Skip to main content
Dryad logo

Data from: Bumble bees damage plant leaves and accelerate flower production when pollen is scarce

Citation

Pashalidou, Foteini et al. (2020), Data from: Bumble bees damage plant leaves and accelerate flower production when pollen is scarce, Dryad, Dataset, https://doi.org/10.5061/dryad.9ghx3ffdv

Abstract

Maintaining phenological synchrony with flowers is a key ecological challenge for pollinators that may be exacerbated by ongoing environmental change. Here, we show that bumble bee workers facing pollen scarcity damage leaves of flowerless plants and thereby accelerate flower production. Laboratory studies revealed that leaf-damaging behavior is strongly influenced by pollen availability and that bee-damaged plants flower significantly earlier than undamaged or mechanically damaged controls. Subsequent outdoor experiments showed that the intensity of damage inflicted varies with local flower availability; furthermore, workers from wild colonies of two additional bumble bee species were also observed to damage plant leaves. These findings elucidate a feature of bumble bee worker behavior that can influence the local availability of floral resources.

Methods

Plants and insects: Bombus terrestris colonies were obtained commercially from the Biobest group. These included queenless microcolonies and founding queenright hives. Each queenless “mini-hive” box contained approximately 30 workers. 
Brassica nigra, B. oleracea, Solanum elaeagnifolium, S. melongena and S. lycopersicum plants were grown from seed in a climate chamber (Kälte 3000, RH 60–80%, LD 16:8). 
Flowering time studies: Plants (B. nigra and S. lycopersicum) in each flowering-time experiment were assigned to one of three treatment groups: control (undamaged), bee-damaged, and mechanically damaged (S. lycopersicum, n=20 per treatment; B. nigra, n=10 per treatment). At the start of the experiment plants were of uniform age (S. lycopersicum, 6 weeks old; B. nigra, 9 weeks old). Plants in the bee damaged treatments were placed together with a pollen-deprived B. terrestris colony inside a
mesh enclosure  inside a climate chamber (Kälte 3000, RH 60–80%, LD 16:8). To standardize damage, plants were regularly monitored and removed when bees had made a target number of leaf holes (5 for S. lycopersicum; 5–10 for B. nigra). Each plant in the mechanical damage treatments was then paired with a plant in thecorresponding bee-damage treatment, and we used a metal forceps and razor to replicate the damage pattern observed in the bee-damaged plant as closely as possible. Plants from both damage treatments, as well as undamaged controls, were then placed at randomly selected positions within a climate chamber (Kälte 3000, 60–80% RH, LD 16: 8) and moved to new
random positions every two days. Plants were monitored daily, and flowering time was reported as days elapsed since treatment (i.e., bee-damage, mechanical damage, no-damage). 
Laboratory pollen deprivation study: We assessed damage behavior by Bombus terrestris microcolonies that were either given abundant pollen resources within the hive (pollensatiated) or deprived of pollen (pollen-deprived). These microcolonies were queenless, but contained psuedoqueens that produced haploid larvae. Queenless microcolonies (with psuedoqueens) are commonly used for behavioral experiments with B. terrestris and have been shown to serve as good models for investigating a variety of behavioral, developmental, and ecological questions. Colonies were then given three days (days 2–4) to adjust to the treatments prior to being exposed to plants. During
this adjustment period (and throughout the experiment). On day 5, two non-flowering B.nigra plants (5-7 weeks old) were placed into each cage. Plants were replaced every 24h over three days (days 5–7) and we recorded the proportion of leaves damaged for each plant. On day8, all plants were removed and the treatment (pollen satiated vs pollen deprived) for each colony was reversed. Colonies were then given three days (days 9–11) to adjust to the new treatments, before again being exposed to plants (on days 12–14, during which we replaced plants and recorded damage as described above). Hives were weighed on days when the treatments were implemented or reversed (days 1 and 8) and monitored during the subsequent adjustment periods. At the end of the experiment (Day 15), final measurements were taken and hives were frozen and kept for dissection.
Roof study 2018:
Phase one (March 26th – May 25th): Bombus terrestris microcolonies were placed on a rooftop (on the Zentrum campus of ETH Zurich; Zurich CH) near a focal patch of 36 non-flowering plants of 6 different plant species (initially, Alliaria petiolata, Alyssum montanum, Aurinia saxatilis, Brassica nigra, Brassica oleracea, and Isatis tinctoria) (Fig. S5a). A. montanum was replaced by Fragaria vesca early in the experiment (13th April 2018) because bee-inflicted damage was not visible on the very small leaves of A. montanum. Bumblebee colonies were deprived of pollen for 3 days prior to the experiments, and received no supplemental pollen during the experiment, but had access to the Biogluc sugar solution within the hive throughout the study. Colonies were replaced approximately every 3 weeks, in accordance with the timeframe for effective pollination estimated by the commercial distributor. Plants were replaced at the same time. A total of three colonies were used during phase one. Damage was scored as the number of new leaf-holes produced each day (to facilitate tracking, damaged leaves, having known numbers of pre-existing holes, were individually marked with small green metal rings).
On sampling days (all weekdays except when it rained) we also recorded the number of bumblebees entering and exiting the hive and the number of returning foragers with and without pollen during three 60-minute periods per day (beginning at ~09:00, ~13:00, and ~16:00).
Phase two (June 4th- July 20th): Starting on June 4, we repeated the experimental design of phase one, except that we additionally placed a patch of 100 flowering plants, comprising four species (Brassica nigra, Fragaria vesca, Isatis tinctoria, Alliaria petiolata), separated by ~1 m
from the focal patch of 36 non-flowering plants. Additionally, on sampling days we recorded the number and species identity of any bees visiting our flowering patch, as well as those visiting the non-flowering patch, during three 60-minute periods per day (beginning at ~09:00, ~13:00, and ~16:00), along with the other observations described for phase one. Plants in the flowering patch that started fruiting were replaced. As in phase one, bee colonies (and plants in the non-flowering patch) were replaced every three weeks; two colonies were employed during this phase.
2019 Rooftop experiment (May 29th – July 13th):
Sixteen founding queenright Bombus terrestris colonies were equally divided between two treatments (non-flowering and flowering) and placed on separate rooftops, including the rooftop used during the 2018 study (Roof 1) and another nearby (Roof 2; ~200m from Roof 1). Colonies on both Rooftops (n = 8 per roof) were placed near focal patches of 300 non-flowering plants comprising 7 different species (Armoracia rusticana, Aurinia saxatilis, Brassica nigra, Fragaria vesca, Isatis tinctorial, Solanum lycopersicum and Solanum melongena). No flowering plants were present on Roof 1, but Roof 2 had a rooftop garden planted with wildflowers (4.5 x 7m; ~20m from the focal patch); in addition, we placed 30 additional flowering border plants (Antirrhinum ca. 3 months old) near the focal patch on Roof 2 (Fig. S5b). Colonies were not given access to any supplemental pollen or nectar resources, except on June 17th and 27th, when all colonies on both rooftops were provided with Biogluc sugar solution within the hive for 24h
to mitigate the effects of heavy rain and hot weather (respectively). Additionally, colonies were provided with external water feeders to provide relief to the hive during hot weather. Colonies were weighed and measured twice a week (after sunset) to monitor queen presence and hive development. The rooftop garden (on Roof 2) was mowed on June 29, and we also removed the other flowering plants on Roof 2 on this date, after which point no flowering plants were present on either roof. Colonies and focal (non-flowering) plant patches were kept in place until the onset of the reproductive switch point (measured as the appearance of the first drone). All hives were removed for continued monitoring in climate chambers on experimental day 45. Damage was monitored, as in the 2018 experiment, for 150 randomly selected non-flowering plants in
each focal patch. Plants in the focal patches were replaced after three weeks to ensure they remained in a non-flowering state. Concurrently, we conducted a transect study between 11th March- 26th July to roughly estimate surrounding flowering resources. Flowering plants were identified along two 1 km long transects (with North-South and East-West axes) originating at Roof 1. Surveys were conducted every two weeks, during which we identified all flowering
plants visible within 5m of the transect (using the iNaturalist app). 


Statistical analyses: All statistical analyses were carried out using R: A language and Environment for Statistical Computing version 2.15.3 (R Core Team 2013). To test for effects of the three damage treatments on flowering time, we used generalized additive models (GAMs) using the “gam” function in the R package mgvc. To account for the
fact that six plants in the B. nigra experiment (4 in the undamaged control treatment and 2 in the mechanically damaged treatment) did not flower by day 40 (termination of experiment), we converted flowering data into a binary response variable, with each plant classified as “flowering” or “not-flowering” on each experimental day. After initial data exploration, we decided to model flowering time as a function of treatment, time itself, and plant ID, with an interaction between time and plant. Treatment and time were entered as parametrically estimated explanatory variables, with the smoothed term including time by plant (Table. S3). We set the data family to binomial and the basic dimension of the smoothing function to ‘re’ in order to account for random factors (34). Additionally, we used the package itsadug to handle the autocorrelation within the time series. Models were validated by plotting Pearson versus Fitted residuals for response variables and covariates, to ensure that homoscedasticity and normality assumptions were met. Model selection was made using the ANOVA function and AIC to get the best fit. This approach allowed us to explore the simultaneous effects of treatment and time and thereby explain a high proportion of deviance within the model. Following this analysis, we used a generalized linear model (GLM) with a binomial error distribution and logistic-link function to calculate the least square means comparisons between the treatments (35). Additionally, we used the mean least squares coefficients to estimate the estimated time to reach 50% of flowering plants for each treatment. To test for effects of pollen availability on damaging behavior, we used a generalized linear
model with discrete error distribution (Poisson), with diet as a fixed effect and colony as a 7 random effect. We used backwards stepwise model simplification based on likelihood ratio tests to reduce model complexity as far as possible. To test for differences among damage treatments over the course of our rooftop studies, we
used a series of generalized linear mixed effect models (GLMMs). For the 2018 rooftop study, we used a GLMM with Poisson error fitted by maximum likelihood, in order to look at differences between the phases. Date was included to account for temporal non-independence of data. We used backwards stepwise model simplification based on likelihood ratio tests to reduce model complexity as far as possible . We also calculated a rolling 7-day average for plant damage, using weekday data to estimate the weekends, using the “zoo” package. In 2019, we also used a GLMM to look at the effect of roof treatment, in addition to using a general linear model to test for differences before and after the roof treatment was cut. These models were also validated using backwards stepwise model simplification.

Funding

Eidgenössische Technische Hochschule Zürich