Diets maintained in a changing world: Does land‐use intensification alter wild bee communities by selecting for flexible generalists?

Abstract Biodiversity loss, as often found in intensively managed agricultural landscapes, correlates with reduced ecosystem functioning, for example, pollination by insects, and with altered plant composition, diversity, and abundance. But how does this change in floral resource diversity and composition relate to occurrence and resource use patterns of trap‐nesting solitary bees? To better understand the impact of land‐use intensification on communities of trap‐nesting solitary bees in managed grasslands, we investigated their pollen foraging, reproductive fitness, and the nutritional quality of larval food along a land‐use intensity gradient in Germany. We found bee species diversity to decrease with increasing land‐use intensity irrespective of region‐specific community compositions and interaction networks. Land use also strongly affected the diversity and composition of pollen collected by bees. Lack of suitable pollen sources likely explains the absence of several bee species at sites of high land‐use intensity. The only species present throughout, Osmia bicornis (red mason bee), foraged on largely different pollen sources across sites. In doing so, it maintained a relatively stable, albeit variable nutritional quality of larval diets (i.e., protein to lipid (P:L) ratio). The observed changes in bee–plant pollen interaction patterns indicate that only the flexible generalists, such as O. bicornis, may be able to compensate the strong alterations in floral resource landscapes and to obtain food of sufficient quality through readily shifting to alternative plant sources. In contrast, other, less flexible, bee species disappear.

Insect pollinators, such as wild bees, are endangered by multiple factors, of which agricultural intensification is one of the most severe (Ghazoul, 2013;Goulson et al., 2015;Quintero et al., 2010;Raven & Wagner, 2021;Steffan-Dewenter & Tscharntke, 1999;Stout & Morales, 2009;Winfree et al., 2009). Intensified land use is typically associated with strongly reduced floral diversity and abundance and altered plant community composition (Blüthgen & Klein, 2011;Kaluza et al., 2016Kaluza et al., , 2018Newbold et al., 2015;Requier & Leonhardt, 2020;Weiner et al., 2011). At the extreme, this can lead to largely homogenous landscapes with similarly homogenous plant and animal communities of very low diversity (Dormann et al., 2007). Here, monocultures of mass-flowering crops, such as oilseed rapes, can-where present-provide timely restricted large amounts of food for bees, but do typically not compensate for the lack of flowers and thus food resources before and after mass-flowering (Holzschuh et al., 2016;Riedinger et al., 2015). However, negative effects of such intensively managed homogenous landscapes on, for example, trap-nesting bees appear to be (at least partly) mitigated by additional floral resources as, for example, provided by semi-natural habitats in the surrounding landscape (Dainese et al., 2018;Persson et al., 2018) and even at smaller experimental scales (Ebeling et al., 2012). Notably, both positive effects of additional floral resources and negative effects of land use appear to strongly depend on the scale considered and the bee species studied (Dainese et al., 2018;Fabian et al., 2014;Hopfenmüller et al., 2020;Steckel et al., 2014;Weiner et al., 2011).
In general, changes in the floral resource landscape can reduce the quantity, taxonomic diversity (Carvell et al., 2006;Potts et al., 2010), and nutritional quality of food with still poorly understood effects on bee populations (Vaudo et al., 2015). In fact, larvae of both specialist and generalist bee species show reduced performance or fail to develop on inappropriate pollen diets (Dharampal et al., 2020;Eckhardt et al., 2014;Moerman et al., 2017;Sedivy et al., 2011), highlighting the role of diverse pollen sources to support diverse bee communities (Waser & Ollerton, 2006). As bees are centralplace foragers and are therefore restricted in their foraging ranges (Greenleaf et al., 2007), the composition of floral communities surrounding nesting sites strongly determines whether or not bees can access the pollen sources required to successfully raise offspring (Wilson et al., 2020). How precisely changes in the resource landscape as a consequence of, for example, intensified land use affect the taxonomic and chemical composition of pollen used by bees to provision their offspring has so far been little investigated (Filipiak & Weiner, 2017;Watrous et al., 2019). Recent advances in molecular and analytical methods, such as pollen DNA metabarcoding or new analytical protocols to analyze pollen nutrients (Danner et al., 2017;Sickel et al., 2015;Vanderplanck et al., 2011), provide novel tools to address these knowledge gaps. Unraveling the relationship between land-use-induced changes in plant community composition and bee species-specific patterns in taxonomic and chemical resource use is, in turn, essential for understanding the effect of land use on bee pollinator population dynamics and thus declines.
In this study, we investigated how land-use intensification affects the diversity and composition of trap nesting solitary bee species communities as well as the spectrum of pollen collected for larval provision in managed grasslands across three bioregions in Germany.
We expected to find more bee species and a more diverse composition of plant species in larval pollen provisions at field sites with comparatively low agricultural intensification. We also expected the complexity of bee-plant interaction networks to decrease with increasing land-use intensity as a consequence of impoverished bee and plant communities. We found only one bee species to be present across all three investigated bioregions and along the entire land-use intensity gradient, Osmia bicornis. For this species, we additionally assessed fitness relevant factors, that is, nutritional quality of pollen provisions and numbers of brood cells per nest, to better understand how this species managed to thrive across the entire land-use gradient.

| Sampling area and study design
The study was carried out from March to October in 2017 and 2018 on 27 grassland plots in three geographically separated regions as part of the German Biodiversity Exploratories ( Figure SM1). In each of the three regions, nine experimental grassland plots were selected to cover different intensities and combinations of land-use management, including meadows, pastures mowed or grazed by livestock as well as fertilized and unfertilized plots (Blüthgen et al., 2012) (Table   SM1). Each experimental plot covers and area of 50 × 50 m, but grasslands and respective land-use span beyond this area (Fischer et al., 2010).
In early spring 2017, four artificial perpendicular solitary bee trap nests with hollow reed internodes (Staab et al., 2018), pointing in every compass direction, were installed at the fence of a weather station located on each of the 27 plots ( Figure SM1). We collected samples (i.e., occupied reed internodes) five times in 2017 and three times in 2018.
In the laboratory, reed cane internodes were opened lengthwise.
Bee species were identified by reed nest closures and bee morphology according to (Amiet et al., 2017). Brood cells per species and site were counted as a proxy measure for reproductive fitness. To investigate how land-use and subsequent changes in plant species diversity and composition affected the taxonomic composition of larval pollen provisions, we collected pollen provisions from nests of bee species which were found at at least three sites across bioregions.
Per reed, a maximum of 3-5 pollen provision samples were collected with sterile forceps, pooled, and weighed to measure the total amount of wet pollen provision per reed. We used left-over pollen provisions in reed cells with well-developed larvae and pupae. Thus, we made sure to only sample minute amounts of pollen per cell as to not severely impair larval development.
In total, 150 pooled pollen provision samples of nine bee species

| Metabarcoding
Genomic DNA isolation was conducted with the Macherey-Nagel Nucleospin (Düren, Germany) kits for food and according to the supplementary protocol for pollen (Keller et al., 2014). We followed the dual-indexing strategy based on (Sickel et al., 2015) in order to generate a pooled amplicon library based on the ITS2 rDNA region used for pollen metabarcoding for the Illumina platform (Illumina, 2017) (see SM Analytical details for Metabarcoding and bioinformatics workflow).

| Nutritional analyses
To assess how land-use-induced differences in the diversity and intert XL MSD, Agilent Technologies) following Brückner et al. (2017) with procedures adapted as detailed in SM (Analytical details). Total fat content was calculated as the sum of all fatty acids.
We investigated how several explanatory variables related to land use affected our response variables. Explanatory variables were land-use intensity (LUI), an index used as standard measure in the Biodiversity Exploratories framework (Blüthgen et al., 2012), its components grazing, mowing, and fertilization, as well as flowering plant species richness (dataset provided by Biodiversity Exploratories). Flowering plant species richness was included as it correlates with LUI and the number of flower visitor-plant interactions (Weiner et al., 2011(Weiner et al., , 2014. Flowering plant species was assessed at the same plots as used in our study (also see details in Data availability). We composed a rank correlation matrix (psych package) to determine significant correlations between all our response and explanatory variables (Supplemental material, Table SM2).
We always conducted two separate generalized linear mixedeffect model analyses (GLMMs, lme4 package) to assess if individual (non-correlating) components of LUI or LUI itself better explained our results. Thus, one GLMM comprised flowering plant species richness and land-use intensity (LUI) as fixed factors (Supplemental material, Table SM1). The other model comprised flowering plant species richness, grazing, and fertilization as fixed factors. We did not include mowing to avoid multicollinearity, because mowing was significantly positively correlated with fertilization and significantly negatively correlated with grazing for our study grassland plots (see Table SM2). Plot (field site, which was nested in bioregion, i.e.,

| Effects of land use on solitary bee species communities
We investigated how parameters related to land use and biogeographical region (Biodiversity Exploratories) affected trap nesting bees, that is, overall bee species diversity, richness, and abundance.
Data of overall bee abundance were log-transformed and bee species richness arcsine square-root transformed to achieve statistical requirements.
To analyze how bioregion and/or LUI affected the community  (Blüthgen et al., 2006). We composed Patefield's null models to assess levels of specialization (H2') of our observed networks in relation to random networks.

| Effects of land use on pollen foraging, nutrition, and reproductive fitness in Osmia bicornis
We investigated how parameters related to land use and biogeo- Effects of all land-use variables on pollen provision nutrients, that is, total FA, total AA, and essential AA content as well as the AA:FA ratio were analyzed using separate GLMMs as described Finally, land-use-induced differences in reproductive fitness (i.e., number of O. bicornis larval brood cells) were analyzed using separate GLMMs as described above followed by correlation tests.

| Effects of land use on solitary bee species communities
In total, we found 285 occupied reed sticks and 1544 well-developed bee individuals, that is, larvae which were not affected by fungi, parasitic insects, or nematodes, across all three bioregions. Overall bee species abundance, bee species richness, and Shannon diversity of bee species differed between bioregions (Table SM3). Some bee species were only found in specific regions, but not in others Variation in bee species abundance, richness, and diversity were best explained by land-use intensity (LUI) (  Figure SM2). Both richness and abundance of bees significantly increased with increasing flowering plant species richness and with grazing intensity (Table 1). The Shannon diversity of bees also tended to increase with increasing flowering plant species richness (Table 1).

| Effects of land use on pollen foraging and bee-pollen plant networks
Sequencing of pollen samples generated on average 18,672 qualityfiltered ITS2 reads (range from 6429 to 98.819), in total 2.083.229 reads for the whole study. We found 267 taxonomic assignments on plant species level and 177 on genus level (Table SM4) Figure SM5) and also significantly different from random interactions (Table SM7)

| Effects of land use on pollen foraging, nutrition, and reproductive fitness in Osmia bicornis
The  high: 10.09%) increased with increasing land-use intensity (Table   SM5 and SM6, Figure SM7).
When analyzing the nutritional composition of O. bicornis larval pollen provisions, we found that both total FA concentrations and total AA concentrations significantly decreased with increasing LUI (Figure 2, Table 2, Table SM2), while the AA:FA ratio remained relatively constant across the entire land-use intensity gradient ( Figure 2). Additionally, total essential AA concentrations significantly decreased with increasing LUI (Table 2). Furthermore, total AA and total essential AA were positively affected by plant species white bars indicate that a species was absent from a specific bioregion or land-use intensity category richness and grazing, while total FA decreased with grazing intensity (Table 2).
While not all identified FAs were found in every pollen sample, 16 FAs were found across larval pollen provisions (Table SM9). Concentrations of six (out of the 16 analyzed) single FAs significantly decreased with increasing LUI (Table SM8 and

| DISCUSS ION
Our study shows that land use can severely affect diversity, composition, and fitness parameters of trap-nesting solitary bee species through altered interactions between bees and pollen resource plants and thereby their nutritional intake and reproductive success.

| Effects of landuse on solitary bee communities and bee-pollen plant networks
Bee species richness and Shannon diversity decreased with increasing land-use intensity as did overall plant species richness in larval pollen provisions. In fact, we found bee species-specific pollen compositions of larval provisions, indicating that diverse bee communities need diverse floral communities to meet all bee species-specific dietary needs. Intensive land use can negatively impact on plant species richness as shown for the Biodiversity Exploratories (Weiner et al., 2014), where 34% of grassland plant species responded negatively to intensive land use (Busch et al., 2019), in particular to frequent mowing and fertilization . Reduced floral resource abundance and diversity, in turn, correlated with reduced bee species richness and diversity (Biesmeijer et al., 2006;Weiner et al., 2011Weiner et al., , 2014. Our results indicate that the composition and diversity of trap nesting bee communities depended not only on overall resource diversity but also on the availability of specific pollen source plants. For example, H. truncorum and M. versicolor, which are all specialized on Asteraceae (Michener, 2007), mainly used pollen of Crepis sp. (>50%) for larval provisions and were almost exclusively present at sites where Crepis biennis was recorded, that is, on lowintensity sites. The specialist C. florisomne was found across bioregions and mainly foraged on Ranunculaceae (79.22%) which were present at all study plots. Interestingly, C. florisomne was absent from field sites with intermediate or high land-use intensity despite the occurrence of Ranunculus species, indicating that this bee species may require additional resources, for example, for nesting, or is additionally influenced by other factors, like, for example, pesticide exposure or competition (Centrella et al., 2020). By contrast, the two TA B L E 2 Results of two separate generalized mixed effect models (GLMMs, F-and p-values) analyzing the effect of (i) land-use intensity (LUI) and flowering plant species richness (PSR), and (ii) grazing (G), fertilization (F), and plant species richness (PSR) on taxonomic Shannon plant diversity of O. bicornis larval pollen provisions, the abundance of O. bicornis larval brood cells, as well as total fatty acids, total amino acids and the sum of total essential amino acids in O. bicornis larval pollen provisions. We did not include mowing to avoid multicollinearity, because mowing was significantly positively correlated with fertilization and significantly negatively correlated with grazing for our study grassland plots (see Table SM2). Year (2017 and 2018) was included as additional fixed factor and plot nested in bioregion as random factor in all models.  (Michener, 2007) may more easily find plant resources across landscapes differing in plant community composition and diversity (Mallinger et al., 2016). Such flexibility in foraging allows bee individuals to plastically respond to spatiotemporal changes in plant community composition and availabilities and to switch to alternative resources when necessary (Pornon et al., 2019). The less flexible specialist bees, for example, H. truncorum or M. versicolor (Michener, 2007), may in turn be restricted to specific landscapes and habitats providing suitable resources (Mallinger et al., 2016). This likely explains why intensively managed agricultural grasslands with severely reduced floral diversity harbor fewer bee species and impoverished F I G U R E 2 Effect of land-use intensity (LUI) on (a) plant taxonomic Shannon diversity, (b) total fatty acid (FA) concentration (c) total amino (AA) acid concentration, and (d) total essential AA concentration and (e) the ratio of total FA to total AA in Osmia bicornis larval pollen provisions sampled from nests installed at plots differing in land-use intensity (LUI) in three biogeographical regions in Germany (Exploratories: Swabian Alb, Hainich-Dün and Schorfheide-Chorin). Plant diversity is based on revealed ASVs (Amplicon sequent variants) per bee nest bee communities (Grab et al., 2015;Mallinger et al., 2016), which does not only negatively affect bee populations (Renauld et al., 2016;, but also services provided as for example pollination (Goulson et al., 2015;Jauker et al., 2016;Potts et al., 2010).
Interestingly, we were able to observe effects of landuse on solitary bee communities at relatively small (plot/grassland site) scales, while previous studies did not find such effects at plot level (Fabian et al., 2014;Steckel et al., 2014). One limitation of our study is that we did not analyze our findings at larger scales. We can therefore not rule out that effects differ at different scales or that our results were additionally affected by variation in floral resource diversity and composition at the landscape scale. However, many (~85%) plant species revealed by metabarcoding were actually also found at plots or in the respective grasslands, indicating that female bees readily restrict pollen foraging to small scales if adequate pollen host plants are available (Ebeling et al., 2012).

| Effects of land use on pollen foraging, nutrition, and reproductive fitness in Osmia bicornis
Interestingly The contents and ratios of these two macro-nutrients are known to directly affect brood rearing success and immune defense in bees (Genissel et al., 2002;Human et al., 2007;Pirk et al., 2010;Ruedenauer et al., 2020;Tasei & Aupinel, 2008;Vaudo et al., 2015. It is therefore likely that O. bicornis regulates their intake and shows clear preferences for specific P:L ratios, as indicated by the relatively constant AA: FA (a proxy for P:L) ratio of 2:1 in larval provisions. In fact, O. bicornis appears to be able to maintain a specific nutritional intake (Vaudo, Stabler, et al., 2016) (Nicolson, 2011). In particular, PSMs strongly vary qualitatively and quantitatively among plant species and can both positively and negatively affect bee performance depending on their type and concentration (Palmer-Young et al., 2019;Stevenson, 2020). There is some evidence that bees may be more likely to mitigate negative effects of PSMs through dilution when they have access to a broader spectrum of suitable pollen host plants (Eckhardt et al., 2014), which may further explain why some bee species were absent at sites with lower floral diversity. Besides pollen quality, pollen quantity can also increase with increasing floral diversity in the surrounding landscape, thus providing overall more resources to feed bees (as shown for the stingless bee species Tetragonula carbonaria: (Kaluza et al., 2018)).
Moreover, reduced overall floral abundance as frequently observed at sites with high land-use intensities (Newbold et al., 2015) may force adult female bees to increase foraging trip duration to obtain sufficient resources (Danner et al., 2017;Westphal et al., 2006).
This can in turn increase the vulnerability of nests to parasitism and predation (Goodell, 2003). Together, these factors may explain the observed decrease in the number of larval brood cells per nest. In fact, reduced floral diversity can reduce both the quantity and quality of allocated food (Kaluza et al., 2018;Trinkl et al., 2020). Changes in the nutritional quality of pollen diets may subsequently impair larval development, because plant pollen differ in their nutritional suitability even for generalist bees (Haider et al., 2013;Sedivy et al., 2011). Besides decreasing plant diversity and nutritional quality as well as shifting pollen composition, agri-environmental factors may also indirectly impact on bee reproduction by increasing the risk of pesticide exposure (Centrella et al., 2020). Consequently, even highly flexible generalists, such as O. bicornis, may only be able to partly compensate for the loss of floral diversity.

ACK N OWLED G EM ENTS
We thank the managers of the three Exploratories, Kirsten Voulgari-Kokota for support with sequencing technologies. We are also very grateful for the constructive comments provided by three anonymous reviewers. Funding was provided by the Deutsche Forschungsgemeinschaft (DFG project: LE 2750/6-1 to SDL and KE1380/6-1 to AK). Fieldwork permits were issued by the responsible state environmental offices of Baden-Württemberg, Thüringen, and Brandenburg.

CO N FLI C T O F I NTE R E S T
The authors declare no conflict of interest.

DATA AVA I L A B I L I T Y S TAT E M E N T
The following data sets are available at BEXIS (https://www.bexis.