Recipient Organization
AUBURN UNIVERSITY
108 M. WHITE SMITH HALL
AUBURN,AL 36849
Performing Department
Biological Sciences
Non Technical Summary
This research focuses on the 3-dimensional organization and development of honey bee nests, and how nest organization contributes to colony function. Honey bee nests exhibit stereotypical organization, the result of over 34 million years of evolutionary pressure. Our understanding of this 3-dimensional organization, however, is based on a single-snapshot study of mature colonies, andno study has investigated the impact of disrupting nest organization on colony function. This is a critical gap, because beekeeping practices regularly and explicitly break nest organization at different points in a colony's life, the impacts of which are unknown.Understanding both how and whynaturalcolonies organize their nests, will have immediate implications for how we can bettermanagethese critical pollinators within the modern agricultural landscape.InAim 1 we perform a two-year study tracking the 3-dimensional nest throughout the process of colony growth, development, and reproduction. To collect and process these data at-scale, we have twomethodological advancements: (1) a portable field-photography rig, and (2) an automated classifier of comb contents.InAim 2, wetest the impact of disrupting nest organization in both incipient and mature colonies, to determine how nest organization contributes to colony function at different stages of development. This aim is independent of Aim 1, because it manipulates global nest organization (parallel combs), not comb contents.
Animal Health Component
(N/A)
Research Effort Categories
Basic
100%
Applied
(N/A)
Developmental
(N/A)
Goals / Objectives
Specific Goals and ObjectivesThe goal of this research is to describe the 3-dimensional organization of honey bee nests, to track its development over time, and to experimentally test its importance for colony function.- In Aim 1, we will perform a two-year study tracking the 3-dimensional nest structure throughout honey bee colony growth, development, and reproduction.- In Aim 2, we will test the impact of disrupting the 3-dimensional nest organization in both incipient and mature honey bee colonies. This will provide an understanding of how global nest structure may contribute to colony function at different stages of colony development.- In Methodological Advancements, we detail how we will adapt new and existing technologies to advance both the collection and analysis of these data.Together, this research forms a complementary and robust investigation into how and why honey bee nests are organized in 3-dimensions.
Project Methods
AIM 1: Description of the 3-dimensional development of a honey bee colonyAIM 1 Methods: To track the 3-dimensional growth and development of honey bee colonies, we will begin with swarms of honey bees, which will be installed into 40L hive boxes containing 10 empty wooden bee frames, but no comb (a "bee frame" or "frame", is a rectangular piece of wood upon which bees build comb). These colonies will be free to build and use their nest naturally. Each swarm will contain 10,000 workers, a single locally-sourced queen, and will be installed into the hive boxes when swarms typically occur in the Auburn AL area. Given a choice, honey bee swarms choose to inhabit 40L cavities, and so the space provided matches their natural-history (Seeley, 2010).From the installation date through October (see green box in Fig 3), these colonies will be inspected weekly (n=10 colonies). During each inspection, we will measure: the number of workers in the colony (Liebefeld method: Imdorf et al., 1987; Ainat et al., 2019), the number of drones in the colony (if any), and whether the queen has been replaced (swarming or supersedure). For monitoring the nest contents, we will use a portable photography rig (see Methodological Advancements, below) to photograph each side of each frame, to map out the following: capped honey, uncapped honey/nectar, pollen stores, uncapped brood (eggs/larvae), capped brood, queen cells (if any), and empty comb. This will be done for both worker comb, and drone comb. Finally, each colony's weight and brood nest temperature will be monitored using BroodMinder data loggers. During the colder months (Nov-Feb), we will inspect these colonies twice per month, or as weather permits. In total, these colonies will be tracked over two years, to describe the nest organization both during the incipient stages of colony growth and development (first year), as well as when the colonies are mature and reproducing (second year) (Fig 3). As a control for the weekly disturbance. additional colonies will be installed in identical conditions, but only inspected every 4 months ("inspection control" colonies; n=10).AIM 2: Testing the impact of disrupting nest organization in honey bee coloniesAIM 2 Methods: The general premise for this experimental manipulation is simple - we will regularly disrupt the 3-dimensional nest organization of honey bee colonies by shuffling the frames upon which the combs are built. Colonies will be monitored weekly, and shuffled, as in Aim 1. This includes monitoring the colony's progress via weekly inspections, as well as colony weight and brood nest temperature using BroodMinder data loggers (see methods for Aim 1).Incipient Colonies: The incipient colonies in the "nest disruption" treatment group (n=10 colonies) will be installed at the same time as the colonies used for describing the 3-dimensional nest structure (Aim 1, n=10 colonies), and will be monitored in the same way. Therefore, the colonies used for accomplishing Aim 1 (describing the 3-dimensional nest structure) will serve as the control for the colonies in Aim 2 (incipient colonies whose nests are disrupted). The only difference between the two groups is that the colonies in the nest disruption treatment group will have their frames shuffled after each weekly inspection, whereas the colonies in the control group will have their frames put back into the same positions in the bee box after inspection (i.e. the nest is restored to its same 3-dimensional state as before). Note that we also have an "inspection control" group: colonies that are only inspected every 4 months, to serve as a control for the weekly inspections. In addition to monitoring the nest contents, we will also track the ability of colonies to survive their first winter (control vs disrupted nest). Incipient colonies in the disrupted nest group will be monitored for 1 year, to focus specifically on the impact of disrupting the nest organization during early colony development.Mature Colonies: The next stage of examining the impact of nest organization on colony survival, growth, and reproduction will be conducted with mature colonies (i.e., colonies that have a fully-built nest). We will not be using the incipient colonies whose nests were shuffled in the first year, because the goal of this part of the study is to investigate the impact of disrupting nest organization at a different developmental stage, not the long-term effects of nest disruption.For this experiment, we will begin with 30 full-sized colonies in 10-frame equipment, with standardized numbers of workers, worker comb, drone comb, brood, and honey stores. In early March (approx. 6 weeks before swarming season), colonies will be inspected, and randomly assigned to one of three treatment groups: 1-control, 2-nest disruption, 3-inspection control (n=10 colonies per group; initial inspections ensure that all three treatment groups have colonies of similar condition). Using the same methods as Aim 1, these colonies will be inspected weekly, to measure the number of workers, drones (if any), and whether the queen has been replaced (swarming or supersedure). We will also monitor nest contents using the portable photography rig (see Methods below). At the end of each weekly inspection (March through October), colonies in the nest disruption treatment group will have their frames randomly shuffled to maximize mixing. This will be the only difference between colonies in the control and nest disruption treatment groups. We will monitor survival through the winter and into the spring, at which point the experiment ends (Fig 3). Colonies in the "inspection control" treatment group serve as a control for the weekly inspections, and will only be inspected every 4 months.METHODOLOGICAL ADVANCEMENTS:Aim 1 and 2 both reply on mapping the honey bee nest. To streamline this process, we will (1) build a portable photography rig which can take high-quality photographs in the apiary, and (2) use computer vision and machine learning to build an automated classifier of comb contents.Portable photography rig: Traditionally, bee frames would be photographed in the lab. However, this can damage comb contents, and cause undue stress on colonies. The portable photography rig brings key improvements into the apiary: controlled LED lighting, a camera at fixed distance/aperture, and diffusive fabric to eliminate shadows from ambient light. In pilot trials, we were able to photograph both sides of a 10-frame colony in under 15 minutes, a significant improvement over bringing frames into the lab. These images were of sufficient quality for a human to identify all types of comb contents, including eggs within cells.Automated classifier of comb contents: The second key improvement is to automate image classification. This is a well-established "image segmentation" problem, where the goal is to label every pixel in an image as belonging to one of a set of predefined classes. Honey bee nests are a particularly tractable, because the contents of the cells are inherently categorical (worker cell, drone cell, queen cell), and each cell contains only one type of contents (capped honey, uncapped honey/nectar, pollen stores, uncapped brood - eggs/larvae, capped brood, queen cells, and empty cells). To build the automated classifier, we will use the open-sourced software CVAT (Computer Vision Annotation Tool) to annotate a subset of images by hand. Pairing the images with the per-cell annotations, we will then train a model to automatically classify test images, using Python 3, the PyTorch library, and associated packages (Scikit-image, OpenCV, SciPy) (Bradski, 2000; Walt et al., 2014; Virtanen et al., 2020). Of course, to build a robust model that will be applicable to all the images that will eventually be collected, we must annotate sample images from across the entire dataset (approx. 1% of all images will be hand-annotated).