Location: Carl Hayden Bee Research Center
2025 Annual Report
Objectives
Objective: Benefit commercial beekeeping through understanding of the honey bee microbiome, colony communication, physiology, nutrition and behavior in health and disease. This effort will define the social aspect of known disease states, reveal novel states of disease or opportunism and their extended social influence, and provide targeted hypotheses for disease testing, diagnosis, and treatment. Using a combination of laboratory and field approaches, we will further our understanding of the functional capacities of the microorganisms typical of the hive environment, the alimentary tracts of queens, workers and developing larvae. We will apply this information to the management of disease and colony loss associated with commercial beekeeping. Finally, we will collaborate with industry to produce an artificial intelligence driven diagnosis of brood disease using a combination of high-resolution photographs and associated microbiome and viral signatures. This diagnostic tool will benefit the apiary inspectors and beekeeping community by providing a quick and reliable method for the diagnosis of brood disease, curtailing the misuse of antibiotics.
Objective 1: Conduct research to quantify the potential benefits of various queen gut microbiomes to colony health in a laboratory setting and compare with a commercial beekeeping operation.
Sub-obj. 1.A: Determine the tissue specific microbiome structure of the queen gut by age and sub-species.
Sub-obj. 1.B: Investigate host-microbial function of the aging queen gut.
Objective 2: Determine the effects of queen and worker microbiota on queen quality, queen productivity, and worker-queen interactions, including semiochemical signaling.
Sub-obj. 2.A: Determine the effects of social resource space and opportunistic bacteria on worker-queen interactions, queen health, and signaling.
Sub-obj. 2.B: Determine whether queens selectively avoid trophallactic feeding exchanges with workers infected by known pathogens or opportunists.
Objective 3: Conduct research to assess the effects of microbiota on worker-brood interactions in health and disease.
Sub-obj. 3.A: Determine host brood resilience to pathogens relative to microbiota character and social variables.
Sub-obj. 3.B: Determine host brood resilience to various EFB strains with exposure to propolis, honey and royal jelly in vitro.
Objective 4: Use artificial intelligence to develop a rapid and accurate brood disease diagnosis tool.
Sub-obj. 4.A: Expand our AI-training set to include a greater variety of larval disease symptomology from around the USA.
Sub-obj. 4.B: Produce an AI-driven brood disease diagnosis tool.
Approach
To explore relationships of health and disease in greater depth, we merge the techniques of metagenomics, chemical ecology and artificial intelligence, with a focus on microbiomes and disease states that dominate the aerobic to microaerophilic antimicrobial niches associated with colony function and social nutrient processing. This approach will determine how different microbial species or groups of native microbiota, including opportunistic and disease causing species respond to, or cause stress, and influence key social interactions, largely unknown.
We will select particular queens to examine in molecular detail. Two highly informative factors we consider are known age, and carbonyl accumulation; a proxy for biological age often considered "molecular mileage". Microbiome co-factors include the absolute abundance of C. melissae, the queen gut bacterium associated with youth, fecundity and colony size. Concurrent with this economic colony-level assessment, we will examine how selective interactions contribute to the queen’s gut microbiota and queen quality, examining both physiological indicators of queen reproductive and nutritional physiology, pheromone signatures and odors.
We will assess queen reproductive quality, physiology, and productivity to determine impacts of worker-queen and worker-larval interactions. Representative queens, workers, and brood will be sampled and analyzed for microbiota, semiochemicals, and physiological state and queen productivity assessed as before. We will compare across treatments and time points by repeated measures analysis. Queen and worker-queen interaction metrics will be checked to determine if nest worker microbiota composition impacts queen care or productivity. We will compare statistically the number of successful, unsuccessful, and total queen feedings across treatments by Pearson’s chi square test.
To improve brood disease diagnoses, we use a combination of machine learning, high throughput analyses, and molecular diagnostics. Using SCINet resources and open source software we extract features from high-resolution digital images of larval disease phenotypes that have been paired with corresponding microbiome data. Our preliminary machine learning algorithms have been successful in predicting the microbiome result when supplied with novel image symptomology. We propose that an AI application can be trained to identify and predict disease agents with great certainty when working from a representative and comprehensive set of pathogen-defined imagery.
Progress Report
This is the initial report for new project 2022-30500-002-000D, titled, “The Honey Bee Microbiome: Social and Reproductive Functions in Health and Disease” which replaces expired project 2022-21000-021-000D, titled, “The Honey Bee Microbiome in Health and Disease”.
For Sub-objective 1A, ARS researchers in Tucson, Arizona, performed transcriptomic and metagenomic analyses on queens of differing age from a commercial beekeeping operation in the Midwest. The microbial community analysis showed that two major bacterial species; Commensalibacter melissae and Lactobacillus panisapium dominated the queen microbiome hindgut, comprising approximately 73% of the total hindgut microbiota. Transcriptomic analysis uncovered substantial differences in host gene expression between young and old queens, as well as between queens with high and low C. melissae abundance. A consideration of colony variables and queen quality supports the inferred and complex relationship between queen age and C. melissae abundance as reflected in the observed gene expression patterns. Statistically supported, the presence of young queens with low C. melissae abundance exhibiting gene expression patterns similar to older queens suggests that the loss of this symbiont can accelerate the aging processes. The observations of three statistical outliers with youthful gene expression: two older queens with low C. melissae abundance and one old queen with high C. melissae abundance, suggests that a minority of queens can maintain youthful physiology even in the face of symbiont loss or advanced age. The aging physiology and gut microbiome of queens is a rich source of information for commercial beekeepers and queen breeders.
For Sub-objectives 2A and 2B, research efforts and samples derived from the original study design (seasonal changes in worker-queen interactions, microbiota and microbe transmission in Midwest commercial colonies) were redirected to an emergency research priority: catastrophic overwintering colony losses experienced nationwide by beekeepers in 2024-2025. Our beekeeper collaborator from the Midwest seasonal study lost over 80% of overwintering colonies in 2023-2024 in a manner similar to 2024-2025 nationwide losses. In collaboration with response teams across the United States including the Beltsville and Baton Rouge bee labs, researchers performed colony level and molecular analyses on colonies that survived and perished during overwintering/cold storage in 2023-2024. Longitudinal colony samples are being examined for stress factors (pathogens and microbiomes, Varroa mites, mite resistance to miticides, reduced worker and queen quality, and nutrition) commonly associated with reduced colony survival and colony productivity. Samples remaining after the catastrophic loss study will be redirected back toward the original study design examining the microbiomes of queens by genotype and age, and the role of queen-worker feeding exchanges in disease/microbial transmission. The original study design focuses on understanding bee-microbial interactions in particular and disease states and their transmission in general.
In support of their emergency research priority (catastrophic colony loss 2024-2025), ARS researchers identified mite loads and viral patterns associated with honey bee colony winter mortality. They analyzed longitudinal colony samples from a commercial beekeeping operation that suffered severe winter losses in the winter of 2023-24, comparing colonies that failed in October against those that survived through February. Failed colonies showed dramatically higher Varroa mite loads with an average of 6.32 mites per 100 bees in October compared to only 2.21 mites per 100 bees in surviving colonies, which further decreased to 1.17 mites per 100 bees by February. Preliminary analysis of viral loads using pooled colony samples revealed a number of trends. Considered a major pathogen, Deformed Wing Virus B was most prevalent and abundant regardless of season. Deformed Wing Virus A was slightly less abundant than DWV-B, but showed a similar pattern by season. Although not typically considered a major driver of colony loss, Lake Sinai Virus was the next most abundant virus. The fast-acting viral diseases Israeli Acute Paralysis Virus and Acute Bee Paralysis Virus showed the greatest increase overwinter suggesting a potential contribution to colony loss. While the pooled colony data is suggestive of major trends, individual worker bees, queens and Varroa mites remain to be screened for pathogen loads to explore transmission and co-infection and examine longitudinal data in detail. We will also screen the collected mites for molecular resistance to the acaricide amitraz, as well as measure viral loads and bacterial communities within the Varroa mites themselves to understand disease transmission patterns. Mitigating the specific factors that drive colony collapse or colony dwindling are still under investigation, but crucial to reduce heavy economic losses in commercial beekeeping.
In support of Sub-objective 3B, ARS researchers continued to investigate the effects of propolis on social immunity and the aerobic (social) and microaerophilic/ anaerobic (gut) microbiota. Investigating the most prevalent and destructive bacterial pathogen, they determined European Foulbrood (M. plutonius) strain resilience to hive products propolis, honey and royal jelly by culturing microbes in a laboratory setting. While all six tested strains of M. plutonius were resistant to honey, they found variation within species for resistance to royal jelly and propolis. Only two of the six EFB strains were resistant to all three hive products. However, not one of the tested strains had developed resistance to the antibiotics used by beekeepers: tylosin and oxytetracycline. Moreover, we tested other pathogenic bacteria often found in association with outbreaks of brood and adult disease, including Enterococcus, Paenibacillus, and various species/strains of Enterobacteriaceae, demonstrating that a minority of these bacteria have developed antibiotic resistance and are likely hive residents.
In support of Sub-objectives 4A and 4B, ARS researchers have organized a nation-wide collaboration to understand the variety of brood disease afflicting honey bees, and more quickly and accurately diagnose causative agents of larval disease. Progress this year includes multi-locus sequence typing of various brood disease bacteria found throughout the United States, more detailed characterization of the nature of disease progression throughout larval development, and distinguishing the variety and prevalence of virus that can inflict brood. Using high-resolution photos and collected microbiome data, they created proof-of-concept machine learning models to predict diseased microbiomes based on digital imagery. With the help of Convolutional Neural Network (CNN) and SCINet’s high performance computing (HPC) clusters, they successfully generated several artificial intelligence (AI) models that can distinguish between European Foulbrood, viral infection and healthy larvae and are now working toward expanding our training dataset to include all larval disease. Their proof-of-concept models achieved an average of 80% accuracy on the training/validation sets. When tested on an independent dataset from Illinois containing additional viral pathogens not present in training data, the models showed higher accuracy for EFB (72-88%) than viral infections (28-68%), highlighting both the promise and current limitations of this approach. Implementing AI-based diagnostic tools can reduce unnecessary antibiotic treatments and help maintain the microbiome integrity critical to colony health. However, expanding training datasets to include all major pathogens, healthy larvae, and diverse geographic regions will be essential for developing field-ready diagnostic tools.
Accomplishments
1. Royal gut check: key bacteria linked to honey bee queen health and longevity. Honey bee queens can live up to eight years compared to worker bees' six weeks, but queen failure remains a leading cause of colony loss in commercial beekeeping operations, making understanding of queen longevity mechanisms critically important. The biological factors that enable some queens to remain productive while others fail prematurely were poorly understood, limiting development of methods to predict or improve queen performance. ARS researchers in Tucson, Arizona, discovered that healthy, productive queens have higher levels of the beneficial gut bacteria Commensalibacter melissae in their digestive systems compared to underperforming queens. Using advanced genetic analysis, ARS researchers found that queens with more of this bacteria showed enhanced expression of over 1,400 genes associated with stress resistance, protein maintenance, and longevity. This is a stronger correlation than age alone, which was associated with only 719 genes. This finding provides the first evidence linking specific gut microbiota to gene expression patterns that influence queen bee health and longevity. The research opens new possibilities for improving queen quality through probiotic treatments or the use of gut microbiome profiles as biomarkers for selecting superior queens in commercial queen breeding programs.
2. A picture paints a thousand words: honey bee disease diagnosed by artificial intelligence (AI). Accurate field diagnosis of brood disease requires years of specialized training, leading many beekeepers to apply antibiotics prophylactically across entire apiaries when uncertain about the specific pathogen involved. When the pathogen is a virus, rather than the bacteria that antibiotics are active against, this practice promotes antibiotic resistance, disrupts beneficial gut bacteria essential for bee health, and proves completely ineffective. ARS researchers in Tucson, Arizona, developed an artificial intelligence system that can distinguish between bacterial European Foulbrood and viral infections by analyzing photographs of infected bee larvae. Final models achieved overall accuracy rates between 73-88%, with stronger reliability in detecting bacterial infections (72-88% accuracy) compared to viral infections (28-68% accuracy). Further training on other honey bee-associated diseases is ongoing and should greatly enhance the predictive accuracy and utility of this diagnostic system. By enabling beekeepers to quickly and inexpensively make targeted treatment decisions, this tool can help prevent the unnecessary application of broad-spectrum antibiotics that damage the gut microbiome and breed pathogen resistance.