Does a strong reduction of colony workforce affect the foraging strategy of a social pollinator?

The way pollinators gather resources may play a key role for buffering their population declines. Social pollinators like bumblebees could adjust their foraging after significant workforce reductions to keep provisions to the colony optimal, especially in terms of pollen quality, diversity, and quantity. To test what effects a workforce reduction causes on the foraging for pollen, colonies of the bumblebee Bombus terrestris were experimentally manipulated in field by removing half the number of workers. The pollen pellets of the workers were taxonomically identified with DNA metabarcoding, a ROC approach was used to filter out underrepresented OTUs, and video cameras and network analyses were employed to investigate foraging strategies and behaviour. The results suggested that the plant diversity in the pollen pellets was high but plant identity and pollen quantity traits were influenced mainly by plant phenology. During the experiment, although the treated colonies increased foraging effort in relation to control nests, only minor changes in the diet breadth and in the other node-level and network-level indices were observed after workforce removal. Therefore, a consistency in the bumblebees’ feeding strategies emerges despite the lowered workforce, which questions the ability of social pollinators to adjust their foraging in the field.

macronutrients of the resources it collects in order to provide a balanced and optimal diet to the developing brood 21 . However, it is not known how population declines could modify the way pollinators gather resources. According to the Optimal Foraging Theory, it can be expected that individuals are influenced not only by the reward's nutrients but also by competitive interactions, so a higher forager density causes faster depletion of the resources and triggers a wider diet breadth in the foragers as consequence (i.e. densitydependent mechanisms 22,23 ). On the other hand, a significant loss of workforce in a social pollinator colony could lead to a sudden decrement of diversity and quantity of resources incoming to the nest. As a consequence to this, the colony should react in an adaptive and/or optimal way (i.e. gathering resources for enhancing and/or maximizing fitness) 24 . In other words, according to the Optimal Foraging Theory 25 , the colony could respond by augmenting the overall foraging effort to increase the amount of incoming resources, or the foragers could favour plants with high pollen production; alternatively, after workforce reduction, the foragers could enhance resource heterogeneity in order to assure the nutritional value given to the larvae by compensating for the resource types that had previously been brought into the nest by the foragers that went missing.
In the case of pollen collected by pollinators, studying insect-plant interactions is complicated by several methodological aspects. In addition to the direct observation of an insect's behaviour 26 , the analysis of pollen on an insect's body can reveal the interactions that happened during a pollinator's trip and can also yield the rarest interactions that normally remain undetected during observational surveys 27,28 . Yet, morphology-based identification of pollen lacks a uniform discriminatory power and requires great taxonomical knowledge [29][30][31] .
However, the potential benefits of pollen studies highlight the need to improve methods that are alternative to the morphological analyses. In this context, DNA-based approaches, such as DNA barcoding and DNA metabarcoding, represent reliable approaches 32,33 .
In other words, by using integrative approaches (e.g. DNA metabarcoding applied to ecological questions), methodological issues can be overcome and the interactions and the resource usage by declining pollinators can be explored in more depth.
In this work, we tested the possible expectations about changes in foraging preferences due to colony workforce reduction by experimentally inducing a sudden decline in the colony size of commercial colonies of the bumblebee Bombus terrestris (Linnaeus, 1758) and by investigating consequent changes in foraging. We intended to recreate a situation of workforce loss due to natural or human-based environmental conditions (see [34][35][36] ). We explored the foraging behaviour before and after the manipulation with video recordings and also resource utilization by identifying the pollen with a DNA metabarcoding approach. We focused on bumblebees, because (a) they are among the most effective  18,[39][40][41] . In each pair, a colony was used as a control, and it was not treated during the length of the experiment, while a second colony was used to apply a treatment of diminishing the worker population, which in practice consisted of manually removing 50% of the workers relative to the number of workers present in the period before removal in that colony. This removal threshold was inspired by studies reporting mortalities or worker losses up to 50% with respect to control colonies due to multiple stressors (see [34][35][36]. For removing the workers, as we used nest boxes with a way-in and a way-out holes, the way-in was left open for an entire afternoon so that workers could return to the nest but none could leave it, and then the nest was completely closed during the following night. Early in the next morning, light anaesthetization with CO 2 was applied to the nest for a very short time, workers were counted and half of the worker amount was removed from the nest. . Four days after placing the colonies in the field, the workers' pollen pellets were collected from the corbiculae of the legs just before entering the nest and after light anaesthetization with CO 2 42 (the workers were afterwards released outside their nest to avoid immediate complications for the larvae related to workers being anesthetized 43 ). The pollen of 18 bumblebee workers for each nest were surveyed before workforce halving ("before" phase, 6 th -11 th July). In the period after removing the workers ("after" phase, 20 th -23 rd July), pollen pellets of 18 workers for each colony were collected in the same way as the "before" period (17 workers for one of the nests).

DNA analyses and taxonomical assignments
Reference ITS2 DNA barcodes for the sampled plant species were obtained as described in 46

Selection of OTUs
Sorting false positives from data produced with DNA metabarcoding has been recently underlined 53 Table S3).

Networks of foraging
For each nest and at each experimental phase (time Firstly, we tested several node-level indices (where a "node" is either a foraging bumblebee or a plant).
Specialization was investigated using: (a) the "degree", that is the number of plant species found in a pollen pellet; (b) RR, the "resource range", that estimates the fraction of used resources to the total available 59

Traits of pollen production
Values of the plant trait of pollen production ("pollen quantity") were assigned to both used and unused plants flowering at the study area during the experimental time (Supplementary Information Table S1 and Table S2). These values were extracted from 71 , which ranks plants from low ("P0") to high ("P5") levels of pollen quantity in several European plant species.
Specifically, this ranking is based on the amount of pollen produced by the plant species, it is coherent with other rank-based pollen databases (see 72 ) and it also is incorporated in the plant trait databases provided by the TR8 package for R 73 . We deem that pollen production is a suitable plant trait for the comparative purpose of testing changes in food preference before-after a treatment in bees.
The probability of collecting pollen of a species was analyzed using logistic regression (generalized linear mixed-effect models) with presence/absence of a plant in a sample for a given colony as a response variable, treatment in interaction with (numerical) pollen quantity as predictors and nest identity as a random intercept, with binomial distribution and logit as link function, with the lme4 package for R. Confidence intervals were estimated with 1000 bootstrapping using the function bootMer.
Control and treated nests were analyzed separately. samples of more than two taxa).

Pollen plant diversity, node-and network-level responses to the treatment.
Taxa composition of the pollen samples changed over the study period, both in the control and in the treated nests (Supplementary Information Figure S1). In both the treated and control colonies, the experimental phase (before/after workforce reduction) predicted the plant identity of the pollen samples better than the nest identity, although both variables were significant ( Table   1).
The node level network analyses in the phase before removal revealed that Degree and PG were low but the plant's Closeness Centrality was high and PDI and RR were both low, while d' spanned over a wide range of the specialization-generalization gradient (Fig.   1). Changes after treatment were not significant, except for the quantitative PG index which changed significantly only in the control nests (Table 2 and Fig.   1).
The binary indexes of the network-level analyses didn't change significantly after treatment in the treated colonies. On the other hand, only two of the quantitative indexes (i.e., the Link Density and Vulnerability of plants) changed significantly over the study period (Table   3 and Fig. 2).

Foraging rate
After removing the workforce, the proportion of workers leaving the treated nests relative to the control nests' foraging rate increased (Fig. 3, the trend without the proportion to control's foraging is in Supplementary   Information Figure S2). Specifically, the treatment was a significant predictor of the number of workers leaving in the GLMM with an offset of the control's leavings (β after -β before = 0.40, likelihood ratio test χ 2 =14.945, df=1, p < 0.001). The probability of collecting plants of high pollen quantity decreased during the phase after the workforce removal with respect to the phase before, in both control and treated colonies (Fig. 4). The interaction of treatment and pollen quantity well predicted the collection probability (β after -β before = -0.55 and likelihood ratio test χ 2 =19.351, df=1, p < 0.001in the treated colonies; β afterβ before = -0.35, χ 2 =6.67, df=1, p < 0.01in the control colonies).

Discussion
Previous studies on the foraging activity of bumblebees mainly focused on altering a diet and investigating adjustments in foraging in laboratory conditions 22,39,40 . A novel aspect of our study is that we investigated how reductions in colony size would affect the resource utilization and the foraging behaviour of these key pollinators when free to forage in the field. To our knowledge, only two studies have previously investigated the effect of experimentally removing the bumblebees' workforce exclusively on colony fitness 74 and on the feeding of larvae 75 . Nevertheless, in our study we have focused on aspects related to foraging in the field and to the resource utilization by bumblebees when collecting plant pollen.
We acknowledge that our experimental design could have included more replicates (e.g., a higher number of colonies) and more samples of pollen and that these could strengthen the results. Even though our sample size and number of replicates was similar to other studies (see methods) and even though accumulation curves revealed that an acceptable level of plant diversity was yielded from the pollen samples, we encourage researchers to employ experimental designs that are replicated more. In our experimental design, we chose to  Figure 2 -Bumblebee-plant networks during the experimental phases of before (plot's panels "a" and "c", first and second nest respectively) and after (plot's panels "b" and "d", first and second nest respectively) the workforce removal. For each plant species a colour is given and the plant's full name is provided in Table S2. That the vegetation phenological changes played a role is also supported by the fact that the workers from all nests, both treated and control ones, collected a diversity of plants that was different between the "before-removal" phase and the "after-removal" phase (Table 1, Supplementary Information Figure S1)   . 3), which suggests an increase in the foraging effort of the colonies. Increased foraging rates were recorded also in honeybee colonies after reductions in the amount of stored pollen 42,82 , which suggests a link between the foraging rate and the amount of pollen stored in the nest.
Furthermore, the higher foraging rate we have found could either indicate that foragers made more foraging bouts per time unit and thus used more energy in travelling, or the alternative hypothesis of an increase in the number of foraging workers relative to colony size.
Furthermore, the increased foraging rate could result in storing a higher amount of pollen in the nest or in storing an overall wider plant diversity, to compensate for the missing workers. A limitation of our study is that the pollen stored in the nests was not evaluated in spite of its potential of revealing deeper colony-level responses, because we did not want to cause disturbance to the treated or control colonies other than by removing workforce. Furthermore, we expected that bumblebees will collect more heterogeneous resources after workforce reduction and that this will impact the feeding networks of the bumblebees and the plants they collected pollen from. Conversely, our results suggest that the diet breadth did not expand and the bumblebee-plant network was not impacted by the workforce manipulation, as suggested by the indices of binary networks (based on presence/absence of interactions), and by the minor changes found in quantitative network indices related to the quantity of the resources used by foragers (Table 3, Fig. 2). This constancy in the foraging networks after workforce reduction can be explained by some aspects of bumblebees' biology. In contrast to honeybees, the bumblebees are primitively eusocial which implies that colonies' performance tends to rely more on individual choices of single foragers than on social information 40,83 (the latter being the case of honeybees). This results from workers of Bombus terrestris having almost no contact with the larvae during their development 84 . Thus, foragers of our treatments might not have acquired any information on the development of the larvae and this could have prevented the expected foraging adjustment to take place. The lack of direct feedback between larvae and forager could uncouple the foraging choices and the colony's growth rate, as it was clearly shown that removing workforce results in having less progeny and of smaller size 74 .

Conclusions
By using DNA metabarcoding of pollen samples to overcome limitations of the morphological identification, this study investigated the effect of workforce decreases on the bumblebee foraging dynamics, on the chosen plant's pollen-production traits and on the foraging rate, using an experimental manipulation in the field.
After applying a reduction of pollinator's workforce, the bumblebees' foraging strategies and the heterogeneity of collected resources were mostly constant, except for the increase in the colony's foraging rate. If our results of a limited adaptation of foraging were confirmed by further studies and more replicated field-experiments, then these pollinators would have a limited ability to adapt to a decreased colony size as those that occur after multiple stressors (e.g., pesticide exposure, parasites, and diseases 85 ).