We demonstrate that habitat fragmentation and its associated reductions in site level pollinator diversity alter the structure of plant-pollinator interaction networks. However, these changes ran counter to our expectations in several respects, and suggest that the structure and function of networks in our system may be more or less robust to the level of pollinator diversity loss documented here. The robustness of network structure to species loss resulted in part from the high numerical dominance of the nonnative honey bee, a super-generalist that contributed the majority of floral visits in every study plot. Despite the finding that our networks exhibited notable inter-annual variation with respect to many of the statistics we measured, the majority of our analyses yielded qualitatively similar conclusions in the two study years, bolstering our confidence that our findings resulted from real biological phenomena rather than chance. While we predicted that networks in fragments should exhibit reduced nestedness relative to reserves, we found the opposite pattern. This surprising finding runs counter to other studies that have found nestedness to be positively related to pollinator species richness . Two mechanisms could lead to an increase in nestedness: the addition of species, individuals, large plastic pots or links that contribute to nested patterns, or the removal of those that reduce nestedness .

In our system,there is no evidence that fragments experienced relative gains in species whose interaction patterns would contribute to nestedness, such as specialists that visit the most generalized plant species or generalist species that visit the majority of plant species. Thus, the increased nestedness in fragment networks is likely attributable to their experiencing a combination of removal of species whose interaction patterns do not conform to nestedness . Whatever the mechanism, higher nestedness in fragment networks may buffer the structure and function of these networks from further perturbation , especially if drivers of pollinator species loss tend to extirpate specialists or rare species first . The increased nestedness in fragment networks may be related to another unexpected finding, that pollinator niche overlap did not respond to pollinator diversity loss. Niche overlap measures the overall similarity of interaction patterns among pollinators in a network . In this theoretical framework, extirpating species whose interaction patterns are similar to those of extant species will tend to decrease niche overlap, while extirpating species that occupy uncommon niches will increase niche overlap . Thus, for niche overlap to remain relatively unaffected by pollinator species loss, the extirpated species must have intermediate levels of overlap with other pollinators. Pollinator species with an intermediate number of plant partners that are not proper subsets of more generalized pollinator species may fall into this category of intermediate niche overlap; the extirpation of these species, as discussed above, would also increase network nestedness.

The results of our analysis of pollinator niche overlap deviate from two different, but ecologically reasonable, expectations regarding patterns of species loss in modified habitats. The first expectation is that highly connected generalists persist when more specialized species are extirpated , thereby increasing niche overlap. The second expectation is that habitat alteration reduces the functional redundancy of biological communities , thereby reducing niche overlap. Given that our results deviate from these expectations, our findings represent an interesting ecological phenomenon worthy of further investigation. A number of other studies have described the degree of niche overlap in plant-pollinator interaction networks , but few have empirically investigated this metric in the context of pollinator species loss , and thus general patterns relating niche overlap to pollinator diversity remain to be uncovered. Our third expectation regarding the impact of pollinator species loss on network structure—that links will be lost at faster rates than species—was also not supported by our data. While the minimum number of links in a network could be as low as half of the number of species present , the generalized and asymmetric structure of most networks generally yields a larger number of links than species . Empirical studies have also found that links may increase at a faster rate compared to the accumulation of pollinator species, both due to sampling effects and to the underlying biology of interaction networks . In our case, the number of links per species is indeed higher than the number of species present, but not as high as reported in other studies .

The relatively low number of links per species in our system may be driven by the large number of rare species whose addition to the network adds only one additional link each, thereby shifting the average number of links per species closer to one. From a conservation standpoint, our finding that the number of links per species was not altered by species loss represents another line of evidence that network structure in our system is robust to habitat alteration, at least when plant assemblages remain intact. As with our analyses of network structure, our analyses of pollinator generalization received yielded unexpected results. Network-level interaction selectivity was indeed higher in reserves than in fragments, but it was negatively, not positively, related to pollinator diversity. Since reduced selectivity by pollinators may negatively impact the reproductive success of plants they visit via reducing conspecific pollen transfer , plants in our fragment plots may suffer reduced reproductive success as a result of reduced H2′ relative to reserves. On the other hand, enhanced network-level interaction selectivity in species-poor networks may, to some degree, buffer the erosion of pollination services in networks that have suffered the greatest extent of pollinator diversity loss. Interestingly, Burkle and Knight found a similar pattern in which habitat size was positively related to species richness, but negatively related to H2′; the authors attributed their finding to an increase in the selectivity of numerically abundant generalist pollinators in smaller habitats. In our system, there was no evidence of a negative relationship between the selectivity of numerically abundant generalist pollinators and pollinator species richness or habitat size . However, it is interesting to note that patterns in H2′ resembled patterns in honey bee proportional abundance, which likewise exhibited opposing responses to habitat fragmentation and pollinator diversity loss. While elucidating the mechanisms underlying these intriguing findings is beyond the scope of this study, we can infer from these results that habitat fragmentation impacts plant-pollinator mutualisms above and beyond the effects of removing pollinator species. Habitat modification has also been documented to strongly alter networks of interactions among organisms independently of impacts on species richness ; these findings underscore the complexity of ecological interactions and highlight the need to take into account the natural history of organisms when predicting how the structure and function of biological communities may respond to anthropogenic impacts. Unlike our predictions regarding network structure, our hypothesis regarding the role of honey bees in networks was wholly supported by our analyses . Our results corroborate the findings of other studies that the super-generalist honey bee behaves as a ubiquitous and highly connected network “core” species that enhances network nestedness, increases overall network generalization, large pots plastic and contributes a disproportionate number of links. As such, it also likely performs the majority of pollination services in our system , at least to the plant species it visits frequently and effectively . While we found that the honey bee is by far the most numerically dominant pollinator in our system, it is also important to note that the exclusion of honey bees from our analyses did not qualitatively alter our findings regarding the impacts of habitat fragmentation and pollinator diversity loss on properties of networks .

Thus, our conclusions do not appear to be driven by the finding that honey bee abundance varied with pollinator richness and differed between habitat categories. While comparisons of network metrics with versus without the inclusion of honey bees yielded results consistent with theoretical predictions regarding the loss of highly connected generalists , this analysis does not necessarily provide insight into the consequences of physically removing honey bees from the ecosystem . In real-world systems, the presence or absence of a numerically dominant pollinator species can elicit profound behavioral and numerical responses in other pollinator taxa, and links among plants and pollinators may shift in response to species removals so as to maintain network robustness . However, this analysis does shed light on the honey bee’s current role in structuring networks and its potential to impact the fitness and evolution of co-occurring plants and pollinators. While numerically abundant generalists are thought to be relatively resistant to extirpation , there is at least one report of precipitous declines in unmanaged honey bee populations in the past . Given that factors related to increased mortality in honey bee populations remain pervasive in many ecosystems worldwide , it is essential for conservation efforts to secure the structure and function of plant-pollinator interaction networks irrespective of current contributions by honey bees. As with other studies, we found notable year-to-year variation in the structural properties of our networks , such that data from the two years were uncorrelated with respect to three of the metrics we calculated . However, perhaps more surprising is our finding that nestedness , the proportion of interactions attributed to honey bees, and the number of singleton species in each network were largely consistent across the two years of sampling, despite the fact that the two years differed markedly in their temperature profile and the timing and quantity of precipitation . Pollinator assemblages are known to be highly variable from year to year , as well as exhibit time lags in their response to environmental conditions . The network metrics that exhibit high consistency from year to year in spite of fluctuations in plant and pollinator diversity, distribution, and phenology may thus provide insight into how patterns of interactions between mutualists structure communities at a locality over longer timescales. Additionally, metrics that are robust to inter-annual variation in plant and pollinator assemblages at a locality may be candidate metrics that enable quantitative comparisons between networks with different spatial and temporal origins .We discovered that the structure of plant-pollinator networks in our system remained robust to the loss of pollinator species richness. In fact, networks appeared to gain resistance against further loss of structural integrity as pollinator species are lost due to habitat fragmentation. Our finding of multiple counterexamples to the predictions of prevailing theory also underscores the need for more research examining plant-pollinator interaction network structure and function in species-rich systems experiencing pollinator species loss and integration of novel pollinators. Lastly, while networks provide an excellent glimpse into patterns of interactions between plants and pollinators, more research is needed to mechanistically map the relationship between the myriad of network statistics available today to empirical measures of plant and pollinator fitness and population dynamics. Plants requiring animal pollination are diverse, comprising an estimated 87.5% of plant species worldwide . Animals engaged in pollination are also diverse . While the diversity of pollinators is important for maintaining both the short-term functionality and the long term stability of pollination services , at the local scale, the majority of pollination services may be performed by pollinator taxa that are numerically abundant and exhibit generalist foraging behavior . Thus, understanding the population dynamics of taxa that occupy central roles in pollinator communities can provide insight into the fate of plant pollinator interactions in a changing environment. The western honey bee is recognized as the most important pollinator species for many crops , though the role of non-honey bee pollinators in augmenting crop pollination has recently been emphasized . Conversely, the importance of honey bees as pollinators in natural habitats remains a matter of debate . Given the large native range of the honey bee , its long history of naturalization in its worldwide introduced range , and rather early recognition of its potential to impact other pollinator taxa and plants , it is surprising that only recently have researchers begun to develop a broader understanding of the role of honey bees in non-agricultural ecosystems worldwide . The prevalence of honey bees as potential pollinators in natural habitats is important for several reasons. First, factors that contribute to recent increases in the mortality of managed honey bee colonies may also affect feral or wild honey bees . If honey bees are important pollinators of plant communities in natural systems, their decline could have implications for pollination services. Second, non-honey bee pollinators are declining in many parts of the world, largely driven by habitat loss and degradation, as well as other contributing factors such as pesticides, pathogens and parasites, invasion of non-native species, and climate change .