes of plants are believed attractive to pollinators, but this assumption has not been extensively tested . Hicks et al. (2016) showed that nectar and pollen quantity and quality varied among 23 pollinator friendly plants, but they did not assess bee visitation, while Garbuzov and Ratnieks (2014) found highly fluctuating bee visitation rates among 32 UK garden plants. Therefore, the attractiveness of pollinator friendly plants for bees deserves further exploration.
Many commercially available pollinator friendly plant mixes often include both native and non-native plants (Fowler, 2016;Morandin & Kremen, 2013). Selecting plants for pollinators based on the quality or quantity of floral rewards, rather than their native range, may benefit bees and support conservation efforts. However, non-native plants may adversely affect native bee communities by providing floral resources of inappropriate nutritional value or by benefiting generalist over specialist bee species. The (long-term) effect of non-native plants on bee communities warrants further investigation Klein et al., 2003;Morandin & Winston, 2005;Ollerton et al., 2011).
Flowers of non-native plants are foraged on by native bees (Drossart et al., 2017;Tepedino et al., 2009;Williams et al., 2011;Figure 1), but whether they prefer native or non-native plants is still unclear. While Williams et al. (2011), Nienhuis et al. (2009), and Morales and Aizen (2006) found no differences in bee visitation between native and non-native plants, other studies report either higher bee abundances (Bartomeus et al., 2008;Gibson et al., 2013;Lopezaraiza-Mikel et al., 2007;Matteson & Langellotto, 2010;Vilà et al., 2009) or lower bee abundances for non-native plants Menz et al., 2011;Morandin & Kremen, 2013;Moroń et al., 2009;Pardee & Philpott, 2014). Similarly, studies found bee species richness either higher (Bartomeus et al., 2008;Lopezaraiza-Mikel et al., 2007;Vilà et al., 2009), lower Morandin & Kremen, 2013), or the same (Pardee & Philpott, 2014) on non-native plants versus native plants. The majority of this research has been conducted in Europe and the United States, while studies from the southern hemisphere are still scarce (but see Morales and Aizen (2006) and Gibson et al. (2013) for examples from Argentina and South Africa) despite the often negative effects of non-native (invasive) plant species on native flora and fauna (Bellard et al., 2017;Van Kleunen et al., 2015). Most of these studies focused on invasive non-native plants instead of directed pollinator friendly plantings and were observational, rather than the result of hypothesis-driven field trials. An exception was the study by Morandin and Kremen (2013) who analyzed hedgerows planted for natural habitat restoration along agricultural fields in the United States. However, in their study, native and non-native plants were mixed in the same treatments and were not explicitly pollinator friendly. To date, no study has experimentally assessed whether flowers of planted native and non-native pollinator friendly seed mixes differently affect pollinator communities.
Several studies document that the visitors of non-native flowers are less specialized than those visiting native flowers (Grass et al., 2013;Lopezaraiza-Mikel et al., 2007;Memmott & Waser, 2002;Schweiger et al., 2010;Stout & Morales, 2009), suggesting that non-native plants support generalist bees and disproportionally affect specialist bee species, which are overall more sensitive to land use change than generalists . This can alter the community composition and the structure of the bee visitation network (Bartomeus et al., 2008;Olesen et al., 2002;Vanbergen et al., 2018), while overall bee abundance or species richness may not be affected.
In this 2-year study, we experimentally tested how flowering plants grown from seed mixes composed of either native or non-native seeds affected the abundance, species richness, and community structure of bees. We also investigated the specialization of visitation networks and the specialization of individual bee species.
As both native and non-native seed mixes were comprised of beefriendly plant species, we expected similar bee abundances and species richness for native and non-native plantings. Furthermore, we expected bees to forage in a more specialized manner on native than non-native plants. Our study is the first to experimentally compare native versus non-native pollinator friendly plant communities planted in plots of the same size in a common garden experiment.

| Study system
The study took place at the Beltsville research farm of the University of Maryland, USA. The Beltsville facility spans 116 ha in a rural landscape with forest fragments, creeks, ponds, and agricultural fields.
At the research farm, we established three experimental sites which were 1-1.7 km apart from each other (Figure 2). Each site had two F I G U R E 1 Native bumble bee (Bombus impatiens/bimaculatus) visiting a non-native cosmos plant (Cosmos bipinnatus), Maryland, USA plots: one sown with a native pollinator friendly seed mix and one sown with a non-native pollinator friendly seed mix (Figures 2 and   3). Plots were 10 × 10 m in size, and the distance between the two treatment plots was 20 m. The central coordinates for the three sites Both the native and non-native seed mixes contained 20 different flower species and two grass species each (Tables 1 and 2).
Grasses are typically added to pollinator friendly meadows as they also provide nesting or overwintering sites for insects and can protect wildflowers by preventing soil erosion (Lee-Mäder et al., 2013). In 2017, Solidago odora seeds were not available and were replaced with Solidago nemoralis seeds. Throughout both years, the plots were continuously hand-weeded.

| Bee sampling
Bees were sampled by hand netting and with pan traps between April and October in 2016 and 2017. Sampling took place approximately every 2 weeks (13.3 ± 5.7 days (mean ± SD) between consecutive sampling events) on rain-free and (mostly) sunny days. Each year, hand netting started as soon as the first plants started to bloom (resulting in different starting points for native and non-native plots) and was performed during 30-min random walks through plots: All bees observed to touch reproductive flower parts were captured from flowers with a ziplock plastic bag. Nectar-or pollen-gathering behavior was not distinguished. We stored bees in 70% ethanol for identification. All plots were sampled within 1 day between 9:00 and 18:00 in alternating random order. Hand netting was always performed by the same person (N. Seitz). Note that Cichorium intybus closed its flowers very early in the day, which is why these plants could not be observed later in the day. We therefore conducted additional sampling of C. intybus in the morning of the next day (2 August 2017 and 25 August 2017). Furthermore, sampling on 1 August 2017 was interrupted by rain and therefore continued the next day. Over 2 years, 21 hand netting events were conducted across sites. In 2016, non-native plots were sampled eight times and native plots five times due to a later onset of flowering at native plots. In 2017, non-native plots were sampled 13 times and native plots 11 times at site B and 10 times at sites A and C due to a later onset of flowering at these sites. We at least partly accounted for these differences in flowering time by calculating bee numbers per sampling event (see 2.3 Statistics for details).
On each day of hand netting, bees were also sampled with pan traps (except for 15 July 2016). Additionally, we sampled bees with pan traps once in April 2016 and once in May 2016 before plants started flowering. We followed guidelines by Droege (2015) for pan trap sampling. We placed twelve pan traps (3.5 oz) filled with soapy water and colored in fluorescent blue, yellow, and white along plot edges in the morning, before the hand netting started. We recollected F I G U R E 2 Overview of the three experimental sites (A, B, and C)

| Statistics
All statistical analyses were performed with R version R 3.5.1 (R Core Team, 2018)  We differentiated between three partly overlapping seasons, that is, spring to early summer (April 1-July 15), midsummer (June 1-August 31), and late summer to fall (July 16-October 3). For each season, data were pooled across years. Midsummer overlapped with the early and late seasons in order to smoothly incorporate the phase of transition from early to late season, capturing the late species of the early season and the early species of the late season. It is nevertheless important to treat midsummer as a distinct season, as many studies often focus exclusively on this time period Cusser & Goodell, 2013;Fründ et al., 2010;Hegland et al., 2010).
We visualized and analyzed plant-bee visitation networks with the "bipartite" package (functions "plotweb," "networklevel," and "specieslevel" (Dormann et al., 2008)). Species that could be identified to genus level only were included in networks if no other species of the same genus was present (i.e., Melissodes and Sphecodes).
The H2′ index within the function "networklevel" was used to compare the community-level specialization of each network (Blüthgen et al., 2006). At the species level, we used the d′ index of the function "specieslevel" to compare the specializations in plant visitation of single bee species within a network (Blüthgen et al., 2006). We assessed species specificity (d′) of the five most abundant bee species and only included those d′ values for these species that were based on more than two observations per network.
We analyzed differences in bee species richness per sampling event, abundance per sampling event, network specificity (H2′) per season, and species specificity (d′) per season between native and non-native plots using linear mixed-effect models (LMM; "lmer" function, "lme4" package (Bates et al., 2015)). Data were log-transformed where necessary (i.e., bee species richness and abundance) in order to achieve a normal distribution (as visually assessed with histograms and tested with Shapiro tests). We calculated models for species richness and abundance (per sampling event) for each season and included plant nativity as a fixed effect and site and Julian date as random effects. As floral variables were measured in 2017 only, we did not include them in our main models. However, we composed additional models for the data of 2017 which included floral abundance and floral richness as fixed effects to test for their effect on bee diversity and abundance (see Table S3).
Due to the earlier flowering of non-native plants resulting in a higher number of sampling events at non-native plots, we based our comparisons of bee abundance and richness on visitations per sampling event. To additionally account for the uneven sampling, we performed supplementary analyses for spring/early summer (the only season were sampling days differed) which were restricted to those sampling events where plants of both types were flowering.
For network and species specificity, we obtained always one H2′ value and one d′ value (per species) for each of the 18 networks (i.e., each native and non-native plot at each of the three sites for each of the three seasons pooled across years). Our models for network and species specificity included plant nativity and season as fixed effects and site as a random effect. To ensure that trends in network specificity were not heavily influenced by the behavior of the non-native and domesticated honeybee (Apis mellifera), we performed additional network analyses excluding this species.
Significance of fixed effects (plant nativity and season) was assessed with the "ANOVA" command of the "car" package. Multiple comparison of means for differences between seasons was analyzed using Tukey's post hoc tests ("glht" function, "multcomp" package (Hothorn et al., 2008)). To finally assess the overall variance explained by models, we calculated R 2 -values with the pseudo-R-squared ("r. squaredGLMM") function of the "MuMIn" package (Bartoń, 2018).
To analyze bee community composition, we visualized data using non-metric multidimensional scaling (NMDS; function "metaMDS," "vegan" package) with a Bray-Curtis distance matrix based on the abundances of different bee species (Oksanen et al., 2018).
Differences in bee community composition between native and non-native plots were then assessed using a permutational multivariate analysis of variance (PERMANOVA; "adonis" function in the "vegan" package; 100 000 permutations) also based on Bray-Curtis distances between abundances of bee species ("vegdist" function, "vegan" package (Oksanen et al., 2018)). As only bees were included in this analysis and not their temporally changing interactions with plants, we only included data of the early and late seasons here and excluded the midseason to avoid pseudoreplication.
Network analyses as well as statistical analyses on differences of bee abundance, species richness, Shannon diversity, and bee community composition between native and non-native plots were based on data from hand netting, while data from pan traps were only used to gain information on the potential pool of visitors to experimental plants. Data from pan trap sampling were not used to test for effects of plant nativity on bee community composition due to the close proximity of native and non-native plots. Bees with uncertain identification were included in analyses of abundances, but not in analyses of bee diversity or bee community composition.

| RE SULTS
Over the 2-year study period, 17 out of the 20 initially seeded pollinator friendly plants came to flower within native plant plots and 18 within non-native plant plots (Tables 1 and 2). Non-native plants started flowering about 5 weeks earlier than native plants in both years (Tables 1 and 2). In the second year, many plant species of both plant types started flowering about 2 months earlier and also flowered longer. In both years, both native and non-native plants were still flowering when observations ceased.
A total of 3,744 bees representing 31 genera and 120 species were recorded in this study. Pan traps sampled 2,036 bees, representing 30 different genera and 107 different species (see Table S1 for complete species list). With hand netting, 1,708 bees were sampled, comprising 25 different genera and 72 different species (Table   S1). Some of the bee species were only recorded with one of the two sampling methods: Pan trapping sampled 89% of the total of 120 bees, and 48 species (40%) were sampled exclusively. Hand netting sampled 60% of all species seen in this study, 13 species (11%) exclusively.
On native Table S1). Apis mellifera was the only non-native bee species visiting our experimental plants.
We recorded eight oligolectic bee species (Fowler, 2016)  Bee abundance and species richness recorded per hand-netting event differed between native and non-native plots in some but not all seasons ( Figure 5, Table 3). In spring and early summer, both abundance and species richness were significantly lower at native plots than at non-native plots ( Figure 5, Table 3). Species richness was also significantly lower at native plots in late summer and fall ( Figure 5, Table 3). These differences of bee abundance and species richness F I G U R E 4 Differences in bee community composition between native plant plots (pink symbols) and non-native plant plots (dark red symbols) displayed by non-metric multidimensional scaling (NMDS, stress value = 0.14). Sites are plotted for spring/ early summer (circles) and late summer/fall (triangles) separately, resulting in two data points per site. Each symbol represents one site in one season. Corresponding native and non-native plots of the same site in the same season are connected with gray lines

F I G U R E 5 Bee abundance (a) as number of individuals (n) and bee species richness (b) per sampling event at native and nonnative plots by season (data pooled for both observation years).
Statistical differences between native versus non-native plants per season were assessed with LMMs (Table 3) and are indicated here with asterisks when significant: *p < .05, **p < .01. This and the following box plots show the interquartile range (IQR) of data, which extends from the 25th percentile to the 75th percentile. The line within the box indicates the median. The lower whiskers extend to the smallest observation, but not further than 1.5 * IQR from the 25th percentile. The upper whiskers extend to the largest observation, but not further than 1.5 * IQR from the 75th percentile. Outliers are plotted as individual points at native and non-native plots per sampling event also remained when restricting analyses of spring and early summer to sampling events where both plant types were flowering (Table S2, also see

Statistics).
When we analyzed the subset of data for which floral cover and floral richness were available (year 2017), we found that bee abundance and bee species richness were best explained by floral cover and increased with increasing proportions of floral cover (Table S3).
Native plants were part of more specialized networks than non-native plants, and early season networks were more specialized than late season networks (Table 4). These trends also remained when removing honeybees (A. mellifera) from networks (Table S4, also see 2.3 Statistics). Early season networks included fewer plant and bee species (Figures S1-S6).
The five most abundant bee species across both native and non-native plant plots were A. mellifera, H. poeyi/ligatus, X. virginica, B. impatiens, and Lasioglossum pilosum, in order of decreasing abundance (Table S1). Xylocopa virginica, B. impatiens, and L. pilosum showed a more specialized foraging behavior at native than at non-native plants ( Figure 6 and Table 4). Apis mellifera became less specialized over the entire flowering season with the highest specialization early in the year and lowest specialization later in the year ( Figure 6 and Table 4). The other species' specificity remained similar across seasons, but X. virginica and B. impatiens were absent in the early season ( Figure 6 and Table 4).

| D ISCUSS I ON
Our study showed that a seed mix of non-native pollinator friendly plants was well accepted and frequented by a diverse bee community. Depending on the season, non-native plants attracted either similar or higher numbers of individuals and species compared to native plants. Particularly in the early and late season, many bees chose non-native over native plants, while no differences were found in the middle of summer. Generally, bee visitation increased with increasing floral resource coverage, which has also been shown by previous studies (Banaszak, 1996;Batáry et al., 2010;Baude et al., 2016)

TA B L E 3
Results of the linear mixedeffect models (LMM) for bee species richness and abundance in the three different seasons with plot type (native/ non-native) as explanatory variable and site and date as random effects One possible explanation for this difference is that late flowering non-native plants offered higher quantities of nutritionally more at- ing bee species-specific responses to non-native plants (Pardee & Philpott, 2014;Schweiger et al., 2010;Urbanowicz et al., 2020), which may be due to species-specific nutritional requirements (Leonhardt & Blüthgen, 2012;Nicolson, 2011;Vaudo et al., 2016), previous experiences (Harmon-Threatt & Kremen, 2015;Vaudo et al., 2015), or competition (Somme et al., 2015;Wilms et al., 1996).
Additionally, plant-bee networks differed between native and non-native plants. As expected, networks associated with native plants were more specialized than networks associated with non-native plants. Native plants share a longer evolutionary history of interactions with native bees than non-native plants, which likely increased the chance of more specialized interactions evolving (Fenster et al., 2004). Other studies found that non-native plants were predominantly visited by more generalized insects (Aizen et al., 2008;Lopezaraiza-Mikel et al., 2007;Memmott & Waser, 2002;Olesen et al., 2002;Schweiger et al., 2010;Stout & Morales, 2009 TA B L E 4 Results of the linear mixed-effect models (LMM) for the plant-bee network specialization (H2′) and for individual specialization (d′) of the five most abundant bee species with plant nativity and season as explanatory variables and site as random factor; and results (direction of seasonal difference and p-values) of the Tukey post hoc tests on differences between seasons  Figures S1-S6. The marginal R 2 -value gives the variance explained by the fixed effects and the conditional R 2 -value variance explained by both fixed and random effects. Species specialization of Bombus impatiens and Xylocopa virginica only included data of two seasons, summer and late summer to fall, because these species were not present in spring to early summer. Seasons are abbreviated as follows: early = spring to early summer; mid = summer; and late = late summer to fall. Asterisks indicate significant effects (in bold): *p < .05, **p < .01, ***p < .001. Winfree et al., 2011). In our study, we recorded only two oligolectic bee species visiting flowers, preventing more robust inferences on disproportional effects on specialists. In the mid-Atlantic region, oligolectic bee activity peaks in summer (with up to 57 spp. active in June), followed by spring and fall (Fowler, 2016 The occurrence of more specialists in a plant-bee network can be one reason for an increased network specialization. Another reason can be more specialized interactions of generalist species. In our study, most bee species were generalists. Although generalist bees typically feed on a broad range of floral resources, their foraging patterns and thus their level of specialization can vary with the spectrum of plants available (Cook et al., 2003;Somme et al., 2015;Vaudo et al., 2016). For example, three out of the five most abundant generalist species in our study showed more specialized foraging when feeding on native plants. While we did find that non-native networks were comprised of slightly more plant species than native networks, this difference should not have affected network or species specialization, as both d′ and H2′ are robust against variations in network size, shape, and sampling intensity (Blüthgen et al., 2006). One reason for a more specialized foraging behavior of generalists on native plants could be overall stronger differences in chemical profiles between native plants, with specific plant species better meeting the nutritional requirements of specific bee species.
Native generalist bees could also have evolved a higher degree of specialization on certain native plant species, through adapting to their chemical repertoire.
Network specialization also changed over the seasons for interactions involving both native and non-native plants. Specialization was highest in the early season and decreased toward fall. The higher degree of specialization coincided with a lower number of plant species as part of the networks in spring, which is not uncommon for grasslands (Mallinger et al., 2016). The d′ index provides information on the degree of specialization of individual species and ranges from 0 (not specialized) to 1 (highly specialized). It is based on interaction links between bee species and plant species, but also takes into account the dominance of each linked plant species in the interaction network. Thus, bee species visiting the most dominant plant species are considered less specialized than bee species visiting plants that are rarely visited by other bee species. Statistical differences between plant types and seasons per bee species were assessed with LMMs (  . To conclude, our study suggests that non-native plants can complement native pollinator friendly plantings, because they are vis- Furthermore, our results highlight the importance of analyzing entire flowering periods instead of focusing solely on specific seasons (e.g., summer) and of taking into account phenological matching for network analyses (see also Kantsa et al. (2017)). More research that experimentally compares native versus non-native pollinator friendly plant mixes in different regions of the northern and southern hemisphere, at different scales and with other plant species would be helpful in order to best support local bee communities and to globally improve conservation strategies for bees.

ACK N OWLED G M ENTS
We would like to thank Sam Droege for identifying our specimens and for sharing his profound knowledge on native bees. We are thankful to Ernst Conservation Seeds for sponsoring the seed mixes for this study. We specially thank Mark Fiely at Ernst Conservation had no involvement in the study design; in the collection, analysis, and interpretation of the data; and in the writing of the report. Open access funding enabled and organized by Projekt DEAL.

CO N FLI C T O F I NTE R E S T
The authors declare no competing interests.

DATA AVA I L A B I L I T Y S TAT E M E N T
All data used in analyses can be found in Dryad at https://doi. org/10.5061/dryad.pzgms bcj8.