We constructed models with raw network values as the response variables and constructed models with Z-scores from observed vs. null comparisons as response variables. Each model included honey bee abundance as a fixed effect and site and sample round as separate random effects. For models of PAC and d’, which were measured at the pollinator species level, we also included native bee taxon as a fixed effect. For models predicting network metrics in in the Central Valley, we included year as an additional fixed effect. We fit all models using the lmer function in the lme4 package and tested for significance using likelihood ratio tests. All analyses were conducted in R . We assessed whether honey bee abundance in wildflower plantings was associated with wild bee pollen fidelity and wild bee pollen diet diversity using separate GLMMs for each response and ecosystem. Each model included honey bee abundance, bee taxon, and the plant taxon from which the specimen was caught as fixed effects and site and sample round as separate random effects. We fit models using the lmer function in the lme4 package and tested for significance using likelihood ratio tests. We assessed whether pollen species composition varied as honey bee abundance increased using permutational MANOVAs in R, vertical tower for strawberries using the adonis function in the vegan package with separate models for each ecosystem. We tested the effect of honey bee abundance, native bee taxon, and plant taxon visited.
For bees caught in the Central Valley, we also tested the effect of year. Bee specimens were collected from different sites, and we accounted for nestedness using ‘stata = site’ in all models. Statistical results obtained from the adonis function depend on the order in which variables are added so we ran multiple permutations and report the most conservative results . We evaluated how pollen and nectar availability responded to honey bee introductions using separate GLMMs for each ecosystem and reward type. All models included as fixed effects: the abundance of honey bees, the plant species sampled, and, to control for baseline pollen and nectar resources, the mean pollen and nectar availability in unvisited bagged flowers. In the Central Valley, we also included year as an additional fixed effect. In the Sierra, nectar measurements varied by data collector, so we added data collector as a random effect. All models also included site and sample round as separate random effects. Pollen and nectar data were both zero-inflated. We modeled pollen availability as a binary response where successes were dehisced anthers with visible pollen and failures were dehisced anthers without visible pollen. Nectar availability was also modeled as a binary response where successes were flowers with measurable nectar and failures were flowers with no measurable nectar.Honey bee abundance in meadows, measured as the total number of honey bees visiting flowering plants during morning and afternoon netting transects, ranged from 9 – 2,363 bees per m2 per hour in the Central Valley and 0 – 184 honey bees per m2 per hour in the Sierra .
In the Central Valley, we recorded 1,082 native bees comprising 57 native bee morphospecies. In the Sierra, we recorded 2,329 native bees representing 116 native bee morphospecies. Apparent competition between honey bees and native bees was higher at sites with more honey bees in both the Sierra and the Central Valley . However, when comparing raw values from observed networks against null networks, there was no change in apparent competition in the Central Valley and apparent competition decreased with increasing honey bee abundance in the Sierra. Raw values for native bee specialization decreased as honey bee abundance increased in the Central Valley, but there was no relationship between honey bee abundance and d’ in the Sierra nor when comparing observed networks from either system against null networks. Raw values for complementary specialization decreased as honey bee abundance increased in the Central Valley but not the Sierra. However, when compared to null networks, increased honey bee abundance was associated with an increase in H2’ relative to null expectations in both the Sierra and the Central Valley. Neither the pollen fidelity of individual visitors nor the species richness of pollen carried on native bee bodies varied as honey bee abundance increased . The species composition of pollen was best explained by the plant taxon a bee had been visiting and bee taxonomic identity . Honey bee abundance was not a significant predictor of pollen composition in the Central Valley. In the Sierra, honey bee abundance was associated with a subtle shift in pollen species composition but explained less than 1% of variation across bee specimen.
Pollen availability in flowers, measured as the proportion of dehisced anthers with visible pollen, declined as the number of honey bees visiting flowers increased in both the Sierra and the Central Valley . Likewise, the probability of observing measurable nectar in flowers declined sharply as honey bee abundance increased in both the Sierra and the Central Valley. Pollen and nectar availability also varied among plant species and in response to baseline pollen and nectar availability in both ecosystems.Across two California ecosystems, increased honey bee abundance decreased floral resource availability, leading to shifts in native bee floral visitation patterns. Perceived apparent competition , a measure of niche-overlap, increased in both systems. However, when compared to randomly re-assembled null networks, honey bee abundance was associated with a decrease in apparent competition in the Sierra, suggesting native bees altered their interaction patterns to escape competition. These seemingly contradictory conclusions highlight the value of using null models to understand ecological data. For example, perceived apparent competition increases when one of the competing species is disproportionately abundant. Our null models conserved total numbers of honey bee and native bee visits but randomly redistributed them to different plants. Thus, deviation from null networks suggests that species are non-randomly shifting their visits to minimize niche overlap. Similarly, increasing abundance of generalist honey bees may decrease raw values of network-level complementary specialization as they account for an increasingly large share of all interactions. Again, null models help account for this bias to reveal shifts in interactions patterns across the full bee community. Complementary specialization at the network level increased in both systems, suggesting plant-pollinator interactions became more specialized as honey bee abundance increased. In contrast, there was no change in specialization at the species level , however, this metric is sensitive to small samples sizes and thus it is not surprising that we detected significant changes at the network level but not the species level. These findings align with those from a similar study in Spain which also found that complementary specialization increased as honey bee abundance increased in natural habitat neighboring orange groves. Such changes in network structure may affect the robustness of communities to species loss with potential implications for community functioning . Observed changes in complementary specialization reveal that pollinators can adapt to minimize competition in the short term and such adaptive foraging may allow species and communities to persist . On the other hand, specialized diets pose greater extinction risk for species and the more specialized a network, the greater the extinction risk for interacting partners , Thus, when there are diverse floral resources, container vertical farming native bees may be able to shift their visitation patterns to avoid competition with honey bees. However, in a world with decreasing floral abundance and diversity , adaptive foraging may not always be possible and there could be delayed effects of competition on the ability of all plants and pollinators to persist across longer time scales.
Although we observed shifts in visitation patterns at the species level, the pollen fidelity of individual foraging bees, as well as the diversity and composition of pollen grains on their bodies were relatively unchanged by honey bee abundance. Furthermore, the absence of a significant relationship between honey bee abundance and pollen fidelity persisted even when we restricted data to single plant species . These results are, to our knowledge, the first test of whether honey bee competition might alter the composition of pollen carried by individual bees. Brosi and Briggs found that removal of a dominant bumble bee species led to a decrease in the pollen fidelity of bumble bees visiting Delphinium barbeyi, suggesting competition and high species diversity maintain high levels of niche segregation. In contrast, we find that honey bee competition is not a major driver of native bee pollen fidelity. Instead, most variation in pollen fidelity and pollen composition was explained by bee taxon and the plant species the bee had been visiting when it was captured. For example, in the Sierra, honey bee abundance was associated with a decrease in visits to Camassia quamash with parallel declines in C. quamash pollen carriage . Although some plant species are over- or under-represented in the pollen data when compared to the visitation data , likely reflecting differences in pollen production among plant species and pollen vs. nectar collection by bees, shifting visitation patterns explain variation associated with changes in honey bee abundance. As such, while the pollen diets of wild bees were altered by honey bee competition, visitation data would have sufficiently documented this change. By simultaneously documenting declines in floral resource availability and shifts in resource use we demonstrate that native bees are being competitively displaced by honey bees and are thus likely to collect fewer resources or collect different resources. Decreases in resource availability could decrease native bee reproduction by limiting pollen collection and offspring provisioning . Indeed, although parasitism and nest site availability are sometimes more limiting than flowers , floral resources almost universally increase bee reproduction and flower availability is often a key limiting factor for population growth . Collecting different resources may also decrease reproduction if resources are of lower nutritional quality or otherwise unsuitable replacements for preferred host plants. For generalist feeders, having a large set of diet choices allows for maximum caloric and nutrition intake , and pollen and nectar quality influence bee health and reproduction . As such, changes in native bee diets and floral resource availability are likely to have negative consequences for native bee populations. If honey bee competition reduces resource availability, wildflower plantings may fail to benefit native bee populations, as has been shown in other systems . However, in our Central Valley wildflower plantings, the benefit of augmenting floral availability seems to outweigh any negative effects of bee-bee competition. This is confirmed by work from a separate project, conducted at these same sites over the same time-period which showed that wildflower plantings enhanced O. lignaria and B. vosnesenskii reproduction when compared to un-enhanced control sites. As such, wildflower plantings remain a valuable conservation tool despite honey bee competition, in agreement with studies showing overall benefits of wildflower plantings for native bee populations . Nonetheless, understanding how to improve wildflower plant mix selection to minimize negative effects of honey bee competition remains a key conservation objective. In the Sierra, honey bee abundance was more than twenty times lower than it was in the Central Valley and native bees were able to shift resource use to minimize niche overlap. Yet, resource availability sharply declined in both systems and the observed increases in network specialization may make the native bee community more susceptible to species extinction . Thus, even “low” levels of honey bee abundance may disturb ecosystems and future hive placements in sensitive habitat should be approached with extreme caution. More generally, this study contributes to our ecological understanding of competition. We document compelling evidence that honey bee competition increases niche overlap among species, alters native bee resource use, and decreases floral resource availability, broadly meeting the definition of exploitative competition . Yet, the age-old question of whether such competition might drive future extinctions remains unresolved. Honey bees have been implicated in the extirpation of native bee species , but there are also cases where honey bees and native bees coexist without one species fully displacing the other . Our findings suggest that native species can adapt to honey bee competition by shifting floral visitation pattens but declines in resource availability imply there is a limit to coexistence. Understanding what that limit is and how to sustainably manage honey bees in a way that reduces risk of native bee extinctions remains a key ecological and ethical question moving forward.Maureen Page and Neal Williams conceived and planned the study.