Abstract
Variation in diet breadth and specialization stems from fundamental interactions species have with their environment1-3. Consequently, understanding the drivers of this variation is key to understanding ecological and evolutionary processes, and will facilitate the development of predictive tools as ecological networks respond to environmental change4,5. Diet breadth in wild bees has been an area of focus due to both their close mutualistic dependence on plants, and because both groups are under threat from global biodiversity loss6. Though many of the principles governing specialization for pollinators have been identified7,8, they remain largely unvalidated. Using mechanistic models of adaptive foraging in pollinators9,10, we show that while temporal resource overlap has little impact on specialization in pollinators with extended flight periods, reduced overlap increases specialization as pollinator flight periods decrease. These results are corroborated empirically using pollen load data taken from bees with shorter (genus Andrena) and longer (genus Lasioglossum) flight periods across environments with both high and low temporal resource overlap. This approach reveals how interacting phenologies structure plant-pollinator networks and drive pollinator diet breadth via the temporal overlap of floral resources.
The dietary specialization of an organism sits at the nexus of numerous fundamental processes in ecology and evolutionary biology. Diet breadth helps define organismal niches1, mediates migration and dispersal potential2,11, shapes inter- and intraspecific competition3,12, influences species persistence in the face of environmental disturbance4, and can affect rates of speciation and extinction5. Even the simple assembly of the collection of dietary ranges of organisms in a community defines one of the major concepts in ecology, the food web. Therefore, identifying the drivers influencing diet breadth and specialization provides fundamental steps towards understanding a multitude of essential biological questions.
Much of the progress in understanding specialization in consumers comes from studies of herbivory1, particularly on insects13. Using wild bees and plant-pollinator networks as our study system, we extend the scope of those studies to include mutualistic interactions. A mutualism is a ubiquitous ecological interaction in which participant species benefit each other14, and pollination therefore differs from other insect herbivory because of the reproductive benefit offered to both consumer and consumed species. Bees depend on removing pollen from flowers to rear their offspring, providing a pollination service at the same time15. Pollination is also of particular interest given its role in supporting terrestrial biodiversity6 and agricultural output16, a role threatened by widespread declines in both pollinators and insect pollinated plants6,17.
Research on insect herbivores has largely focused on how adaptation to plant traits and defenses drives consumer concentration on single plant families18,19. However, other drivers of diet breadth have been identified (ref 7,8, see Table S1). Here, we expand on the fundamental effect of resource density on a consumer’s diet breadth and specialization7, not in a spatial context, but temporally. This resource density is important in its interaction with consumer phenology. For long-lived organisms (e.g., predatory vertebrates), opportunities to access a single resource type principally depend on the resource’s spatial distribution. If a single resource type is physically dense, then a predator can specialize on it without needing to turn to alternatives, whereas if it is patchily distributed then a strategy of specialization is less optimal. However, for a comparatively short-lived organism like a bee whose adult flight period can be measured in days, the distribution of a resource type in time is the principal determinant of accessibility. A plant species that blooms for 30 days is de facto dense, continuously available resource for a bee species whose flight period lasts for only 25 days, but sparse and patchily distributed for a species which flies for 90 days. The temporal pattern of flowering can therefore be considered the functional equivalent of resource density for short-lived organisms. These ideas have proven influential but require concrete theoretical and empirical validation to solidify their conclusions.
Expanding upon this idea of temporal density, we hypothesize that the degree of temporal overlap in the availability of different resources (flowering overlap among co-occurring plant species) in combination with the fundamental life history phenologies of bees (adult flight period) determines realized diet breadth. Applied to plant-pollinator systems, we hypothesize that reduced flowering overlap will reduce the diet breadth of short-lived but not of long-lived pollinators (Fig. 1). Shorter-lived pollinators will experience markedly different resource availability based on the degree of temporal overlap of co-flowering species. High flowering overlap will allow shorter-lived pollinators a greater number of options while low overlap will restrict options. Longer-lived pollinators will be less constricted by the temporal overlap of co-occurring resources, often having equal number of options regardless of the degree of overlap (Fig. 1). Importantly, our hypothesis does not indicate that flight period will necessarily correspond with being more or less specialized. Instead, it describes how temporal resource overlap will affect diet breadth as a function of a pollinator’s flight period.
We employed a multi-faceted approach to test our hypothesis. First, we leverage advances in modeling adaptive foraging in ecological networks9,10 to directly address the phenological mechanism driving diet breadth across different pollinator flight periods. Second, we utilize our intercontinental pollen load dataset of shorter lived Andrena bees and longer lived Lasioglossum bees from both a highly seasonal and a less seasonal environment which function as low and high temporal resource overlap treatments, respectively, to empirically validate our model outputs.
Results
Overview
Plant-pollinator dynamics are modeled using a dynamic consumer-resource approach that incorporates adaptive foraging of pollinators to mechanistically model pollinators’ consumption of floral rewards and reproductive services to plant species9,10 (see Supplementary Methods Table S2). The model’s adaptive foraging mechanism serves as a useful tool to implement and measure changing pollinator diet breadth (see Supplementary Methods). Phenology is integrated into the base model through modified sinusoidal wave functions which produce unique phenologies for every plant and animal pollinator . These functions and regulate floral reward production, adaptive foraging rates, pollinator visitation rates, and consequently, plant and pollinator reproduction across time (Tables S3-S4). The malleability of our phenology functions facilitates direct control over the availability of specific resources and the activity patterns of the animal pollinators over time in simulations (see Supplementary Methods, Fig S1-S5). The quantitative degree of overlap in co-occurring floral resources can be measured as the Total Resource Overlap (TRO) of the entire plant community, and the Averaged Resource Overlap (ARO) per plant species. Activity patterns of animal pollinators are set up to produce a range of different flight period lengths separated by differing lengths in between flight periods (Fig S6). Simulations use three fully connected bipartite network frameworks across 2072 combinations of plant and pollinator phenologies to produce 62160 unique networks used to model over 22 million plant-pollinator interactions (see Supplementary Methods).
Our pollen load data (see Supplementary Methods) comprises a large sample set of two bee genera, Andrena (Andrenidae) and Lasioglossum (Halictidae), collected both in the state of Michigan, USA (Nearctic) and the United Kingdom (Palearctic). Restricting our geographic scale to these two specific Holarctic locations aids in our ability to compare across both bee and plant communities due to their shared evolutionary history and overall similarity (see Supplementary Methods). The highly seasonal continental climate of Michigan20 produces lower degrees of overlap and highly seasonal flowering communities, while the more mild oceanic climate of the British Isles20 and its consistent temperatures produces longer, overlapping flowering times among plant species21.
Diet breadth or the degree of specialization in pollinators is measured both in-model with output on pollinator foraging effort and empirically with relative abundance of different pollen in bee pollen loads. We used two metrics that were applicable to both theoretical and empirical results for better direct comparisons, because simply counting unique taxonomic groups in the pollen data will not be applicable to theoretical results. We developed the first metric and labeled it Deviation from Generalism (DFG) (see Supplementary Methods, Table S5). Briefly, DFG is the normalized summation of all pairwise differences of pollinator foraging effort (or relative pollen load) on potential floral resources (see Supplementary Methods, Table S5). It ranges from 0 (perfect generalist) to 1 (perfect specialist). The second metric was the Coefficient of Variation (CV) of a pollinator’s foraging effort (or pollen load). Higher CV values occur with outlying values indicating specialization. With model results, both DFG and CV were applied to final foraging levels at the end of simulations as well as the average foraging levels across the last 1000 time steps (see Supplementary Methods, Fig S7).
Model Results
Our simulation results (Figs. 2, 3) support our hypothesis (Fig. 1) demonstrating strong effects of temporal resource overlap on the diet breadth of short-but not long-lived pollinators. Fig. 2 shows the results of one network as an illustration of how resource overlap explains large portions of the variation in diet breadth of pollinators with shorter flight periods (Fig. 2a, c), but little for the diet breadth of pollinators with longer flight periods (Fig. 2b, d). The transition from high explanatory power to low explanatory power is apparent as pollinator flight period increases regardless if specialization was measured at the end of simulations or averaged across the last 1000 model time steps (Fig 3). This result is also consistent across varying proportional lengths of time between pollinator flight periods using either the DFG or CV metric across all networks tested (Fig S8-Fig S10). As predicted, low levels of resource overlap drive pollinators with short flight periods to specialize because potential resource options are limited at any given moment in time. As temporal resource overlap increases, potential options increase resulting in a more generalized diet breadth (Fig. 2a, c). Long-lived bees, on the other hand, do not experience the same limitation of potential diet options due to temporal resource overlap, because they are active most of the entire flowering season and can potentially access most or all of the flowering plants. Results were consistent regardless of whether resource overlap was measured as total resource overlap (TRO) or average resource overlap (ARO).
Pollen data results
Our empirical results support our hypothesis, showing more specialized diets of short-lived bees in more seasonal environments (Michigan) with lower flowering overlap, than in less seasonal environments (UK) with higher flowering overlap. Our pollen data from Michigan presents more single family specialists than the less seasonal UK dataset (Fig S11). Dietary specialization between bee genera and regions varied significantly at both the level of botanical family (χ2=27.1, p<0.001, Fig 4a) and genus (χ2=25.3, p<0.001, Fig 4b). In accordance with our hypothesis, there was an effect of region at the botanical family level, but only for the short-lived Andrena where species in Michigan were significantly more specialized than their British counterparts. In contrast, there were no differences for long-lived Lasioglossum. At the botanical genus level, the same trend was more strongly expressed, with Andrena in the UK showing a more generalized diet than their Michigan counterparts, but with no differences from Lasioglossum in either Michigan or the UK. Results were consistent when analyzed using the DFG metric (Fig S12).
Discussion
Our model results mechanistically demonstrate how differing phenologies interact with varying degrees of temporal resource overlap to alter pollinator diet breadth. These results are corroborated through our regional pollen load data which also presents the first empirical validation of benchmark theories on adaptive foraging by consumers in general7 and pollinators in particular8. Furthermore, our results suggest broader implications regarding the drivers of bee diet breadth and diversification.
Global patterns of herbivory, particularly in insects, show strong trends of specialization at the botanical family level13. These patterns have been linked to phytochemical diversity19,22,23 which consequently drives an increase of both dietary specialization and species diversification in many herbivorous insect groups across a latitudinal gradient towards the tropics, most clearly seen and well-studied in Lepidoptera22,24. However, bees break sharply from this pattern, with a relatively low diversity in the tropics when compared to Mediterranean and xeric environments8,25,26. Tropical environments are dominated by highly generalized, often social species that visit a huge variety of botanical families27,28. Tropical areas have the highest levels of global phytochemical diversity, but their flowering patterns have wider phenological variation than temperate areas at both the inter- and intraspecific levels. This results in tropical floral communities that de facto show higher resource overlap in any particular year29,30 than temperate communities. Our theoretical and empirical results suggest that bees’ strong departure from the typical global pattern in herbivory13 may be due to a relatively stronger effect of phenology on their diet breadth than the phytochemical restrictions suggested for other insect herbivores. In fact, tropical bees have not evolved anything approaching the same high degree of dietary specialization as seen in the tropical Lepidoptera despite existing in the same extremely phytochemically diverse landscapes. For further discussion of this point, see Supplementary Discussion.
Our modeling developments present a flexible operational basis going forward. Integration of time dependent functional components into species behavior and traits show that even high dimensional network models can be further expanded in tractable ways to include the dynamics of organismal phenology. Additionally, our metrics for diet breadth complement each other (see Supplementary Methods) and reflect past methods (Fig S10). These metrics also integrated well with empirical pollen load data sets that can be used to vet theoretical predictions. By expanding pollen datasets globally and incorporating empirically vetted plant-pollinator phenology, future research can address how foraging competition occurs across overlapping phenologies and what effects changing climatic conditions can have on network stability.
Conflict of interest statement
We declare no conflict of interest.
Author Contributions
F.S.V. conceived the project. T.J.W., P.G., F.S.V., and J.R.M. developed the conceptual basis for the project. T.J.W. identified suitable collection sites and bee genera for the empirical study design, collected bee and pollen data, and identified specimens. P.G. and F.S.V. developed dynamic model, simulation design, code, and diet breadth metrics. P.G. implemented simulations and analyses. T.J.W. and P.G. wrote the first draft, and all authors edited and revised the manuscript.
Supplementary Information
Supplementary Methods, Supplementary Discussion, Supplementary Figures and Tables.
Data Availability
Pollen load data is available as Supplementary Table S9.
Code Availability
Simulation code and simulation data are available at the repository: https://github.com/fsvaldovinos/Phenology. Phenology parameters used in the simulation portion of our study are available in Table S10.
Acknowledgements
Part of this project was supported by the National Science Foundation grant DEB-1834497 awarded to F.S.V.
Footnotes
OrcidID of co-author was added.