Hot and Bothered: The interactive effects of heat and pathogens on Bombus impatiens
Author
Maya Frey, Micah Lohr, and Tali Szulanksi
Introduction
Bees are important pollinators that are at risk due to climate change. Larger species of bee such as the native common Eastern bumblebee, Bombus impatiens, are particularly vulnerable to population loss due to climate change. While heat is, of course, a stressor to our pollinators; there are many stressors at work. Another major stressor for B. impatiens is Crithidia bombi, a gut pathogen that reduces bees’ ability to forage and reproduce - and therefore their ability to pollinate. The research collected for this study focuses on the interactive effects between C. bombi infections and heat stressors on B. impatiens. The data has major implications for pollinator health, which in turn has major implications on our ability to continue eating agricultural products as the climate continues changing. As part of this project, we took behavioral data from the microcolonies– boxes with 10 bees each– over the course of around five months. To get this information, we took a “snapshot” of the microcolony in an attempt to get the behavior of each bee in around 2 seconds. We recorded the date, time of day, temperature of the room, infection status of the colony, and the following behaviors: feeding on 30% sucrose, incubating the brood, moving, remaining stationary, and fanning.
In this project, we will focus on these stressors’ effect on measurable behavioral outputs, namely feeding, incubating, and fanning behavior. We also looked at discrepancy in observed fanning behavior by observers. This is interesting because collecting behavioral data can quickly become subjective, and fanning behavior is particularly challenging to accurately measure. Fanning behavior typically occurs when a bee becomes overheated and attempts to cool down by fanning their wings very rapidly. Looking at the biology involved led us to generate the following three hypotheses that we proceeded to test and analyze.
Hypotheses
1.) We hypothesized that temperature and infection status have an interactive effect on incubating behavior by our replicate bees. Bees incubate their brood to keep it warm enough to survive, and as temperatures increase this becomes less necessary since the ambient temperature is warm enough for the brood. Additionally, as the temperature nears the bees thermal tolerance limits, they allocate their energy to immediate survival over maintaining the brood. We also expect infection status to have an effect on the behavior of the bees, as it may change their tolerance to high temperatures and what activities they allocate their energy towards.
H0: There is no effect of infection status and temp on incubating behavior, and no interactive effect of the two on behavior.
HA: There is an interactive effect of infection status and temp on incubating behavior.
2.) We further hypothesize that there is a difference in mean fanning behavior based on the observer. Since we are humans, there are likely differences in how each of us recorded the behavior. We chose to examine fanning in particular because it is a more difficult behavior to discern when you’re staring at bees through a plastic box in a dark room– their wings are small, and they move very quickly when fanning. It can be difficult to determine whether the wings are folded against their backs, or moving so fast you are not able to see them.
H0: There is no difference in mean fanning behavior based on the observer.
HA: There is a difference in mean fanning behavior based on the observer.
3.) Finally, we hypothesize that there is an interactive effect of infection status and temperature on feeding behavior. Past research has already indicated that infection with Crithidia bombi has sublethal effects on bumblebees, such as reduced foraging behavior and a shorter lifespan. Furthermore, heat stress at temperatures near or above their critical thermal limit will likely impact what the bees choose to allocate their energy towards. Thus, it might be possible that infection and heat stress together will impact feeding if the stress has a metabolic cost. We think that this could go either way in terms of feeding: they could feed more to accommodate for the increased energy needed to thermoregulate and fight infection, or they could feed less if they succumb to these stressors.
H0: There is no effect of infection status and temp on feeding behavior, and no interactive effect of the two on behavior.
HA: There is an interactive effect of infection status and temp on feeding behavior.
Statistical Analysis
Data Collection
Each data point represents a 2 second “snapshot” of bee behavior. We recorded the counts of bees exhibiting behavior in the categories of moving, feeding, stationary, incubating, and fanning. Bee colonies were in rooms with temperatures of 23°C, 30°C, 34°C, 37°C, and 40°C, and colonies as a whole were either infected or uninfected. In this report, we look at both uninfected and infected bees in rooms at every temperature, but only look at the feeding, incubating, and fanning behaviors. Feeding was identified as bees standing on the sucrose wick with their tongue clearly on the wick. Incubating as identified as bees standing stationary on the brood. Fanning was identified as bees moving wings quickly without flying.
Data Processing
After loading in the data, we dropped all of the NAs from the dataset. We also mutated one variable, observer_initials, to combine the data from “JM” and “JLM” because they are the same observer that just recorded their name differently.
Since the incubation and feeding data were count data and did not reflect the proportion of bees alive at the time of data collection, we converted our incubation and feeding variables to the proportion of live bees exhibiting that behavior in each microcolony. Three feeding proportions were over 1 and we removed these observations as the data were likely recorded incorrectly. We also converted our infection status to a factor with two levels and temperature to a character. These variables are discrete and not continuous, and we treated them as such in our analysis.
Statistical Analysis
Hypothesis 1
To assess our first hypothesis, we used an interactive ANOVA with a post-hoc Tukey HSD test to determine the interactive effect of temperature and infection status on incubating behavior. We ran a model check to test the assumptions of this ANOVA. According to this check, we have questionable homogeneity of variance and normality of residuals. We also made a boxplot of the data to assess outliers, and found that there are some; however, we feel it is ecologically to include this data because they represent unusual behavior. Based on our experimental design, we may not be able to ensure independence due to multiple observations of the same replicates; however, the large sample size accounts for this. Since the assumptions are not fully met for this analysis, we should be cautious interpreting our results. This concept applies for all three of our hypotheses as they are all from the same data set and the data was collected in the same way.
Hypothesis 2
To assess our second hypothesis, we used an ANOVA and a post-hoc Tukey test to determine the effect of the observer on the proportion of bees feeding. We also ran a model check on this ANOVA to test the assumptions for running this type of test. According to this check, the homogeneity of variance and normality of residuals are questionable. We also ran a Levene’s test for the homogeneity of variance, which we did not find significance for, meaning that this assumption is met enough to proceed. We also made a boxplot to assess outliers. Once again, we did find them, but we applied the same rationale as stated above.
Hypothesis 3
To assess our third hypothesis, we used an interactive ANOVA and a post-hoc Tukey test to determine the interactive effect of temperature and infection status on feeding behavior. We ran a model check to test the assumptions of this ANOVA. According to this check, we have good homogeneity of variance, but questionable normality of residuals. We also made a boxplot to assess outliers. Though there were some, we still included them as per the previously stated reasoning.
Loading required package: Matrix
Attaching package: 'Matrix'
The following objects are masked from 'package:tidyr':
expand, pack, unpack
library(glmmTMB)
Warning in checkDepPackageVersion(dep_pkg = "TMB"): Package version inconsistency detected.
glmmTMB was built with TMB version 1.9.1
Current TMB version is 1.9.2
Please re-install glmmTMB from source or restore original 'TMB' package (see '?reinstalling' for more information)
library(effects)
Loading required package: carData
lattice theme set by effectsTheme()
See ?effectsTheme for details.
library(performance)library(car)
Attaching package: 'car'
The following object is masked from 'package:dplyr':
recode
The following object is masked from 'package:purrr':
some
library(rsample)library(data.table)
Attaching package: 'data.table'
The following objects are masked from 'package:dplyr':
between, first, last
The following object is masked from 'package:purrr':
transpose
library(rstatix)
Attaching package: 'rstatix'
The following object is masked from 'package:stats':
filter
Table 1. Summary for interactive ANOVA for the effects of temperature and infection status on proportion of bees incubating. * represents significance for an alpha of 0.05.
Table 2. Tukey post-hoc results for the ANOVA examining the interactive effect of temperature and infection status on the proportion of incubating bees.
`summarise()` has grouped output by 'temp'. You can override using the
`.groups` argument.
pd<-position_dodge(width=0.2)ggplot(data=behavior_meanincubating, aes(x=temp, y=meaninc, color=infected)) +geom_point(position=pd) +geom_errorbar(data=behavior_meanincubating, aes(x=temp, ymin=meaninc-se,ymax=meaninc+se),size=0.5,position=pd) +theme_bw() +labs(x="Temperature", y="Average Bees Incubating", color="Infected",caption="Figure 1. Average proportion of bees incubating by temperature and infection status. \n Error bars represent the standard error the the mean.") +theme(plot.caption=element_text(hjust=0.5)) +scale_color_manual(labels=c("No","Yes"), values=c("red","blue"))
Warning: Using `size` aesthetic for lines was deprecated in ggplot2 3.4.0.
ℹ Please use `linewidth` instead.
Table 4. Tukey post-hoc results for the ANOVA examining the effect of observer on proportion of bees fanning.
behavior_meanfan<- behavior %>%group_by(observer_initials) %>%drop_na(fanningprop) %>%summarize(meanfan =mean(fanningprop), sd=sd(fanningprop),n=n(),se=sd/sqrt(n))behavior_meanfan<- behavior_meanfan[-c(9,12),]behavior_meanfan$n=as.factor(behavior_meanfan$n)ggplot(data=behavior_meanfan, aes(x=observer_initials, y=meanfan,color=n)) +geom_point(position=pd) +geom_errorbar(data=behavior_meanfan, aes(x=observer_initials, ymin=meanfan-se,ymax=meanfan+se),size=0.5,position=pd) +theme_bw() +labs(x="Observer", y="Average Bees Fanning",color="Number of Observations",caption="Figure 2. Average proportion of bees fanning recorded by each observer.\n Error bars represent the standard error of the mean.") +theme(plot.caption=element_text(hjust=0.5))
Table 5. Summary for interactive ANOVA for the effects of temperature and infection status on proportion of bees feeding. * represents significance at an alpha of 0.05.
`summarise()` has grouped output by 'temp'. You can override using the
`.groups` argument.
ggplot(data=behavior_meanfeed, aes(x=temp, y=meanfeed, color=infected)) +geom_point(position=pd) +geom_errorbar(data=behavior_meanfeed, aes(x=temp, ymin=meanfeed-se,ymax=meanfeed+se),size=0.5,position=pd) +theme_bw() +labs(x="Temperature", y="Average Bees Feeding", color="Infected",caption="Figure 3. Average proportion of bees feeding by temperature and infection status.\n Error bars represent the standard error of the mean.") +theme(plot.caption =element_text(hjust =0.5)) +scale_color_manual(labels=c("No","Yes"), values=c("red","blue"))
Interpretation
Hypothesis 1
For our first hypothesis, we found a significant effect of temperature and infection status on the proportion of bees incubating, and also a significant interactive effect of temperature and infection status. The P-values are all less than 0.05 (Table 1), which indicates that the results we found are not due to chance and are statistically significant. The F-value of the effect of temperature is 219.58, the F-value for infection status is 178.42, and the interactive effect has an F-value of 24.23 (Table 1). The F-values indicate how much more substantial the difference between groups is than the difference within a group, and therefore shows us how different the proportion of bees incubating was across the different temperatures, across infection status, and across the interactive groups of the two.
In the post-hoc Tukey HSD test, we found that each of the temperatures had significantly different bee incubation behavior from each other (Table 2). Incubation behavior in infected and uninfected colonies was also significantly different from each other (Table 2). These relationships can also be seen in Figure 1. The relationship became more complicated when we looked at the interactive effects of temperature and infection on incubation. Interactive effects in general were significant, however we also found three trends of insignificance. When the lower temperature colony was infected and the higher temperature colony uninfected at 30°C and 34°C, 30°C and 37°C, 34°C and 37°C, 30°C and 40°C, and 37°C and 40°C, incubating behavior was not significantly different (Table 2). A potential explanation for this is infection by the pathogen causing bees at lower temperatures to behave similarly to healthy bees at higher temperatures, perhaps because they are experiencing similar overall levels of stress. At both 34°C and 40°C, the incubation behavior of infected bees and uninfected bees were not significantly different from each other (Table 2). We also found that there was not a significant difference in incubation of uninfected bees at 37°C and 34°C, and infected bees at 34°C and 30°C, 37°C and 30°C, and 40°C and 37°C (Table 2). More research is needed to arrive at potential explanations for the second two trends we found.
Hypothesis 2
For our second hypothesis, even though the ANOVA is significant (Table 3), we do not have the statistical power to determine where the statistically significant effect is in a post-hoc test. Our sample sizes are inconsistent between observers: while some have over 200 observations, others have below 20. The small sample sizes are likely the reason why although we have a significant p-value, our F-value is close to one, demonstrating that the test’s statistical power is low (Table 3). However, the graph shows that WR may be driving the differences, as their mean and standard error are visually lower than the other observers (Figure 2). Still, we would need a larger sample size of observations from WR to determine if this disparity is due to the small sample size (n = 16) or real differences.
Hypothesis 3
For our third hypothesis, there is a significant effect of temperature and infection status on the proportion of bees feeding, and there is also a significant interactive effect of temperature and infection status. The P-values are all less than 0.05 (Table 5), which indicates that the results we found are not due to chance and are statistically significant. The F-value of the effect of temperature is 544.745, the F-value for infection status is 5.873, and the interactive effect has an F-value of 3.357 (Table 5). The F-values indicate how much more substantial the difference between groups is than the difference within a group, and therefore shows us how different the proportion of bees feeding was across the different temperatures, across infection status, and across the interactive groups of the two. The effect of infection status and the interactive effect are orders of magnitude smaller than in our first hypothesis ANOVA, but the values still indicate a significant difference.
In the post-hoc Tukey HSD test, we found that each of the temperatures had significantly different bee feeding behavior from each other except between 23º and 30ºC (Table 6). These are both below the ambient temperature for bumblebees, so it is likely that they were not experiencing any effects of heat stress here. Feeding behavior in infected and uninfected colonies are also significantly different from each other (Table 6). The overall interactive effect of temperature and infection status on feeding behavior was significant; however, we also found some trends that were not statistically significant. At 23, 30, and 37ºC, there was no significant difference in feeding between infected and uninfected bees (Table 6). There is also no difference between infected bees at 23ºC and bees regardless of infection status at 30ºC and 34ºC (Table 6). Overall, the trend appears to be that feeding behavior is not significantly different at lower temperatures; that is, temperatures at 34ºC and lower. At higher temperatures, the interactive effect becomes much more pronounced, likely because these temperatures require much greater thermoregulation. Interestingly, we see the highest average proportion of feeding at 40ºC (Figure 3), which may be due to the increased metabolic cost of thermal regulation at such a high temperature. This could also be due to the lack of a water source for the bees. Further research would be needed to determine the causes of this trend.
Conclusion
Overall, the results of our analyses indicate that heat stress and infection with C. bombi have an interactive effect on bumblebee fitness. Furthermore, the results also suggest that variables such as observers should be taken into account as a random effect in future analyses. The reduced focus on incubation at higher temperatures suggests that heat and sickness impact reproduction. This is especially compelling when combined with additional reproduction data we have from this experiment, which shows that at higher temperatures such as 37ºC and 40ºC, the bees do not produce any offspring whatsoever. Furthermore, the increased feeding behavior as temperature increases speaks to the metabolic cost of these two stressors on bees. These results are important outside of the context of the experiment. As climate change increases the frequency and magnitude of thermal events, we need to be looking at how pollinators are impacted. This is especially true as pathogens such as C. bombi often experience range shifts due to climate change, enabling them to impact bees in more places and for more of the year.
