Study system
In both 2016 (June) and 2017 (July), eight 2-frame observation hives (53 cm × 48 cm × 5 cm) were filled with honey bees and installed in a temperature-controlled facility at the Starks Lab Apiary on the Tufts University campus in Medford, MA, USA (see Bonoan et al., 2014 for diagram of set-up). All observation hives were queen-right, contained two frames of mixed brood and food, and had similar brood area. After installation, the facility was kept at a constant early New England spring-like temperature of approximately 19 °C and was continuously monitored (HOBO UX100-011 Temperature/Relative Humidity data logger). Following installation, bees were left alone to acclimate to the new location for 3 days. During this time, each colony was fed 150 ml 1:1 sucrose water. Upon inoculation, colonies in 2017 were significantly smaller than in 2016 (ANOVA on LM, F = 9.42, df = 1, p = 0.022) (Figure 1).
Inoculation with A. apis
Following acclimation, half the colonies (N2016 = 4, N2017 = 4) were fed inoculated 1:2 sucrose water while the other half (N2016 = 4, N2017 = 3) were fed 1:2 sucrose water without the pathogen. For each inoculum, one chalkbrood mummy (i.e., bee overtaken by the fungus, obtained from the USDA-ARS Bee Research Laboratory) was ground and added to 150 ml 1:2 sucrose water (Jensen et al., 2013). All colonies were fed the corresponding sucrose solution in inverted 1-pound queenline jars each day for 3 days. This method was used in both 2016 and 2017. To carry out field inoculations safely and correctly, we obtained approval (registration #2016-MIA18) and proper certification from the Tufts University Biosafety Office and worked closely with the Massachusetts Department of Agricultural Resources Apiary Program. To ensure the inoculation was successful, we counted the total number of mummies in each hive on days 4, 7, and 9 post-inoculation (PI) at 08:00 and 16:00.
Collection of temperature data and colony size
In 2016, we used the FLIR ONE (Gen 2 for Android, emissivity = 0.95) personal thermal imager to collect temperature data. Using the FLIR ONE, we recorded the temperature at the central point of each side of each frame in the observation hive for 5-day pre-inoculation and ten-day PI. A healthy honey bee colony should develop a colony-level fever (> 30 °C) to fight the disease and clean out any mummified larvae within 10 days (Jensen et al., 2013). Following Starks et al. (2000), we collected temperature data at 00:00, 08:00, and 16:00 each day. We estimated the adult population three times a week (Monday, Wednesday, Friday) after thermal imaging at 08:00. Colony size was estimated according to Sammataro and Avitabile (2011): a standard deep frame entirely covered by one layer of bees is roughly 2000 adult individuals, estimates were taken in increments of 250 bees.
In 2017, we used the FLIR E6 thermal imager (emissivity = 0.95). Using the FLIR E6, we recorded the average temperature of each side of each frame in the observation hive at 08:00, and 16:00, and 00:00 each day for 3-day pre-inoculation and 10-day PI. We estimated the adult population (see above) every day after thermal imaging at 08:00. In 2017, we used in-hive sensors (BroodMinder) to validate hive surface temperatures determined via thermal imaging as a proxy for internal colony temperature (Supp. Figure 1). In 2019, we used both thermal imaging cameras and protocols to record temperatures of non-experimental colonies. We confirmed a significant correlation (Spearman correlation, S = 364.28, df = 38, p < 0.001, rho = 0.963) between the two cameras, and thus, we confidently compare temperature data collected in 2016 with data collected in 2017 (Supp. Figure 2).
In both years, the order in which data were collected from each colony was randomized (using random.org to generate random lists) for each collection period.
Data analysis
All analyses were done using car, MASS, plyr, lme4, and glmmTMB in R version 3.3.2 (2016-10-31) (R Core Team 2018).
Infection status was confirmed using a zero-inflated generalized linear mixed model (GLMM) with a Poisson distribution in glmmTMB( ) (Brooks et al., 2017) for each year. The models tested for independent and interaction effects of treatment and day PI (mummies = treatment × day PI).
Colony size in 2016 was compared with colony size in 2017 using a linear mixed model (LM) with fixed effects of year and colony on number of adult bees (number adult bees = year + colony) at the onset of inoculation. To determine significance, we used marginal hypothesis tests, implemented with the Anova() function. Data fit a normal distribution.
In both years, we used the temperature collected during the five (2016) or three (2017) day pre-inoculation to calculate an average baseline temperature for each colony. We then used this baseline temperature to calculate the mean temperature change from baseline for each colony. Thus, each colony served as its own internal control. We ran LMMs that tested for independent and interaction effects of treatment and day PI on temperature change from baseline (temperature change = treatment × day PI), with colony added as a random effect. We ran one model for all data pooled, data collected in the morning (08:00), data collected in the afternoon (16:00), and data collected at night (00:00) for a total of four models per year. Again, we used marginal hypothesis tests, implemented with the Anova() function. Data fit a normal distribution.
We also examined the “per bee” effort of warming the hive by dividing the average temperature of the hive by the number of adult bees estimated in the hive that morning. To determine if there was a significant difference in the “degrees per bee”, we ran Gaussian family, log link, and generalized linear models (GLMs) that tested for independent and interaction effects of treatment and day PI (degrees per bee = treatment × day PI). Again, to determine significance, we used marginal hypothesis tests, implemented with the Anova() function. As above, we ran four GLMs: one for all the data together and one for each time of day separately.