Recurrent honey bee losses make it critical to understand the impact of human interventions, such as antibiotics use in apiculture. Antibiotics are used to prevent or treat bacterial infections in colonies. However, little is known about their effects on honey bee development. We studied the effect of two commercial beekeeping antibiotics on the bee physiology and behavior throughout development. Our results show that antibiotic treatments have an effect on amount of lipids and rate of behavioral development. Lipid amount in treated bees was higher than those not treated. Also, the timing of antibiotic treatment had distinct effects for the age of onset of behaviors starting with cleaning, then nursing and lastly foraging. Bees treated during larva-pupa stages demonstrated an accelerated behavioral development and loss of lipids, while bees treated from larva to adulthood had a delay in behavioral development and loss of lipids. The effects were shared across the two antibiotics tested, Terramycin^R (oxytetracycline) and Tylan^R (tylosin tartrate). These results on effects of antibiotic treatments suggest a role of microbiota in the interaction between the fat body and brain that is important for honey bee behavioral development.

Antibiotic treatment

A total of twelve (N=12) colonies of A. mellifera were selected for the experiment. The experiment was conducted during the summer seasons of 2013 (trial 1 n=6) and 2014 (trial 2 n=6) with different hives. The apiary is located at the University of Puerto Rico’s Agricultural Experimental Station in Gurabo, PR, (18°15'26.6"N 65°59'11.5"W). Colonies were screened and health assessed. Colonies that had received previous treatment, sickly or weak were excluded. Colonies were paired based on hive population and composition. Selected control colonies were treated with powdered sugar, the vehicle of antibiotic. Experimental colonies were treated with the commercial Oxytetracycline “Terramycin” (Terra-Pro; Mann Lake Hackensack, MN, USA) or with Tylosin tartrate “Tylan” (Elanco, Greenfield, IN, USA), following the recommended commercial dose of the powdered antibiotic.

Cross-fostering design

Setup: To examine effects of exposure to antibiotics in physiology and behavioral development, we used twelve colonies (N=12) in a general cross-fostering design (Fig. 1) divided into two trials: trial 1 treatment with Oxytetracycline (N=6 colonies, 3control:3antibiotic) and trial 2 treatment with Tylosin tartrate (N=6 colonies, 3control:3antibiotic). Colonies were paired based on composition and randomly assigned a treatment; control or antibiotic. Initial treatment was performed in standard (40.6cm W x 50.4cm L x 24.4cm D) one story wooden hive boxes with eight frames. After three weeks of treatment, brood frames were collected from each pair and placed in an incubator (Percival, Perry, IA, USA) at 33°C for 24 hours. From each colony approximately 500 emerging bees were marked on the thorax with numbered color tags (BioQuip, Compton, CA. USA) to indicate source colony and treatment (N=5,786 total bees from both trials; See Table 1). Marked bees were divided by equal numbers and introduced either to their original colony or the cross-fostered alternate colony background with continued treatment. Whole colonies content (including frames) were moved to observation hives. This approach resulted in four different treatment groups: (1) no exposure (-/-) bees raised in control colony, kept in control colony. (2) developmental exposure (+/-) bees raised in treatment colony, introduced into the control colony. (3) adult exposure (-/+) bees raised in control colony, introduced into the treatment colony. (4) developmental and adult exposure (+/+) bees raised in treatment colony and kept in treatment colony.

Additionally, for further fat content analysis, twenty newly emerged bees (not numbered tagged) were collected at random from the control (n=10) and the antibiotic colonies (n=10) for each of the trials, prior cross-fostering. Also, an additional 100 bees from each colony were marked with paint of different colors in the thorax to indicate source colony and treatment and were cross-fostered; eg. 100 bees from C1 and 100 bees from A1 were marked with paint, 50 from C1 stayed in C1 the other 50 cross-fostered with A1 and same manner with the 100 bees from A1, to be later collected for the fat content analysis. This was performed for each pair of colonies.

Colony forming units: We examined the efficacy of antibiotics by measuring colony forming units (CFU) (Fig. S1). Bees collected from the -/- and +/+ groups at one and seven days of age, were washed in 70% ethanol (Sigma-Aldrich, ST. Louis, MO, USA) and had their digestive tracts removed. The digestive tracts were homogenized by vortex mixer in 10mL of nutrient broth medium (Himedia, West Chester, PA, USA) followed by dilution plating on nutrient agar (Himedia) and incubated at 35°C under aerobic conditions for 48 hours (VWR, Radnor, PA, USA). Individual bacterial colonies were counted and multiplied by the degree of dilution to obtain CFU of the original sample (Olsen and Bakken, 1987). This CFU study is a control that demonstrates antibiotic applied to the colony reached bees and bacteria that can be cultured aerobically are reduced after antibiotic treatment.

Dissections and fat measurement: extraction and weighing

Sample collection: Since bees present a lipid profile that follows worker tasks and age stages (Toth and Robinson, 2005), previously painted marked bees were collected at 1 (collected during cross-fostering setup), 7 and 14 days of age from each treatment group (10 bees per treatment group by age, n=100 from each trial). Of note, day 1 bees were collected only from the -/- and +/+ groups. This is due to bees being newly emerged and thus not cross-fostered at this age. Age of collection was selected to match onset of behaviors as described by Seeley, 1982 and Moore et al., 1998. Dissections: Collected bees were dissected by separating the abdomen from the thorax. The entire digestive tract, along with the sting apparatus and any wax scales observed on the outside were removed from the abdomen. Weighing and lipid extraction: To measure lipid content of bees by age and treatment, we used the total body fat extraction method as described by O’Donnell and Jeanne, 1995. Briefly, fresh weight (weight of abdomen after dissection) was obtained using a Fisher 11 analytical balance (Fisher Scientific, Hampton, NH, USA) accurate to 0.1 mg. The abdomens were then placed in a drying oven at 70°C and dried for three days. Next after the abdomens were dried and weighed (dry weight), they were placed in 5mL of extraction solution (2:1 chloroform:methanol; Sigma-Aldrich) on a rotary shaker (Thomas Scientific, Swedesboro, NJ, USA). The solution was replaced every 24hrs for three days. After the three days of extraction period, the abdomens were placed in the drying oven at 70°C for two days. Abdomens were weighed after the second drying period (extracted weight). Abdomen water content was obtained by subtracting fresh weight and dry weight of each abdomen (data not shown). Total fat content was obtained by subtracting the dry weight and the extracted weight of each abdomen.

Behavioral development assay

To examine differences in rate of behavior development, the four treatment groups were followed and recorded in glass-walled observation hives with eight frames, in two daily scans of two hours each (at 10:00 and at 14:00 hrs.) until the beginning of foraging activities. We quantified cleaning behaviors (removing debris from honeycomb cells), brood cell visits (nursing), and foraging behavior during daily behavioral scan samples (Giray and Robinson, 1994; Moore et al., 1998). Age at onset of foraging was determined by two different criteria; bees observed bringing in pollen and time spent outside the hive. More than 10 minutes outside the hive was considered a foraging flight, and less than that time was considered as orientation flights and therefore not marked as foraging. These criteria are based on studies by Robinson 1987, Calderone and Page 1991, Winston and Katz 1992, and Moore et al. 1998.

To measure rate of development, we focused on earliest performance of behaviors by focal group of individuals. Bees were divided by age groups, representative of the age cohorts by which tasks; cleaning, nursing, and foraging, should be performed or start at (Seeley, 1982): 3->6, 7->10, 11->14 days of age. Relative probability of task was examined as proportion of the number of bees performing cleaning, nursing or foraging task at the different age groups to the total number bees from each treatment group in the colonies. Although this proved to be an effective method, mean of onset of behaviors for all three tasks observed could not be determined, therefore we focused on comparative information based on proportions of task performed. 

Statistical analysis

CFU: We performed a Shapiro-Wilk test to determine if the distribution of the CFU counts was significantly different than normal. After confirming the data was not normal, we performed a Mann-Whitney U test to determine whether the CFU count was significantly different between -/-  and +/+ at day one and seven of age.

Lipids: Since trial 1 (Oxytetracycline) and trial 2 (Tylosin tartrate) were conducted at two different summer seasons and with different colonies, data from both antibiotics were analyzed as independent factors. We performed a Shapiro-Wilk test to determine if the distribution of lipid content was significantly different than a normal distribution. After confirming the data did not showed a normal distribution, we performed a square-root transformation to ensure our data followed a normal distribution. After the assumption of normality was met, we performed a three-way ANOVA with type of treatment, age and antibiotic as independent factors to determine their interaction with lipid content. A Tukey test was used as a Post-Hoc analysis.

Behavior counts: We determined that the behavior count data was over dispersed, therefore we used negative-binomial generalized-linear mixed model (GLMM) to determine if job count intercept and slope varied by hive. We determined that colony effects explained very little of the variation (only a 0.002%) for each of the worker counts. Therefore, we utilized a negative-binomial generalized-linear model (GLM) to examine the number of workers performing each task using type of treatment, age and type of antibiotic as independent factors.

Data was analyzed using the statistical program R (R Core Team 2014) v. 3.5.2 (2018-12-20). Packages: glmm (generalized-linear mixed models) v. 1.3.0, lme4 (lineal mixed-effects models) v. 1.1-20.