Pollinator individual‐based networks reveal the specialized plant–pollinator mutualism in two biodiverse communities

Abstract Generalization of pollination systems is widely accepted by ecologists in the studies of plant–pollinator interaction networks at the community level, but the degree of generalization of pollination networks remains largely unknown at the individual pollinator level. Using potential legitimate pollinators that were constantly visiting flowers in two alpine meadow communities, we analyzed the differences in the pollination network structure between the pollinator individual level and species level. The results showed that compared to the pollinator species‐based networks, the linkage density, interaction diversity, interaction evenness, the average plant linkage level, and interaction diversity increased, but connectance, degree of nestedness, the average of pollinator linkage level, and interaction diversity decreased in the pollinator individual‐based networks, indicating that pollinator individuals had a narrower food niche than their counterpart species. Pollination networks at the pollinator individual level were more specialized at the network level (H′2) and the plant species node level (d′) than at the pollinator species‐level networks, reducing the chance of underestimating levels of specialization in pollination systems. The results emphasize that research into pollinator individual‐based pollination networks will improve our understanding of the pollination networks at the pollinator species level and the coevolution of flowering plants and pollinators.


| INTRODUC TI ON
The interactions between plants and their pollinators are considered to be one of the fundamental drivers of angiosperm diversity (Stebbins, 1970). The specialization of the pollination system is important for a successful pollination, since a high level of specialization would assure conspecific pollen transfers within plant species and a high energy intake rate of pollinators (Johnson & Steiner, 2000). Local populations of entomophilous plant species would benefit from specializing foraging bouts of pollinator individuals in a short time with no temporal changes caused by adjacent factors. Reducing the interspecific pollen transfer would ensure seed production via outcrossing and species integrity by reducing pollen flows among multiple species (Morales & Traveset, 2008;Waser, 1978). Specialization of the pollination system could improve individual pollinators' energy intake rates by reducing the energy costs of learning new foraging behaviors in adapting new flower structures during shifts to new plant species (Chittka et al., 1999;Waser, 1986). Although the specialized pollination systems are of ecological advantage for both plant species and pollinator individuals (Armbruster, 2017), the number of pollinator individual-based interaction network studies on a community-wide level has so far been very scarce.
Studies of plant-pollinator interaction networks play an important role in understanding the generalization and specialization of pollination systems (Bascompte et al., 2003;Olesen et al., 2007).
The generalized character of pollination networks has been widely accepted (Blüethgen et al., 2007); however, most studies on pollination networks lack biological details that describe flower-visiting insects as pollinators (de Santiago-Hernández et al., 2019;Guimarães, 2020). This is acceptable for food webs, but inappropriate when assessing generalized versus specialized pollination in communities. In addition, most pollination networks are constructed based on flower visitor species Memmott, 1999), but these species-based pollination networks may not reflect the true pollination interactions between plants and their pollinators (de Santiago-Hernández et al., 2019;Guimarães, 2020). Indeed, in traditional species-based pollination networks, pollinator species are aggregates of pollinator individuals, which are the true pollination links observed in nature. In addition, pollinator species-level interaction networks are aggregates of different plants and pollinators species over a long period, which may obscure the pollinator individual behavior and increase forbidden links between interacting mutualists (Ings et al., 2009;Olesen et al., 2011). In this context, plant-pollinator relationships at the community levels could be more accurately represented by combining pollinator individual-based networks with pollinator' foraging behavior.
As species are assemblages of individuals, pollination networks are organized hierarchically and can be scaled down from speciesbased pollination networks to individual-based ones (Dupont et al., , 2014Lucas et al., 2018;Olesen et al., 2010;Tur et al., 2014).
Then, pollination networks can be built at two levels of resolution: pollinator individual-plant species network and pollinator speciesplant species network (Figure 1). Individuals-species networks represent interactions between pollinator individuals and plant species (Figure 1a,b), and species-species networks represent interactions between pollinator species and plant species (Figure 1c,d). However, to date, few empirical studies have attempted to explore the interaction networks between flowering plant species and pollinators at the individual level. For example, the network of the individual thistles Cirsium arvense and the honeybee Apis mellifera was more closely linked than previous knowledge of species pollination networks indicated . Tur et al. (2014) suggested that individual flower visitor-plant species pollen transport networks were more specialized than species-species networks, and generalist pollinator species often comprised specialist pollinator individuals. The two limited cases strongly suggest that pollination networks based on pollinator individual level could be more specialized than those based on pollinator species level (Araújo et al., 2011;Des Roches et al., 2018;Tonos et al., 2021), indicating that pollination networks based on pollinator individual level could fill the gap between high pollinator fidelity and the highly generalized pollination networks (Arceo-Gómez et al., 2016;Lázaro et al., 2008).
It is widely believed that the relationships between plants and pollinators are highly generalized because most pollinator species F I G U R E 1 Pollinator individual-based network and species-based network. At the pollinator population level, different pollinator individuals (squares) can interact with one (a) or lots of flowering plant species (b) (circles; different colors represent different flowering plant species), leading to individualbased networks within pollinator populations. At the pollinator species level, these individual-based networks result in a species-species subnetwork (c, hexagons; different colors represent different pollinator species). Finally, at the community level, the species-species subnetworks combine with each other to form species-based networks (d) Species-species subnetwork visit many plant species across broad time scales at the community levels and vice versa (Waser et al., 1996). This has been strongly supported by community studies on pollination networks over the past two decades (Blüethgen et al., 2007;Memmott, 1999;Petanidou et al., 2008). Building pollination networks based on flower visitors that contact the reproductive organs (Kaiser-Bunbury et al., 2017;Memmott, 1999), identification of pollen carried by flower visitors (Lucas et al., 2018;Tur et al., 2014), pollen on stigmas (Fang & Huang, 2013), and the direct assessment of pollinator effectiveness (de Santiago-Hernández et al., 2019;Traveset et al., 2015) can provide accurate assessments of the specialization of pollination networks.
Therefore, in the present study, we structured pollination networks

| Experimental sites
This study was conducted in 2019 at two alpine communities on the

| Tracking pollinators
Because effective pollen dispersal distance on our alpine meadows was less than 8 m (Wang et al., 2021), we built four small plots (10 m * 10 m) in the large plots (100 m *100 m) to observe the foraging behavior of pollinator (Appendix 1). We chose the same 4 small plots (10 m *10 m) to observe all the flowering plants in the large plots as much as possible. Quantifying the movements of pollinators was practical in relatively small quadrat, as pollinators tend to forage on nearby plant species (Fang & Huang, 2016;Waser, 1986). Pollinator visita- We tracked flower-visiting insects between 0900 and 1900 hours on clear days with no strong wind. We did not perform any observations outside of this period and conditions as pollinator activities are limited due to the low temperature at high altitudes (Fang & Huang, 2016). All pollinators were insects in the orders: Diptera, Hymenoptera, and Lepidoptera. We collected all the pollinators that continuously visited flowers when they left the selected plot, and identified them with the help of taxonomist experts. We observed a total of 28 flowering species (Appendix 2) over 64 h (0.5 h *4 sessions *4 quadrats * 4 small plots * 2 large plots), which were visited by 43 pollinator species (Appendix 3).

| Statistical analysis
To In addition, we use a one-tailed Z test to quantify the possibility of randomly getting higher NODF values than the experimental matrix networks.
To examine whether the network structure changes due to the change in network size when switching from the pollinator species-

| RE SULTS
In two alpine meadows, we tracked 208 pollinator individuals, which visited at least two flowers. Hymenoptera, Diptera, and Lepidoptera consecutively visited 51.2 ± 70.9 (n = 133), 14.6 ± 16.7 (n = 62), and 7.1 ± 5.6 (n = 13) (mean ± SD) flowers, and accounted for 87.3%, 11.6%, and 1.1%, respectively, of 7839 visits. In the Menyuan plot,  (Table 1). Network connectances at both Menyuan and Huangyuan sites in empirical pollinator individual-based networks were less than half compared with the null hypothesis (Table 1). Furthermore, interaction diversity and interaction evenness in empirical pollinator individual-based networks were significantly reduced compared with the null models due to the differences in the number of interactions (Table 1). Therefore, such differences between pollinator species-based and individual-based networks can be attributed to a significant decrease in the average number of links of pollinator nodes in empirical pollinator individual-based networks (Table 1), rather than to an impact of increasing network size. Both pollinator species-based and individual-based networks at two study sites were insignificantly nested, except the pollinator species-based network at Menyuan site. However, the NODF values were significantly lower in empirical pollinator individual-based networks than in null model networks (Table 1). The mean degree pollinator linkage level in pollinator individual-based networks was about 50% lower than that predicted by the null model. The mean interaction diversity for pollinators was significantly less when downsizing from pollinator species-based networks to individual-based networks, suggesting that pollinator individuals had a narrower food niche than their counterparts (Table 1).
In both study sites, the network specialization metrics (H′ 2 ) showed that the pollinator individual-based networks were highly specialized, with values of the network specializations at two sites were greater than 0.9 (Table 2). Downsizing from pollinator speciesbased networks to individual-based networks increased the network specializations (H′ 2 ) 1.7-fold and 2.1-fold at Menyuan and Huangyuan sites, respectively (Table 2). Mean plant specialization (d′p l ) of the pollinator individual-based networks was higher than that of the pollinator species-based networks at both sites ( Table 2).
The mean value of d′ for all plant species was 0.85, indicating that most plant species were the unique interactions in the pollinator individual-based networks. In addition, the mean value of d′ for all pollinators was 0.44, indicating that about half of the interactions were unique for pollinator individuals (Table 2).

| D ISCUSS I ON
In this study, using potential legitimate pollinators who continuously to transfer conspecific pollen between individuals to produce seed (Brosi, 2016;Darwin, 1876), and to ensure high foraging rates, most pollinator individuals restrict their visits to flowers of a single species or species morph (Chittka et al., 1999;Waser, 1986). In the cur- high degree of specialization due to floral fidelity over the short term (Brosi, 2016;Brosi & Briggs, 2013). Our results showed that most pollinator individuals had a high degree of floral fidelity over pollinator individual foraging bouts in the alpine grassland. This finding was consistent with that of Tur et al. (2014), who found that network downscaling indicated high specialization of pollinator individuals.
However, both of the two pollinator species-based networks in our study sites were moderately specialized (0.54 at Menyuan; 0.47 at Huangyuan sites), which are comparable to the degree of specialization (from 0.24 to 0.85) in previous pollination network researches over the long term (Blüethgen et al., 2007). Our results support the view that pollination systems might be specialized in the short term to ensure conspecific pollen transfer but generalized over the long term to ensure the robustness of pollination system (Brosi, 2016).
Moreover, our results also showed that the mean value of d′ for all plant species was 0.85, indicating that the most pollinator individuals visited a few numbers of plant species in the alpine communities ( Figure 2), ensuring effective pollen transfer within the species (Fang & Huang, 2016). It is worth noting that our research only focused on large flying insects (i.e., bees, flies, and butterflies), but some small insects with poor mobility (ants, beetles, thrips, etc.) can also provide pollination services for some alpine plants that are not easily visited by large insects. Future research needs to pay attention to the pollination ability of these insects at the community level.
All pollination network studies combined data on a specific time scale. For example, numerous pollination networks are constructed from highly aggregated information from daily field samples to weeks or entire seasons (Olesen et al., 2010;Petanidou et al., 2008;Schwarz et al., 2020). The structure of pollination networks could change significantly from 1 day to many years due to the temporal dynamics of species diversity, species turn-  season can consequently prevent the establishment of a highly connected network core and thus limit the specialization of the entire network and nodes (Schwarz et al., 2020;Seifert et al., 2021). For example, species can appear as specialists in networks that are combined over narrow time scales, such as 1 day or week. In contrast, they can appear as generalists in networks

CO N FLI C T O F I NTE R E S T
The authors declare no competing interest.  Note: Independent t tests were used to compare the specialization of nodes between pollinator species-based and individual-based networks at Menyuan and Huangyuan sites, respectively. Different letters indicate significant differences at .05 level.