Sulfoxaflor is a globally important novel insecticide that can have negative impacts on the reproductive output of bumblebee (Bombus terrestris) colonies. However, it remains unclear as to which life-history stage is critically affected by exposure. One hypothesis is that sulfoxaflor exposure early in the colony’s life cycle can impair larval development, reducing the number of workers produced, and ultimately lowering colony reproductive output. Here we assess the influence of sulfoxaflor exposure on bumblebee larval mortality and growth both when tested in insolation and when in combination with the common fungal parasite Nosema bombi, following a pre-registered design. We found no impact of sulfoxaflor (5ppb) or N. bombi exposure (50,000 spores) on larval mortality when tested in isolation but found an additive, negative effect when larvae received both stressors in combination. Furthermore, we found that sulfoxaflor and N. bombi exposure individually impaired larval growth but also found possible antagonistic effects when used in combination. Ultimately, our results suggest that colony-level consequences of sulfoxaflor exposure for bumblebees may be mediated through direct effects on larvae. As sulfoxaflor is licenced for use globally, our findings highlight the need to understand how novel insecticides impact non-target insects at various stages of their development. 

Experiment 1: Does sulfoxaflor exposure influence bumblebee larval mortality and development?

Methods

Sulfoxaflor exposure

Data from the United States Environmental Protection Agency (EPA) has shown that the sulfoxaflor residue levels in the nectar of a cotton crop sprayed twice with 0.45 pounds of sulfoxaflor per acre over an 11-day period did not fall below 5ppb, with pollen levels higher by a factor of approximately 10 [35]. It should be noted that spraying flowering crops is prohibited in Europe [59,60] but this is not the case globally [61–63] and recent legislative changes in the USA means that sulfoxaflor can now be sprayed on numerous bee attractive crops during flowering (including, with restrictions, cucurbits, strawberries and ornamental plants) [64]. Based on the EPA data above, we chose to expose the larvae to sulfoxaflor at a concentration of 5ppb, which is the same concentration used in previous work [17,19]. We also included a treatment group that were exposed to 0.28ppb, based on data from the Pest Management Regulatory Agency Canada [65] that demonstrated that sulfoxaflor residue levels in the nectar of seed-treated crops may be significantly lower than in sprayed crops. A higher concentration of 500ppb was also included as a positive control. Fresh treatment solutions were made every 3-4 days and solutions were stored at 4 degrees Celsius in glass, tin-foiled covered containers to reduce the potential degradation of the active ingredient [A. Linguadoca, personal communication].

Experimental protocol

Eight commercially-obtained bumblebee colonies (Bombus terrestris audax; Biobest, Belgium), with approximately 150 workers each, were housed in a room at 26°C (50-60% humidity) with ad libitum access to sucrose solution. 5 workers per colony were arbitrarily removed from the comb of the colony with forceps and were faecally screened for common bumblebee parasites (Apicystis bombi, Crithidia bombi, Nosema spp.) [43,66]. None of the colonies were found to contain any of these parasites.

We removed all living early larvae (n = 692, instar stages 1 & 2; fewer than planned on pre-registration because fewer were present in the colonies) and placed each one in an individual well lined with filter paper (24 wells per plate; 4 rows, 1 row per treatment). Plates were then incubated (Sanyo MIR-554; 32°C; approx. 60% humidity [67]). Larvae were starved for an hour, and then fed untreated sucrose solution (50% w/w) before examination under a dissection microscope (Nikon SM2800) to confirm (through observation of movement) that the larva was still alive. 28 larvae died in transit. The larvae were then left overnight, during which time 14 more died, resulting in a final sample size of n = 650 (control n = 166, 0.28ppb n = 162, 5ppb n = 157, 500ppb n = 165). Based on the results of a pilot experiment that aimed to establish a feeding regime that minimized mortality (Experiment S1; Figure S1), early larvae were fed pollen (honeybee collected pollen, Biobest, Belgium) suspended in sucrose solution (35.12g pollen per litre of 50% w/w sucrose solution) and containing the relevant concentration of sulfoxaflor, for 10 days [68] with each larva receiving 4 feeds of 2 µl a day. The nutritional composition of the pollen was unknown but consistent across treatment groups; likewise, and the likelihood that this pollen contained trace levels of other insecticides was unknown, but consistent across treatment groups. Given that outdoor use of sulfoxaflor is not permitted in the EU, we can be confident that it did not contain our focal insecticide. After the last feed of each day we observed each larva under a dissection microscope (Nikon SM2800). If the larva did not respond with movement to (a) the feeding solution alone or (b) subsequent touch with forceps, it was categorised as dead. Otherwise, pictures (iPhone 7) were taken for image J analysis to record growth (days 1, 5 & 10). After day 10, the larvae were frozen at -20 degrees Celsius.

Statistical analysis

We used an information theoretic approach based on AICc values. For each response variable tested we created a full model containing all fixed and random measured factors, for comparison with all subsets of that full model (retaining all the random factors in each case) and a null model containing just the intercept and random factors (see Table S3). We selected a 95% confidence set of models based on Akaike weights derived from AICc values, and parameter estimates, and confidence intervals are based on model averaging of this set.

Larval mortality was analysed via survival analysis (mixed effects Cox model) with treatment, size at the start of the experiment and the interaction between them included within the model, and with colony of origin and plate included as random factors. As larval size varied considerably between individuals, we analysed larval growth during the experiment (rather than absolute larval size, see pre-registration). Larval growth (day 5 growth = surface area on day 5 – surface area on day 1; day 10 growth = surface area on day 10– surface area on day 5) was analysed with a linear mixed effects model (lmer) with treatment, day (day 5 or 10), size at the start of the experiment and two interactions (day and treatment; size and treatment) included within the model [69]. Colony, plate and Individual ID were also included as random factors.

We made 2 deviations from the original pre-registered analysis plan (PDF attached with submission); (i) here, and with Experiment 2 (below), we pre-registered that we would consider larval growth at day 10 as (larval growth = larval surface area on day 10 – larval surface area on day 1). However, we realised that this approach did not allow us to understand larval growth at different ages, and thus chose to analyse growth at Day 10 as (larval growth = larval surface area on day 10 – larval surface area on day 5). (ii) We did not specify in our pre-registered design that we would include the interaction between day (the day the measurement was taken) and treatment within the analysis. However, we realised that including this interaction could provide information about differences in growth trajectories across treatments, and therefore in all growth analyses we considered this interaction with treatment within the analysis. Note that excluding this interaction does not qualitatively change the main effects.

We used the R packages Hmisc, lme4, coxme & MuMIn [69–72].

Experiment 2: Do Sulfoxaflor and N. bombi influence bumblebee larval mortality and growth when in combination?

Methods:

Parasite preparation

A wild bumblebee queen (Bombus terrestris) infected with N. bombi (determined through faecal examination, as described above) was collected from Windsor Great Park in 2016. The infected queen was dissected, and the fat body and gut were homogenized in 0.01M NH₄Cl. Then, as described in Rutrecht & Brown [73], the spore solution was placed in a centrifuge set to 4^°C and 5000 rpm for 10 minutes to isolate and purify the spore pellet. The spore solution was then resuspended in 0.01M NH₄Cl and the concentration of N. bombi spores was calculated using a Neubauer improved haemocytometer. This inoculum was used to infect 3 bumblebee colonies (Bombus terrestris audax) from which we sampled bees to create the inoculum used in the present experiment.

Experimental protocol

The same basic experimental protocol was used as in experiment 1. We used a fully crossed design that included 4 treatment groups, (control (no sulfoxaflor or N. bombi), N. bombi alone, sulfoxaflor alone, N. bombi and sulfoxaflor). Larvae that were allocated to receive sulfoxaflor exposure were fed a 5ppb sulfoxaflor in sucrose/pollen solution (see experiment 1) throughout, and the control and N. bombi larvae were fed a sucrose/pollen solution containing just acetone.

Following Folly et al. [74] we combined our N. bombi stock solution with 1000 µl of a sucrose/pollen mixture to make a stock solution of 50,000 spores per µl for larval inoculation. In the first feed of the experiment, each of the larvae in the parasite treatment groups were fed 2 µl of the N. bombi solution (paired with either control or sulfoxaflor laced sucrose/pollen solution respectability), and from this the bee received approximately 50,000 spores, a quantity that is known to infect 45 % of larvae [74]. 50,000 spores is well within the range of exposure that would be expected within an infected colony, when exposure occurs through faecal contamination [74]. After the experiment (10 days after inoculation) all surviving larvae were frozen (-80°C) and we later counted N. bombi spores in each surviving larva. We found no extracellular spores, in line with previous work which demonstrates it takes bumblebees between 2-3 weeks to develop extracellular spores [75,76] – due to the process required to count extracellular spores, we were not able to assess the presence of intracellular infections in our larval material.

The rest of the experiment used identical methodology to Experiment 1.

We were able to graft 768 larvae from 8 colonies. Seven larvae died during the plating process and 15 died over night and were thus not included in the experiment. 8 larvae were removed due to experimental error, so our final sample size was 738 (control n = 186, N. bombi n = 187, sulfoxaflor n = 182, N. bombi & sulfoxaflor n = 183).

Statistical analysis

Our statistical analysis followed the same approach as described above (pre-registration PDF provided), whereby each treatment group was compared to the negative control (for both larval mortality and growth). However, since this approach simply treats the combined stressor group as an extra level in the factor “treatment”, it provides no information as to whether any interaction is antagonistic, additive or synergistic. We therefore also conducted an additional, post-hoc analysis (not pre-registered), to confirm whether our results provided support for antagonistic, additive or synergistic effects of the two stressors (see table S3D & S3F). For the mortality data we used a survival analysis (mixed effects Cox model) with sulfoxaflor, N. bombi, N. bombi:sulfoxaflor & larva initial size included within the model, and colony and plate included as random factors (see table S3D for full model). For the growth data we used a linear model with sulfoxaflor, N. bombi, N. bombi *sulfoxaflor, day, initial size, N. bombi*day, sulfoxaflor*day and N. bombi*sulfoxaflor*day included within the model, and colony, larva and plate included as random factors (see table S3F for full model).