The Pennsylvania State University
The Graduate School
Department of Entomology
ASSESSING IMPACTS OF PESTICIDES AND OTHER STRESSORS ON HONEY BEE
COLONY HEALTH: EXPERIMENTAL AND MODELING APPROACHES
A Dissertation in
Entomology and Operations Research
by
Wanyi Zhu
© 2013 Wanyi Zhu
Submitted in Partial Fulfillment of the Requirements
for the Degree of
Doctor of Philosophy
May 2013
The dissertation of Wanyi Zhu was reviewed and approved* by the following:
James L. Frazier Professor of Entomology Dissertation Co-Advisor Co-Chair of Committee
Michael C. Saunders Professor of Entomology Dissertation Co-Advisor Co-Chair of Committee Advisor of Operations Research Major Christopher A. Mullin Professor of Entomology Ottar Bjornstad Professor of Entomology, Biology, and Statistics Timothy Reluga Associate Professor of Mathematics Gary Felton Professor of Entomology Head of the Department of Entomology
*Signatures are on file in the Graduate School
ii
ABSTRACT
A healthy honey bee colony is a population of closely interacting individuals that form a
highly complex society. However, the combinational energy-draining stresses of illness from
environment, nutrition, and human migratory and cultural practices strike honey bee populations
day after day, depriving them of long-term health. The possibility of a multi-factorial cause is one
of the problems that make investigating colony declines especially complex. Pesticides are a
major concern due to their widespread distribution within the hive. Beyond the effects of acute
toxicity, pesticides are also likely to cause sublethal effects that result in behavior alteration or
disorder of individual bees, together with the synergistic interactions among various pesticides in
hive matrices, which eventually trigger serious harm to colony health. Therefore, a combination
of mathematical modeling and experimental approaches were proposed to study the honey bee
colony dynamics and quantify the colony-level effects of nutritional disturbance due to pesticides.
First, a larval rearing method was adapted to assess experimentally the chronic oral
toxicity to honey bee larvae of the four most common pesticides detected in pollen and wax -
fluvalinate, coumaphos, chlorothalonil, and chlorpyrifos - tested alone and in all combinations.
All individual or combined pesticides at hive-residue levels triggered a significant increase in
larval mortality compared to untreated larvae by over two fold, with a strong increase after 3 days
of exposure. Among these four pesticides, honey bee larvae were most sensitive to chlorothalonil
compared to adults. Synergistic toxicity was observed in the binary mixture of chlorothalonil with
fluvalinate at the concentrations of 34 mg/L and 3 mg/L, respectively; whereas, when diluted by
10 fold, the interaction switched to antagonism. Chlorothalonil at 34 mg/L was also found to
synergize the miticide coumaphos at 8 mg/L. The three and four component mixtures of tested
pesticides have mostly demonstrated additive effects in larval bees. One exception is that the
addition of coumaphos significantly reduced the toxicity of the fluvalinate and chlorothalonil
iii
mixture, the only significant effect in all tested ternary mixtures. We also tested the common
‘inert’ ingredient N-methyl-2-pyrrolidone at seven realistic concentrations, and documented its
unexpected high toxicity to larval bees compared to adults.
Considering the extensive detection of chlorothalonil, its coexistence with other
pesticides in diverse combinations especially in hive pollen and wax, and its substantial larval
toxicity alone and in mixtures shown here, the potential impacts of fungicides on colony survival
and development needed further investigation. Thus, we explored the potential hazard to honey
bee larvae of frequently-found fungicides at environmentally relevant levels. Bravo®, its active
ingredient (AI) chlorothalonil, and the formulations Nova® and Pristine® at environmentally
realistic levels all triggered a significant increase in larval mortality through 6-d continuous
dietary exposure. We also found a significant difference in larval toxicity of the fungicide
formulation and its AI. Bravo® exhibited a monotonic and positive dose response for larval
mortality, with hazard ratios increasing with concentrations; however, chlorothalonil showed a
complex nonmonotonic dose response for larval mortality. A critical concentration of 3 mg/L was
most toxic to honey bee larvae among those tested. Bravo® EC50 was significantly lower than the
AI by a factor of 4.6 to 18.3. Enhanced toxicity of this formulation positively correlated with the
length of exposure or the stage of larval development. The pairing of Bravo® and Nova® is the
only mixture inducing significant synergism on mortality of larvae older than 3 days, with the
mixture eliciting 2-times greater lethality than the expected concentration additive toxicity. This
is the first study to report synergism for developing honey bee larvae between the non-systemic
fungicide chlorothalonil and systemic EBI fungicide myclobutanil at environmentally relevant
dietary levels.
For further testing of pesticide impacts at the colony level, and linking of larval responses
to effects on later honey bee life stages and colony health, we developed a worker-based, stage-
structured model of honey bee population dynamics. This model was formulated with combined
iv
difference and differential equations, consisting of six discrete stages based on honey bee
temporal polyethism: egg, larva, pupa, nurse, house bee and forager stage. It is unique in
capturing the adaptive feedback mechanisms in the population and resource dynamics in a
healthy bee colony, including the comb pattern formation, brood maintenance and collective
foraging behavior. By validating with independent data sets for colony numbers at different
latitudes and under different conditions, our model represents the most advanced population
model for integrating the top-down differential model and bottom-up difference model that gives
the most refined and realistic details of colony population and resource dynamics to date. Two
variance-based sensitivity analyses of the model suggested that the disruption of the numerical
basis of colony population dynamics has only delayed impacts on colony survival compared to
minor changes in colony social structure, especially the nurse-to-forager transition, which can
have immediate and drastic consequences. The model simulations also indicated that a balanced
allocation of workers with respect to dynamic changes in colony task demands, particularly
during the fall season, which is the sensitive stage for colonies to prepare for entering the winter
season, is the key to sustain the colony survival. This colony‐level simulation model represents a
useful tool that can be used to integrate exposure and effects data at individual bee levels of
potential stressors within the social dynamics of a honey bee colony. Based on our modeling
results of the most influential role of nurse-to-forager transition in maintaining colony
homeostasis, the future risk assessments should include testing pesticide impacts on life-history
transition in honey bees.
Lastly, as a further step to extend our current modeling efforts, a decision support system
(DSS) model (also called knowledge-based system model, expert system) was developed using
the NetWeaverTM software, as a new and innovative method of transferring timely, up-to-date
decision support to non-specialists and stakeholders for colony management and conservation.
This expert system was built upon recent modeling work that determined, described and
v
quantified the major indicators of the colony dynamics including: 1) reproduction quality of the
queen, 2) brood-caring quality of nurse bees, 3) foraging quality of foragers, 4) quality of the
nectar and pollen resource in the environment, 5) quantity of the honey and pollen storage in the
hive, and 6) disease levels of the hive (the levels of Varroa mite infection). Using associated
reference conditions and thresholds for each indicator based on peer-reviewed literatures and
model simulations, the DSS model can be developed to compare the current condition (user
input) with reference conditions or threshold values, based on a range of arguments and logical
relationships, which depict how the key factors characterize the colony dynamics and is defined
by our existing model and domain experts. This system will offer effective directions for
beekeepers and other stakeholders to determine the condition of honey bee colony, diagnose the
potential stressors affecting colony health, and prioritize management plans based on relevance to
colony health.
vi
TABLE OF CONTENTS
List of Figures ......................................................................................................................... vii
List of Tables ............................................................................................................................ xiii
Acknowledgements .................................................................................................................. xv
Chapter 1 Introduction ............................................................................................................. 1
Literature Review ............................................................................................................. 5 Background ............................................................................................................... 5 Honey bee colony as a superorganism ..................................................................... 8 Colony Collapse Disorder ........................................................................................ 15 Pesticide Impacts ...................................................................................................... 18
Research Rationale ........................................................................................................... 21 Current Honey bee Population Models .................................................................... 22
Research Objectives ......................................................................................................... 25
Chapter 2 Pesticide Impacts on Honey Bee Larval Health ...................................................... 29
Introduction ...................................................................................................................... 30 Materials and Methods ..................................................................................................... 33 Results .............................................................................................................................. 39
Single pesticide toxicity ........................................................................................... 39 Two-component mixture toxicity ............................................................................. 42 Three-component mixture toxicity ........................................................................... 49 Four-component mixture toxicity ............................................................................. 52 ‘Inert’ ingredient toxicity ......................................................................................... 52
Discussion ........................................................................................................................ 53 Chronic toxicity ........................................................................................................ 53 Mixture toxicity ........................................................................................................ 56 ‘Inert’ toxicity ........................................................................................................... 58
Conclusions ...................................................................................................................... 59 Acknowledgements .................................................................................................. 60
Chapter 3 Common Fungicides and Their Formulation Impacts on Honey Bee Larval Health ............................................................................................................................... 61
Introduction ...................................................................................................................... 61 Materials and Methods ..................................................................................................... 65
Test Fungicides ......................................................................................................... 65 Test Organisms ......................................................................................................... 65 Fungicide Chronic Toxicity Tests ............................................................................ 65 Statistical Analysis ................................................................................................... 67 Formulation vs. Active Ingredient Toxicity ............................................................. 68 Pesticide Interaction Determination ......................................................................... 69
vii
Results .............................................................................................................................. 70 Control Toxicity ....................................................................................................... 70 Pesticide Formulation vs. Active Ingredient Toxicity .............................................. 70 Bravo and Nova Mixture Toxicity ........................................................................... 77
Discussion ........................................................................................................................ 80 Chlorothalonil Toxicity ............................................................................................ 80 Fungicide Formulation Toxicity ............................................................................... 83 Synergistic Toxicity .................................................................................................. 85
Conclusions ...................................................................................................................... 86 Acknowledgements .................................................................................................. 87
Chapter 4 A Stage-structured Honey Bee Population Model .................................................. 88
Introduction ...................................................................................................................... 89 Model Development ......................................................................................................... 91 Results .............................................................................................................................. 106 Discussion ........................................................................................................................ 121 Conclusions ...................................................................................................................... 131
Acknowledgements .................................................................................................. 132
Chapter 5 Conclusions ............................................................................................................. 133
Summary .......................................................................................................................... 133 Implications of this Study ................................................................................................ 134 Future Work ..................................................................................................................... 138
Experimental and Modeling Research ...................................................................... 138 Development of A Decision Support System for Colony Management .................. 140 Acknowledgements .................................................................................................. 152
REFERENCES ................................................................................................................. 154 Appendix A The Development of Honey Bee Expert System ........................................ 168 Appendix B The Mixture Toxicity of Fungicide Formulations ...................................... 178
viii
LIST OF FIGURES
Figure 1-1. A conceptual diagram of the dynamic networks of individual-level nutrient sensing pathway and social-level pheromone communication pathway and its role in regulating division of labor and maintaining the colony homeostasis. Nutrient-sensing pathways in the social and physical development of worker bees all converge on the major social pathway at the individual bee level, which involves the antagonistic interaction of juvenile hormone (JH) and vitellogenin (vg): JH level is positively correlated with foraging behavior and therefore enhances the food dynamics and egg production; vg level in the ovary tissue is a proposed negative regulator of insulin/insulin-like signaling (IIS). It can suppress the transition from nurse to forager bees and therefore is elevated in nurse and overwintering bees. The regulation of JH and vg levels at the individual levels may thus trigger the colony-level transition from the field-season reproductive mode to winter survival mode. ........ 14
Figure 2-1. Larval survival during the 6-d development stage reared using artificial diet contaminated with four pesticides at the selected concentrations and a 1% solvent control. 1A: the cumulative mortality of honey bee larva through 6-d development continually exposed to 34 mg/L Chlorothalonil, 3 mg/L Fluvalinate, 8 mg/L Coumaphos, 1.5 mg/L Chlorpyrifos and 1% solvent; 1B: the conditional mortality for different development stages of bee larva. Asterisks denote significant difference from the respective solvent controls (Log-rank test, p < 0.0001). ................................... 42
Figure 2-2. Synergistic interactions for two pairs of pesticide mixtures: 8 mg/L Coumaphos, 34 mg/L Chlorothalonil and the mixture; 3 mg/L Fluvalinate, 34 mg/L Chlorothalonil and the mixture. 2A,2B: the respective Kaplan-Meier survival plots for honey bee larvae reared for each pair of pesticide mixture; 2C,2D: the interaction determination based on the deviation of observed mixture toxicity (black bar) from the expected additive toxicity (stacked bar). .................................................................... 45
Figure 2-3. Additive effects for three pairs of pesticide mixtures: 3 mg/L Fluvalinate, 1.5 mg/L Chlorpyrifos and the mixture; 8 mg/L Coumaphos, 1.5 mg/L Chlorpyrifos and the mixture; 8 mg/L Coumaphos, 3 mg/L Fluvalinate and the mixture. 3A,3B,3C: the respective Kaplan-Meier survival plots for honey bee larvae reared for each pair of pesticide mixture; 3D,3E,3F: the interaction determination based on the deviation of observed mixture toxicity (black bar) from the expected additive toxicity (stacked bar). .................................................................................................................................. 49
Figure 2-4. Antagonistic interactions for two pairs of pesticide mixtures: 0.3 mg/L Fluvalinate, 3.4 mg/L Chlorothalonil and the mixture; 3 mg/L Fluvalinate + 34 mg/L Chlorothalonil mixture, 8 mg/L Coumaphos and the three-component mixture. 4A,4B: the respective Kaplan-Meier survival plots for honey bee larvae reared for each pair of pesticide mixture; 4C,4D: the interaction determination based on the deviation of observed mixture toxicity (black bar) from the expected additive toxicity (stacked bar). ....................................................................................................... 51
Figure 2-5. The estimated time to cause 50% larval mortality by seven nominal concentrations of N-Methyl-2-pyrrolidone mixed in larval diet. ..................................... 53
ix
Figure 3-1. Kaplan-Meier survival plots for honey bee larvae reared for 6-days with eight concentration levels each of (A) Bravo and (B) chlorothalonil. ...................................... 71
Figure 3-2. The dose-response relationship of Bravo® (solid line) and chlorothalonil (dashed line) during 6-d larval feeding: Y-axis is the accumulative mortality at each time interval; X-axis is the formulation or AI concentration. .......................................... 74
Figure 3-3. EC0.01, EC0.1, EC1, and EC50 values calculated for each developmental day for in-vitro reared honey bee larva exposed to Bravo® and its active ingredient chlorothalonil. Panel A: Estimated larval LOEC of chlorothalonil based on calculated EC0.01, EC0.1, and EC1 values; Panel B: Estimated larval LOEC of Bravo®
based on calculated EC0.01, EC0.1, and EC1 values; Panel C: Comparison of time-dependent effective concentrations between Bravo® and chlorothalonil that caused 50% death of the honey bee larvae. .................................................................................. 76
Figure 3-4. The hazard ratio for honey bee larvae exposed to six concentrations of Bravo with respect to its corresponding active ingredient chlorothalonil. .................................. 77
Figure 3-5. Synergistic interaction for the 24 mg/L Bravo and 21 mg/L Nova larval diet mixture. Panel A shows the Kaplan-Meier survival plots for honey bee larvae reared on fungicide mixture and each component. Panel B shows the interaction determination for Bravo/Nova mixture based on the deviation of observed mixture toxicity (black bar) from the expected additive toxicity (stacked bar). ............................ 79
Figure 4-1. A stock-and-flow diagram of the most important components of our model. The diagram depicts the flow of materials (brood, adults, nectar, honey, and pollen) through our model: grey boxes indicate model variables that indicate one kind of material. Arrows indicate possible flows of material. Arrows with a plus sign or a minus sign indicates whether an increase in one variable causes an increase or a decrease of another variable. Grey circles indicate sources and sinks, through which material enters or leaves the model. Diamonds indicate the important social signal involving homeostatic regulations in nursing, pollen foraging and nectar foraging in a healthy hive. Black solid line indicates environmental factors involved in social regulation of the colony. Please note that we showed only the most important feedback loops in this diagram and that in the model the influence of one variable onto another variable often acts through several “intermediate” variables that are not shown in this diagram. ..................................................................................................... 93
Figure 4-2. A theoretical life history diagram for honey bees with a six stage structured life history. Circles represent stage-specific classes. The arrows between the stages are called transitions, indicating the probability P of transitioning from one class to the next (horizontal arrows) or of remaining in the same class (lower curved arrows). The curved arrows between Queen and Egg, labeled β, represent fecundity. .................. 94
Figure 4-3. Numerical simulation of population dynamics of honey bees at six stages, pollen stores (in cell), honey stores (in cell), and egg production rate of a simulated healthy honey bee colony throughout two years. ............................................................. 107
x
Figure 4-4. Model validation by comparing simulated healthy honey bee population dyanmcis with the historical colony data. Grey line shows intracolonial population dynamics of adult bees (a) and brood population (b) in our simulated honey bee colony throughout a year (maximum, minimum and normal population). Historical colony data from Bretschko (Bretschko, 1985), Bühlmann (Bühlmann, 1985), Fukuda (Fukuda, 1983), Omholt (Omholt, 1986), Bodenheimer (Bodenheimer, 1937), and Kunert and Crailsheim (Kunert and Crailsheim, 1988) are normalized to the maximum median value of the model to allow comparisons of the seasonal dynamics. .......................................................................................................................... 110
Figure 4-5. Model validation by comparing simulated healthy honey bee pollen dynamics with the historical pollen data. Green dashed line shows intracolonial pollen dynamics in our simulated honey bee colony throughout a year. Red solid line shows empirical pollen data from Jeffree and Allen (1956), allowing for comparisons of the seasonal dynamics. ........................................................................................................... 111
Figure 4-6. Model validation by comparing the simulated honey bee colony weight changes during the field season with the field data under controlled healthy and realistic pesticide exposures. Black dotted lines with different markers show the actual colony weight data (A-healthy colony, B-fungicide-exposed colony, C-dimethoate-exposed colony as positive control). The red line shows the simulated mean weight of honey bee colony during with the historical pollen data throughout the 2012 field season. ....................................................................................................... 113
Figure 4-7. Morris screening results for model outputs of colony population growth rate during the first, second and third years. The mean of distribution of absolute values of EEs, µ, estimates the overall effect of the parameter on a given output; while the standard deviation of EEs, σ, estimates the higher order characteristics of the parameter. The higher the value µ is, the more important factor is. The higher the value σ is, the stronger influence of the values of other input parameters have. Factors u, v, st4 (nurse bee survival rate), associated with the behavioral transition between nursing and foraging, are important for colony population growth. .................. 115
Figure 4-8. Local sensitivity analysis of colony dynamics under the potential disrupted scenarios: A. the reduced egg laying rate (field experimental results: 50% reduction per day when queen exposed to the fungicide-contaminated pollen diet); B. the reduced larval survival rate (larval rearing experiemntal results: least 50% reduction at the larval stage when exposed to the realistic exposure of a common fungicide chlorothalonil; C. the simulated combined effect of reduced egg laying rates (50% reduction) and larval survivorships (50% reduction); D. the 9% precocious foraging behavior. ........................................................................................................................... 119
Figure 4-9. The ratio of forager and brood population versus in-hive bees population in a healthy honey bee colony (maximum-dashed line, minimum-solid line) throughout the field seasons over three years. .................................................................................... 120
Figure 5-1. Two Group windows for the honey bee knowledge base given the data depicted in Table 5-1. This first window A shows the evaluation results of a honey bee colony health during the spring season and the second window B shows the
xi
level of TRUEness of this colony’s health in each of the various stressor categories. The color bar represents the risk evaluation results: the green means the argument of the dependency network is firmly true; however, the red means the goal is completely wrong. For this colony, it shows stable growth during the spring and there is no evidence that the colony is under any disturbances/impacts form all the stressors listed above. Yellow means the goal is under precautious condition. The left bottom section shows data that are not used for the colony evaluation during the spring. The right section shows all the data required for the spring evaluation. .............. 146
Figure 5-2. Three Group windows for the honey bee knowledge base given the data depicted in Table 5-2. The first window A shows the evaluation results of a honey bee colony health in October; the second window B shows the level of TRUEness of this colony’s health in each of the various stressor categories; the last window C shows the suggestions of colony management to help restore the colony health. The color bar represents the risk evaluation results: the green means the argument of the dependency network is firmly true; however, the red means the goal is completely wrong. Yellow means the goal is under a precautious condition. For this colony, there is strong evidence that this colony has been experiencing serious nursing and mite problems, resulting in the considerable disturbance in the colony survival and health. Therefore, appropriate management actions are needed to improve the nursing efficiency and control the Varroa mite infection in the hive. The left bottom section shows data that are not used for the colony evaluation during the fall. The right section shows all the data required for the fall evaluation. ...................................... 150
Figure 6-1. A dependency network for evaluating queen health as represented in NetWeaver™. In this dependency network, there are three items of data represented by the squares at the bottom of the figure: critical time of the year, the age of the queen, the ratio of egg laying rate of queen compared to that of healthy queen. Each of the data items is evaluated relative to the degree to which it satisfies its arguments. This network can be read as a rule as follows: “IF the time of year satisfies the specified argument. AND queen age does not satisfy the specified argument. OR Ratio of queen egg production compared to the optimal egg laying rate of a healthy queen (derived from the stage model) does not satisfy the argument. THEN the assertion of “queen is a concern” is true”. The degree to which the assertion is true is a function of the degree(s) to which the individual data satisfy their arguments and the types and arrangements of the logical nodes used within the network. ............................................................................................................................ 172
Figure 6-2. A dependency network for evaluating food quality in the hive as represented in NetWeaver™. In this dependency network, there are four items of data represented by the squares at the bottom of the figure: critical time of the year, the foraging frequency, the foraging time per day when outside temperature is suitable for foraging trips, and ratio of inadequate egg production compared to that of a healthy queen. Each of the data items is evaluated relative to the degree to which it satisfies its arguments. This network can be read as a rule as follows: “IF the time of year satisfies the specified argument, AND SOR node satisfies the three specified arguments related with foraging frequency, foraging time, and queen health. THEN the assertion of “food quality is a concern” is true”. The degree to which the assertion is true is a function of the degree(s) to which the individual data satisfy
xii
their arguments and the types and arrangements of the logical nodes used within the network. The SOR node is used to pick among these three competing evaluation methods where the methods are arranged in descending order of preference. Often the first choice method using foraging frequency of determining some property is not available, but another, less desirable method of foraging time during the day is available. The third method is using queen egg production, which is time-consuming evaluation. The SOR node provides a way of switching between these competing methods based on how well each method is currently functioning based on the sufficiency of the data driving the method. ...................................................................... 173
Figure 6-3. The first task module-testing the seasonal colony health. It examines the colony health at each season: Spring colony health is determined by the queen reproduction (dependent on the queen’s age, the hive temperature, and the queen egg laying rate compared with optimal rate) and food quality and quantity; Summer colony health is determined by the foraging frequency, pollen quantity; Fall colony health is determined by the nursing quality represented by brood vs. nurse ratio and hive temperature; Winter colony health is determined by the adult bees number, the frequency of Varroa Mite infection, and amount of honey stores. ................................... 174
Figure 6-4. The second task module-diagnosing the potential stressors in a disrupted hive. There are six candidate stressors that can all weaken the colony fitness. This task is designed to check whether the colony has any problems of queen, pollen stores, honey stores, food quality, nursing or Varroa mite infection. .......................................... 175
Figure 6-5. The third task module-predicting the survival of a disrupted colony. There are three major threats affecting colony survival: single threat caused by single stressors listed above; combinational threat caused by the synergistic or additive effects of multiple stressors; the overwinter collapse caused by insufficient adult population, or inadequate honey storage, mite infection and pollen stores before winter. This task is designed to evaluate whether the weakened colony is in an emergency situation and should be prioritized for restoration. ................................................................................ 176
Figure 6-6. The last task module-making suggestion of colony management for restoring the weakened colony. There are eight groups of actions: if the queen is a concern, then requeen the colony; if nursing is a concern, then improve hive temperature/ supply brood pheromone/ provide diverse fresh pollen to stimulate egg production and rearing, or combine colonies with strong ones before winter; if the pollen/honey quantity is a concern, then feed the colony with supplemented pollen/honey; if the mite is a concern, then treat the colony; if the colony has extra honey, then determine the amount of honey ready for harvesting. ...................................................... 177
Figure 7-1. Additive effects of the binary mixture of common fungicide formulations. Panel A shows the Kaplan-Meier survival plots for honey bee larvae reared on 24ppm Bravo/ 30ppm Pristine fungicide mixture and each component (ppm: mg/L). Panel B shows the Kaplan-Meier survival plots for honey bee larvae reared on 21ppm Nova/ 30ppm Pristine fungicide mixture and each component. .......................... 178
xiii
LIST OF TABLES
Table 2-1. Comparison between the predicted adult mortality rate (PM, %) for each tested concentration (Conc., mg/L) of four pesticides using a probabilistic toxicity model and the observed brood mortality rate (AOM, %) for bee larva from the 6-d in-vitro rearing experiments. ......................................................................................................... 40
Table 3-1. Fungicide formulations tested for honey bee larval toxicity. ................................. 66
Table 3-2. Results of Cox proportional hazards regression of survival in bee larvae exposed to different concentrations of Bravo® or chlorothalonil. .................................... 71
Table 3-3. Summary of the lowest observable effect concentrations (LOECs) and median effective concentrations (EC50s) for honey bee larvae exposed to Bravo® and its AI chlorothalonil (in mg/L), calculated from Finney's harmonic-mean formula. Larval LOEC or EC50 ratios for the active ingredient vs. formulation (adjusted to AI concentration in formulation) are presented for each time interval. Ratios marked by an asterisk indicate a significant difference in AI and formulation toxicity. ................... 74
Table 4-1. The model structure ................................................................................................ 94
Table 4-2. The summary of the definition and the baseline values of all parameters (Sakagami and Fukuda, 1968, Winston et al., 1981) in the honey bee stage-structured population model. ............................................................................................ 96
Table 4-3. The four ordinary differential equations model of nectar foraging dynamics (at second time scale) (Edwards and Myerscough, 2011). .................................................... 102
Table 4-4. The summary of variables and parameters in the differential equations of nectar dynamics (Edwards and Myerscough, 2011). ....................................................... 103
Table 4-5. Summary of simulation results for a healthy honey bee hive located at in the typical northern temperate region. ................................................................................... 108
Table 4-6. Sensitivity indices for individual and social parameters on honey bee population and their corresponding ranking obtained with EFAST. ............................... 115
Table 4-7. Sensitivity indices for individual and social parameters on pollen dynamics and their corresponding ranking obtained with EFAST. .................................................. 116
Table 5-1. The summary of simple data values (user-input initial data), calculated data through arguments and functions, and the level of Trueness for each dependency networks in the honey bee colony knowledgebase. ......................................................... 144
Table 5-2. The summary of simple data values (user-input initial data), calcuated data through arguments and functions, and the level of Trueness for each dependency networks in the honey bee colony knowledgebase. ......................................................... 147
Table 6-1. The simple data requirements (user input) identified by the domain expert panel as sufficient for characterization of honey bee colony health ................................ 169
xiv
Table 6-2. The summary of calculated data (based on the stage model calculation) sufficient for characterization of honey bee colony health .............................................. 170
xv
ACKNOWLEDGEMENTS
Working on the Ph.D. especially on interdisciplinary research has been a wonderful and often overwhelming experience. It is hard to say whether it has been grappling with the topic itself which has been the real learning experience, or grappling with how to write papers and proposals, give talks, work in a group, learn not to be easily discouraged by reviewers, stay up until the birds start singing, start early before the bees start working, and stay focus after getting multiple painful stings…
In any case, I am indebted to many people who have generously helped me throughout years and make the time working on my Ph.D. an unforgettable experience.
First and foremost, I would sincerely thank all my graduate committee members, James L. Frazier, Michael C. Saunders, Christopher A. Mullin, Timothy Reluga, Ottar Bjonstad for their advice, guidance, and encouragement during my graduate program at Penn State. I am deeply grateful to my advisor Professor James Frazier. Your kindness, understanding, encouragement, and patience made this thesis possible. You have been a steady influence throughout my Ph.D. career; you have oriented and supported me with patience and care, have always been understanding and encouraging in times of new ideas and difficulties. The brainstorming discussion with you frequently led to key insights. Your ability to select and to approach compelling research problems, your appreciation towards the interdisciplinary research, and your courage when faced with difficulties, all set an example for me. I feel extremely lucky to have a caring advisor like you and I appreciate all your help from my heart.
I want to thank Professor Mike Saunders for inspiring me to study operations research, taking the time to help developing the honey bee expert system with NetWeaverTM software, and giving many insightful comments in my work. His enthusiasms and inspirations immensely magnified the excitement science brought me.
I want to express my sincere gratitude to Professor Chris Mullin. I want to thank him for constructive comments in my pesticide studies due to his broad knowledge of toxicology, for his great patience in helping write papers, for many motivating discussions in my research, and for the prompt sharing of his extensive understanding of toxicology and latest bee research. His high scientific standards and hard work inspired me to develop a right research ethic.
Furthermore, I have been very privileged to get to know and to collaborate with Professor Timothy Reluga. I want to thank him for teaching me so much in developing the honey bee model and devoting tremendous time in helping tackle many technical problems throughout the model development. This population model would not have been possible without him. Many long discussions with him have significantly improved my work and inspired many new research directions. I also owe my thanks to his patience for pushing me forward for the modeling work after my long and intense field experiment periods.
I would also pay my tributes to the extensionist Maryann Frazier. She provided invaluable assistance with the colony field experiments, and generously shared her extensive understanding of honey bee systems and provided precious suggestions from the colony management perspective. Her technical excellence, tremendous field experiences and great enthusiasm of bees had a great impact on me.
I want to sincerely thank Bruce Miller, who is the co-developer of NetWeaverTM software and helped me with promptness and care. His technical excellence has significantly improved the work on the development of bee expert system.
This thesis would not have been possible without the encouragement and help from past and present lab members. Over the years, Daniel Schmehl has been a faithful friend and helped me with many personal challenges. He taught me so much in our joint research on larval rearing.
xvi
The larval toxicity study would not have happened without him. His positive and caring attitudes toward the research and life inspired me to find a balance in life. I am also indebted to Sara Ashcraft, who I had the pleasure to work with. She has been an invaluable support in the lab and field. Her tremendous grasp of experimental issues and positive attitude towards life were a valuable source of motivation in both research and life. I have greatly enjoyed the opportunity to work with Lauren Rusert, Stephanie Mellott, Sara Ziegler, Samuel Gruneberg, and Tim Ciarlo. I am so lucky and blessed to meet these intellectual and caring people, who have provided me a stimulating and fun environment in which to learn and grow over these years.
I would also pay my tributes to many excellent faculty instructors in Department of Entomology, to name a few, Consuelo De Moraes, Diana Cox-Foster, Gary Felton, Kechung Kim, Kelly Hoover, Mark Mescher, Michael Saunders, Shelby Fleischer, James Frazier, Chris Mullin and Tom Baker. Their dedication to the courses and professional development guidance has shaped my understanding of entomology.
As a student majoring in operations research, I owe my thanks to those great professors, Terry Harrison from Department of Business, Jose Ventura, M. Jeya Chandra and Yao Tao from Department of Industrial Engineering, John Fricks and Steven Arnold from Department of Statistics, Christopher B. Byrne from Department of Mathematics, and many more, for their efforts on teaching and exposing me to the exciting field of operations research.
Special thanks go to my friend Chen Shi, whose understanding of entomology and mathematics is tremendous and whose scientific work inspired me a lot. From the very beginning of my application to Penn State, he has been consistently supportive and helpful. My work has greatly benefited from his suggestions and kind encouragement. I would thank many fellow students at and outside Penn State for the discussion on many research ideas and projects as well.
I would like to thank Penn State entomology staff, Ellen Johnson, LuAnn Weatherholtz, Roxie Smith, and Karen Dreibelbis, for their kindness and help over all five years. I want to thank Professor Gary Felton for offering me the fellowship opportunity to pursue a Ph.D. degree at Penn State. I am very grateful for my research and salary support from Penn State Department of Entomology for all these years. This research would not have been possible without the financial support of the two funded grants from National Honey Board and Penn State Center for Pollinator Research.
Deep gratitude goes to my best friends, Ye Wang, Zheng Liu, HuiFang Yu, Na Yi, Changcheng Liu, who have been supporting and helping me selflessly for over 10 years and make me feel that I am never far away from home. Thank you for sharing these years with me!
Last but not least, I want to thank my parents Shengwen Zhu and Xiuzhen Wang, and my grandparents Fengming Wang and Wenying Chang for raising me to appreciate the beauty of nature and the beauty of humanity. Their unconditional love and courage in the dark days teach me how to live with gratitude! Finally, I am grateful and blessed for having known my husband, Zhengwen Pu. Life may be like a roller coaster with its joys and sorrow, but thank you for always standing by me with hope, strength, belief and peace!
Chapter 1
Introduction
In a healthy honey bee colony, individuals act as cooperative vehicles for colony growth
and development and their actions often rely on a network of self-organizing behaviors via a
series of highly elastic feedbacks to achieve stability and optimal fitness. This system is driven by
the initial population size, queen fertility, demographic structure of workers and social
organization (Seeley, 1995). Over the past several years, honey bee (Apis mellifera L.)
populations worldwide have experienced a large decline with no clear causative agent, also
known as colony collapse disorder (CCD) or honey bee depopulation syndrome (HBDS)
(National Academy of Sciences, 2007, VanEngelsdorp et al., 2009). There is currently no
straightforward answer for the mysterious honey bee die-off, increased overwintering losses, or
declines in honey bee health in general. The combinational energy-draining stresses of illness
from environment, nutrition, and human migratory and cultural practices strike honey bee
populations day after day, depriving them of long-term health.
Our dependence on honey bees in agriculture has brought a collective effort from
researchers to discover the root cause of this decline; however, the possibility of a multifactorial
cause is one of the problems that make investigating the cause of recent colony declines
especially complex. Current lines of investigation include diseases (e.g. Israeli Acute Paralysis
Virus, nosema infection) (Cox-Foster et al., 2007), environmental factors (e.g. global warming,
monoculture agriculture) (Potts et al., 2010), and pesticides (in-hive miticides and agricultural
pesticides) (Johnson et al., 2010).
Pesticides are a major concern due to their widespread distribution within the hive. Given
the current findings of widespread occurrence of pesticides and multiple residues within the hive
2
including brood nest wax, foundation, beebread, trapped pollen, adult bees and brood (Mullin et
al., 2010), the adverse effects of pesticides on honey bee colony health can potentially happen in
all aspects of colony life. Prior to adult emergence, eggs and developing bees may be exposed to
pesticide residues through contaminated wax cell walls or food sources provided by contaminated
nurse bees. Younger adult bees in the hive may be exposed to incoming contaminated pollen and
nectar, as well as beekeeper-applied pesticides frequently used in the hive to control common
pests and parasites such as Varroa mites, tracheal mites, small hive beetles, and wax moth. Older
foragers may be directly exposed to agricultural pesticides during flight and foraging on common
blooming crops, such as oilseed rape (Brassica napus L.), maize (Zea mays L.), or sunflower
(Helianthus annuus L.) that are now routinely treated against insect pests (Mullin et al., 2010).
Queen bees can be exposed to pesticides by contact with contaminated bees, wax, and in-hive
food stores, as well as royal jelly food fed by contaminated bees. Studies have demonstrated that
a prolonged persistence of two commonly used in-hive miticides fluvalinate and coumaphos
could negatively impact the queen reproductive capability, delay brood development and adult
emergence, reduce brood survival, shorten adult longevity, and disrupt proper performance of
emerged adult bees (Haarmann et al., 2002, Desneux et al., 2007, Johnson et al., 2010, Mullin et
al., 2010, Wu et al., 2011). Beside the lethal effects, sublethal impacts of pesticides on forager
bees have also been reported both in the laboratory and field studies including abnormal foraging
activity, reduced olfactory memory and learning performance, impaired orientation and
navigational skills, and disrupted homing flights (Yang et al., 2008, Henry et al., 2012, Schneider
et al., 2012). However, most studies currently focus on studying pesticide effects at the individual
bee level rather than at colony level (Desneux et al., 2007), largely because that significant
challenges remain in assessing pesticide impacts on the ultimate colony survival and development
including: 1) considerable uncertainty in the translation of the loss of individual bees into effects
at the entire colony level, which is complicated by various hive functions performed by different
3
bees, the flexibility of the colony to compensate for the reduced functions served by these bees,
and the dynamics of pesticide circulation in the hive; 2) substantial differences in pesticide
exposures in terms of the magnitude, timing and duration, frequency and medium of exposure
among and within different castes of bees; 3) limited understanding of the interaction of
pesticides with other stressors in the hive and the dynamics of the colony while reacting to abiotic
and biotic disturbances (Desneux et al., 2007, Johnson et al., 2010, Thompson, 2010).
To address these multiple open questions on honey bee colony losses, the development of
a practical honey bee risk assessment approach is imperative. Moreover, while generally
acknowledging the associations among potential environmental and anthropological stressors
with CCD, the mechanism(s) by which a colony either succumbs to or overcomes individual or
combinational effects of anthropogenic and environmental threats remains largely unstudied.
Such questions cannot usually be answered by reductionist experimental approaches alone, but
require the help of simulation models; we thus proposed to take an integrative mathematical
modeling and experimental research approach to investigate whether inadequate nutrition in both
quality (pesticide contamination) and quantity (inadequate food) level can progressively
destabilize one or more components of this system, eventually altering its dynamics. This
research project consists of two parts: initially, we performed laboratory and field experiments to
examine how commonly encountered pesticides individually and in combination, at the
environmentally realistic levels, will impact queen fertility and fecundity, as well as the
demographic structure of workers including brood survival and adult longevity; subsequently, a
honey bee intracolonial population dynamics model was designed to translate these individual-
level effects into the colony fitness by integrating exposure and effects data of potential stressors
with the complexities of the social structure and biology of a honey bee colony. Ultimately, the
proposed modeling study is designed to answer the following questions:
§ What are the stable population dynamics of a honey bee colony population?
4
§ Can the system continue proper functioning if negative stressors from nutrition and/or
pesticides singly or simultaneously take a toll on some stages of the system?
§ Can this destabilized system retrieve its stability by supplemental nutrition?
We think that before beekeeper friendly “tools” can be developed to monitor and use
nutritional supplements to maintain healthy colonies, one must understand how multiple factors
interact at the colony level. Therefore, we submit that a modeling approach is a necessary first
step yielding results that will allow informed nutritional changes to be made. The honey bee
population model represents a powerful tool to: describe the intracolonial population dynamics of
honey bees, quantify the colony-level effects of nutritional disturbance at the quality level
(pesticide residues in diet), and predict the colony response to supplemental nutrition (pollen
supplement). These results will help researchers and beekeepers to decide: (1) the critical stages
of the honey bee life history in determining population fitness; (2) how to design validating
experiments and colony management techniques, especially protein supplements, towards the
most important demographic parameters for honey bee conservation. This modeling approach not
only offers direction to experimental and management efforts, but also saves time and resources
overall.
As a step forward, we proposed to develop a decision support system (DSS) model (also
called knowledge-based system model, expert system) (Shu-Hsien, 2005) based on the
quantitative mathematical model of the honey bee colony, as a new and innovative method of
transferring timely, up-to-date management decisions to farmers, growers and beekeepers. The
complex and large amount of technical information in the mathematical model presents a difficult
hurdle for adequately communicating model outputs to beekeepers, growers and other
stakeholders. Additionally, extension meetings and scientific publications are a slow route of
education. With the NetWeaverTM knowledge base development tool (Saunders et al., 2005), our
DSS model is designed to: 1) initially, assist beekeepers as a diagnostic tool to determine the
5
condition of colony health; 2) secondly, help them to identify stressors impacting colony health;
3) thirdly, assist them as a prediction tool to determine the potential of colony collapse; 4) further,
provide them with cost-effective decision support needed to manage bees, help
disrupted/weakened colonies recover, and honey to be harvested; 5) develop criteria for
determining the success of restoration management 6) evaluate the capability of our existing
model in predicting colony dynamics; and 7) be used as a valuable educational tool to assist the
traditional educational methodologies.
The significant portion of our ecosystem depends heavily on the presence and service of
healthy honey bees. This service can be maintained only if we understand the inner life and
functions of this “superorganism” so well that we are able to support and protect them in a
focused and sustainable way. We hope this avenue of research, integrating experimental,
mathematical modeling and knowledge-based modeling approaches, can help researchers,
beekeepers and other stakeholders to understand how the healthy honey bee system behaves, what
contributes to its success, how to quantify the effects of potential disturbance on colony fitness,
and what management plans should be taken to restore the colony health.
Literature Review
Background
Honey bees (order: Hymenoptera) are oviparous, holometabolous insects that live in large
colonies usually containing one queen and her progeny, some 20,000–40,000 female workers and
200–300 male drones (Winston, 1987). Honey bees have a haplodiploid sex determination
system: females, queens and workers arise from fertilized (diploid) eggs laid by the queen -
diploid eggs become queens or workers depending on which cell they are laid in and whether the
6
resulting larvae are fed royal jelly or worker jelly by the bees performing brood care; In contrast,
males arise from haploid unfertilized eggs (Wilson, 1975). Male and female larvae undergo a
series of larval stages followed by pupation and a full metamorphosis within a cell in the
honeycomb, which go through egg, larva, pupa until adult stage. The length of the egg stage (3
days) is the same for all three castes, but the larval and pupal stages (also called brood) are
shortest for the queen (13 days: larval-6 days, pupa-7days), followed by worker (18 days: larva-7
days, pupa-11days) and longest for the drone (21 days: larval-9 days, pupa-12 days) (Winston,
1987).
Honey bees are obligate social bees, which represent a large and successful branch of the
insect’s family tree: more than 20,000 species have been described worldwide (Seeley, 1995,
Camazine et al., 2001). Although most bees are solitary or subsocial, the family Apidae contains
three distinct groups that exhibit eusocial behavior: stingless bees, bumble bees, and honey bees
(genus: Apis). Eusociality is rare, but highly successful. Honey bee is one of the best-known
social insects that exhibit an advanced level of eusociality, which is the highest level of social
organization in a hierarchical classification and characterized by communal nest, cooperative care
of juveniles (individuals care for brood that is not their own), reproductive division of labor (not
all individuals get to reproduce), and overlap of generations (Wilson, 1975). Species diversity
amongst the genus Apis is remarkably small, although the question of the number of Apis species
is still debated among taxonomists. Currently, there are nine species known based on where and
how they nest: the giant open-nesting honey bees Apis dorsata and Apis laboriosa; the dwarf,
single-combed honey bees Apis florae and Apis andreniformis; and the cavity-nesting honey bees
Apis cerana, Apis koschevnikovi, Apis nuluensis, Apis nigrocincta, and Apis mellifera (Winston,
1987)..These highly adaptive species thrive wonderfully in environmental extremes such as
deserts, rain forests and tundra for hundreds of millions of years. The success of a honey bee
colony relies on two principal attributes: a plastic division of labor among the worker bees, which
7
contribute to their inclusive fitness by performing all tasks necessary for colony growth and
development; and a single reproductive queen bee, the “life-central” of the entire colony, which is
primarily specialized for laying eggs to ensure the quantity and survival of a colony, as well as
producing queen pheromones to aid in the social organization (Seeley, 1995).
Most people only know the European/Western/African honey bees Apis mellifera, the
agricultural darling. It is the most commonly managed bee in the world and the third most
important domestic animal in Europe and the USA, after cattle and pigs and before poultry (Tautz
et al., 2008). As a highly adaptable species, Apis mellifera has a native range that stretched from
the southern Scandinavia to Central Asia and throughout Africa (Seeley, 1985). Before 2600
BCE, ancient Egyptians were the first from record to domesticate A. mellifera. The practice of
beekeeping and honey bees were eventually spread around the world by European beekeepers.
Since the 1600s, as the result of deliberate human transport for their ability to produce honey and
bees wax, A. mellifera’s range has expanded to nearly all-habitable corners of the globe
(Sheppard and Meixner, 2003).
Honey bees are important to both society and the environment in various ways. Apis
mellifera is the third most important domestic animal in the US and Europe, after cattle and pig
and before poultry (Tautz et al., 2008). They are enormously important for pollination, mainly
due to their fascinating ability to exploit food sources rapidly by recruiting masses of foragers
(Tautz et al., 2008). From the ecological perspective, honey bees provide pollination, which
enable the transfer of pollen from one plant to the stigma of another plant, thereby enhancing the
density and diversity of crops and wild plants (Morse and Calderone, 2000, Aizen and Harder,
2009). From the economic view, the direct or indirect contribution made by managed honey bees
hired by U.S. crop growers to pollinate crops amounted to just over $18.9 billion. If including the
global cash-crop yields and apiculture markets (honey production, wax, propolis, pollen and royal
jelly), the more recent estimates range up to $217 billion (Bauer and Sue Wing, 2010).
8
This literature review explores the issues surrounding the survival and development of a
honey bee colony. It is divided into four sections: factors contributing to the success of honey bee
colony life; potential stressors associated with Colony Collapse Disorder; stage-specific impacts
of pesticide contamination on honey bees; and an overview of the existing honey bee population
models. A summary of research rationales and questions are discussed in the end.
Honey bee colony as a superorganism
The organization in a honey bee colony is characterized by the cooperation of tens of
thousands of bees that are contiguously active and respond to the conditions of their
surroundings, towards the fulfillment of various functions crucial for the welfare of the colony.
Two organizational principles are the key to the success of the honey bee colony.
Coordination
One key feature is group-level coordination and colony integration through self-
organization and local decisions, facilitated by flexible communication within and among distinct
groups through cues and chemical signals. The coordination among honey bees follows adaptive
patterns to a homeostatic life, involving three crucial features: comb pattern formation consisting
of three distinct regions--a compact, centrally-located brood area, a surrounding rim of pollen,
and a large peripheral region of honey (Camazine and Sneyd, 1991); brood maintenance
involving collaborations between nurse bees, the honey bee queen and the available pollen stores
(Seeley, 1995, Crailsheim and Hrassnigg, 1998); and the collective decision making in foraging
behavior-for instance, 1) the nectar foraging process involves interactions between the nectar
foragers and food receiver bees to collaboratively direct the colony commitment to nectar
9
foraging (Fewell and Winston, 1996, Edwards and Myerscough, 2011) ; 2) the homeostatic
regulation of the pollen foraging process by balancing the pollen need in the hive and foraging
workforce (Schmickl and Crailsheim, 2004, Sagili and Pankiw, 2007).
Nectar foraging is a particularly well-studied task of the honey bee colony (Camazine and
Sneyd, 1991, Dyer, 2002). The nectar foraging is considered as a collective decisive-making
process, which needs the interaction of both in-hive bees (or nectar receivers who process and
store the nectar within the hive) and the nectar foragers (who are knowledgeable about and
working in the external environment). The recruitment of an active nectar forager population is
determined primarily by three factors: the available/experienced forager bees in the hive ready for
nectar foraging; the quality of the nectar outside the hive, which is signaled by the waggle dance
(positive information sent by foragers) and shaking signal (foragers dorso-ventral abdominal
vibration), tremble dance (negative information by returned foragers) and stop signal (audible
piping sound) (Anderson and Ratnieks, 1999); and the colony nutritional’s status, which is
reliably cued by the time/ difficulty each returned forager spent searching for an available nectar
receiver in the hive for unloading (Seeley, 1989). When nectar is abundant in the foraging
environment, a healthy colony can adjust its selectivity among different nectar sources in relation
to its rate of nectar intake and its current nectar/carbohydrate need: when a colony's rate of nectar
intake and the need of nectar are high, its thresholds for accepting and recruiting to a nectar
source are higher than when its intake rate and need is low (Seeley, 1989, Fewell and Winston,
1996). Therefore, in the face of a fluctuating and unpredictable nectar resource in the
environment, a healthy honey bee colony can adapt with changes via a combinational system of
signals and cues to fine-tune the worker allocation system and balance the work capacities of
nectar foragers and receivers, thereby exploiting the resources in a timely and efficient way
(Anderson and Ratnieks, 1999, (Seeley, 1995).
10
The allocation of the foraging force, especially pollen foraging, is critical in regulating
colony growth and development. Several hypothetical mechanisms have been proposed for pollen
foraging regulation including: the stimulus-response threshold hypothesis involving the quantities
of brood, pollen stores, bees’ age, and resource reward levels, which converge on the sucrose
threshold levels of individual forager. Given the fact that returning pollen foragers have direct
access to brood and stored pollen, the pollen stores and brood can therefore directly and
independently decide the tendency of individual foragers to perform certain tasks related to
foraging-foragers with a medium sucrose threshold will tend to forage for pollen and those with a
high threshold will tend to forage for nectar (Ricarda et al., 2004, Oldroyd and Thompson, 2006);
and the brood food hypothesis, which is the indirect regulation of brood and stored pollen on
pollen foraging either through a single inhibitory signal distributed through trophallaxis
interactions between nurse and forager bees, or positive social pheromones secreted by brood or
queen bee to stimulate pollen foraging (Sagili and Pankiw, 2007). Both the direct and indirect
hypotheses predict a similar outcome; that is the pollen foraging is regulated through stored
pollen and brood population towards a homeostatic set point.
Differentiation
The other key principle is the specialization and differentiation of worker bees for
flexible divisions of labor, which are co-determined by temporal polyethism (Winston, 1987,
Beshers et al., 2001), genotypic variability (Robinson, 1992, Smith et al., 2008) and the social
context (Ament et al., 2010, Amdam, 2011, Sagili et al., 2011). The temporal castes are thought
to be relevant to the physiology-dependent tasks, for which a physiological specialization such as
the activation and deactivation of glands is required for efficient performance of the task and
makes individual workers temporarily non-interchangeable (Seeley, 1995). This age-related
11
process can be illustrated by the behavioral maturation of worker bees from in-hive nursing
behavior for the first 2-3 weeks of the adult life to the foraging behavior outside the hive for the
final 1-2 weeks. These two task groups define the most distinct and highly stable states in a bee’s
life (Winston, 1987). The nurse bees have enlarged jelly-synthesizing hypopharyngeal glands,
hypertrophied abdominal fat bodies and a high number of circulating hemocytes (immune cells).
However, the foragers are characterized by atrophy and apoptosis of the hypopharyngeal glands
(Silva de Moraes and Bowen, 2000, Jeroen and Johan, 2005) and the abdominal fat (Toth and
Robinson, 2005, Chan et al., 2011). Since the forager bees are not efficient to digest the stored or
collected pollen due to their lack of digestive endopeptidases, the intake of dietary protein and
lipid of a forager is primarily dependent on feeding by nurse bees in proportion to her foraging
activity level (Crailsheim, 1992).
In addition to the age-dependent and physiological maturation, the plastic differentiation
in a healthy honey bee colony has been proposed to emerge through social signals, like
pheromones (queen/brood pheromones), brain hormones (dopamine, octopamine) and conserved
nutrient-related signals, produced by workers themselves, to fine-tune the mechanisms regulating
divisions of labor and to promote social homeostasis (Ament et al., 2010, Amdam, 2011). At the
physiological level, these signals modulate a positive regulatory feedback loop, which is mediated
by the mutually inhibitory interaction between juvenile hormone (JH) and the vitellogenin (vg)
genes (Corona et al., 2007). In worker bees, vitellogenin, an egg yolk protein, is normally
synthesized and released from the trophocyte cells of fat body, where it can be a general nutrient
signal, and it has pleiotropic effects on multiple physiological processes, such as suppressing life-
history transition, prolonging adult longevity by reducing oxidative stress, and stimulating
foraging specialization in pollen (Toth and Robinson, 2005). JH is synthesized by the corpora
allata glands behind the insect’s brain, and is central to the regulation of behavioral development
and adult longevity as well as stress responses (Sullivan et al., 2000, Sullivan et al., 2003). During
12
honey bee behavioral transitions, these endocrine signals can feed back to the regulatory system
to further inhibit the vitellogenin synthesis (Pinto et al., 2000), shift the gene expression pattern of
the hypopharyngeal glands (Ohashi et al., 1997), and suppress immunity (Amdam et al., 2004,
Amdam et al., 2005). Therefore, this cross-tissue feedback loop is proposed to serve as a central
timing mechanism in regulating behavioral differentiation and transition in the colony (Toth and
Robinson, 2005).
Furthermore, advances in sociogenomics have recently led to major progress in decoding
the genetic and molecular bases of social organization. A hierarchically structured regulatory
network integrating genomic, physiological, behavioral components, and social context have been
hypothesized to explain the complex phenotypes and its considerable flexibility in the honey bee
colony (Johnson, 2010). The major components of this regulatory network are the nutrition-
sensing pathways, including IRS (insulin receptor substrate), TOR (target of rapamycin) and
Egfr (epidermal growth factor receptor-mediate the reaction to royalactin, which is the key
protein component of royal jelly), as well as the reproductive-related protein vitellogenin. These
nutrient cascades act through the PI3K/Akt signaling pathway, which has been a major focus of
attention since its initial discovery as a proto-oncogene, because of its critical regulatory role in
diverse cellular processes and developmental pathways in many model organisms, including
cancer progression, apoptosis regulation and insulin metabolism(Patel et al., 2007, Kamakura,
2011, Mutti et al., 2011, Wolschin et al., 2011). Additionally, PTEN, which is a tumor/growth
suppressor gene and high in forager brain, is recently found partly linked to the nutrition-sensing
pathway via its antagonistic role in the PI3K/Akt signaling pathway by disrupting insulin/insulin-
like signaling (IIS) and Egfr signal transduction. These cross-tissue components from brain, fat
body, to ovary tissue can act interactively or independently to converge onto one central
downstream signal integrator juvenile hormone (JH) to regulate the caste differentiation (Mutti et
al., 2011). A recent epigenetic study showed substantial and reversible DNA methylation changes
13
corresponding to the reversible behavioral phenotypes of nurse and forager subcastes in honey
bees (Herb et al., 2012). This is the first study suggesting the subcaste-species DNA methylation
signature can assist in forming worker phenotypes.
In addition to the flexible feedback system at the individual bee level, the complex
chemical communication system is a fundamental component of social life, which maintains the
integrity and function of a colony. The nutrient stimuli, which flow from the nurse bees to the
queen and brood through direct feeding, and to the forager bees through trophallaxis interactions,
is suggested as the crucial social factor in regulating social organization, due to its role in
transmitting chemical messengers through the colony (Ament et al., 2010, Amdam, 2011). The
intricate systems of chemical messengers in the social context are pheromones including: Queen
pheromone (QP), brood pheromone (BP), and (E)-ß-ocimene (EO). QP is a blend of chemicals
produced by multiple glands of queen bee. It can elicit both primer (elicit long time behavioral
responses) and releaser response (elicit immediate behavioral responses). They act both as: 1) sex
pheromone - maintain the queen reproductive status in the hive, inhibit worker ovary
development and rearing of replacement queen, and attract drones during mating flights, and 2)
social regulator- influence comb construction, affect worker brain gene expression, inhibit
workers’ juvenile hormone synthesis, delay nurse-to-forager transition, coordinate and stabilize
swarming (Dyer, 2002, Maisonnasse et al., 2010). Brood pheromone, a 10-component mixture of
fatty acid methyl and ethyl esters produced by older brood, can act on the nurse bees’
neurosensory system, thereby encouraging the brood-rearing behavior of nurse bees, delaying the
nurse-to-forager transition and stimulating foraging preference for pollen (Camazine and Sneyd,
1991, Sagili et al., 2011, Sagili and Breece, 2012). (E)-ß-ocimene, another brood pheromone
component produced by young brood, can inhibit worker ovaries and accelerate workers’
behavioral maturation (Maisonnasse et al., 2010, Edwards and Myerscough, 2011) (Figure 1-1).
14
Figure 1-1. A conceptual diagram of the dynamic networks of individual-level nutrient sensing pathway and social-level pheromone communication pathway and its role in regulating division of labor and maintaining the colony homeostasis. Nutrient-sensing pathways in the social and physical development of worker bees all converge on the major social pathway at the individual bee level, which involves the antagonistic interaction of juvenile hormone (JH) and vitellogenin (vg): JH level is positively correlated with foraging behavior and therefore enhances the food dynamics and egg production; vg level in the ovary tissue is a proposed negative regulator of insulin/insulin-like signaling (IIS). It can suppress the transition from nurse to forager bees and therefore is elevated in nurse and overwintering bees. The regulation of JH and vg levels at the individual levels may thus trigger the colony-level transition from the field-season reproductive mode to winter survival mode.
15
Perturbations in either pathway are sufficient to alter the caste development (Edwards and
Myerscough, 2011); for example, experimental colonies treated with JH, JH mimic, or JH analog
(Robinson, 1987), or insulin treatment (Mott and Breed, 2012), or artificial suppression of
vitellogenin (Nelson et al., 2007) or PTEN (Mutti et al., 2011), or primer pheromone treatment
(Leoncini et al., 2004), were found to modulate the normal rise in juvenile hormone and result in
either precocious behavioral development (marked by the significant acceleration of foraging
onset by 2 weeks earlier) or retarded development (the continuation of brood care despite
increasing chronological age).
The complete picture of the regulatory interactions between genomic, cellular,
physiological, and behavioral factors involved in the development and evolution of social
phenotypes have not yet been established; however, the current understanding of neurochemical
pathways suggests the importance of nutrition in triggering behavioral and physiological changes
in the honey bees, regulating the reproductive division of labor, maintaining the plasticity in task
allocation, ultimately contributing to the overall colony growth and development.
Colony Collapse Disorder
Though the remarkably coordinated and manifold activities of honey bee societies have
long been subject to curiosity and awe (Fewell and Winston, 1996); the widespread colony
disorder has received unprecedented attention recently. Sustained honey bee declines have been
documented worldwide with no clear causative agent, also known as colony collapse disorder
(CCD) or honey bee depopulation syndrome (HBDS). Some beekeepers in the United States
reporting CCD have lost 50-90% of their colonies, often within weeks. Despite these high losses,
the average number of colony losses has been ~30% since CCD was first reported in mid-
16
November 2006, by a Pennsylvania beekeeper overwintering in Florida (VanEngelsdorp et al.,
2009, VanEngelsdorp and Meixner, 2010).
Generally, the symptoms of CCD-caused collapsed colonies are defined as follows: 1)
complete absence of adult bees in colonies, with no or little build-up of dead bees in the colonies
or at the hive entrances, 2) the presence of capped brood, and 3) the presence of food stores in
hives (both honey and bee bread) that are not immediately robbed by other bees or typical colony
pests (small hive beetles, wax moths, etc.). Colonies with CCD can appear healthy just weeks
prior to collapse, but still showing early signs of collapsing: 1) an insufficient workforce to
maintain the brood in the colony, 2) the workforce is composed largely of younger adult bees, 3)
the queen is present, appears healthy and is usually still laying eggs, 4) the cluster is reluctant to
consume supplement food provided by beekeepers, and 5) foraging populations are greatly
reduced/non-existent (VanEngelsdorp et al., 2009, VanEngelsdorp and Meixner, 2010).
Colony Collapse Disorder may not be a new disorder. In fact, symptoms similar to CCD
have been described over the past 50-60 years, and been labeled many different names including
spring dwindle disease, fall dwindle disease, autumn collapse, May disease, and disappearing
disease (VanEngelsdorp and Meixner, 2010). Whether or not these historic colony losses share a
common cause with modern-day CCD or if any new factors are involved is not yet clear.
Regardless, this higher-than-normal colony declines translates into a significant loss in viable
pollinating or honey producing services provided by millions of honey bees each year (National
Academy of Sciences, 2007, Bauer and Sue Wing, 2010). This elicits a range of human responses
for caring about bee health and welfare including every private stakeholder and public concerned
about conservation. The benefit contributed by bees goes beyond the mere dollars and cents for
agricultural production worldwide to the point that bees are determiners of the diversity and
quality of human everyday diets. However, the number of managed beehives has been shown to
not keep pace with the global demand for the pollination services (Aizen and Harder, 2009). Our
17
dependence on honey bees in agriculture has brought a collective effort from researchers to
discover the root cause(s) of this decline. Current lines of investigations on the causes of CCD
include environmental and anthropogenic stressors (National Academy of Sciences, 2007,
VanEngelsdorp et al., 2009): 1) traditional bee pests and diseases (including American foulbrood,
European foulbrood, chalkbrood, nosema, small hive beetles, and tracheal mites), though these
traditional maladies have not been documented to promote CCD-like symptoms, there is potential
that these diseases may exacerbate the disorder; 2) colony management practices (such as
splitting hives, swarm control, chemical use, migratory beekeeping); 3) agricultural practices-
genetically modified crops, habitat loss and fragmentation, and climate change; 4) queen source
lacking of genetic diversity and vitality, potentially making bees increasingly susceptible to other
stressors; 5) chemical residue/contamination in the hive matrices (wax, food stores, bees);
different broad-spectrum pesticides applied in agricultural systems or used inside the hive for
controlling bee pests could have both lethal and sub-lethal effects on the survival, longevity,
behavior and cellular physiology of non-target bees; 6) Varroa mites and associated pathogen
load in the bees and brood; Varroa mite itself is damaging to bees, transmits viruses, and elicits
chemical uses from beekeepers; 7) novel pests and diseases such as Nosema ceranae and Israeli
Acute Paralysis Virus; 8) malnutrition weakening the bees’ immune system; and 9) synergisms
between the above stressors. Despite comprehensive research efforts on colony losses, no single
driver in the above list has emerged as the definitive cause of the destructive phenomenon.
Instead, interactions between multiple drivers are the most probable explanation for elevated
mortality in honey bee colonies. It has been proposed that these multiple drivers may initiate the
infection cascade by suppressing the immune defense of individual bees, directly/ indirectly
causing the significant energetic stress and causing positive chilling cascade, subsequently
leading to the substantial starvation cascade by forcing premature foraging and thereby thwarting
18
the nursing behavior, ultimately contributing to the irreversible demise of collapsing colonies
(Potts et al., 2010, VanEngelsdorp et al., 2010).
Pesticide Impacts
Pesticides are a major concern due to their widespread distribution within the hive.
Recently, one hundred and twenty one different pesticides and pesticide metabolites were
identified in the hive with an average of seven pesticides per pollen sample, including miticides,
insecticides, fungicides, herbicides, and insect growth regulators (Mullin et al., 2010).
Furthermore, one of the most striking features of the Apis mellifera genome is the significant
deficit of detoxification genes, which would then explain the species’ unusual sensitivity to a
wide range of pesticides (Claudianos et al., 2006). Therefore, the extensive occurrence of
pesticides and multiple residues detected within the hive matrices such as brood nest wax,
foundation, beebread, trapped pollen, adult bees and brood (Mullin et al., 2010) has raised
alarming concerns over honey bee health and behavior. Not only may these single or
combinational pesticides cause direct toxicity to honey bees, but can also cause chronic
consequences by disrupting the task performance and development of honey bees including queen
bee, brood and adult bees, ultimately decreasing the colony fitness. For instance, since growers of
many bee-pollinated crops routinely apply fungicides during bloom, an accumulation of
fungicides in beebread occurs by bees foraging either directly by picking up pollen-sized particle
formulations or through their presence in contaminated pollen, nectar, or water. Aside from the
direct toxicity to honey bees by fungicide-sprayed blossoms, fungicides in stored pollen are found
to inhibit the growth of beneficial fungi that are necessary to convert pollen into beebread, thus
reducing the nutritional value of the pollen to bees (DeGrandi-Hoffman et al., 2009). Given the
critical role of the hive matrices in queen, brood and adult bee health, the potential impacts of
19
pesticides on colony survival and development have elicited the following in-depth
investigations.
Queen Bee Health
Development of the honey bee colony is closely tied with the queen and her reproductive
physiology (Winston, 1987). The potential circulation of pesticides in the colony through nurse
bees’ tending has raised concerns over adverse effects of pesticides on queen health. For instance,
several studies demonstrated that a prolonged persistence of two commonly used in-hive
miticides fluvalinate and coumaphos could negatively impact the queen survival, individual
weight, ovary size, egg laying capacity, sperm numbers in the spermatheca, sperm viability, the
developmental rate of queen honey bees, and increase the occurrence of queen rejection (Wallner,
1999, Haarmann et al., 2002, Pettis et al., 2004, Collins and Pettis, 2012). Whether the significant
impacts of pesticides on queen reproductive physiology could result in reduced queen
performance, thereby affecting colony dynamics, is currently unclear.
Brood Bee Health
Since pesticide contaminated pollen is ubiquitous with an average of 7 pesticides (range
of 2-31) in a pollen sample (Mullin et al., 2010), chronic larval exposure to in-hive pesticides
through contaminated larval diet or direct contact may have detrimental impacts on the survival
and development of brood and their ability to properly develop to adults. Honey bee larvae are
particularly sensitive to pesticides for a number of reasons including: 1) the contaminated pollen
food flowing from the digestion and secretion of nurse bees to the larvae could contain persistent
pesticides or metabolites, which could be more toxic than the original chemicals; 2) the larval
bees retain within their cell the contaminated wax comb and bee bread as well as metabolic
waste, both transdermally, orally and internally, throughout the larval stage up to the pupal molt
after which they defecate (Desneux et al., 2007); and 3) pesticide disruption of the beneficial
20
mycofloral community in the colony may thwart the processing of pollen into beebread and allow
undesirable pathogens to thrive, therefore impacting the brood development (DeGrandi-Hoffman
et al., 2009). Several studies have demonstrated that insecticides ranging from insect growth
regulators and encapsulated organophosphate formulations to systemic insecticides have lethal
and sublethal impacts on larval growth and development. For instance, the exposure of larval bees
to fenoxycarb (IGR, insect growth regulator) showed malformation at larval and pupal stages,
indirectly suppressing the adult emergence (Tasei, 2001). When exposed to non-IGR substances,
in particular systemic insecticides, encapsulated insecticide formulations, or antibiotics
(chlortetracycline), the bee brood stage showed increased mortality and disrupted development
such as retarded development, precocious pigmentation and an elevated level of apoptosis in the
midgut (Atkins and Kellum, 1986, Davis et al., 1988, Peng et al., 1992, Aupinel et al., 2007,
Gregorc and Ellis, 2011). Our recent work shows that bees are readily exposed to numerous
pesticides in their pollen food as larvae and this may cause both neural and hormonal changes
that may alter development (Frazier et al., 2008, Zhu et al., 2013). Conceivably, these impacts
on early life stages of honey bees due to inadequate nutrition and/or direct poisoning could lead to
weakening of the colony structure over time. However, to date, few studies have examined the
effects of pesticides at environmentally realistic levels on larval honey bee development and
how this exposure may translate into a decreased fitness once reaching adulthood, with a
corresponding disrupted colony age demography.
Adult Bee Health
The dependence of worker bee health, and division of labor on their protein diet has been
repeatedly demonstrated (Toth et al., 2005, Rueppell et al., 2007). In addition to the direct lethal
toxicity of pesticides to adult bees, pesticides can induce the sublethal effects, mainly affecting
physiology and behavior of honey bees (Desneux et al., 2007). The negative impacts at the
21
physiological level include disrupting: the important biochemistry and enzymatic processes such
as the Na +/K+ ATPase and acetylcholinesterase (AChE) activities, which may relate to many
hidden damages in the colony such as disease outbreak and disturbance in foraging behavior
(Bendahou et al., 1999); the neurophysiological-related cytochrome oxidase activity in the honey
bee brain, which is important to the olfactory learning and memory (Decourtye et al., 2004, Yang
et al., 2012); reduced longevity (Schmuck, 2004, Wu et al., 2011); and immune system
functioning (Desneux et al., 2007, Pettis et al., 2012). The behavioral changes include: impairing
foraging behaviors such as mobility, orientation capacity, communicative capacity and learning
performance of forager bees (Desneux et al., 2007, Thompson, 2010); disrupting nursing
behaviors due to the suppressed development of adult bees’ hypopharyngeal glands thereby
resulting in undernourished larvae, (Johnson et al., 2010); interfering with the task-related
development of worker bees (Thompson, 2003, Wu et al., 2011) - for instance, treating bees with
juvenile hormone analogues can result in a shift in worker bee activity from the brood nest to
food handling and early foraging (Robinson, 1987, Wu et al., 2011).
However, most of these impacts were observed in laboratory studies, which might
potentially underestimate or overestimate the level of impact on colonies in the field. Honey bees
in the field are doubtless routinely exposed to other stressors, such as disease, starvation, or the
dietary presence of diverse agrochemicals (Mullin et al., 2010); therefore, pesticide impacts are
likely to be exacerbated by the occurrence of other stressors that were not considered in the
laboratory.
Research Rationale
The combinational energy-draining stresses of illness from environment, nutrition, and
human migratory and cultural practices strike honey bee populations day after day, depriving
22
them of long-term health (National Academy of Sciences, 2007, VanEngelsdorp et al., 2009,
Johnson et al., 2010). The possibility of a multi-factorial cause is one of the problems that make
investigating CCD especially complex. While generally acknowledging the associations among
these possible factors with CCD, the mechanism(s) by which a colony either succumbs to or
overcomes individual or combinational effects of anthropogenic and environmental threats
remains largely unstudied. These research gaps are largely because of the limited understanding
of honey bee colony population dynamics while reacting to disturbances. For instance, pesticide
impacts discussed in the previous section involve all aspects of colony life; however, how to
translate these impacts seen at the individual bee level and controlled laboratory studies into the
reduced colony-level fitness in the actual changing environment are still open questions. Such
questions cannot usually be answered by reductionistic experimental approaches alone, but
require the help of mathematical models. Therefore, this research is aiming to take an integrative
experimental and modeling approach to explore the processes leading to colony failure, with the
focus on the pesticide impacts. We hope the integrative approach could further our
understanding of honey bee population dynamics, make predictions of the resulting colony
disturbance in response to environmental and anthropogenic stressors, and ultimately aid in
colony management.
Current Honey bee Population Models
Several honey bee population models have been developed with different foci using
various mathematical modeling approaches. Attempts to model the colony dynamics were begun
by Ebert (Ebert, 1922), who first divided the colony into 3 subgroups (juveniles, hive bees and
foragers) rather than considering the entire population as a single entity. Based on the improved
understanding of the consecutive division of labor in a social bee colony, several models,
23
including Nolan (Nolan, 1932), Bodenheimer (Bodenheimer, 1937, Bodenheimer and Ben-Nerya,
1937), and Fukuda (Fukuda, 1971), were developed subsequently to refine Ebert’s model;
however, these models were largely generated by directly measuring honey bee populations with
experiments and field observations.
The formal mathematical modeling approach was first implemented by McLellan
(McLellan et al., 1980) and Harris (Harris, 1985), using linear regression models and was able to
generate some realistic predictions. These models required the actual natality and mortality rates
generated from the colony it was trying to estimate. Dolejsky and Schley (Dolejsky and Schley,
1980) were the first to study the populations of bees and Varroa mites using a simple differential
equations model with time-dependent parameters. This modeling approach allows for observing
the colony-level influence of a simulated reduction of mites.
Omholt (Omholt, 1986) built the first dynamic model incorporating a feedback
mechanism with regard to the queen’s egg-laying rate. The model considered a worker-density
related ‘switching term’ in determining whether the influence on queen egg production from the
workers is inhibitory or stimulative. Due to the paucity of quantitative experimental data and the
computational cost, this model did not consider the extra-colonial factors such as external
temperature and field conditions. The more complex intracolonial factors such as cannibalism,
pheromone-based interactions, and division of labor have also been disregarded in this model.
DeGrandi-Hoffman et al. developed the BeePop model, which concentrates on the queen’s egg
laying and the derived age-based population structure (DeGrandi-Hoffman et al., 1989). BeePop
is the first model designed for researchers and beekeepers to conduct “what if” scenarios on the
intracolonial dynamics by providing for user-specified parameter values. They extended
Omholt’s model on the queen’s egg laying by introducing the impacts of extracolonial factors
including the daily temperature and the photo-period; however, most of these models simulated
the division of tasks purely based on the fixed age of the bees. They ignored the influence of
24
external factors (weather, foraging resource) and inside nutrition-related feedbacks in regulating
the flexibility of task allocations in the bee colony.
The application of individual-based modeling promoted a new level of understanding of
honey bee population dynamics. The object-based model “AHBsim” developed by Makela
(Makela et al., 1993) incorporated the external (resource availability and other physical factors in
the environment) and internal genetic factors (the genotype of the queens and drones) in
estimating the birth, growth, reproduction and death processes of a colony (Makela et al., 1993).
They also were the first to introduce the priority-based system in colony task allocation, with the
assumption that there are no age requirements for job performance in a honey bee colony. The
recent “HoPoMo” model leads to an outstanding improvement in modeling the detailed aspects of
colony dynamics. It is the first in the literature that incorporate the important feedback loops that
link pollen supply and nursing workforce to brood cannibalism, and to division of labor
(Schmickl and Crailsheim, 2007); however, this priority-based division of labor does not fully
reflect the colony organization, particularly with regard to nectar foraging behavior. It assumes
two tasks of high priority: brood nursing and pollen foraging, expressed by assigning specific
ratios of workforce (the available adult bee population) to workload (the least number of adult
bees required by nursing and pollen foraging). Both these agent-based models consider that nectar
foraging has the lowest priority.
A key feature of social organization in social insects involves behavioral plasticity. It is
the result of the integration of acquired internal and external information, coupled with behavioral
biases associated with worker genotype, temporal-specific castes, physiological status, and prior
experience (Johnson, 2010). Previous honey bee models have not incorporated any informational
pathways that the colony task allocations are based on, partially because the mechanisms of
behavioral flexibility of honey bees has not been clearly defined until recently (Mutti et al.,
2011). Currently all honey bee population models concentrate on queen egg production, fixed
25
temporal polyethism or the dynamic of high-priority-based task divisions between brood care and
pollen foraging. There is no actual demographic or mechanistic model yet to explain CCD from
the perspective of nutritional deficiency at the quality and quantity levels.
Research Objectives
Most of today's honey bee colony research could be called 'negative biology'. It is
conducted in an intellectual framework that presumes that the most important question to answer
is: what causes colony disorder? Disease is its central focus and is mainly concerned with trying
to understand, prevent and treat specific diseases.
Rather than making pathology and disease the central focus of intellectual efforts,
positive biology seeks to understand positive phenotypes: why do some honey bee colonies live
more than 2 or 3 years without ever suffering from the chronic or acute collapse that afflict most
colonies today? Which biological mechanisms would explain these exemplars of health and
super-organized honey bee colonies? Observations and analyses of the healthy colonies with
superior structure and exceptional longevity from the historic data and controlled field data
therefore present fascinating puzzles for positive biology.
The key to explain the exceptional organization in honey bee society is nutrition-sensing
pathways, which direct complex phenotypes and the considerable flexibility to maintain colony
homeostasis, and promote growth and development in the honey bee colony (Johnson, 2010).
Therefore, the overall goal of this research was to explore: what mechanisms contribute
substantially to the stable population dynamics of a healthy honey bee colony; and can the system
continue proper functioning if negative stressors from nutrition and/or pesticides singly or
simultaneously take a toll on some stages of the system? The following section describes the
hypotheses and objectives utilized by this research to examine how the characteristics and
26
mechanisms of the population and resource dynamics in a healthy honey bee colony are disrupted
by pesticide-contaminated nutrition.
Objective 1: Determine if and how pesticide exposure will impact the colony fitness through its
effect on the demographic structure of workers.
Research hypothesis: If the demographic parameters of workers (survival rate, growth rate,
transition rate) under pesticide exposures are decreased, then the honey bee population will be in
decline as well.
• Measure the impacts of individual and combined pesticides at environmentally
relevant levels on honey bee brood development, adult bee longevity
• Expand these changes in vital rates at the individual worker stage to the colony-
level effects by integrating our experimental data of the brood survival, workers’
longevity and the timing of the transition to foragers under pesticide stressors to
the stage-based simulation analysis of honey bee population model
Objective 2: Determine if and how pesticide exposure will impact the colony fitness through its
effect on queen fertility and fecundity.
Research hypothesis: If the queen egg-laying rate under pesticide exposures is decreased, then the
colony will be in decline as well.
• Measure the difference of the queen egg-laying rate with and without individual
and combined pesticide exposures at environmentally realistic levels
• Extrapolate the changes in the Queen’s egg laying rate into the stage-based
population model
• Assess how this local change affects the population survival and growth rate.
Objective 3. Determine if and how nutrition deficiency will impact the colony fitness through its
effect on task organization in the colony.
27
Research hypothesis: If task organization in the colony mediated by the nutrient signaling is
unsustainable, the colony fails to maintain a viable population and collapses.
We assume that social regulation is achieved by the brood maintenance behavior
specifically the balance between nurse bees (food provider) and brood (food consumer). We will
investigate the long-term behavior of the honey bee system using the mix of numerical simulation
and sensitivity analysis.
Objective 4. Determine if and how the colony fitness could benefit from nutritional supplements.
Research hypothesis: If the vital rates (worker-age-specific survival rate and growth rate) of each
stage or the task organization are changed by altered nutritional supply at different times of the
year and to different extents, the stage-based simulation analysis of honey bee populations will
show different rates of achieving stability.
We will simulate how each compartment would benefit from additional nutrition by
changing the vital rates (stage-specific survival rate and growth rate) or the task organization
(ratio of brood, nurse bees, forager bees distributed in the colony) by a biologically reasonable
amount in response to pollen or potential nutrient supplements, and extrapolate the individual
responses to the annual population growth rates. Since fungicides are largely metabolic inhibitors
that exhibit synergistic toxicity with many pesticides, we think that added nutrition can likely
mitigate some or potentially all of these effects. The proposed simulation modeling will
determine how much change is needed on specific life stages to actually translate into increased
colony growth dynamics. With this knowledge specific nutritional experiments can be undertaken
with specific goals, rather than just seeking “improvements” as is currently done.
This combinational modeling and experimental approach will help us explore which
factor(s) contributes most to the colony homeostasis, assess how pesticides affect honey bee
colony dynamics, evaluate the potential effects of supplementary nutrition on colony fitness,
ultimately achieving the proposed research objectives and providing suggestion to management
28
actions that are likely (or unlikely) to promote recovery of honey bee population under pesticide
or other stressors.
Chapter 2
Pesticide Impacts on Honey Bee Larval Health
Recently, the widespread distribution of pesticides detected in the hive has raised serious
concerns about pesticide exposure to European honey bee (Apis mellifera) health. A larval rearing
method was adapted to assess the chronic oral toxicity to honey bee larvae of the four most
common pesticides detected in pollen and wax - fluvalinate, coumaphos, chlorothalonil, and
chloropyrifos - tested alone and in all combinations. All pesticides at hive-residue levels triggered
a significant increase in larval mortality compared to untreated larvae by over two fold, with a
strong increase after 3 days of exposure. Among these four pesticides, honey bee larvae were
most sensitive to chlorothalonil compared to adults. Synergistic toxicity was observed in the
binary mixture of chlorothalonil with fluvalinate at the concentrations of 34 mg/L and 3 mg/L,
respectively; whereas, when diluted by 10 fold, the interaction switched to antagonism.
Chlorothalonil at 34 mg/L was also found to synergize the miticide coumaphos at 8 mg/L. The
addition of coumaphos significantly reduced the toxicity of the fluvalinate and chlorothalonil
mixture, the only significant effect in all tested ternary mixtures. We also tested the common
‘inert’ ingredient N-methyl-2-pyrrolidone at seven nominal concentrations, and documented its
high toxicity to larval bees. We have shown that chronic dietary exposure to fungicides, pesticide
mixtures, and a formulation ingredient has the potential to impact honey bee populations, and
warrants further investigation.
.
30
Introduction
Recently, one hundred and twenty one different pesticides and metabolites were
identified in the hive with an average of seven pesticides per pollen sample, including miticides,
insecticides, fungicides, herbicides, and insect growth regulators (Johnson et al., 2010, Mullin et
al., 2010). Feeding on pollen and nectar in the larval diet directly exposes honey bee larvae
transdermally, orally and internally (Desneux et al., 2007); therefore, the potential for chronic
toxicity and synergistic interactions at the brood stage seems likely to occur, especially
considering the fact that early life stages might be much more sensitive to certain contaminants
relative to the adult stage. Several studies have demonstrated that insecticides ranging from insect
growth regulators and encapsulated organophosphate formulations to systemic insecticides are
more toxic to larvae than to adult bees (Atkins and Kellum, 1986, Davis, 1989, Tasei, 2001,
Heylen et al., 2011). Moreover, because beebread serves as an absolute requirement for
developing bee larvae, pesticide disruption of the beneficial mycofloral community in the colony
may thwart the processing of pollen into beebread and allow undesirable pathogens to thrive,
therefore impacting the brood health (Babendreier et al., 2007, DeGrandi-Hoffman et al., 2009).
Indeed, chronic exposure to pesticides during the early life stage of honey bees may thus
contribute to inadequate nutrition and/or direct poisoning with a resulting impact on the survival
and development of bee brood (Becher et al., 2010). Conceivably, these impacts on the larval
phase could lead to weakening of the colony structure over time. To date, only a few peer-
reviewed pesticide toxicity studies assess the risks of oral toxicity of pesticides to honey bee
larvae at environmentally relevant levels (Johnson et al., 2010, Mullin et al., 2010). Therefore, a
goal of our study was to assess the chronic and mixture effects of common pesticides at realistic
exposure concentrations on larval honey bee development. In order to mimic realistic exposure
scenarios of honey bee larva to contaminated pollen food, we chose the four most frequently
31
detected pesticides in the hive - fluvalinate, coumaphos, chlorothalonil, and chlorpyrifos, and
tested them alone and in all combinations via chronic dietary exposure, at concentrations found in
the hive.
The pyrethroid tau-fluvalinate and organophosphate coumaphos have been used widely
for Varroa mite control, and found highly persistent in the hive with an estimated half-life in
beeswax of about 5 years. These compounds have shown evidence of synergistic toxicity on adult
honey bees at the level of cytochrome P450-mediated detoxification (Johnson et al., 2009).
Chlorothalonil, a broad-spectrum agricultural fungicide with an unclear mode of action (Caux et
al., 1996), is often applied to crops in bloom even when honey bees are present for pollination,
because it is currently deemed safe to bees. However, some fungicides have shown direct toxicity
to honey bees or solitary bees at field use rates (Ladurner et al., 2005) and fungicides in stored
pollen are known to inhibit the growth of beneficial fungi thereby reducing the nutritional value
of the pollen to bees (DeGrandi-Hoffman et al., 2009). Chlorpyrifos is a widely employed
organophosphate in crop management (Donovan, 2006) and its residues were frequently found in
honey, propolis and dead bees. These in-hive (beekeeper applied) varroacides and out-of-hive
(farmer applied) insecticides and fungicides may act alone or in concert, in ways currently
unknown, to create a toxic environment for honey bee growth and development.
Another goal of this study was to examine the effect of an ‘inert’ ingredient on brood
survival. Little data exists concerning the toxicity of ‘inert’ ingredients on honey bees, likely
because bee toxicity information for pesticide formulations is not currently required by the U.S.
Environmental Protection Agency as part of the pesticide registration process in contrast to the
European Union where toxicity for representative formulations is mandatory (European
Commission, 2009). Pesticide risk assessment is largely stymied by lack of public access to
product-specific information of ‘inerts’ or co-formulants (Cox and Surgan, 2006). Some ‘inert’
ingredients such as those in formulations of the herbicide glyphosate are more toxic than active
32
ingredients when tested on aquatic organisms (Kudsk and Mathiassen, 2004). That ‘inert’ more
than active ingredients dominate pesticide formulations and spray tank adjuvants so to increase
efficacy and stability of the pesticide makes it important to examine the role of ‘inerts’ on honey
bee toxicity. Here, we studied the chronic toxicity of N-methyl-2-pyrrolidone (NMP, CAS 872-
50-4) to bee brood development. The co-solvent NMP is used extensively in chemical processing
and agricultural chemical formulations (Health and Safety Executive, 1997). The NMP tested
alone or in formulations has demonstrated developmental toxicity in rats by various routes of
administration (Saillenfait et al., 2007) and also has shown high toxicity potential for aquatic
invertebrates (Lan et al., 2004). There is presently no information in the published literature
regarding toxic effects of NMP to honey bees. Our study will be the first to test if this common
‘inert’ ingredient is toxic to honey bee larva by continuous dietary exposure, and will serve as a
foundation for future studies exploring ‘inert’ toxicity. Specific objectives of the present study
using the standardized in vitro larval feeding method developed by Aupinel et al. (Aupinel et al.,
2007) are to: (i) assess possible toxic effects of single pesticides on the survival of individual A.
mellifera larvae during a 6-d continuous feeding with contaminated diet; (ii) compare the
sensitivity difference between larval and adult bees to the same pesticide exposure; (iii) determine
whether the selected pesticides in all combinations at realistic concentrations have any synergistic
effects; and (iv) examine the toxicity of environmentally realistic levels of the formulation
ingredient NMP on larval survival. Measurable impacts on larvae should demonstrate the need to
extend pesticide risk assessment for honey bees from primarily acute effects on adults to chronic
impacts on brood survival and development, and of the need to consider both active and ‘inert’
ingredients in formulations, so that more informed decisions can be made by governments,
beekeepers and growers about pesticide application inside and outside the hive.
33
Materials and Methods
Acquisition of 1st instar larvae
Honey bee (Apis mellifera L.) 1st instar larvae were collected from colonies of A. ligustica
strain reared in our experimental apiary (GPS Coordinates: 40° 49' 20"N, 77° 51' 33"W). In order
to collect newly emerged larvae, a honey bee queen was confined in the queen excluder cage and
placed in the 2nd super from the bottom of the hive and positioned in the center of the super to
allow for proper incubation of the newly laid eggs. After being caged for 30 h, the queen was
released from the cage and eggs were incubated in the hive for 3.5 days. Frames of newly-hatched
1stinstar larvae were taken to the laboratory in a pre-warmed chamber (~35°C).
Diet preparation
Honey bee diet was prepared using 50% royal jelly (Beenatura.com), 12% D-glucose
(Fischer Chemical, Fair Lawn, NJ, USA), 12% D-fructose (Fischer Chemical, Fair Lawn, NJ,
USA), 2% yeast extract (BactoTM, Sparks, MD, USA), and distilled water (24%). Royal jelly was
preserved at -80° C until use. Ingredients minus royal jelly were completely dissolved and filtered
through a 0.2 µm membrane (Corning) to remove particulate matter and bacteria. This solution
was poured onto royal jelly that was free of wax particles, and mixed thoroughly at room
temperature using a spatula. Diet was stored at 4° C for a maximum of three day s prior to use.
Pesticide application
The concentrations of applied pesticides were selected based on our previous laboratory
findings of commonly found pesticides in pollen (Johnson et al., 2010, Mullin et al., 2010).
34
According to the survey of pesticide residues conducted on bee-related product samples from
migratory and other beekeepers during the 2007–08 growing seasons, the most prevalent
detections at 95th percentile values (levels at which only 5% of detections are higher) in trapped
pollen samples were 0.3 mg/L fluvalinate, 0.8 mg/L coumaphos, 0.15 mg/L chlorpyrifos, and 3.4
mg/L chlorothalonil (unpublished data). Foraging bees may avoid and dilute contaminated pollen
with that from alternative hosts; therefore, the level of contamination found in the trapped pollen
pellets varies in relation to the foraging environment of the colony (Winston, 1987, Johnson et al.,
2010, Mullin et al., 2010). We have observed that apple pollen contributes approximately 10% of
overall trapped pollen samples from hives placed in apple orchards during a 10-d pollination
event (unpublished data). In addition, these pesticides have also been detected in other hive
products at even higher levels including beebread, wax comb, foundation, and more rarely in
bees. Developing bees are exposed to pesticide residues by contact with the wax, bee bread and
contaminated bees, so the level found in trapped pollen or royal jelly is not fully representative of
actual exposure of larval bees to pesticides. For example, pollen residues of fluvalinate and
coumaphos primarily originate by transfer from the contaminated comb wax, which contains
much higher levels (e.g. 100-times) of these miticide residues (Johnson et al., 2010, Mullin et al.,
2010). Therefore, in the absence of exact measures of pollen residues in larval foods, we chose to
test at 10 times the levels of these four pesticides found in pollen samples. We mixed fluvalinate
(purity, 95%), coumaphos (purity, 99%), chlorpyrifos (purity, 99%), and chlorothalonil (purity,
98%) purchased from Chem Service (West Chester, PA, USA) in the larval diet at nominal
concentrations of 3, 8, 1.5, and 34 mg/L, respectively.
Pesticide treatments included four pesticides tested alone and in two, three, and four-
component mixtures. To prepare stock solutions, each technical grade pesticide was individually
dissolved in acetone and methanol, respectively. Each test solution was mixed thoroughly into the
artificial diet at specific concentrations and stored in 2 mL sterile glass vials (Corning, USA). We
35
monitored three control groups in the study: untreated diet, one solvent-treated diet containing 1%
methanol and another solvent control containing 1% acetone. We also tested the toxicity of N-
methyl-2-pyrrolidone at various environmentally realistic concentrations on larval survival. NMP
is usually used in cosmetic and pesticide formulations with an application concentration limit of
5% (Saillenfait et al., 2007); therefore, we tested seven nominal concentrations of NMP including
0.01%, 0.02% , 0.05% , 0.1% , 0.2% , 0.5% and 1%.
Each experiment was repeated twice including control (3 groups), single (6 treatment
groups), mixture (binary mixtures: 6 treatment groups; ternary mixtures: 6 treatment groups; four-
component mixtures: 2 treatment groups), and ‘inert’ toxicity tests (seven concentrations of
NMP). Sample size for each treatment starting from the same experimental day is 3 replicates
with 24 larvae per replicate.
In vitro larval rearing technique
Newly hatched 1st instar larvae were transferred from hive frames into sterile, 48-well
culture plates (Corning, USA) for the in vitro rearing technique with 24 larvae per plate. Larval
transfers were done in the lab without the use of a sterile hood. The sterile, push-in queen cups
(B&B Honey Farm, USA) were placed in every other well. Diet was warmed to ~34° C in a
heating block prior to larval transfer. Using an Eppendorf 10-100 µl variable volume pipette, 10
µl of each diet treatment was placed per queen cup. A 00 camel hair paintbrush was used to
transfer each larva from the cell on the frame to the cup. The paintbrush was dipped into distilled
water between each larva to aid in a smooth transfer, and was sanitized by dipping in 95% ethanol
after every four to five transferred larvae. Larvae were placed directly on top of the diet and
inspected for mobility to ensure a quality transfer. Four additional queen cups were equally
spaced in four of the remaining open wells before placing the lid on the culture plate, allowing for
36
adequate ventilation of the larvae throughout the experiment. Each plate was placed in a humidity
chamber and kept at 95% relative humidity with a 10% aqueous solution of sulfuric acid being
used at the base of the chamber to maintain humidity. Humidity chambers were placed in an
incubator at 34° C in the dark and were not disturbed throughout the experiment, except when
replacing the diet for ~15 min/d.
For this study, only the survivorship of honey bees during the larval stage was monitored
to evaluate the impacts of selected pesticides. Larval mortality was recorded daily by probing the
larvae with sanitized forceps. The dead larvae were removed daily. Diet for each larval bee was
replaced daily. Old diet was removed using a glass disposable pipette and new diet was
immediately placed in each queen cup according to the following schedule to account for larval
growth: day 1- 10 µl, day 2- 10 µl, day 3- 20 µl, day 4- 30 µl, day 5- 40 µl, and day 6- 50 µl.
Kaplan-Meier survival analysis
The 6-d larval survival data were segregated by pesticide treatment and analyzed using
Kaplan-Meier survival analysis (Kalbfleisch and Prentice, 1980). This estimate generally assumes
independence among the individual death events and randomization within the treatment group.
The hazard rate h(t) is the conditional probability of failure or death in a small time period given
that the subject has survived up until a specified time t. The greater the value of the hazard rate,
the greater the probability of impending failure. The null hypothesis of no difference between
survival curves of treatment and control groups was tested by the Log-rank test that weights each
death by the square root of the total number of individuals at risk per time interval, placing less
emphasis on deaths occurring later in the experiment. All the survival analyses were implemented
in SAS survival program (SAS/STAT® 9.2 User’s Guide).
37
Comparison between adult and larval sensitivity
The difference in sensitivity to the same pesticide between adult bee and larva can be
quantitatively evaluated by comparing the actual larval mortality per day from the in vitro test
with the predicted mortality for adult bees if exposed to the same concentrations of pesticides.
The larval mortality data was corrected with Abbott’s formula beforehand. Here, the impacts of
pesticide treatments on adult bees were estimated from the adult acute topical LD50 data
converted to whole-bee LC50 values, because neither the chronic nor acute oral toxicity data of
adult bees is currently available for all pesticides selected for this study. Predicted adult toxicity
can be estimated as a function of the magnitude of toxicant exposure and the individual's
sensitivity to a toxicant, which is generally characterized by the probit model (Atkins et al.,
1981). The predicted proportion of insects killed (p), in probit transformed units, calculated as
where a = intercept and b = slope from the regression of the transformed data and x is
the log-transformed concentration or time. Results of probit analyses are reported typically as a
concentration or time required to kill a certain proportion of the test insects (e.g., LC50). Table 2-1
shows the average LC50 values from the literature (Johnson et al., 2010, Mullin et al., 2010) and
probit slopes from other sources (Atkins et al., 1981, DeGrandi-Hoffman et al., 2009). One
exception is chlorothalonil, which is estimated using the default probit slope of 4.5 because its
mortality levels under topical or oral applications to honey bees are found to be insufficient to
establish a dose-response relationship. Therefore, the probit function for each pesticide to adult
honey bees can be inferred from the LC50 values (x), probit mortality ( =5) and probit slope (b)
(Atkins et al., 1981). Then, the probit model can be extrapolated to predict the probability of an
impact of each pesticide on adult bee survival for a specified concentration. Using the Probit
program in SAS 9.2 (SAS/STAT® 9.2 User’s Guide), the predicted probit-type mortality can be
transformed to the original percent units and compared with the actual larval percent mortality
p̂ = a + bx,
p̂
38
data. Using the compilation of acute data from different sources may complicate the accurate
estimation of the adult toxicity because of the heterogeneity introduced by differences among the
studies; however, given the limitations we felt this was a reasonable approach to obtain a first
approximation of the differences in adult and larval sensitivity to the same pesticide exposure.
Pesticide interaction determination
We used significant departures from additive toxicity to define antagonistic and synergistic
interactions between pesticides in mixtures (Hertzberg and MacDonell, 2002). The expected
additive toxicity for the chemical mixture is the sum of each chemical’s toxicity to larval survival,
calculated as , where n is the number of chemical components in the pesticide
mixture and hi is the hazard rate for a specific component estimated from the laboratory bioassay
data. The sum of the responses (Ehn) to the individual components is estimated based on the
assumption that the selected pesticide mixtures are the combination of substances with
independent modes of action or similar modes of action. The mixture toxicity can be predicted as
follows: Additive interactions-- Simultaneous action of components in which the observed
response of honey bee larva to a mixture (hn) is equal to the sum of the responses (Ehn) to the
individual components; Synergistic interactions--Simultaneous action of components in which hn
is significantly higher than Ehn; Antagonistic interactions--Simultaneous action of components in
which hn is significantly less than Ehn.
We did not test different concentrations of each pesticide component and of the
combinations to fit dose-response curves. Neither food intake nor concentrations of pesticides
consumed by each larva were measured during the oral feeding. Therefore, this method does not
allow exact quantification of the level of interaction but makes only an initial qualitative
assessment of synergism or antagonism.
Ehn = hi ni
n∑
39
Results
No significant differences in larval mortality were observed when larvae were reared on
untreated artificial diet or diet mixed with 1% methanol or 1% acetone (Log-rank test, p > 0.05)
(data not shown). These three control groups showed an accumulative 6-d percent mortality of
approximately 17.2% (Figure 2-1), which is within the normal range observed for control
mortality using the in-vitro larval rearing protocol (Aupinel et al., 2007). Because control
mortality exceeds 10%, the larval mortality data from treatment groups were corrected with
Abbott’s formula.
Single pesticide toxicity
Chronic exposure of bee larvae to each of the four pesticides at tested concentrations
showed significant toxic effects to larval survival (Log-rank test, p < 0.0001), resulting in an
overall 2- to 4-fold reduction in the total 6-d percentage survival compared to the control
mortality (Figure 2-1A). Based on age-specific toxicity data, mortality rates for each pesticide
were uneven across different larval stages (Figure 2-1B). For 1-day-old larva, 8 mg/L coumaphos
and 3 mg/L fluvalinate were more toxic than the other two pesticides. The 2 and 3-day-old larvae
showed similar sensitivity to different pesticide exposures, approximately 10% mortality per day.
The 4 and 5-day-old larvae were most sensitive to 1.5 mg/L chlorpyrifos, causing more than 32%
larval death each day (Table 2-1). A dramatic increase in larval mortality for 6–day-old larvae
was observed in 34 mg/L chlorothalonil and the two miticide groups, ranging from 53.7% to
68.8%. Using the probit model, notable differences were found in pesticide sensitivity between
the adult bee and larvae (Table 2-1). Among the four pesticides tested, 1.5 mg/L chlorpyrifos was
the only treatment that adult bees were more susceptible to than the larvae. For the other
40
pesticides, the larvae showed increased sensitivity over that of adult bees. Notably, chlorothalonil
at the sublethal concentration of 34 mg/L was least toxic to adult bees, however most toxic to
larvae followed by 8 mg/L coumaphos and 3 mg/L fluvalinate. On average, coumaphos was the
least toxic to larval bees among the four pesticides.
Table 2-1. Comparison between the predicted adult mortality rate (PM, %) for each tested concentration (Conc., mg/L) of four pesticides using a probabilistic toxicity model and the observed brood mortality rate (AOM, %) for bee larva from the 6-d in-vitro rearing experiments.
Pesticide βα LC50
b Conc. PMc 1-dd 2-dd 3-dd 4-dd 5-dd 6-dd AOMe
Fluvalinate 2.5 15.86 3 3.6 3.13* 8.06 12.28 10.00 11.11 68.85** 11.72
Coumaphos 2.9 46.3 8 1.4 6.25* 1.67 8.47 5.56 3.92 53.73** 8.60
Chlorothalonil 4.5 1110 34 4 E-10 0.00 8.93 7.84 12.77 7.32 56.60** 9.82
Chlorpyrifos 10 1.22 1.5 82 0.00 4.17 8.70 33.33** 32.14** 0.00 10.07
a β is the slope of the probit function for different pesticides (Atkins et al., 1981, Johnson et al., 2009). b LC50 is the median lethal concentrations of each pesticide to adult honey bees (Mullin et al., 2010). c.PM=predicted adult mortality rate (%) for each pesticide at the tested concentrations using inverse prediction of the probit function. d 1,2,3,4,5,6-d is the observed conditional mortality rate (%) for larval bees at each age (in day) in the in vitro rearing process. e AOM=average daily mortality rate (%) for larval bees in the in vitro rearing process. * significant at p < 0.05; ** significant at p < 0.001.
Adult honey bee Honey bee larva
Inverse probit prediction
In-vitro brood test
41
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
1 2 3 4 5 6
Cum
ulat
ive
Mor
talit
y
Larval development in days
Chronic cumulative toxicity
34 mg/L Chlorothalonil
3 mg/L Fluvalinate
8 mg/L Coumaphos
1.5 mg/L Chlorpyrifos
1% Solvent control
*
*
*
*
*
*
*
*
*
*
*
*
*
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
Chlorothalonil Fluvalinate Coumaphos Chlorpyrifos Solvent
Con
ditio
nal M
orta
lity
Chronic conditional toxicity 1-d larva 2-d larva 3-d larva 4-d larva 5-d larva 6-d larva
B).
AA).
42
Figure 2-1. Larval survival during the 6-d development stage reared using artificial diet contaminated with four pesticides at the selected concentrations and a 1% solvent control. 1A: the cumulative mortality of honey bee larva through 6-d development continually exposed to 34 mg/L Chlorothalonil, 3 mg/L Fluvalinate, 8 mg/L Coumaphos, 1.5 mg/L Chlorpyrifos and 1% solvent; 1B: the conditional mortality for different development stages of bee larva. Asterisks denote significant difference from the respective solvent controls (Log-rank test, p < 0.0001).
Two-component mixture toxicity
Chronic toxicity of chlorothalonil and coumaphos
The effects of chlorothalonil (34 mg/L), coumaphos (8 mg/L), and their mixture on larval
survival through the 6-d development are shown in Figure 2-2A. In the first 3 days of larval
rearing, these three groups exhibited similar survival curves (p = 0.1988, Log-rank test).
Subsequently, the larvae reared on the diet contaminated with the chlorothalonil/coumaphos
mixture died most quickly. The risk of 4-day-old larvae being killed by the mixture was higher
than for the other stages of larvae and the single pesticide groups. The hazard rate of the
combination group (hn(4) = 0.523) was 9-times higher than the coumaphos group (hCM(4) =
0.057) and 3-times higher than the chlorothalonil group (hCL(4) = 0.136). The conditional
probability of 4-day-old larvae being killed by the mixture treatment was 5-times higher than that
of expected additive toxicity (Figure 2-2C, Ehn(4) = 0.0965, p < 0.0001, Mann–Whitney test).
Therefore, the pairing of chlorothalonil and coumaphos produced a significant synergism on
mortality of larvae older than 4 days.
Chronic toxicity of chlorothalonil and fluvalinate
For the 4-day-old larvae, the hazard rate of the mixture (hn(4) = 0.78) was the highest
during the 6-d larval development, which was 7-times higher than the fluvalinate (3 mg/L) group
43
(hFlu(4) = 0.105) and 5-times higher than the chlorothalonil (34 mg/L) group (hCL(4) = 0.136)
(Figure 2-2B). The chlorothalonil/fluvalinate mixture at the tested concentrations gave a
synergistic interaction, which significantly magnified the hazard rate by 7 fold over the sum of
the individual effects (Figure 2-2D, Ehn(4) = 0.121, p < 0.0001, Mann–Whitney test).
44
45
Figure 2-2. Synergistic interactions for two pairs of pesticide mixtures: 8 mg/L Coumaphos, 34 mg/L Chlorothalonil and the mixture; 3 mg/L Fluvalinate, 34 mg/L Chlorothalonil and the mixture. 2A,2B: the respective Kaplan-Meier survival plots for honey bee larvae reared for each pair of pesticide mixture; 2C,2D: the interaction determination based on the deviation of observed mixture toxicity (black bar) from the expected additive toxicity (stacked bar).
Chronic toxicity of fluvalinate and chlorpyrifos
Larval survival on fluvalinate (3 mg/L) and chlorpyrifos (1.5 mg/L) declined the fastest
among pesticide mixture treatments, ranging from 4.17% to 70.83% (Figure 2-3). No significant
0
0.1
0.2
0.3
0.4
0.5
0.6
Expected Observed
Con
ditio
nal m
orta
lity
at d
ay 4
C)
8 mg/L Coumaphos 34 mg/L Chlorothalonil
0
0.1
0.2
0.3
0.4
0.5
0.6
Expected Observed
Con
ditio
nal m
orta
lity
at d
ay 4
D)
3 mg/L Fluvalinate 34 mg/L Chlorothalonil
46
differences were found in larval survival between single component groups through the 6-d
development (Figure 2-3A, Log-rank test, p = 0.1711). This binary combination produced
additive toxicity. The 6-d cumulative percent mortality caused by this mixture (hn = 71%) was
slightly higher than the sum of the response to single components, but not at a significant level
(Figure 2-3D, Ehn = 48.96%, p = 0.171, Mann–Whitney test).
Chronic toxicity of chlorpyrifos and coumaphos
The larval chronic toxicity of this combination treatment was the highest among tested
pesticide mixtures causing from 10.4% to 79.2% mortality during the 6 days. Survival was least
affected by the diet with 8 mg/L coumaphos (Figure 2-3B). The interaction between these
pesticides showed an additive effect. The 6-d cumulative percent mortality of larvae reared on the
mixture (hn = 79.2%) did not differ significantly from expected additive toxicity (Figure 2-3E,
Ehn = 56%, p = 0.558, Mann–Whitney test).
Chronic toxicity of fluvalinate and coumaphos
The survivorship of larval bees on the combination and fluvalinate alone treatments
exhibited a similar gradual declining trend, achieving the highest cumulative mortality at the end
of the 6-d development (Figure 2-3C). Both showed more toxicity to larval bees than coumaphos
alone (Figure 2-3C, p = 0.0425, Log-rank test). Fluvalinate and coumaphos, mixed at 3 mg/L and
8 mg/L respectively, showed an additive effect. The accumulative percent mortality in the
mixture group (hn = 68.75%) did not vary significantly from the expected additive toxicity
(Figure 2-3F, Ehn = 60.94%, p = 0.052, Mann–Whitney test).
47
48
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
Expected Observed
6-d
cum
mul
ativ
e m
orta
lity
D)
3 mg/L Fluvalinate 1.5 mg/L Chlorpyrifos
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
Expected Observed
6-d
cum
mul
ativ
e m
orta
lity
E)
8 mg/L Coumaphos 1.5 mg/L Chlorpyrifos
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
Expected Observed
6-d
cum
mul
ativ
e m
orta
lity
F)
8 mg/L Coumaphos 3 mg/L Fluvalinate
49
Figure 2-3. Additive effects for three pairs of pesticide mixtures: 3 mg/L Fluvalinate, 1.5 mg/L Chlorpyrifos and the mixture; 8 mg/L Coumaphos, 1.5 mg/L Chlorpyrifos and the mixture; 8 mg/L Coumaphos, 3 mg/L Fluvalinate and the mixture. 3A,3B,3C: the respective Kaplan-Meier survival plots for honey bee larvae reared for each pair of pesticide mixture; 3D,3E,3F: the interaction determination based on the deviation of observed mixture toxicity (black bar) from the expected additive toxicity (stacked bar).
Chronic toxicity of fluvalinate and chlorothalonil at low concentrations
The 3.4 mg/L chlorothalonil and 0.3 mg/L fluvalinate mixture showed the least toxicity to
larval development among pesticide combinations tested (Figure 2-4A). Especially, for the 4-day-
old larva, the hazard rate of individual component groups (hCL(4) = 0.214, hFlu(4) = 0.259) was
greater than twice the mixture treatment (hn(4) = 0.088). This mixture showed antagonistic
interaction, significantly reducing the hazard rate of 4-day-old larva by three-fold from the
expected additive toxicity (Figure 2-4C, Ehn(4) = 0.2365, p < 0.0001, Mann-Whitney Test).
Three-component mixture toxicity
All six possible pairings were selected to determine the toxicity for three-component
mixtures including chlorothalonil/fluvalinate/coumaphos and fluvalinate/coumaphos/chlorpyrifos.
The only significant difference found was when coumaphos (8 mg/L) was added to the two-
component mixture of fluvalinate (3 mg/L) and chlorothalonil (34 mg/L), giving a 3% reduction
in the 6-d accumulative larval mortality (hn = 38%) from the expected additive effect (Figure 2-
4B, 2-4D; Ehn = 41.41%, p = 0.006, Mann-Whitney Test). The other five pairings did not yield
significant changes in larval survival when adding one component into the existing binary
mixtures.
50
51
Figure 2-4. Antagonistic interactions for two pairs of pesticide mixtures: 0.3 mg/L Fluvalinate, 3.4 mg/L Chlorothalonil and the mixture; 3 mg/L Fluvalinate + 34 mg/L Chlorothalonil mixture, 8 mg/L Coumaphos and the three-component mixture. 4A,4B: the respective Kaplan-Meier survival plots for honey bee larvae reared for each pair of pesticide mixture; 4C,4D: the interaction determination based on the deviation of observed mixture toxicity (black bar) from the expected additive toxicity (stacked bar).
0
0.1
0.2
0.3
0.4
0.5
0.6
Expected Observed
6-d
cum
mul
ativ
e m
orta
lity
C)
3.4 mg/L Chlorothalonil 0.3 mg/L Fluvalinate
0
0.1
0.2
0.3
0.4
0.5
0.6
Expected Observed
6-d
cum
mul
ativ
e m
orta
lity
D)
8 mg/L Coumaphos
3 mg/L Fluvalinate+34 mg/L Chlorothalonil
52
Four-component mixture toxicity
Two pairings of mixtures including chlorothalonil added to
fluvalinate/coumaphos/chlorpyrifos and chlorpyrifos added to
chlorothalonil/fluvalinate/coumaphos were tested at the same concentrations as before to
determine toxicity interactions in going from three- to four-component mixtures. There were no
significant changes in larval survival when integrating a fourth component into these three-
component mixtures. The four-component mixture caused 54.17% larval mortality at the end of
the 6-d larval development.
‘Inert’ ingredient toxicity
Chronic exposure of bee larvae to the ‘inert’ ingredient NMP at seven different
concentrations ranging from 0.01% to 1% greatly impacted larval survival (Figure 2-5).
Increasing amounts of NMP correspondingly increased larval mortality. A 1% concentration of
NMP was the most acutely toxic, generating 100% mortality within 24 h after treatment. Even for
the lowest concentration of 0.01%, the estimated time to cause 50% larval mortality was 4 days.
53
Figure 2-5. The estimated time to cause 50% larval mortality by seven nominal concentrations of N-Methyl-2-pyrrolidone mixed in larval diet.
Discussion
Chronic toxicity
Our findings suggest that chronic dietary feeding at realistic levels of common pesticide
ingredients including the fungicide chlorothalonil, miticides fluvalinate and coumaphos, and
insecticide chloropyrifos, individually or in mixtures, have statistically significant impacts on
honey bee larval survivorship. A significant increase in larval mortality was found at or beyond 4-
d of feeding. This is the first study to report serious toxic effects on developing honey bee larvae
of dietary pesticides at environmentally relevant concentrations. This chronic (6-d) toxicity is
likely to be undetected in a conventional acute (24/48 h) toxicity study, resulting in potential
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
0.01 0.02 0.05 0.10 0.20 0.50 1.00
Med
ian
surv
ival
tim
e (D
ays)
N-methyl-2-pyrrolidone concentrations (% in diet)
N-methyl-2-pyrrolidone toxicity to honey bee larvae
Day to 50% Mortality
54
underestimation of pesticidal effects. The lethal effects on honey bee larvae appearing after 4-d
continuous exposure to pesticides at low concentrations are also observed in adult honey bees.
The accumulated dose of the organophosphorus insecticides acephate, methamidophos or
dimethoate resulting in 50% adult bee mortality was over 100-fold lower than the respective acute
24 h oral LD50 (Fiedler, 1987). For these organophosphates and also the pyrethroids tested, their
toxicity to worker bees was significantly increased by continuous versus single ingestion of the
contaminated food. At low doses of imidacloprid, adult bee mortality was observed only 72 h
after onset of feeding in contrast to immediate effects at much higher doses (Suchail et al., 2001).
The causes for chronic larval bee toxicity for 6-d dietary subacute pesticide exposures
remain unknown. It may be associated with the extended time needed to accumulate sufficient
insecticide concentrations internally to exert nerve action at central target sites, which is
consistent with the pharmacological receptor theory; or may reflect variation in honey bee
detoxification capacities from the more peripheral to internal tissue sites. For instance, the results
of high toxicity of low doses of all imidacloprid metabolites suggest the existence of binding sites
with different affinities in honey bees (Suchail et al., 2001). Another explanation may be that
honey bee detoxification mechanisms are not induced by chronic exposure of low concentrations
of active substances, but require higher more acute concentrations to impact honey bee
susceptibilities. In the former case, bee mortality would be latent due to the time needed for
pesticide bioaccumulation, further favored by the more lipophilic pesticides fluvalinate,
coumaphos, chlorpyrifos and chlorothalonil tested here. The latter case of acute higher
concentrations driving induction of detoxification enzymes can result in both antagonistic and
synergistic effects on the target-effective insecticide concentration depending on if the induced
cytochrome P450 first activates (e.g., chlorpyrifos, coumaphos to respective oxons) or detoxifies
(e.g., fluvalinate) the insecticide (Yu, 2008). Other induced enzymes (e.g. hydrolases, glutathione
transferases) will further degrade and detoxify the primary metabolites.
55
It is also plausible that more general stress mechanisms (e.g., altered feeding, suppressed
growth) dominate the chronic response. For example, exposures of some repellent pesticides such
as pyrethroids at sublethal levels have been demonstrated to impair feeding behaviors of honey
bees and bumble bees (Desneux et al., 2007). In the case of honey bee larvae, they retain
internally all metabolic wastes throughout the larval stage up to the pupal molt after which they
defecate a waste pellet called the meconium (Winston, 1987). Concentrations of pesticides and
metabolites within brood tissues may result in continuous pesticide stress (Wu et al., 2011),
which differs from the adult honey bee and most other insects where excretion of toxic wastes
regularly occurs. Little information is available on the distribution of fluvalinate (Bonzini et al.,
2011) and coumaphos (Vanburen et al., 1992) and their degradates in honey bee adults and brood.
Further studies to examine the distribution and accumulation of fluvalinate, coumaphos,
chlorpyrifos and chlorothalonil and their metabolites, in honey bees at different developmental
stages are needed. Meanwhile, how honey bees at different life stages withstand chronic exposure
need more detailed study of metabolic regulation in this social insect.
Remarkably, among the four pesticides tested in the present study, premature honey bees
are highly vulnerable to the common fungicide chlorothalonil (Figures 2-1, 2-2). Dietary
chlorothalonil killed more than 50% of larvae in 6 days at a level of 34 mg/L, a nontoxic dose to
adult bees in acute bioassays (Table 2-1). This difference in larval to adult susceptibility was the
largest among the four pesticides tested. It is unclear why, larval bees exhibited much greater
sensitivity to chlorothalonil compared to adult bees; however, the present results demonstrate that
investigating fungicide impacts on honey bees is particularly necessary for a realistic evaluation
of pesticides impacts on colony health, given the frequent detections of chlorothalonil in pollen
and wax samples. Hence, considering that honey bees are experiencing a diverse array of
agrochemicals in the hive, the chronic toxicity test may better assess pesticide exposure for a
honey bee colony.
56
Mixture toxicity
Currently, studies of mixture toxicity between different classes of pesticides at
concentrations of environmental relevance are rarely available for honey bees. The present study
of four pesticides in all combinations is the first study to investigate the potential synergism of
common pesticides at realistic exposure levels to larval bees. The present results showed
interactions between binary combinations of synthetic pesticides tested were mostly additive,
which can be attributed to the same or independent mode of actions of the pesticides involved
(Yu, 2008). For instance, additivity of the coumaphos/chloropyrifos mixture may be explained by
their identical action as organophosphate inhibitors of acetylcholinesterase. The additive toxicity
of the pyrethroid fluvalinate with either coumaphos or chloropyrifos is probably due to the
independent primary action of the former on nerve sodium channels. Our result with larvae is not
consistent with the adult honey bee study of Johnson et al., where the combination of fluvalinate
and coumaphos was synergistic (Johnson et al., 2009). This discrepancy may be explained by the
different life stage, lower insecticide concentration levels, and longer length of exposure used
here.
The three and four component mixtures of tested pesticides have mostly demonstrated
additive effects in larval bees. This finding is in general agreement with the Funnel Hypothesis
(Warne and Hawker, 1995), which states that the toxicity will tend towards concentration
additivity as the number of components in equitoxic mixtures increases. One exception was the
significantly less than additive response when coumaphos was integrated into the fluvalinate and
chlorothalonil mixture. That coumaphos antagonizes the synergistic effect of fluvalinate and
chlorothalonil may be related to its possible induction of the detoxification of one or both of the
other pesticides. This anomaly may be related to the observation that elevated coumaphos levels
in brood had the highest discriminatory value with regard to healthy bee colonies whereas higher
57
levels of this miticide in the pollen food correlated with colony collapse (VanEngelsdorp et al.,
2010), again indicating that pesticide susceptibilities differ across honey bee developmental
stages.
Remarkably, binary mixtures of chlorothalonil with the miticides fluvalinate or coumaphos
were synergistically toxic to 4-day-old bee larvae. This is the first demonstration for honey bee
brood of a synergistic interaction between dominant in-hive miticides and the frequently-
encountered fungicide chlorothalonil at environmentally relevant concentrations. Synergism with
chlorothalonil and fluvalinate but not coumaphos for adult honey bee toxicity has been noted
previously (Thompson and Wilkins, 2003, Johnson, 2011).
Surprisingly, a significant antagonism was found for larval toxicity from the fluvalinate-
chlorothalonil combination at one-tenth of the concentrations (Figure 2-4) that otherwise
exhibited a five-fold synergism (Figure 2-2). One rationale behind this latter interaction, beyond
the fact that the very diverging pyrethroid-multi-site chlorothalonil mechanisms of action may
alone elicit synergistic effects, is that the high concentrations may directly inhibit detoxification
enzymes. For example, the competitive inhibition of cytochrome P450 monooxygenase enzymes
has been suggested to explain the synergistic interactions among pesticides for adult honey bees
such as pyrethroid insecticides (Johnson et al., 2006) or mixtures of organophosphate insecticides
and ergosterol biosynthesis inhibiting fungicides (Pilling et al., 1995). Also, synergism between
chlorothalonil and the herbicide atrazine has been documented in aquatic species (DeLorenzo and
Serrano, 2003). Modes of action for chlorothalonil range from inhibiting glutathione and other
thiol-dependent enzymes or protein receptors, to disrupting or degrading cell membranes causing
lysis that can enhance penetration of other pesticides (Caux et al., 1996). The tendency toward
antagonism of brood toxicity at the lower dietary chlorothalonil-fluvalinate concentration may be
associated with alternative peripheral mechanisms such as gut microbial detoxification that may
be overwhelmed at higher dosage where more internal neurotoxic effects of the pyrethroid can
58
prevail. The consequence is that biphasic low and high dose response relationships may result
depending on the extent of multiple peripheral and internal sites of action that diverge in
sensitivity to the toxicants as well as to the available detoxification pathways that differ in a
tissue-dependent manner to the concentrations required for their induction.
While the mechanisms of interactions among pesticides with diverse modes of action and
their dynamics in the developing honey bee larvae are not known, application of the
concentration-addition model combined with chronic feeding tests represent a starting point for
investigation of mixture effects at realistic levels and their risks on this pollinator. Considering
that the diverse arrays of chemicals (Johnson et al., 2010, Mullin et al., 2010, Chauzat et al.,
2011) and general additivity exist in the hive environment, examining the toxicity of chemical
mixtures in addition to single toxicants is critical for realistic assessment of pesticide hazards
experienced by honey bees and other non-target organisms. Hence, the dose dependency of the
synergy, the multitude of compounds, the difference in adult bees and larvae, and the possibility
of continuous exposures should be taken into account in the environmental risk assessment.
‘Inert’ toxicity
Another important health issue that involves pesticide formulations and bees is the
consequence of the additives or so-called non-active ingredients. The commonly-used ‘inert’
ingredient N-methyl-2-pyrrolidone was found here to be highly toxic to larval honey bees (Figure
2-5). Unfortunately, despite the potential toxicity of ‘inert’ ingredients and their widespread use
in pesticide products, their testing and risk assessment seems to be inadequate. There is a growing
body of research that has reported a wide range of adverse effects of ‘inert’ ingredients to human
health, including enhancing pesticide toxicities across the nervous, cardio-vascular, respiratory,
and hormonal systems (Cox and Surgan, 2006). However, limited data exists on the potential
59
impacts of ‘inerts’ on non-target pollinators, although recent studies implicate formulation
additives or adjuvants as key risk factors (Ciarlo et al., 2012). As one example, the toxicity of the
fungicide captan to honey bee brood development was attributed to formulation ingredients other
than the active ingredient alone (Everich et al., 2009). The lack of detailed information of the
usage of formulation ingredients greatly impedes appropriate risk assessment of ‘inert’ ingredient
toxicity; therefore, label disclosure of the composition of pesticide formulations would facilitate
this much-needed evaluation.
Conclusions
The current study demonstrates the chronic oral and mixture toxicity of common pesticides
at realistic exposure levels to honey bees at the larval stage. Most notable are the chronic larval
toxicities of the fungicide chlorothalonil and its synergistic combinations with frequently used in-
hive miticides, and the unexpected high toxicity of the formulation ingredient N-methyl-2-
pyrrolidone. Considering the extensive detection of chlorothalonil and its coexistence with other
pesticides in diverse combinations especially in hive pollen and wax, and its substantial larval
toxicity alone and in mixtures shown here, the application of this and other fungicides during crop
bloom cannot be presumed innocuous to pollinating honey bees. Given the critical sensitivity of
larvae to chlorothalonil and its complex interactions with other pesticides, the potential impacts of
fungicides on colony survival and development need further investigation. In the more complex
milieu of this social insect and its aging hive environment, pesticides, formulation additives and
their resulting mixtures may have greater long-term impacts on colony health. Consequently, the
scope of pesticide risk assessment for non-target honey bees should be expanded from the present
emphasis on acute toxicity of individual Apis-cidal pesticides to a priority for assessment of
60
chronic and mixture toxicities that incorporate fungicides, other pesticide pollutants and their
‘inert’ ingredients.
Acknowledgements
This work cannot be done without the contribution of three co-authors: Daniel Schmehl,
who first proposed using the in-vitro larval rearing methods to examine pesticide impacts on
larval health and also provided crucial help with the experimental set-ups; Professor Chris Mullin
and James Frazier, who provided constructive comments and suggestions on analyzing the results
of pesticide toxicity on larval survival, and reviewed the manuscript. I also acknowledge the
contribution of Maryann Frazier, Lauren Russert and Sara Ashcraft who have helped establish
and manage three experimental hives in the forestry building of the Pennsylvania State
University. This work is funded by the 2009 research grant approved by North American
Pollinator Protection Campaign (NAPPC).
Chapter 3
Common Fungicides and Their Formulation Impacts on Honey Bee Larval Health
Recently, unprecedented levels and types of pesticides, especially fungicides coexisting
in diverse combinations, were detected in the honey bee (Apis mellifera) hive environment. We
previously provided evidence of the remarkable sensitivity of larval bees to chlorothalonil. Here
we explore further the potential hazard to honey bee larvae of frequently-found fungicides at
environmentally relevant levels. Thus, a goal of this study was to determine whether the
respective active ingredients are the main drivers of the chronic oral toxicity of environmentally
relevant levels of common fungicide formulations to larval A. mellifera. Our study provides the
first evidence of high susceptibility of honey bee brood to common fungicides at environmentally
relevant levels, nonmonotonic dose response of chlorothalonil, elevated toxicity of the fungicide
formulation, and synergism of co-occurring fungicides. Our findings imply that common
fungicides, particularly chlorothalonil could be a plausible contributor to honey bee colony
declines. Future research is needed to examine ‘inert’ and mixture toxicity that quantifies colony-
level health effects of fungicides.
Introduction
Since the development of modern agriculture, European honey bees (Apis mellifera L.),
one of the world’s most important pollinator species, have regularly encountered a tremendous
diversity of synthetic toxins through pesticide drift, residues in bee-pollinated crops or in-hive
pesticide treatment (Bogdanov, 2006). Of particular concern are fungicides, which are routinely
62
applied during bloom to prevent fungal diseases, when bees are present for pollination. Though
fungicides are generally considered safe for use around adult honey bees (Atkins et al., 1981),
they nevertheless can be brought back to the colony and stored in colony matrices, such as wax,
honey, pollen, and bee bread (Chauzat and Faucon, 2007, Mullin et al., 2010). The levels of
pesticides are unintentionally enhanced in the hive matrices due to the habit of honey bees in
concentrating nectar to make honey and processing pollen into royal jelly diet and bee bread
(Schmidt and Johnson, 1984). Therefore, many routes (e.g., direct contact, oral feeding) exist for
potential exposure of honey bees at different life stages to toxicants.
Recently, unprecedented levels and types of pesticides coexisting in diverse combinations
were detected in the hive environment, particularly the fungicide chlorothalonil in bee collected
pollen and wax. Efforts to restrict fungicide exposures during bloom have been made since
beekeepers frequently report bee losses or morphological malformations in larval and pupal
stages coinciding with fungicide application (Atkins and Kellum, 1986, Everich et al., 2009).
Certain fungicides not only increase adult mortality, but also affect bee physiology such as
thermoregulation (Vandame and Belzunces, 1998), foraging behavior (Mayer and Lunden, 1986)
and detoxification activity (Pilling et al., 1995). Fungicides are also found to disrupt the
composition and organization of the fungal community critical to processing pollen into bee
bread, thereby reducing its nutritional value (DeGrandi-Hoffman et al., 2009).
Given the fact that early life stages are generally more sensitive to contaminants than
adults (Desneux et al., 2007), chronic consumption of contaminated diets, inadequate nutrition,
and/or direct contact of toxicants at honey bee immature stages may be detrimental to the survival
and development of bee brood (Becher et al., 2010). Conceivably, fungicide impacts on larvae
could weaken colony population structures and contribute to the wide-spread honey bee decline
reported annually since 2006. However, to our knowledge, few studies have assessed the chronic
effects of frequently encountered fungicides on larval honey bee development (Frazier et al.,
63
2008), in part because a standardized guideline for pesticide risk assessment has not been
developed that includes chronic testing on larval honey bees (Thompson, 2010, Hendriksma et
al., 2011). We previously provided evidence of the remarkable sensitivity of larval bees to
chlorothalonil (Zhu et al., 2013).
Given the fact that ‘inert’ more than active ingredients dominate pesticide formulations
and spray tank adjuvants, it follows that honey bees are also likely exposed to significant amounts
of ‘inert’ ingredients. A growing body of evidence points to the direct or indirect toxicity of
formulation ingredients to non-target invertebrates. For example, aqueous solutions of trisiloxane
surfactants Silwet L-77, at field realistic levels, were acutely toxic to two spotted spider mites
(Cowles et al., 2000), and the juvenile stage of several arthropod pests of table grapes such as
cotton aphids and grape mealybugs (Tipping et al., 2003). The common inert ingredient N-
methyl-2-pyrrolidone at low concentrations also showed high larval toxicity (Zhu et al., 2013).
However, very few studies have examined the toxicity of formulation ingredients to honey bees
(Mullin et al., 2010). Thus, a goal of our study was to determine whether the respective active
ingredients are the main drivers of the chronic oral toxicity of environmentally relevant levels of
common fungicide formulations to larval A. mellifera.
Assessing the hazards of fungicides to honey bees is complicated further by intricate
interactions in fungicide mixtures. In agricultural practice, fungicides are often applied in
mixtures that are intended to interact in a synergistic manner (Gisi, 1996). Synergists have been
increasingly used to enhance pesticide toxicity and mitigate the development of fungicide
resistance; however, synergistic toxicity for beneficial pollinators may be potentially overlooked
(Pilling and Jepson, 1993, Schmuck et al., 2003). In addition, an extensive literature review
documenting pesticide sensitivity in the honey bee shows that its capacity to metabolize multiple
toxic compounds simultaneously may be limited, even though this species is not uniquely
vulnerable to individual pesticides (Hardstone and Scott, 2010). We previously showed the
64
synergistic toxicity of the fungicide chlorothalonil and two common in-hive varroacides to honey
bee larval development (Zhu et al., 2013). The present study is part of an effort to expand our
knowledge regarding the hazards of pesticide mixtures to honey bee larvae. Test chemicals were
chosen based on their widespread and frequent use in bee-pollinated crops such as almond, grape,
apple and blueberry. Chlorothalonil (TPN; 2,4,5,6-tetrachloroisophtalonitrile), a broad-spectrum
chlorinated aromatic fungicide, is one of the most widely used fungicides in the world and among
the most frequently detected pesticides in apicultural matrices (Mullin et al., 2010). Myclobutanil,
a triazole demethylation inhibitor (DMI) fungicide, has been used on a large number of additional
crops since its initial registration in 1988 for controlling powdery mildew and black rot on grapes
(California Department of Food and Agriculture (CDFA), 1988).
Specific objectives of this study using the standardized in vitro larval feeding method
developed by Aupinel et al. (Aupinel et al., 2007) are to: (i) evaluate the toxic effects of single
fungicide at various realistic concentrations on the survival of individual A. mellifera larva during
a 6-d continuous feeding with contaminated diet; (ii) compare the toxicity of formulation Bravo®
with its AI (technical-grade pesticide: chlorothalonil) to honey bee larvae; (iii) determine whether
a common fungicide mixture at environmentally relevant concentrations has synergistic toxicity
on larval survival. Measurable larval impacts should exhort risk assessors to consider chronic and
mixture impacts of fungicide formulations on honey bee brood survival and development, so that
better informed decisions can be made for improved hive management.
65
Materials and Methods
Test Fungicides
The fungicide formulations Bravo Weather Stik® (54% AI, chlorothalonil; Syngenta,
Greensboro, NC, USA), Nova® 40W (40% AI, myclobutanil, now Rally®; Dow AgroSciences,
Indianapolis, IN, USA) and Pristine® (25.2% boscalid and 12.8% pyraclostrobin, BASF,
Research Triangle Park, NC, USA) were obtained from retail suppliers. Technical-grade
chlorothalonil (purity, 98%) and myclobutanil (purity, 98%) were purchased from Chem Service
(West Chester, PA, USA).
Test Organisms
Honey bee 1st instar larvae were collected from colonies of an A. m. ligustica strain reared
in our experimental apiary (40° 49' 20"N, 77° 51' 33"W). In order to collect newly emerged
larvae, a honey bee queen was confined in an excluder cage and placed in the 2nd super from the
bottom of the hive and positioned in the center of the super to allow for proper incubation of the
newly laid eggs. After 30 hours, the queen was released from the cage and eggs were incubated in
the hive for 3.5 days. Frames of newly hatched 1st instar larvae were taken to the laboratory in a
pre-warmed chamber (~35°C).
Fungicide Chronic Toxicity Tests
We adapted the standardized in vitro larval feeding method developed by Aupinel et al.
(Aupinel et al., 2007) to conduct 6-d toxicity tests with honey bee larvae continually exposed to
technical grade fungicides and their formulations at the selected doses. Test diets were renewed
66
every 24 h and larval survival was recorded daily. Dead larvae were removed daily. For the dose-
response test of chlorothalonil and its formulation Bravo®, the six concentrations used (1.5, 3, 4.5,
9, 18 and 36 mg/L for chlorothalonil; 3, 6, 9, 18, 36 and 72 mg/L for Bravo®) were within the
range of chlorothalonil residues detected in pollen samples (Mullin et al., 2010).
According to our survey of pesticide residues in beebread and floral samples from
migratory beekeepers, the 95%-tile value for the most frequently detected fungicide
chlorothalonil and the maximum residue for a less frequently detected fungicide myclobutanil
were 12 and 8.2 mg/L, respectively (Mullin et al., 2010). Based on the relative percentage of AI
in each formulation, 24 mg/L Bravo® and 2l mg/L Nova® (nominal concentrations) were used
alone and in combination for examining their mixture toxicity on honey bee larval stages (Table
3-1). Maximum pollen residue values for boscalid (7.27 mg/L) and pyraclostrobin (3.48 mg/L)
were also tested in all possible binary combinations by dietary incorporation of the combined
fungicide formulation Pristine® (28 mg/L, Table 3-1) which is also commonly found in hives of
migratory beekeepers (Mullin et al., 2010).
Table 3-1. Fungicide formulations tested for honey bee larval toxicity.
Fungicide AIa MOAb AI Level c
(mg/L) AI Conc.d
(%) Form. Conc.e
(mg/L)
Bravo® Chlorothalonil Multiple sites 12 54.0 24
Nova® Myclobutanil Demethylation inhibitor 8.2 40.0 21
Pristine® Pyraclostrobin Cytochrome bc1
(Complex III) inhibitor 3.48 12.8 28
aAI= Active ingredient . bMOA= Mode of action of each fungicide AI. c AI detection levels found in pollen samples (Mullin et al., 2010) see Materials and Methods. dAI Conc. is the percentage AI found in formulation (http://www.cdms.net/).
Boscalid Succinate-dehydrogenase
(Complex II) inhibitor 7.27 25.2
67
eForm. Conc. is the concentration of formulated fungicide tested in the larval rearing bioassay, based on the calculation: AI Detection level/ AI Conc.
To prepare stock solutions, each technical grade pesticide was individually dissolved in
acetone, and fungicide formulations were dissolved in distilled water only. The stock solutions
were prepared weekly and stored at -20°C. The concentration of acetone in solvent controls was
equal to the greatest concentration of acetone in the treatments (1% for all tests). Each test
dilution was mixed thoroughly into the larval artificial diet prior to in-vitro feeding and stored in
2 mL sterile glass vials (Corning, USA). We monitored two control groups in the study: untreated
diet and the acetone-treated diet. Each experiment was repeated twice. Sample size for each trial
starting at the same experiment day is 3 replicates with 24 larvae per replicate. The test
acceptability criterion was 80% or greater accumulative survival of 6-day-old larval bees in the
control (Aupinel et al., 2007).
Statistical Analysis
Pesticide impacts were estimated based on the survivorship of honey bees during the
larval stage. The 6-d larval survival data were segregated by pesticide treatment and analyzed
using Kaplan-Meier survival analysis (Kalbfleisch and Prentice, 1980). This estimate generally
assumes independence among the individual death events and randomization within the treatment
group. The null hypothesis of no difference between survival curves of the treatment and control
groups was tested by the Log-rank test that weighs each death by the square root of the total
number of individuals at risk per time interval, placing less emphasis on deaths occurring later in
the experiment. We examined proportional hazards assumptions by using Kaplan-Meier curves
and interactions with time for each variable in the model. All variables met the conditions for
proportional hazards assumptions, which assume the individual hazard function (the
68
instantaneous probability of the hazard occurring at time t, given survival to time t) depends on a
common baseline hazard and the values of explanatory covariates (Cox, 1972). Therefore, we
employed Cox proportional hazards regression model to estimate hazard ratios (HRs) and 95
percent confidence intervals for the association of larval survival with each measure of pesticide
use. The hazard ratio is the relative risk of death of treatment individuals compared with controls
(Cox, 1972). Each HR from different treatments were compared using a 95% CI constructed from
the standard error of the difference between HRs. This time-to-event analysis allows for right
censoring (individuals still alive at the end of the observation interval), as needed in this analysis.
All the survival analyses were performed with SAS version 9.2 (SAS/STAT® 9.2 User’s Guide).
In the dose response relationship test of Bravo® and chlorothalonil, the EC50 and Lowest
Observed Effect Concentration (LOEC) were determined by least-squares fit model for larval
survival data regressed over concentrations of treatment and time in 6-d larval development
(Mayer et al., 1994, Lee et al., 1995). Associated with each regression model is a ‘goodness of fit’
statistic that reflects the degree of discrepancy between the actual responses in the data and the
predicted responses. The model with the best ‘goodness of fit’ statistic was chosen to generate
EC0.01, EC0.1, EC1, EC5, EC10 and EC50. EC0.01 was generally chosen to be representative of the
LOEC as it is a small effect value close to zero (Mayer et al., 1994, Lee et al., 1995).
Formulation vs. Active Ingredient Toxicity
The toxicity of pesticide formulation and AI was compared qualitatively using Finney's
harmonic-mean formula:
(Schmuck et al., 1994)
where
Ci = % concentration for each active ingredient in the formulation
Ci Tii
N∑ =100 TM
69
Ti = the toxicity (EC50, LC50 values) of each active ingredient
TM = the expected toxicity of the formulation if the effects of combined AIs are additive.
The ratio of AI/Formulation toxicity was obtained by comparing the expected over actual
toxicity value of the formulation. A ratio greater than 1.0 would indicate an increased toxicity of
the formulation or synergism; and a ratio less than 1.0 would indicate a decreased toxicity of the
formulation or antagonism. Generally, ratios between 0.5 and 2.0 are considered within the range
of normal experimental variation, and thus equal.
Pesticide Interaction Determination
We used significant departures from concentration additive toxicity to define antagonistic
and synergistic interaction between pesticides in mixtures (Hertzberg and MacDonell, 2002). The
expected additive toxicity for the chemical mixture is the sum of each chemical’s toxicity to
larval survival, calculated as , where n is the number of chemical components in
the pesticide mixture, and hi is the hazard rate for a specific component estimated from the
laboratory bioassay data. The sum of responses (Ehn) to individual components is estimated based
on the presumption that selected pesticide mixtures are the combination of substances with
independent modes of action or similar modes of action. The mixture toxicity can be predicted as:
Additive interactions--Simultaneous action of components in which the observed response of
honey bee larva to a mixture (hn) is equal to the sum of the responses (Ehn) to the individual
components; Synergistic interactions--hn is significantly higher than Ehn; Antagonistic
interactions-- hn is significantly less than Ehn.
We did not test different concentrations of each pesticide component and of the
combination to fit dose-response curves for the Bravo® and Nova® mixture. Neither food intake
Ehn = hi ni
n∑
70
nor doses of pesticides consumed by each larva were measured during oral feeding. Therefore,
this method does not allow exact quantification of the level of interaction but makes only an
initial qualitative assessment of synergism or antagonism.
Results
Control Toxicity
No significant differences in larval mortality were observed when larvae were reared on
untreated artificial diet or diet mixed with 1% acetone as solvent control (Log-rank test, p > 0.05).
These two control groups showed a 6-d accumulative percent mortality of approximately 16%,
which is within the normal range observed for control mortality using the in-vitro larval rearing
protocol (Aupinel et al., 2007).
Pesticide Formulation vs. Active Ingredient Toxicity
Exposure to chlorothalonil at a dietary concentration of 0.75 mg/L and Bravo® at
concentrations of 3, 6, 12 mg/L did not result in significant larval mortality compared to control
groups (Figure 3-1). The hazard ratio for each concentration of Bravo® and chlorothalonil relative
to controls (Table 3-2) was consistent with the Kaplan-Meier curves (Figure 3-1). For Bravo® at
mostly high concentrations, hazard ratios of treated versus controls were significantly positive
and increased with increasing concentrations (Figure 3-2). Relative to controls, larval 6-d survival
was significantly reduced by 30-85% in the 9, 18, 24, 36, and 72 mg/L concentrations. The
highest concentration (72 mg/L) of Bravo®, had the highest hazard ratio, and quickly killed all
larvae within 5-d (Figure 3-1A).
71
Table 3-2. Results of Cox proportional hazards regression of survival in bee larvae exposed to different concentrations of Bravo® or chlorothalonil.
Treatment Exposure (mg/L)
Hazard ratioa p 25% Failure
Survival Timeb Mean Survival
Timeb Bravo 3 1.61 0.17 5 5.39
6 1.46 0.29 6 5.58 9 3.25 <.0001* 4 4.96
12 1.32 0.44 6 5.63 18 5.23 <.0001* 3 4.50 24 8.18 <.0001* 2 3.75 36 7.17 <.0001* 1 3.60 72 17.47 <.0001* 1 2.52
Chlorothalonil 0.75 1.70 0.12 5.5 5.54
1.5 2.31 0.01* 5 5.42 3 6.28 <.0001* 2 3.86
4.5 3.37 <.0001* 5 5.21 9 3.69 <.0001* 4 5.06
18 2.35 0.01* 6 5.63 36 2.73 0.001* 6 5.27 72 5.06 <.0001* 4 4.73
aThe hazard ratio is the relative risk of death for larvae exposed to each treatment with respect to control. A hazard ratio of 1 implies no difference between the hazard function of treatment and control groups. bThe 25% and mean survival times are the calculated times at which the cumulative survival function is equal to 25% and 50%, respectively. The higher the values, the slower the individuals in the group die.
Figure 3-1. Kaplan-Meier survival plots for honey bee larvae reared for 6-days with eight concentration levels each of (A) Bravo and (B) chlorothalonil.
72
However, the concentration-mortality relationship of chlorothalonil was more complex
and exhibited a nonmonotonic pattern with low (3 mg/L) and high concentrations (72 mg/L)
causing significantly greater mortality than adjacent concentrations and control groups. Two
73
moderate peaks existed in early exposure, and became steeper as larvae aged (Figure 3-2).
Relative to controls, larval 6-d survival was significantly reduced by 30-60% in the 1.5, 3, 4.5, 9,
18, 36 and 72 mg/L concentrations (Figure 3-1B). Larvae exposed to chlorothalonil at 3 mg/L had
the highest hazard ratio (Table 3-2) and died most quickly, causing approximately 50% death
within 4 days and overall 66.7% larval mortality, which was the second highest 6-d accumulative
mortality observed (Table 3-2). The highest concentration (72 mg/L) of chlorothalonil caused
75% death of larvae, the highest accumulative mortality over 6-d exposure. Based on the
comparison of survival time (Table 3-2), when larvae were continuously exposed to a high
concentration of fungicide treatment (4.5~36 mg/L chlorothalonil or 9~72 mg/L Bravo®), Bravo®
elicited much higher toxicity than chlorothalonil. Except for 3 mg/L AI (6 mg/L Bravo®), hazard
ratios of Bravo® relative to AI also demonstrated that the formulation was markedly more toxic
than AI by a factor of 1.5 to 7 (Figure 3-4).
74
Figure 3-2. The dose-response relationship of Bravo® (solid line) and chlorothalonil (dashed line) during 6-d larval feeding: Y-axis is the accumulative mortality at each time interval; X-axis is the formulation or AI concentration.
The calculated effective concentrations causing 0.01%, 0.1% and 1% larval mortality
during 6-d larval development are shown in Figure 3-3A. There were no significant differences
among the low effect levels (EC0.01, EC0.1, EC1) selected for Bravo® and chlorothalonil at each
time interval (p < 0.01); EC0.01, therefore, can be used to estimate the larval LOEC. The LOECs
and EC50s for honey bee larvae exposed to Bravo® and chlorothalonil are summarized in Table 3-
3. Both LOEC and EC50 of Bravo® during 6-d larval development decreased asymptotically over
increasing time of exposure (Figure 3-3B, C). EC50 values decreased sharply from 394.56 mg/L at
d-1 to 49.18 mg/L at d-3, then tended to stabilize over the last three days of larval development
(Figure 3-3C). Similarly, Bravo® LOEC values dropped markedly from 8.64 mg/L at d-1 to 4.88
mg/L at d-3, and leveled off to approximately 4 mg/L the last three days (Figure 3-3B). As
expected, Bravo® LOEC values were lower than EC50 values by factors of 5–50 over the 6-d
larval development period. This indicates that larval toxicity of Bravo® increases with time of
exposure, and is dose-dependent. Meanwhile, larval susceptibility to chlorothalonil was also time-
dependent, but exhibiting a distinctly different pattern (Figure 3-3A, C). EC50 values were
initially equivalent to approximately 280 mg/L during the first three days of exposure, then
strongly decreased from 251.44 mg/L at d-4 to 34.76 mg/L at d-6 (Figure 3-3C). Similarly, LOEC
values decreased moderately until d-3, but descended more steeply over the remaining days
(Figure 3-3A). EC50 values were over 150-fold greater than the corresponding LOEC values.
LOEC and EC50 values of Bravo® and its AI both demonstrated dose and time dependence for
larval toxicity.
Table 3-3. Summary of the lowest observable effect concentrations (LOECs) and median effective concentrations (EC50s) for honey bee larvae exposed to Bravo® and its AI chlorothalonil (in mg/L), calculated from Finney's harmonic-mean formula. Larval LOEC or EC50 ratios for the active ingredient vs. formulation (adjusted to AI concentration in formulation) are presented for
75
each time interval. Ratios marked by an asterisk indicate a significant difference in AI and formulation toxicity.
Toxicant 1-day 2-day 3-day 4-day 5-day 6-day
LOEC EC50 LOEC EC50 LOEC EC50 LOEC EC50 LOEC EC50 LOEC EC50
Bravo 8.64 394.56 6.05 107.76 4.88 49.18 4.223 29.08 3.81 19.96 3.53 15.04
Chlorothalonil 1.18 293.44 1.12 289.54 0.98 279.05 0.72 251.44 0.41 182.70 0.16 34.76
AI/Formulation 0.27*
1.49
0.37*
5.37*
0.40*
11.35*
0.34*
17.29*
0.22*
18.30*
0.09*
4.62*
76
Figure 3-3. EC0.01, EC0.1, EC1, and EC50 values calculated for each developmental day for in-vitro reared honey bee larva exposed to Bravo® and its active ingredient chlorothalonil. Panel A: Estimated larval LOEC of chlorothalonil based on calculated EC0.01, EC0.1, and EC1 values; Panel B: Estimated larval LOEC of Bravo® based on calculated EC0.01, EC0.1, and EC1 values; Panel C: Comparison of time-dependent effective concentrations between Bravo® and chlorothalonil that caused 50% death of the honey bee larvae.
Ratios of LOEC and EC50 of AI vs. formulation toxicity were calculated for each day
interval during larval development (Table 3-3). All LOEC values of chlorothalonil were
statistically less than the formulation by more than 2-fold. This indicates a higher sensitivity of
honey bee larvae to the AI than to the formulation especially for the 6-day-old larvae. Inversely,
for each stage of larval development except the 1-day-old larvae, Bravo® EC50 was significantly
lower than the corresponding AI by a factor of 4.6 to 18.3. Moreover, this enhanced toxicity of
formulation positively correlated with the length of exposure or the stage of larval development.
The only exception is the 6-day-old larvae. The ratio of AI vs. formulation dropped to 4.6 at d-6
from 18.3 at d-5, which is the highest difference between formulation and AI with respect to the
EC50 value. This change is likely caused by the notably increased susceptibility of 6-day-old
larvae to chlorothalonil (d-6 EC50 = 34.76).
Toxicity differences between Bravo® and its AI, shown by the comparisons of HRs,
LOECs and EC50s, clearly demonstrate the formulation toxicity to developing honey bee larvae
cannot simply be represented by the AI alone. Meanwhile, the variations in their LOECs and
EC50s during 6-d larval development suggest the complex toxic mechanisms of the fungicide
chlorothalonil to honey bee at the larval stage.
77
Figure 3-4. The hazard ratio for honey bee larvae exposed to six concentrations of Bravo with respect to its corresponding active ingredient chlorothalonil.
Bravo and Nova Mixture Toxicity
The effect of Bravo® (24 mg/L), Nova® (21 mg/L), and their mixture on larval survival
through the 6-d development is shown in Figure 3-5. The survivorship of larvae reared on this
binary mixture was significantly different from single component effects (p = 0.0029, Log-rank
test, Figure 3-5A). Bravo®, Nova® and their mixture at environmentally realistic concentrations in
the larval artificial diet caused a total of 89.58%, 72.92% and 93.75% death in the reared larvae
after 6-d exposure, respectively. The risk of larval bees being poisoned by the mixture each day
was higher than by any single fungicide treatment particularly by the third day of exposure. The
hazard ratio for the mixture relative to Nova® treatment is 1.9 (95% CI =1.22-3.01, p = 0.0046),
0
1
2
3
4
5
6
7
3 mg/L FM/AI
6 mg/L FM/AI
9 mg/L FM/AI
18 mg/L FM/AI
36 mg/L FM/AI
72 mg/L FM/AI
Haz
ard
ratio
Bravo (FM)/Chlorothalonil (AI)
*
*
*
78
and was significantly higher. However, the hazard ratio for mixture relative to Bravo® is 1.4 (95%
CI = 0.93-2.16, p = 0.1054), and not statistically higher.
The pairing of Bravo® and Nova® produced significant synergism for the survival of
larvae older than 3 days (Figure 3-5B). The accumulative mortality of larvae killed by the mixture
was approximately 2 times higher than the expected additive toxicity. The other two binary
mixtures including Bravo®/Pristine® and Nova®/Pristine®, exhibited additive effects (Appendix
B). The accumulative mortality of these mixtures during 6-d larval development was not
significantly different from the expected additive toxicity.
79
Figure 3-5. Synergistic interaction for the 24 mg/L Bravo and 21 mg/L Nova larval diet mixture. Panel A shows the Kaplan-Meier survival plots for honey bee larvae reared on fungicide mixture and each component. Panel B shows the interaction determination for Bravo/Nova mixture based on the deviation of observed mixture toxicity (black bar) from the expected additive toxicity (stacked bar).
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Exp
ecte
d
Obs
erve
d
Exp
ecte
d
Obs
erve
d
Exp
ecte
d
Obs
erve
d
Exp
ecte
d
Obs
erve
d
Exp
ecte
d
Obs
erve
d
Exp
ecte
d
Obs
erve
d
Day 1 Day 2 Day 3 Day 4 Day 5 Day 6
Acc
umul
ativ
e M
orta
lity
Bravo/Nova Mixture 24 mg/L Bravo 21 mg/L Nova
B.
80
Discussion
Chlorothalonil Toxicity
The objectives of this study included measurements of toxicity elicited from exposures of
the fungicide Bravo® and its AI chlorothalonil to honey bee larvae. The results based on the in-
vitro larval rearing bioassay provided clear evidence of the significant oral toxicity of the
environmentally realistic levels of formulation and AI to early life stages of the honey bee.
Continuous 6-d exposure of this fungicide cause more than 60% accumulative larval mortality,
significantly differed from controls by two to three-fold. However, chlorothalonil is claimed by
EPA to be relatively non-toxic to adult honey bees. Nevertheless, the unprecedented levels of
chlorothalonil (up to 99 mg/L) detected in hive matrices (Mullin et al., 2010) leads us to question
whether it could be partially responsible for recent colony declines. Some fungicides show direct
toxicity to honey bees and solitary bees at field use rates (Ladurner et al., 2005). Chlorothalonil is
also potentially associated with entombing behavior in weak bee colonies, which may be a new
defensive behavior of bees faced with large amounts of potentially toxic food stores
(VanEngelsdorp et al., 2009).
Little is known about the toxicity of chlorothalonil to honey bees at juvenile stages.
Gregorc and Ellis (2011) found that 400 mg/L chlorothalonil could induce a significantly elevated
level of apoptosis in the midgut of honey bee larvae reared in vitro (Gregorc and Ellis, 2011). Our
chlorothalonil dose-response results reinforce the significant lethality to honey bee immature
stages due to chlorothalonil at low concentrations (> 1.5 mg/L). The pivotal concentration of 3
mg/L chlorothalonil was the most toxic among tested concentrations in causing larval mortality.
Our finding that larval bees are markedly sensitive to chlorothalonil is consistent with toxicity
studies of chlorothalonil on early life stages of aquatic invertebrates. Chlorothalonil LC50 values
81
ranged from 3.6 µg/L (Eastern oyster larvae, 96 h) to 837 µg/L (hard shell clam larvae, 96 h)
(DeLorenzo and Fulton, 2012). Given the fact that detections of chlorothalonil at 95th percentile
values (levels at which only 5% of detections are higher) are 3.4 mg/L in trapped pollen samples
(Mullin et al., 2010) and our low honey bee larval LOEC estimates for chlorothalonil (Table 3-3),
exposure to this fungicide has the potential to directly and/or indirectly contribute to honey bee
declines, although additional work at the colony-level is needed to demonstrate a causal link.
Furthermore, we provided the first evidence of a nonmonotonic dose–mortality response
of A. mellifera larva to chlorothalonil. The low (3 mg/L) and high concentrations (72 mg/L) cause
significantly greater mortality than did intermediate concentrations and controls (Figure 3-2).
This nonmonotonic dose response for mortality to chlorothalonil has been observed previously
for three amphibian species (Osteopilus. septentrionalis, Hyla cinerea, Rana sphenocephala)
(McMahon et al., 2011). Here the dose response of endogenous stress hormone corticosterone
was bimodal with increasing fungicide exposure, underlining nonmonotonic responses for both
survival and immune function in common amphibian species due to chlorothalonil (McMahon et
al., 2011). Nonmonotonic dose responses defy the toxicological dogma that “dose makes the
poison” and present a critical challenge to the traditional approaches in regulatory toxicology that
generally perform the high-dose testing for risk assessment.
Several mechanisms have been identified and suggested as the cause for nonmonotonic
responses including cytotoxicity, cell- and tissue-specific receptors and cofactors, receptor
selectivity, receptor down-regulation and desensitization, receptor competition, and endocrine
disruption (Vandenberg et al., 2012). However, the causes for the complex and multiphasic
response of chlorothalonil are currently unknown, likely because its mode of action on non-target
organisms has not been completely characterized (Caux et al., 1996). It could be suggested that
chlorothalonil affects different target systems or receptors with a different affinity for each of
them. The lowest concentration of chlorothalonil could bind to a more peripheral sensory receptor
82
in larval bees, triggering behavioral effects; while a higher concentration could bind to or block
an independent but more critical receptor, which would override the effects of the first
interaction. The respiratory/oxygen stress system and intestinal systems of developing larval bees
could be potential candidates. The action of chlorothalonil in inactivating key respiratory proteins
and disrupting normal metabolic functions of exposed organisms resulting in mortality has been
reported for both target and non-target aquatic species (Davies, 1987, Gallagher et al., 1992,
Sherrard et al., 2003). With the rapid uptake and bioconcentration of chlorothalonil due to its high
lipophilicity and sufficient duration of exposure, honey bee larvae would be expected to readily
exhibit mortality effects of chlorothalonil at low concentrations. Other potential targets could
involve internal systems. For example, chlorothalonil can cause acute necrosis of the intestinal
systems resulting in the mortality after sufficient exposure (Gallagher et al., 1992, Pariseau et al.,
2009). This more toxic mechanism could partially or completely mask the more sensitive
peripheral effects, possibly explaining the positive correlation between chlorothalonil toxicity and
dietary concentrations within the range of 18 to 72 mg/L.
Alternatively, the nonmonotonic dose response effect could also be triggered by
concentration-dependent metabolism of chlorothalonil. Chlorothalonil produces degradation
products that can be more or less toxic than the parent compound (Caux et al., 1996). Moreover,
chlorothalonil has been suggested as the potential endocrine disruptor to amphibian species based
on the evidence of elevated corticosterone levels. This further implicates potential adverse effects
of chlorothalonil on immune systems and disease resistance (McMahon et al., 2011). Although it
is not certain whether the nonmonotonic dose response for larval survival to chlorothalonil
observed in this study would also associate with immunological effects, we strongly advocate
future efforts of complete risk assessment on the chlorothalonil toxicity, in particular its specific
mode of action and degradation pathways in honey bees.
83
Fungicide Formulation Toxicity
In contrast, formulation Bravo® exhibited monotonic dose response with larval mortality
(Figure 3-2). Hazard ratios exhibited a similar increasing pattern with concentrations (Table 3-2).
Effective concentration (LOEC, EC50) estimates decreased exponentially with the length of
exposure. The AI/formulation toxicity ratios for EC50 demonstrated that Bravo® was generally
more toxic than chlorothalonil to honey bee larvae at each stage. The highest ratio was 18.30 at 5-
d. Conversely, the AI/formulation LOEC ratios showed an inverse trend with chlorothalonil
LOEC estimates lower than Bravo® by 2-10 fold, mostly due to the nonmonotonic dose response
of the AI. However, because of the lack of the chemical information on inert ingredients within
Bravo®, these inconsistencies in formulation and AI comparisons of EC50 and LOEC cannot
presently be critically evaluated.
‘Inert’ ingredients are generally considered as non-toxic and are added to help improve
the solubility, stability, spreading, binding capacity, and penetration of pesticide AIs (Surgan et
al., 2010). A growing body of research has shown their adverse effects to human health, including
enhancing neurotoxicity, cardio-vascular, respiratory, genetic and hormonal effects (Cox and
Surgan, 2006). However, there are still too few studies focused on the potential impacts of ‘inerts’
(co-formulants) on non-target pollinators, likely because toxicity information for pesticide
formulations on honey bees is not currently required by the U.S. Environmental Protection
Agency (EPA) as part of the pesticide registration process (Surgan et al., 2010). Everich et al
(Everich et al., 2009) reported that ‘inert’ ingredients within the fungicide Captan® contributed to
overall toxicity for honey bee brood development. Commercial formulations often exhibit higher
toxicity to other non-target organisms than the corresponding AI (Oakes and Pollak, 2000, Krogh
et al., 2003, Solomon and Thompson, 2003, Cedergreen and Streibig, 2005, Kitulagodage et al.,
2008, Pereira et al., 2009). Hence, this study reinforces concerns that the risk of formulated
84
pesticide products on target or non-target organisms may not be fully reflected in assessment
conducted on the AI alone.
Notably, Bravo® after day one showed increased toxicity over chlorothalonil during
larval development. Six-day honey bee larvae exhibited less tolerance to pesticide treatment
compared to younger larvae, which is contrary to many herbivorous insect species, where older
larvae, due to increased detoxification or decreased cuticular penetration, tend to be less
susceptible to insecticides (Bouvier et al., 2002). Given that larvae accumulate their wastes
throughout development until released at pupal molt as a meconium, increased mortality with age
is likely related with larval continuous exposure to increasing concentrations of parent pesticides
and their degradates, although total diet consumption was not measured and leftover diet was
removed every day; additionally antifeedant or other sensory effects on consumption, or an age-
dependent toxicity cannot be excluded. Another possible explanation could be that the tested
concentrations of both Bravo® and chlorothalonil, particularly in the low range, were not
sufficient to activate pesticide detoxification in honey bees. Mao et al. (2011) demonstrated that
in-hive miticides and diet-derived phytochemicals like flavonoids share common detoxificative
cytochrome P450-mediated monooxgenases, suggesting the possibility of synergistic interaction
between dietary phytochemicals and pesticides in honey bees (Mao et al., 2011). Since older
larvae consume larger amounts of pollen as well as pollen-derived royal jelly diet, we may
speculate that natural dietary toxicants including phenolics may overwhelm the rate at which
additional pesticides can be metabolized.
Another common fungicide formulation Nova® at environmental realistic concentrations
was also shown to cause significant larval mortality. Myclobutanil, the AI of Nova® widely used
to control powdery mildew on grapes, almonds, apples and many other crops, is a demethylation
inhibitor (DMI) fungicide that disrupts ergosterol biosynthesis principally involved in fungal cell
wall formation (EPA, 2007). Gregorc and Ellis (2011) demonstrated elevated apoptotic cell death
85
in the midgut, salivary glands and ovaries of developing honey bee larvae exposed to 400 mg/L
myclobutanil (Gregorc and Ellis, 2011). Though unclear how fungicides specifically affect
individual bees or their colonies, our laboratory results of fungicide toxicity at environmentally
realistic concentrations could be relevant to effects in nature. Understanding how A. mellifera
metabolizes fungicides is, therefore, crucial for assessing the limits of fungicide tolerance and for
evaluating the long-term impacts of commonly used fungicides on colony health.
Synergistic Toxicity
Further, the binary mixture of fungicides Bravo® and Nova®, at the concentrations of 24
mg/L and 21 mg/L, respectively, is the only one of three tested binary mixtures exhibiting
significant synergistic toxicity to 3-day-old bee larvae. To our knowledge, this is the first study to
report synergism for developing honey bee larvae between the non-systemic fungicide
chlorothalonil and systemic EBI fungicide myclobutanil, or their respective formulation inerts, at
environmentally-relevant dietary levels. The other two mixtures including Bravo®/Pristine® and
Pristine®/Nova® demonstrated additive effects. This is in accordance with our previous findings
that interactions between binary combinations of synthetic pesticides were mostly additive, which
can be attributed to the same or independent mode of actions of the pesticides involved (Zhu et
al., 2013).
A rationale behind the synergism is the multi-site mode of action of Bravo® distinct from
Nova®. Two theories of how EBI fungicides synergize the effects of the pyrethroids and
fungicide Captan® in honey bees have been proposed. Firstly, EBI fungicides inhibit a broad
range of cytochrome P450 monooxygenases, partially or completely preventing detoxification of
co-occurring pesticides (Pilling et al., 1995, Meled et al., 1998). Secondly, EBI fungicides or their
co-formulants may synergize by enhancing penetration rates of other pesticides into organisms,
86
thereby increasing their effective concentrations (Kennaugh et al., 1993). Chlorothalonil has been
proposed to disrupt membranes and cause cell lysis thereby synergizing the toxicity of the
herbicide atrazine in marine phytoplankton species (DeLorenzo and Serrano, 2003). While the
role of formulation ingredients in the synergism was not established here, formulation types such
as wettable powders (WP) may be even more hazardous to honey bee larvae as indicated for
adults (Johansen and Kleinschmidt, 1972). Nova®, but not Bravo®, was a WP formulation. This
study represents a starting point to investigate mixture effects of commonly used fungicides and
their risks for honey bee larvae at environmentally realistic and dietary levels. While underlying
mechanisms behind the observed synergistic interaction between these two fungicide
formulations with different modes of actions and their dynamics in the environment are still not
known, our results emphasize the importance of including chronic and mixture toxicity studies
into pesticide risk assessment for non-target organisms.
Conclusions
Together, given the extensive documented contamination of hive matrices with in-hive
and agricultural pesticides, high susceptibility of honey bee brood to common fungicides at
environmentally relevant levels, nonmonotonic dose response of chlorothalonil, elevated toxicity
of the fungicide formulation, and synergism of co-occurring fungicides, our results are
particularly relevant in wake of the colony collapse disorder. Future testing of fungicide impacts
at the colony level, and linking of larval responses to effects on later honey bee life stages are
needed. Furthermore, more specific label disclosure of pesticide formulation components would
help to inform users of actual risks to pollinator health.
87
Acknowledgements
This work cannot be done without the contribution of other two co-authors-Professors
Chris Mullin and James Frazier who have provided constructive suggestions on selecting the
types and concentrations of fungicides to be tested and also helped with reviewing the
manuscript. I also acknowledge the help of Maryann Frazier, Stephanie Mellott and Sara Ashcraft
who have helped establish and manage three experimental hives in the forestry building at the
Pennsylvania State University. This work is funded by two 2011 research grants approved by the
National Honey Board, and Pennsylvania State Beekeepers Association and the Center for
Pollinator Research.
Chapter 4
A Stage-structured Honey Bee Population Model
As an aid to testing hypotheses for the causes of recent colony failure and providing
suggestions for management actions to promote recovery of honey bee populations, we developed
a worker-based, stage-structured model of honey bee population dynamics. This model was
formulated with difference equations consisting of six discrete stages based on honey bee
temporal polyethism: egg, larva, pupa, nurse, house bee and forager stages. The model is unique
in capturing the adaptive feedback mechanisms in population dynamics and the resource
dynamics in a honey bee colony, including the comb pattern formation, brood maintenance, and
collective foraging behaviors.
Numerical simulation of a healthy colony shows seasonal patterns of adult population,
brood population, pollen and honey stores and provides a good agreement with empirical colony
data from the literatures and from my own experiments. This allows for further simulations of
various scenarios of a weakened colony and of testing whether or not subtle changes at the
individual or social group level by potential stressors could result in substantial outcomes at the
colony level. Local and global sensitivity analyses were performed to assess the relative
importance of input factors associated with the intrinsic processes of honey bee behavioral
development and behavioral change through social interactions on the colony dynamics. Both
sensitivity analyses reached the same conclusions that the colony dynamics are most sensitive to
input factors related to the task-based transition among the workers in a colony, and if the task
transition rate is interrupted beyond the critical threshold such as precocious development and/or
reversed transition from foraging to hive work, a rapid population decline is predicted and colony
failure is inevitable.
89
The simulation results indicate that the ratio between brood, foragers and hive bees in the
late field season, which is typically the preparing period for the coming winter season, can serve
as a direct measure of colony health. The change in this task-based ratio may be either a
consequence or a contributing cause of colony collapse. Our results suggest the importance of a
balanced allocation of workers for nursing and foraging in sustaining the colony survival and
development. Our model presented here can be used for colony dynamics forecasting, offering
directions to hypothesis-driven research, and providing suggestions for colony management
efforts.
Introduction
As a social insect, individual bees act as cooperative vehicles for colony growth and
development via a series of components within a highly elastic feedback system to achieve
stability and optimal fitness (Seeley, 1995). Coordination and differentiation are the key
principles to the success of the honey bee colony. However, currently the combinational energy-
draining stresses of illness, nutrition, and human migratory and cultural practices strike honey bee
populations day after day, depriving them of long-term health (VanEngelsdorp et al., 2009). The
possibility of a multi-factorial cause is one of the problems that makes investigating the recent
extensive colony decline (also called Colony Collapse Disorder, CCD) especially complex. While
generally acknowledging the associations among these possible factors with CCD, the
mechanisms by which a colony either succumbs to or overcomes the individual or combinational
effects of anthropogenic and environmental threats remains largely unknown. These research
gaps are largely because of our limited understanding of honey bee colony population dynamics
in reacting to multiple disturbances (VanEngelsdorp and Meixner, 2010). Such questions cannot
usually be answered by reductionist experimental approaches alone, but require the help of
90
mathematical models. To address the multiple open questions on colony losses, the development
of a practical honey bee risk assessment approach is imperative. Therefore, the objective of this
study was to build a population model of the honey bee colony, which could further our
understanding of population dynamics, make predictions of the resulting colony disturbance in
response to environmental and anthropogenic stressors, and ultimately aid in colony management.
A few population models have been developed to characterize the dynamics of managed
honey bee colonies, with a focus on the queen egg production, and fixed or priority-based stage
structure (DeGrandi-Hoffman et al., 1989, Schmickl and Crailsheim, 2007). Due to the limited
understanding of colony dynamics in honey bees, the influence of external factors (weather,
foraging resources) and inside nutrition-related feedbacks in regulating social organization has
been mostly disregarded. The model presented here is thus unique in capturing the feedback
mechanisms involving internal and external factors in regulating a colony’s age demography,
food collection, variable environmental drivers, and the plasticity of division of labor in the hive.
We constructed a stage-structured population model with the integration of difference and
differential equations model at different timescales, to simulate crucial adaptive features in the
colony (illustrated in Figure 4-1): comb pattern formation; the negative feedback interactions
among queen egg production, brood nursing, and the age demography in the colony; the
collective decision making in nectar foraging behavior involving house bees, forager bees, hive
space, and nectar quality in the environment; the homeostatic regulation of the pollen foraging
process by balancing the pollen need in the hive and foraging workforce; and the age-based
division of labor with the probability-based transition rates to mimic the abnormal behavioral
development (accelerated, delayed, precocious and reverted development). The model is designed
to answer the following broad questions that may assist beekeepers in modifying their
management strategies:
• What are the stable population dynamics of a honey bee colony population?
91
• What are the keys in the colony to maintaining honey bee social homeostasis?
• Can the system continue proper functioning if negative stressors from nutrition and/or
pesticides singly or simultaneously take a toll on some stages of the system?
• Can this destabilized system retrieve its stability by supplemental nutrition?
This modeling approach has allowed us to describe the colony population dynamics of
honey bees and to identify the critical stages of the honey bee life history in determining
population fitness. We also highlight the value of performing a variance-based sensitivity analysis
to investigate the effects of varying the stage-specific survival rates and behavioral changes of
honey bees to simulate various scenarios of a weakened colony. We aim to assess the relative
risks of potential stressors on the colony health--hence offering direction to design validating
experiments and colony management efforts, towards important demographic parameters for
honey bee conservation. The model presented here can be used as the central structure of a model
for colony dynamics forecasting, and is described in enough detail to allow the necessary
extensions to be implemented easily.
Model Development
A worker (female)-based, stage-structured population model was constructed with
difference equations to project population growth and food dynamics over discrete time. The
model is implemented in Matlab by defining functions and retrieving values in discrete steps
representing each day of the year starting on the first day of queen egg laying. Figure 4-1 gives
the basic model structure with three key adaptive feedback systems that maintain the hive
dynamics in a homeostatic manner. In a managed healthy colony, there are four main components
of the hive dynamics: brood cells including developing eggs, larvae and pupae; pollen cells;
honey cells including open nectar cells and capped honey cells; and empty cells. Initially, this
92
model is built on the following simplified assumptions: the environment is predator and
competition free, containing pollen and nectar of constant quality throughout the field season; the
optimal queen egg production is not affected by her age within 3 years; no swarming occurs; the
intrinsic mortality rate for each stage of bees is not time-dependent. The model has been
constructed by breaking down individual worker bees’ life cycles into six distinct compartments
(Figure 4-2): egg (day 0-2), larva (day 3-10), pupa (day11-26), nurse bee (day 27-42), house bee
(day 43-47), and forager (day 48-60). This life cycle is based on the biological development time
for each stage (Winston, 1987). Each compartment represents a measurable quantity and linked
together with the daily egg production rate and realistic stage mortality and food dependent
survival.
93
Figure 4-1. A stock-and-flow diagram of the most important components of our model. The diagram depicts the flow of materials (brood, adults, nectar, honey, and pollen) through our model: grey boxes indicate model variables that indicate one kind of material. Arrows indicate possible flows of material. Arrows with a plus sign or a minus sign indicates whether an increase in one variable causes an increase or a decrease of another variable. Grey circles indicate sources and sinks, through which material enters or leaves the model. Diamonds indicate the important social signal involving homeostatic regulations in nursing, pollen foraging and nectar foraging in a healthy hive. Black solid line indicates environmental factors involved in social regulation of the colony. Please note that we showed only the most important feedback loops in this diagram and that in the model the influence of one variable onto another variable often acts through several “intermediate” variables that are not shown in this diagram.
94
Figure 4-2. A theoretical life history diagram for honey bees with a six stage structured life history. Circles represent stage-specific classes. The arrows between the stages are called transitions, indicating the probability P of transitioning from one class to the next (horizontal arrows) or of remaining in the same class (lower curved arrows). The curved arrows between Queen and Egg, labeled β, represent fecundity.
Table 4-1 lists all the state variables, the difference functions and dynamic feedback
functions in the model. Table 4-2 summarizes the definitions and baseline values for all the
parameters and the stage specific survival rates of our model.
Table 4-1. The model structure
95
96
Table 4-2. The summary of the definition and the baseline values of all parameters (Sakagami and Fukuda, 1968, Winston et al., 1981) in the honey bee stage-structured population model.
Brood dynamics
The brood component of the model is largely driven by the daily rates of queen egg
production, which is determined by the potential egg production of a healthy queen, the empty
hive space and the available nursing workforce. The potential queen’s egg laying is based on an
empirical function (McLellan et al., 1980):
Rt= τqhatbect
97
where Rt is the number of eggs laid on day t; e is the base of natural logarithms; t is the
time in days; and a, b and c are coefficients affecting the egg production curve shape; and τqh is
the coefficient ranging from 0 to 1.0, representing the influence of the age and health of the queen
bee. If there are sufficient space and nurse bees, this function is the primary factor determining
egg laying rate. If the number of nurse bees in the hive does not meet the optimal nursing
requirement of egg laying potential, the actual number of daily egg-laying will follow the lower
bound of the available cells and the number of eggs corresponding to the available nursing in the
hive.
For this model, we aim to characterize the population dynamics of worker bees and
neglect the potential influence of drone bees. Therefore, only eggs developing into the worker
brood are considered. The surviving 2-day-old eggs develop into the larval stage with a through-
stage transition probability τel. The developmental period of worker brood includes 7 days for
larvae and 16 days for the pupa stage. Therefore, the total number of brood stage is equal to the
number of brood already present in one stage, plus the number entering that specific day
transiting from the previous stage to the current, minus the number of brood developing into the
next stage and brood lost due to mortality factors. Brood cannibalism and early capping of old
larvae performed by middle-aged bees have been often noted in the colony in need of pollen, or
with underfeeding of brood and/or lowered brood-nest temperatures (Schmickl and Crailsheim,
2004). It can impact the colony age demography in a delayed way, eventually influencing age
task allocation (Schmickl and Crailsheim, 2004). Therefore, two categories of mortality factors
for larval stages were computed in this model: the intrinsic factor, which is the time/environment-
independent mortality rate in a healthy colony; and the cannibalism factor, which is driven by the
pollen status of the colony and the nursing quality in the hive. For simplicity, the rate of larval
cannibalism is assumed to not vary with the specific age of larvae. The pollen status is expressed
by the level of the current pollen stores versus the overall colony demand. The nursing quality is
98
the level of current nurse bee population in relation to the total nursing demand of all brood
stages. These two influencing factors are the critical feedback systems that regulate the colony
age demography to adapt to inadequate nursing quality and insufficient food stores, thereby
maintaining the inner nest homeostasis (Figure 4-1).
The pupal stage is relatively stable and its mortality rate is driven consistently by intrinsic
factors (i.e. the virus/disease loads, and the nutritional status during the rearing period); because
they are capped in each cell with little exposure to the external environment (Winston, 1987). The
intrinsic mortality rates for bee brood at larval and pupal stages are derived from the published
life table data from a healthy colony (Sakagami and Fukuda, 1968).
Adult bee dynamics
Two adult worker bee types are suggested to exist in the field: short-lived bees with a
mean longevity of 60 days counting from the first day of egg, which range from the early spring
to late October. Long-lived bees survive through the winter with a mean longevity of 150 days
(Winston, 1987). Therefore, the worker bees are modeled separately for field-season bees and
winter bees. The surviving cohorts of 26-d-old pupa emerging from the cells as adult bees with a
transit probability of τpn, female workers undergo a sequence of behavioral development in an
age-defined fashion. We only consider three types of behaviors characteristic of the specific ages
in the field-season bees: brood nursing performed by nurse bees, hive maintenance and food
receiving by house bees, and food collection by foragers. The cohort of foragers is further
subdivided into pollen collector and nectar collector. The adult bees’ mortality rate is
demonstrated to be task-specific; and in this model, the stage-specific mortality is concluded as
age- and density-independent. It has been demonstrated that increased brood-rearing shortens
nurse bees’ life span, which can be explained by the negative influence of nursing on the
99
vitellogenin titer of the in-hive workers (Amdam et al., 2009). In order to incorporate the dynamic
regulation of nursing, the daily mortality rate of nurse bees is computed by the baseline mortality
rates weighted by the current nursing workload, which is the level of optimal nursing workforce
required by the current brood population in relation to the available nurse bees. The increase in
the physical exhaustion by brood rearing activities can cause an increase in nurse bee’s mortality.
The mechanisms introduced here capture how the quality of brood maintenance in the colony can
regulate the balance of the brood population and nurse population. For older house and forager
bees, the mortality rate is simply considered as constant, regardless of the changing environment.
With the stage-specific mortality rate and the daily number of newly emerging bees, the daily
adult bee demography can be projected through the model for each day.
The winter bees are classified into four stages: egg (age 1 to 3), larvae (age 4 to 11),
pupae (age 12 to 26), and adult bees (age 27 to 150) (Sakagami and Fukuda, 1968). During the
winter, the bee population dwindles and consumes honey without the possibility of foraging. The
winter loss is generally considered as the combined result of the seasonality in the cycle of bees
and the adverse effects of pests and parasites (Schmickl and Crailsheim, 2004). However, in this
model, we assume that the stock of winter bees is instantly lost if the stocks of honey stored in the
hives reaches zero before the end of the winter. The matrix for the intrinsic demographic
parameters including constant stage-specific survival rate, queen fertility rate and the transition
matrix for each stage can be derived from life tables of honey bee populations in published
literatures (Sakagami and Fukuda, 1968, Winston et al., 1981) using Matlab software.
Pollen dynamics
The allocation of the foraging force, especially pollen foraging, is critical in regulating
colony growth and development. Several hypothetical mechanisms have been proposed for pollen
100
foraging regulation including: the stimulus-response threshold hypothesis, which is the quantities
of stored pollen and brood can directly and independently regulate the pollen foraging, given the
fact that returning pollen foragers have direct access to brood and stored pollen (Ricarda et al.,
2004); and the brood food hypothesis, which is the indirect regulation of brood and stored pollen
on pollen foraging through a single inhibitory signal distributed through trophallaxis interactions
between nurse and forager bees (Sagili and Pankiw, 2007). Both the direct and indirect
hypotheses predict a similar outcome; that is that the pollen foraging is regulated through stored
pollen and brood population to a homeostatic set point. We therefore simulated that pollen
dynamics is driven by the continuous variation of the colony need for pollen. The pollen need
increases with increasing amounts of brood and nurse bees, as well as decreasing quantities of in-
hive pollen stores, consequently stimulating the recruitment of pollen foragers and resulting in the
increased pollen collection. The pollen demand is expressed as the sum of the daily pollen
consumption of brood and nurse bees and the pollen reserves based on the current pollen demand
for maintaining the colony with sufficient pollen source for rainy days and food dearth. A
colony's ability to increase the proportion of pollen foragers is suggested to be a consistent
mechanism of increasing pollen intake to satisfy the demand for protein in the colony. Therefore,
in the simulation model, honey bee colonies respond to changes in the quantity or quality of
pollen reserves by changing the proportion of pollen foragers in their foraging populations,
without altering the qualitative parameters of individual pollen foraging, such as the amount,
diversity, and weight of pollen load collected during each foraging trip, as well as the number of
trips made each day. We thus calculate the daily recruitment of pollen foragers by dividing the
colony-level demand of pollen by the average amount of pollen collected per forager per day.
Moreover, considering the possibility of the genetic predisposition of adult honey bees for
hoarding pollen despite the sufficient in-hive pollen reserves (Seeley, 1995), thus a constant
proportion of forager population is set as the lower bound for the pollen forager population. With
101
the daily recruitment of pollen forager population and daily pollen consumption by bee brood and
nurse bee stages, the net pollen storage can be calculated. In our model, parameters representing
pollen forager activity and pollen consumption were obtained from published experimental data
(Table 2) (Seeley, 1995, Rortais et al., 2005).
Honey dynamics
The honey compartment is determined by the nectar foraging behavior and the honey
consumption by workers and brood. Nectar foraging is a particularly well-studied task of the
honey bee colony (Camazine and Sneyd, 1991, Dyer, 2002). Nectar foraging is considered as a
collective decisive-making process, which needs the interaction of both in-hive bees (or nectar
receivers who process and store the nectar within the hive) and the nectar foragers (who are
knowledgeable about and working in the external environment). In the honey dynamics of this
model, we incorporated a simplified ordinary differential (ODE) model proposed by Edwards and
Myerscough (Edwards and Myerscough, 2011). They modeled the recruitment of an active nectar
forager population determined primarily by two factors: the quality of the nectar outside the hive,
and the time returned foragers spent searching for an available nectar receiver in the hive. We
assume that foragers show constancy in forage choice and the switch between nectar and pollen
foraging will not occur. Therefore, with the fixed forager population, the increase in active pollen
foragers will result in decrease in the upper bound of the nectar forager population. Given a
constant time of nectar foraging during each field day, the daily dynamics of nectar forager can
be computed numerically at the second time scale. Each nectar forager can make an average of 10
foraging trips per day and the mean nectar load collected each trip is about 85% of the body
weight of a bee, approximately 82 milligrams, which is equivalent to a 0.164 cellfull load (Fewell
and Winston, 1996). The daily nectar harvest is a function of the number of recruited nectar
102
foragers per day, and the average nectar collection per forager is the slope. Therefore, the nectar
ODE model (see equations at Table 4-3, and parameter values at Table 4-4) can be adapted to
execute implicitly with the stage structured population model in two ways: 1) the stage model at
the day timescale produces the values of two variables: the daily number of house bees, as the
initial condition for the active nectar receivers available at the beginning of each day; and the
daily forager population and the genetic priority for pollen foraging, as the limit for nectar
foragers population each day; and 2) in turn, the nectar model at the second timescale can return
the total amount of nectar collected by the end of day. With the daily consumption of honey and
the constant ratio of processing nectar to honey, the net storage of honey can be calculated.
Table 4-3. The four ordinary differential equations model of nectar foraging dynamics (at second time scale) (Edwards and Myerscough, 2011).
103
Table 4-4. The summary of variables and parameters in the differential equations of nectar dynamics (Edwards and Myerscough, 2011).
104
Numerical Simulation
The honey bee population simulation model was programmed in MATLAB. To initialize
the model and begin a simulation, days of simulation and initial state variables were assigned
according to empirical data for a healthy honey bee colony (Table 4-2). Once the model was
initialized, the dynamic simulation proceeded from day to day numerically. Model output for
each day included a stage-structured matrix and population densities of various life stages, the
daily dynamics of food sources in the hive including pollen and honey storage.
Sensitivity Analysis
Different sensitivity analysis techniques were carried out to determine if the model
resembles the honey bee system and how the variation in the output of the model can contribute
qualitatively or quantitatively to changes in model input factors. For this model with a large
amount of uncertain input factors, therefore, at an initial phase, a Morris screening method was
used to qualitatively determine which input factors may be considered to have effects, which are
(a) negligible, (b) linear and additive, (c) non-linear or involved in interactions with other factors
(Morris, 1991). This method is based on computing for each input Xi, =1,.., k (where k is a
number of model inputs) across p selected levels, called elementary effects (EEs), which are then
averaged to assess the overall importance of a given input factor. All factors are assumed to be
uniformly distributed in [0,1], and then transformed from the unit hypercube to their actual
distributions. EEs are calculated by varying one parameter at a time across a discrete number of
levels (p) in the space of input factors (k-dimensional p-level grid):
EE Xi = Y X1,…, Xi-1, Xi+∆, Xi+1,…,Xk -Y Xi
∆
105
where: EE(Xi)- elementary effect for a given factor Xi, ∆ is a value in [1/ (p-1), …, 1-
1/(p-1)], which defines a change in the parameter distribution between two levels of single input
factor. Two global sensitivity indices are adopted in this test: the mean of distribution of absolute
values of EEs, µ, which estimates the overall effect of the parameter on a given output; the
standard deviation of EEs, σ, which estimates the higher order characteristics of the parameter.
These indices are plotted on a µ-σ Cartesian plane. The higher the value µ is, the more important
factor is. The higher the value σ is, the stronger influence of the values of other input parameters
have.
For non-linear, non-montonic models, the Extended Fourier Amplitude Sensitivity Test
(EFAST) was used to measure what fraction of the output variance can be explained by variation
in each parameter. All input parameter are defined with sinusoidal function, composing of a
multi-dimensional search curve: x = f (j), j = 1, 2, …, NS, that assigns a value to parameter x
based on the sample number 1 through the total number of samples per search curve, NS. Here,
we used the recommended value 65 for NS (Marino et al., 2008). Fourier analysis is then used to
partition the variance of model output caused by the strength of each parameter. A resampling
scheme is implemented for more efficient parameter sampling by: generating different search
curves by introducing a random phase shift into each sinusoidal function; then repeating sampling
NR (the resampling size) times with different search curves; and finally taking the arithmetic
means over the estimates. The total number of model simulations N, is given by N =NS × k × NR,
where k is the number of parameters analyzed. Two global sensitivity indices are adopted in this
method: 1) a first-order sensitivity index, Si, which is calculated as the variance at a given
parameter i’s unique frequency divided by total variance: first, Si2 is calculated from the Fourier
coefficients A, B of the specific input factor at the frequency of interest, j:
Si2=2 Aj
2+Bj2 , where Aj=
1π
f x cos jx dx, π-π Bj=
1π
f x sin jx dxπ-π ,
106
then, the first-order Si is calculated as a fraction of total variance:
Si=Si2/Stotal
2
2) the total-order sensitivity index, STi, which is calculated the remaining variance after the
summed sensitivity index of the entire complementary set of parameters (i.e. all parameters
except i) using their identification frequencies (Marino et al., 2008). All the sensitivity tests were
all implemented in Matlab environment.
Results
Simulation Results of a healthy honey bee colony
Numerical simulation of a healthy colony (Figure 4-3) shows seasonal patterns of the
population of brood and adult honey bees, and the storage of pollen and honey. This model
simulates a honey bee hive in northern temperate areas. Therefore, it is assumed that the queen
initiates the egg production in response to spring nectar flows, which generally begin in early
March in the northern temperate zone.
107
Figure 4-3. Numerical simulation of population dynamics of honey bees at six stages, pollen stores (in cell), honey stores (in cell), and egg production rate of a simulated healthy honey bee colony throughout two years.
Within each year, each stage of bees varies significantly over time on a seasonal
timescale. The maximum egg laying occurs in early June, approximately 2,000 eggs per day
(Table 4-5). The adult bee population hits the maximum peak at the middle of each July, which
approximately 22,000 bees present in the hive. The brood population achieves its maximum at the
middle of each June, which approximately has 36,000 pre-adult bees in the hive. Since July,
brood population starts reducing until becoming broodless in January. During the field season,
foragers collect honey and pollen to maintain colony growth and development, with a maximum
value of 160 and 0.6 kg, respectively. A healthy colony can enter into the winter with
approximately 8,000 bees, 37 kg honey and 0.1 kg pollen. The winter colony will achieve the
maximum size in the middle of November, about 8,900 bees in the hive (Table 4-5). The colony
will resume brood rearing quickly in the late winter if it has sufficient pollen and honey storage.
In March of the second season, most adult bees will become forager bees and start collecting food
immediately.
108
Table 4-5. Summary of simulation results for a healthy honey bee hive located at in the typical northern temperate region.
Population Dynamics
Time Of the Year Colony Component Maximum Population Peak Timef
Field Seasona Egg 5844 June
Broodc 30769 June
In-hive Adult Beesd 22309 July
Foragers 8888 August
Winterb Winter Bees 8339 December
Food Storage (kg)
Field Seasona Honeye 161 May
Pollen 0.6 July
Winterb Honeye 37.4 November
Pollen 0.1 November
a Field season is considered as the time of continuous nectar flow in the environment. Here, it is counted from each March to November, including spring, summer and fall seasons. b Winter season is considered when there is no nectar flow in the environment. Here, it is counted from each December to February. c Brood includes open (larva) and capped brood (pupa) in a hive . d In-hive adult bees includes nurse and house bees in a hive . e Honey represents both open nectar and capped honey storage in a hive . f Peak time depends on the beginning of nectar flow in the environment and the time that takes the specific colony component to reach its maximum in a healthy honey bee colony.
Model Validation simulation
The time series of simulation output of population and food dynamics in a healthy colony
were validated with the historical data from six different peer-reviewed literatures and the field
109
data from our field experiment (Figure 4-4). The published adult bee and brood population data
were collected from healthy hives, which contain disease-free honey bees with queens less than
one-year-old. The actual data were compared with the simulated maximum, minimum and
optimal number of honey bee population in a healthy colony, of which the parameters values are
selected based on the optimal, normal and lowest threshold conditions (Table 4-2). The egg
laying rates from the model produce a good fit to the observational data from colonies located in
northern temperate areas. The dynamics of the pre-adult stages, total worker populations also
closely resemble the output form real healthy hives (Figure 4-4).
110
Figure 4-4. Model validation by comparing simulated healthy honey bee population dyanmcis with the historical colony data. Grey line shows intracolonial population dynamics of adult bees (a) and brood population (b) in our simulated honey bee colony throughout a year (maximum, minimum and normal population). Historical colony data from Bretschko (Bretschko, 1985), Bühlmann (Bühlmann, 1985), Fukuda (Fukuda, 1983), Omholt (Omholt, 1986), Bodenheimer
111
(Bodenheimer, 1937), and Kunert and Crailsheim (Kunert and Crailsheim, 1988) are normalized to the maximum median value of the model to allow comparisons of the seasonal dynamics.
The simulated dynamics of pollen storage also mimic the actual seasonal pattern of
pollen stores (Figure 4-5) (Jeffree and Allen, 1956).The problem with validation of honey
dynamics is that there is no reliable or accurate measurement of the amount of honey stores
throughout the season. Regular uniform weighing of honey bee colonies can generally be used to
indicate the potential honey storage of a hive, therefore, the honey dynamics here were validated
by comparing the simulated colony weight data with the historical weight data documented for
twenty years (Milum, 1956). The daily colony weight data was calculated by considering the
number of adult bees and the brood population, pollen and honey stores (Figure 4-6).
Figure 4-5. Model validation by comparing simulated healthy honey bee pollen dynamics with the historical pollen data. Green dashed line shows intracolonial pollen dynamics in our simulated honey bee colony throughout a year. Red solid line shows empirical pollen data from Jeffree and Allen (1956), allowing for comparisons of the seasonal dynamics.
Further, in order to validate our model in predicting the colony dynamics under
environmental stressors particularly pesticides, we performed a field experiment to measure
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
Pol
len
Sto
res
(cel
ls)
x103
Pollen dynamics in a healthy honeybee colony Field Data (Jeffree et al.,1956)
Predicted data
112
colony weight changes after common pesticide exposures at environmentally relevant
concentrations. We adjusted the baseline matrix model by including the significant reduction in
stage-specific vital rates (queen egg laying rates and adult bee survival rates) resulting from
pesticide exposures. The significant changes were only found in queen’s egg laying rates (data
not published), which were used as the input parameter R(t) to represent the disrupted colony
scenarios. The simulated colony weights during the field season also match closely with the field
data of controlled healthy colonies and pesticide-exposed colonies, respectively (Figure 4-6).
A)
113
B)
C)
Figure 4-6. Model validation by comparing the simulated honey bee colony weight changes during the field season with the field data under controlled healthy and realistic pesticide exposures. Black dotted lines with different markers show the actual colony weight data (A-
114
healthy colony, B-fungicide-exposed colony, C-dimethoate-exposed colony as positive control). The red line shows the simulated mean weight of honey bee colony during with the historical pollen data throughout the 2012 field season.
Sensitivity Analysis
Figure 4-7 shows the graphical representation of Morris sensitivity measures for model
outputs of annual population growth rate. Parameters separated from the origin of the µ-σ plane
are considered important. Parameters located at the origin of the plane are assumed to have
negligible effects on model outputs. Factors u, v, st4, associated with the behavioral transition
between nursing and foraging, are important for colony population growth. The most critical
factors are u and v, which directly determine how much the abnormal behavioral transition
between nurse bees and foragers is shifted. While other factors associated with queen egg
production, the survival probability of bees at other stages except nurse bees and the quality of
nectar foraging, are identified as not individually important for model outputs, they could
potentially interact with other parameters to influence the colony growth.
115
Figure 4-7. Morris screening results for model outputs of colony population growth rate during the first, second and third years. The mean of distribution of absolute values of EEs, µ, estimates the overall effect of the parameter on a given output; while the standard deviation of EEs, σ, estimates the higher order characteristics of the parameter. The higher the value µ is, the more important factor is. The higher the value σ is, the stronger influence of the values of other input parameters have. Factors u, v, st4 (nurse bee survival rate), associated with the behavioral transition between nursing and foraging, are important for colony population growth.
A variance-based sensitivity analysis has been performed on this non-linear model, using
the extended Fourier amplitude sensitivity test with 1040 model evaluations, which is the least
number of iterations required by eFAST for a model with 16 input parameters. The first order
sensitivity indices for population growth rate during the first field season (240 d), presented in
Table 4-6, indicate that v and u contribute with 82% and 16% of the output variability,
respectively. The rest of the input factors contribute with less than 3% in total. For the total
indices, v is still the most important factor, contributing 78% of the output variability. The u
contributes 15% and the rest contribute less than 8%. The results of first and total order effects
obtained for population growth rate at the second and third field season all lead to the same
conclusions: that v is the most important input factor, followed by u, while the importance of
other input factors is almost negligible. All the first order indices do not sum up to 1, thus, there
are interactions between input factors accounting for the unexplained parts. The sensitivity
measures for the three-year dynamics of pollen stores were similar to results obtained for
population growth rate: factors v and u are prominently important for the variations in pollen
stores during each field season (Table 4-7).
Table 4-6. Sensitivity indices for individual and social parameters on honey bee population and their corresponding ranking obtained with EFAST.
116
Table 4-7. Sensitivity indices for individual and social parameters on pollen dynamics and their corresponding ranking obtained with EFAST.
Indi
vidu
al F
acto
r So
cial
Fac
tor
117
With the two sensitivity analysis methods, the results all indicate that the probability of
disruptive behavior during nurse-to-forager transition is critical in determining colony growth and
food dynamics. To further confirm the robustness of these global sensitivity results, we
performed two additional local analyses to measure how sensitive the population model is to
small perturbations of these two most important input factors. First, the local sensitivity analysis
was used to simulate how the change in the probability of precocious or reverted nurse-to-forager
transitions at the individual bee level can progressively influence the colony-level dynamics.
Figure 4-8 shows the simulated colony-level impacts of two factors: the single and combined
individual level factors- reduced egg laying rate when queen bee is exposed to fungicide-
contaminated pollen diet in the field (50%) and reduced survivorship at the larval stage (50%)
when larval bees are continuously exposed to fungicide-contaminated brood diet in the in-vitro
rearing; the social factor-the local threshold probability of precocious foraging each day (9% in-
hive bees at any age within nurse bee stage prematurely develop into forager directly).
Remarkably, only with a low probability of precocious foraging, the colony would suddenly
collapse within the first field season and lack of honey stores; whereas, even if either the queen
egg production or larval daily survival is reduced by 50%, the simulated colony only decreases
population size by over 2-fold and food stores. These comparisons reemphasize the sensitivities
of colony and food dynamics to the disruptive nurse-to-forager transition. Given the current
findings of extensive occurrence of pesticides and multiple residues within the hive matrices
including brood nest wax, foundation, beebread, trapped pollen, adult bees and brood(Mullin et
al., 2010), it is therefore highly likely that pesticide impacts can happen in all aspects of colony
life such as the queen’s egg laying rate, larval survival and adult bees’ longevity. So we looked at
the combined effects of individual factors on colony dynamics (Figure 4-8 C.). The simulated
colony would collapse before winter lacking of enough population size for overwintering, even
though the colony has enough food stores.
118
A.
B.
119
C.
D.
Figure 4-8. Local sensitivity analysis of colony dynamics under the potential disrupted scenarios: A. the reduced egg laying rate (field experimental results: 50% reduction per day when queen exposed to the fungicide-contaminated pollen diet); B. the reduced larval survival rate (larval rearing experiemntal results: least 50% reduction at the larval stage when exposed to the realistic exposure of a common fungicide chlorothalonil; C. the simulated combined effect of reduced egg
120
laying rates (50% reduction) and larval survivorships (50% reduction); D. the 9% precocious foraging behavior.
Subsequently, we investigated how a healthy colony performs division of labor,
especially nursing and foraging. Figure 4-9 depicts the range of the ratios between the sum of
brood and foragers versus in-hive bees in a healthy colony that changes over the three-year field
seasons. The ratio first displays dynamical decay in the summer (exponential decay in the first
summer, multimodal decay in the second and third summer seasons), but finally all attempting to
stabilize within the range of 0.83 and 1 during the fall (from September to November). This result
indicates that a healthy colony is capable of allocating the workforce of in-hive bees via a
dynamic equilibrium of bees switching between nursing and foraging tasks with respect to
variations in task demand. Moreover, our results suggest the end of field season might be the
most sensitive stage in deciding the colony fate.
Figure 4-9. The ratio of forager and brood population versus in-hive bees population in a healthy honey bee colony (maximum-dashed line, minimum-solid line) throughout the field seasons over three years.
121
Results from all sensitivity tests suggest that, the key to colony survival may lie in the
division of labor, particularly nursing and foraging in a hive. Reducing the variability in abnormal
worker transition can make a significant difference between population stability and decline;
however, other parameters including daily variation in bee survival at different stages, queen egg
laying rate and nectar foraging efficiency does not make much of a difference to the overall
population trend.
Discussion
Model Validation and Comparisons
Our model reproduces the historical data of honey bee population and food dynamics
under a healthy condition based on its precise predictions of independent data sets taken from the
literature and our control colonies from our own experimental studies. It also closely reflects the
actual colony weight change under pesticide exposure at environmentally-relevant concentrations
from our own experiments. This is the first application of our stage-structured matrix population
model applied in modeling honey bee populations under various field situations. It allows for
testing how subtle changes at the individual level by multiple stressors can result in substantial
outcomes at the colony level, in stepwise increments. The realistic prediction of pesticide impacts
in the field demonstrates the wide applicability of this model, and gives some precedent to apply
this model for investigating scenarios of pesticide exposures by combining short-term laboratory
measurements of individual-level processes and then validating with field level experiments
Several honey bee population models have been developed with different foci using
various mathematical modeling approaches. Attempts to model the colony dynamics were begun
122
by Ebert (1922), who first divided the colony into 3 subgroups (juveniles, hive bees and foragers)
rather than considering the entire population as a single entity. Based on the improved
understanding of the consecutive division of labor in a social bee colony, several models,
including Nolan (Nolan, 1932), Bodenheimer (Bodenheimer, 1937, Bodenheimer and Ben-Nerya,
1937), and Fukuda (Fukuda, 1971), were developed subsequently to refine Ebert’s model;
however, these models were largely generated by directly measuring honey bee population with
experiments and field observations.
The formal mathematical modeling approach was first implemented by McLellan
(McLellan et al., 1980) and Harris (Harris, 1985), using linear regression models and this was
able to generate realistic predictions. These models required the actual natality and mortality rates
generated from the colony it was trying to estimate. Dolejsky and Schley (Dolejsky and Schley,
1980) were the first to study the populations of bees and Varroa mites using a simple differential
equations model with time-dependent parameters. This modeling approach allows for observing
the colony-level influence of a simulated reduction of mites.
Omholt (Omholt, 1986) built the first dynamic model incorporating a feedback
mechanism with regard to the queen’s egg-laying rate. The model considered a worker-density
related ‘switching term’ in determining whether the influence on queen egg production from the
workers is inhibitory or stimulative. Due to the paucity of quantitative experimental data and the
computational cost, this model did not consider the extra-colonial factors such as external
temperature and field conditions. The more complex intracolonial factors such as cannibalism,
pheromone-based interactions, and division of labor have also been disregarded in this model.
DeGrandi-Hoffman et al. developed the BeePop model, which concentrates on the queen’s egg
laying and the derived age-based population structure (DeGrandi-Hoffman et al., 1989). BeePop
is the first model designed for researchers and beekeepers to conduct “what if” scenarios on the
intracolonial dynamics by providing for user-specified parameter values. They extended
123
Omholt’s model on the queen’s egg laying by introducing the impacts of extracolonial factors
including the daily temperature and the photo-period; however, most of these models simulated
the division of tasks purely based on the fixed age of the bees. They ignored the influence of
external factors (weather, foraging resource) and inside nutrition-related feedbacks in regulating
the flexibility of task allocations in the bee colony.
The application of individual-based modeling promoted a new level of understanding of
honey bee population dynamics. The object-based model “AHBsim” developed by Makela (1992)
incorporated the external (resource availability and other physical factors in the environment) and
internal genetic factors (the genotype of the queens and drones) in estimating the birth, growth,
reproduction and death processes of a colony (Makela et al., 1993). They also were the first to
introduce the priority-based system in colony task allocation, with the assumption that there are
no age requirements for job performance in a honey bee colony. The recent “HoPoMo” model
leads to an outstanding improvement in modeling the detailed aspects of colony dynamics. It is
the first in the literature that incorporates the important feedback loops that link pollen supply and
nursing workforce to brood cannibalism, and to division of labor (Schmickl and Crailsheim,
2007); however, this priority-based division of labor does not fully reflect the colony
organization, particularly with regard to nectar foraging behavior. It assumes two tasks of high
priority: brood nursing and pollen foraging, expressed by assigning specific ratios of workforce
(the available adult bee population) to workload (the least number of adult bees required by
nursing and pollen foraging). Both these agent-based models consider that nectar foraging has the
lowest priority.
A key feature of social organization in social insects involves its behavioral plasticity.
Previous population models did not incorporate any informational pathway regulating the colony
task allocation, partially because the mechanisms of behavioral flexibility of honey bees are not
clearly defined until now. It has been suggested that social organization is the result of the
124
integration of acquired internal and external information, coupled with behavioral biases
associated with worker genotype, temporal-specific castes, physiological status, and prior
experience (Johnson, 2010). Previous honey bee models have not incorporated any informational
pathways that the colony task allocations are based on, partially because the mechanisms of
behavioral flexibility of honey bees has not been clearly defined until recently (Mutti et al.,
2011).
Currently all honey bee population models concentrate on queen egg production and
fixed temporal polyethism, but not the dynamic of high-priority-based task divisions between
brood care and pollen foraging. Comparatively little attention has been given to middle-aged bees
or house bees, which together with external factors govern the rate at which foragers become
committed to nectar collection (Camazine and Sneyd, 1991, Seeley, 1995). Our model, therefore
is unique in combining the temporal-based caste differentiation with three dynamic feedback
systems involving task distribution and resource dynamics including: 1) the interactive regulation
between nursing workload with the survival of brood and nurse bees; 2) the homeostatic
regulation of the recruitment of pollen foragers via nursing workload and pollen stores; and 3) the
dynamic interaction between house bees (middle-aged bees), nectar stores and nectar quality in
the environment. For the demographic structure and age-based division of labor, we took a
bottom-up approach using a stage-structured population model with difference equations, to
simulate detailed aspects of colony dynamics on a daily basis computationally. For the dynamic
feedback system, particularly nectar foraging, we used a different top-down approach with an
ordinary differential equations model to determine the colony’s commitment to nectar collection
at a fine timescale.
No single modeling tool would be able to address all interrelated components in a social
insect colony. The fusion of top-down and bottom-up approaches at different timescales in our
model results in significant advances in providing more refined predictions and thus
125
understanding of honey bee population and resource dynamics. It is also flexible enough to deal
with many of the potential genetic and environmental influences regarding: queen’s health such
as time of initiation of egg laying, maximum laying value, and the rate of attaining the maximum
laying rate; the stage-specific survivorship of bees; the brood rearing efficiency expressed by the
larva-to-nurse ratio; and the potential of developmental disorder, including the rate of accelerated,
delayed, precocious, or reverted maturation.
The Critical Nurse-to-Forager Transition
The detailed sensitivity analysis and further numerical simulations of this complex
population model allowed us to eliminate a number of candidate hypotheses to explain the colony
decline. The use of variance based sensitivity analysis (EFAST used here) can accommodate non-
linearity and interactions within model variables after using the computationally efficient Morris
screening technique. The ranked results from the Morris and EFAST screening techniques
showed close similarity on influential variables. The sensitivity indices were dominated primarily
by the probability of precocious or reverted transition between nurse and foragers. Within a
biologically reasonable range, only increases in the probability of disruptive transition between
nursing and foraging proved influential enough to cause the observed colony collapse during a
single field season. Decreased queen egg production, accelerated/delayed maturation, reduced
stage-based bee survival, insufficient brood rearing, and decreased efficiency of nectar foraging
were all comparatively ineffective at interrupting the stability of this social system. Numerical
simulations also suggested a balanced ratio between brood, foragers and hive bees in the late field
season, which in general is the preparing period for the coming winter season, which could serve
as a direct indicator of colony health. While the current model does not suggest what exactly
causes colony collapse disorder, model simulation of the disrupted nurse-to-forager transition
126
does capture several key features observed in cases of CCD: rapid population decline before the
winter and the abandonment of brood by adult bees. Whether the failure of division of labor
beyond the threshold value could occur in natural colonies is not clear, but our modeling results
indicate that a balanced allocation of workers with respect to dynamic changes in colony task
demands is the key to sustaining colony survival. Sensitivity to changing environmental and
colony conditions within a structured labor system is crucial to social organization, but
underlying mechanisms of this behavioral plasticity are not thoroughly understood; therefore,
whether the change in nurse-to-forager transition is a consequence or a contributing cause of
colony collapse disorder is still an open question.
A hierarchically structured regulatory network integrating genomic, physiological,
behavioral components, and social context have been hypothesized to explain the complex
phenotypes and its considerable flexibility in the honey bee colony (Johnson, 2010). The major
components of this regulatory network are the nutrition-sensing pathways, including IRS (insulin
receptor substrate), TOR (target of rapamycin) and Egfr (epidermal growth factor receptor-
mediate the reaction to royalactin, which is the key protein component of royal jelly), as well as
the reproductive-related protein vitellogenin. These components can act interactively or
independently to converge onto one central downstream signal integrator juvenile hormone (JH)
to regulate the caste differentiation (Mutti et al., 2011). In addition, the potential social pathway
that links the interactions between developing brood and nurse bees has been proposed as a
hypothesis that larval nutritional condition, signaled by glandular secretions such as brood
pheromone, can act through the nurse bees’ neurosensory system and up-regulate the brood-
rearing behavior. Perturbations in either pathway are sufficient to alter the caste development
(Edwards and Myerscough, 2011); for example, experimental colonies treated with JH, JH mimic,
or JH analog (Robinson, 1987), or insulin treatment (Mott and Breed, 2012), or artificial
suppression of vitellogenin (Nelson et al., 2007), or primer pheromone treatment (Leoncini et al.,
127
2004) were found to modulate the normal rise in juvenile hormone and result in either precocious
behavioral development (marked by the significant acceleration of foraging onset by 2 weeks
earlier) or retarded development (the continuation of brood care despite increasing chronological
age).
Forager reversion is not as well documented as is precocious foraging and retarded
behavioral development; however, Hrassnig et al. found that older worker honey bee s in colonies
faced with a severe shortage of nurse bees can revert to that task (Crailsheim and Hrassnigg,
1998). The mechanism for this behavioral reversion is unknown, but Robison et al. proposed that
JH plays a similar role in mediating plasticity in both normal behavioral development (from
nursing to foraging) and retrodevelopment (from foraging to nursing) with low JH titers,
triggering the reverted nurse bees from foragers (Robinson, 1987). A recent epigenetic study
showed substantial and reversible DNA methylation changes corresponding to the reversible
behavioral phenotypes of nurse and forager subcaste in honey bees (Herb et al., 2012). This is the
first study suggesting the subcaste-species DNA methylation signature can assist in forming
worker phenotypes. Though the complete picture of the regulatory interactions between genomic,
cellular, physiological, and behavioral factors involved in the development and evolution of
social phenotypes have not yet been established; the current understanding of neurochemical
pathways suggests the importance of nutrition in triggering behavioral and physiological changes
in the worker honey bee when transitioning from in-nest tasks to foraging. Together with our
modeling results showing the extreme sensitivity of colony fitness to the nurse-to-forager
transition; it seems plausible to speculate that the inadequate nutrition in either quality (pesticide
contamination) or quantity (lack of food) level can disrupt single or multiple components of the
regulatory network involved in social caste development, eventually resulting in an irreversible
alterations in colony homeostasis.
128
Unfortunately, our understanding of how essential nutrition is required for the growth
and development at the individual bee and colony levels are still limited. Carbohydrates, proteins,
free amino acids, lipids, vitamins and minerals are the major nutritional categories to honey bees,
available from floral nectar and pollen. Given that honey bees usually forage in the proximity of
landscapes used extensively for agriculture, bee nutrition is likely to bear many potential threats
including starvation, reduced diversity of diet from monocultures, and pesticides brought to the
colony (Brodschneider and Crailsheim, 2010). For example, fungicides, which are frequently
detected in the hive matrices, are found to disrupt the composition and organization of the fungal
community critical in processing pollen into bee bread, thereby reducing its nutritional value
(DeGrandi-Hoffman et al., 2009). With a better understanding of the molecular mechanisms
underlying pollen's nutritive impact on honey bee health, we can determine whether and how
deficient nutrition affects colony health, ultimately developing "dietary-intervention” strategies to
help a stressed or weakened colony to retrieve its homeostasis. Therefore, the development of a
nutrient-mediated nurse-to-forager feedback module should be considered a high priority in the
future population modeling efforts for honey bees.
It is important to note that diseases can disrupt the phenotypic development of honey
bees; from the point of view of the colony stability, it is still feasible to state that the balanced
task allocation between nurse and foragers is a prominent feature of honey bee colony dynamics.
Further experiments should be designed to answer: (1) how the potential stressors from nutrition,
pesticides, and diseases impact the nurse-to-forager transitions in honey bees?, and (2) can the
colony system continue proper functioning if any negative stressors singly or simultaneously take
a toll on the nurse-to-forager transition in honey bees?
129
Ecological Risk Assessment
Due to the uncertainties embedded in the reductionist field experiments, attempts to
establish a causal link between an adverse ecological impact and exposure to toxic chemicals are
always troublesome and require skilled labor and great expense. Our honey bee population model
using an integrative modeling approach would be a cost-effective tool for estimating and
managing the ecological risks of potential stressors inside and outside the hive. Comparing model
simulations to our field experiments of pesticide impacts demonstrated our model’s potential in
ecological assessments and the guidance it can offer for design of further testing and validation.
Sensitivity analysis helps to determine which demographic parameters are critical to analyze for
colony-level risks. It provides a means of examining the influence of potential stressors on colony
population dynamics, not just the commonly used stage-specific mortality. Given the current
findings of widespread occurrence of pesticides and multiple residues within the hive including
brood nest wax, foundation, beebread, trapped pollen, adult bees and brood (Mullin et al., 2010),
the adverse effects of pesticides on honey bee colony health can potentially happen in all aspects
of colony life; however, most studies currently focus on studying the individual-level pesticide
effects. For example, studies have demonstrated that a prolonged persistence of two commonly
used in-hive miticides fluvalinate and coumaphos could negatively impact the queen reproductive
capability, as well as the survival and development of brood and their ability to properly develop
as adults (Haarmann et al., 2002, Pettis et al., 2004, Mullin et al., 2010). Integrating the various
exposures at different magnitudes, lethal and sub-lethal effects, different bees and their function
in the colony is a pressing challenge for assessing potential impacts of pesticides and other
stressors on the ultimate survival of the colony. Therefore, our colony‐level simulation model
represents a useful tool to integrate exposure and effects data of potential stressors with the
complexities of the social structure and biology of a honey bee colony. In addition, based on our
130
modeling results of the most influential role of nurse-to-forager transition in maintaining colony
homeostasis, the future risk assessments should include testing pesticide impacts on life-history
transition in honey bees.
Model Limitations and future extension
Given our current state of knowledge and colony-level data, the model was developed
initially based on simplified assumptions. For the nectar foraging module, our model assumes an
unchanging, predator and competition free environment with constant forager round-trip times
and single flower source containing nectar of constant quality. The foragers’ tendency to rest, to
return or to recruit was determined by two factors: the quality of the nectar and the time spent
searching for a receiver (Seeley, 1995). In the field, honey bee foragers will simultaneously
exploit multiple sources, each with different nectar and habitat quality. The quality of nectar at a
single source may also change over the course of a day due to changes in weather, inter-species
competition levels, or plant characteristics. Foragers can perform the waggle dance when
experiencing sources with high reward to encourage more foraging activity; while experiencing
long search time of house bees to unload nectar, foragers can discourage others from dancing for
the same source. Inside the hive, recruited house bees may also abandon their nectar-unloading
task and revert to other hive duties (Seeley, 1995). These processes currently are not well
understood over the timescale of a day; therefore, how colonies adapt to dynamic changes of
nectar resources in the environment should be elucidated further which will allow us to model the
dynamics of honey stores inside the colony.
Further, we did not test the factor of honey bee diseases such as those resulting from mite
infection, thus we cannot rule out individual or combinational impacts of bee diseases. The
131
addition of a honey bee disease dynamic module will be needed in order to further clarify its
colony-level impacts and interactions with other hive factors.
For the stage-structured module, our model is developed primarily based on temporal
polyethism. By assigning specific probabilities for accelerated, delayed, or even reverted
transitions within each phenotype, the plasticity during honey bee behavioral development can be
simulated in response to changes in colony conditions or environmental factors. With this better
understanding of the mechanism(s) in regulating these life history transitions and division of
labor in honey bees, a further extension would be to integrate the dynamic feedback of the
behavioral development at a finer timescale into the stage-structured honey bee population model.
Conclusions
As an aid to testing hypotheses for the causes of recent colony failure and for providing
suggestions for management actions to promote recovery of honey bee populations, we developed
a worker-based, stage-structured model of honey bee population dynamics. This model is unique
in capturing the adaptive feedback mechanisms in the population dynamics and resource
dynamics in a honey bee colony, including the comb pattern formation, brood maintenance, and
collective foraging behavior. By validating with various historical colony data and field
experiments, our model represents the most advanced population model for integrating the top-
down differential model and bottom-up difference model that gives the most refined and realistic
details of colony population and resource dynamics to date. Sensitivity analyses suggest the
importance of a balanced allocation of workers for nursing and foraging in sustaining the colony
survival and development. The change in task-based ratios may be a consequence or a
contributing cause of the colony collapse. Our model can be used for colony dynamics
forecasting, offering directions to hypothesis-driven research and for providing suggestions for
132
colony management efforts, as well as justifying the eventual extrapolation of the model to more
generic conditions.
Acknowledgements
This work cannot be done without the contribution of other two co-authors-Professors
Tim Reluga and James Frazier, who have provided constructive help and suggestions with the
model development and biological interpretations. I also acknowledge the help of Maryann
Frazier, Sara Ashcraft and Samuel Gruneberg, who have helped with the field colony experiments
at Wiley apiary during 2011 and 2012 field season, and collected the colony-level data from 36
colonies with different pesticide treatments. This work is funded by three research grants
approved by the National Honey Board, the Center for Pollinator Research and Californian
Almond Board.
Chapter 5
Conclusions
Summary
The significant decline in the population of managed European honey bee colonies has
been observed since 2007. Bees are sensitive and essential indicators of an intact environment
and are essential as a mainstay of crop pollination and honey production. With the increasing
awareness of the significance of honey bees for the maintenance of biodiversity and human well-
being, thorough research for the causative factors of colony losses has been conducted and an
extensive list of detrimental factors, both environmental and human-induced, including
pathogens, parasites and pesticides have been identified. Although no single factor has been
determined to be the sole cause of colony collapse, the combinations of multiple detrimental
factors has been suggested as the potential explanation for honey bee mortality. Rather than
seeking for the answers to what causes colony collapse, this study seeks instead to understand
what biological mechanisms would explain the healthy colonies with superior organization and
exceptional longevity. It has been suggested that the key to explain the exceptional organization
in a honey bee society is via the nutrition-sensing pathways, which direct complex phenotypes
and has considerable flexibility in the honey bee colony to maintain colony homeostasis and
promote its growth and development (Johnson, 2010). Therefore, we hypothesized that pesticide-
caused nutritional deficiency can singly or simultaneously take a toll on some stages of the honey
bee systems thus causing an irreversible disruption of the colony’s proper functioning. To
accomplish this, we took an integrative modeling and experimental approach: to characterize the
population dynamics of a honey bee colony thereby: to determine the critical stages of honey bee
134
life history in maintaining honey bee social homeostasis; to quantify the population-level effects
of nutritional disturbance at the quality level (pesticide residues in diet) by combining the
experimental data with the simulation model; and to make suggestions to design further
validating experiments and colony management techniques, particularly nutrient supplements,
towards the principal stages for honey bee conservation. Lastly, a knowledge-based expert system
has been developed to transform the knowledge of domain experts and the current quantitative
model into a fuzzy expert system. The expert system can provide a comprehensive narrative and
graphic assessment of honey bee colony health and offer a timely, and up-to-date decision
support for colony management to beekeepers and other stakeholders.
Implications of this Study
For the laboratory study: The modified in vitro rearing of honey bee larvae was used to
evaluate the individual and combinational impacts of four most common pesticides (fluvalinate,
coumaphos, chlorothalonil, chlorpyrifos) that have been found extensively in hive matrices,
particularly pollen samples. These pesticides were fed to the first instar larvae throughout the
larval development period and tested individually at environmental realistic concentrations and in
all possible combinations up to the four-component mixture. Firstly, it was found that: these
pesticides and the fungicide chlorothalonil interact in multiple ways that are both concentration
and mixture dependent; the mixture of fungicide and two commonly used miticides interacts
synergistically; there is a unique developmental period in larval development where pesticide
sensitivity is greatest; the fungicide chlorothalonil is far more toxic to larvae than to adult bees;
the common inert ingredient N-methyl-2-pyrrolidone tested at various realistic concentrations is
highly toxic to bee larvae. From this study results, we further investigated the impacts of three
fungicide formulations (Bravo, Nova, Pristine) widely used in bee-pollinated crops on larval
135
honey bee development. This study was also designed to determine whether the respective active
ingredients (chlorothalonil) are the main drivers of the chronic oral toxicity of environmentally
relevant levels of a common fungicide formulation (Bravo) to larval A. mellifera, or if the
formulation ingredients are also of major importance. The results based on the in-vitro larval
rearing bioassay provided clear evidences including: the significant oral toxicity of the
environmentally realistic levels of fungicide formulation and its active ingredient to early life
stages of honey bees; a complex and multiphasic dose–mortality response of honey bee larva to
chlorothalonil, which is significant different from the monotonic impacts caused by its
formulation Bravo; the higher toxicity of commercial formulations to larvae than the
corresponding active ingredient alone; and the synergistic interactions between binary
combinations of tested fungicides. While the mechanisms for interaction among pesticides with
diverse modes of action and its dynamics with the changing environmental scenarios are still not
known with accuracy, this larval study represents a starting point to investigate mixture effects,
formulation and inerts toxicity, and fungicide impacts on non-target pollinators. Our results
emphasize the importance of including the chronic and mixture toxicity studies at realistic levels
into pesticide risk assessment for non-target organisms. Together, given the extensive
documented contamination of hive matrices with both in-hive and agricultural pesticides, high
susceptibility of honey bee brood to common fungicides, miticides and insecticides at
environmentally relevant levels, nonmonotonic dose response of chlorothalonil, elevated toxicity
of fungicide formulations, and synergism of co-occurring fungicides and miticides, our results are
particularly relevant in wake of the colony collapse disorder. In the more complex milieu of this
social insect and its aging hive environment, pesticides, formulation additives and their resulting
mixtures may have greater long-term impacts on colony health. Future testing of impacts at the
colony level of individual and combinations of common pesticides, and linking of larval
responses to sublethal effects on later honey bee life stages are needed; in particular examining if
136
and how newly emerged adult honey bees after chronic exposure of pesticides during larval
development can be impaired at the behavioral level, such as the performance of in-hive tasks
(nursing efficiency and hygienic behavior), the behavioral transition or maturation (transition rate
from nursing to foraging behavior) , and field tasks (olfactory learning and foraging efficiency).
Furthermore, label disclosure of pesticide formulation components would help to inform users of
actual risks to pollinator health.
For the modeling study: A stage-structured honey bee population dynamic model, built
with the integration of difference and differential equations, was designed to capture four key
organizational principles inherent in this social organism: the comb pattern formation, the age-
related behavioral maturation, the brood care behavior and the collective foraging behavior. The
fusion of top-down and bottom-up approaches at different timescales in our model results in
significant advances in providing more refined prediction and understanding of honey bee
population and resource dynamics. The model’s precision has been validated by showing that it
can reproduce independent data set (peer-reviewed historical data and our field experiments)
outcomes for colonies at different latitudes and under different condition such as the healthy and
pesticide-contaminated scenarios. This model is also flexible to deal with many of the potential
genetic and environmental influences regarding: queen’s health such as time of initiation of egg
laying, maximum laying value, and the rate of attaining the maximum laying rate; the stage-
specific survivorship of bees; the brood rearing efficiency expressed by the larva-to-nurse ratio;
and the potential of developmental disorder, including the rate of accelerated, delayed,
precocious, or reverted maturation. It allows for further testing whether or not subtle changes at
the individual level by any potential stressors can result in substantial outcomes at the colony
level. Sensitivity analyses of this model have supported the conclusions that: disruption of the
numerical basis of colony population dynamics has only minimal and delayed impacts on the
colony due to the redundant capacity of the colony as a whole to recover. Most surprisingly,
137
minor changes in colony social structure especially the nurse-to-forager transition can have
immediate and drastic consequences on colony survival. Our modeling results indicate that a
balanced allocation of workers with respect to dynamic changes in colony task demands,
particularly during the fall season which is the sensitive stage for colonies to prepare for entering
the winter season, is the key to sustaining colony survival. A nutrition-sensing pathway, a
hierarchically structured regulatory network integrating genomic, physiological, behavioral
components, and social context, has been hypothesized to explain the complex phenotypes and its
considerable flexibility in honey bee colony. Together with our modeling results showing the
extreme sensitivity of colony fitness to the nurse-to-forager transition; it seems plausible to
speculate that the inadequate nutrition in either quality (pesticide contamination) or quantity (lack
of food) level can disrupt single or multiple components of the regulatory network involved in the
social caste development, eventually resulting in an irreversible alteration in colony homeostasis.
Our modeling results completely change the focus of our research on sub-lethal pesticide impacts
on colonies from looking at reduced numbers of given life stages to the understanding of changes
in hormonal and gene regulatory networks that may produce irreversible changes in colony
functioning and survival.
Moreover, this honey bee population model using an integrative modeling approach
would be a cost-effective tool for estimating and managing the ecological risks of potential
stressors inside and outside the hive. Comparing model simulations to our field experiments of
pesticide impacts demonstrated our model’s potential in ecological risk assessments and the
guidance it can offer for the design of further testing and validation experiments. Sensitivity
analysis helps to determine which demographic parameters are critical to analyze for colony-level
risks. It provides a means of examining the influence of potential stressors on colony population
dynamics, not just the commonly used individual/group-level mortality.
138
Future Work
Experimental and Modeling Research
Our larval toxicity study is the first study to report serious toxic effects of
environmentally realistic levels of common pesticides, formulation ingredients, and their resulting
mixtures on the survival of developing honey bee larvae. However, in the absence of exact
measures of pollen or pesticide residues in larval food, the pesticide concentrations detected in
pollen and beebread samples that were evaluated in this study cannot be fully representative of
the actual exposure of larval bees to pesticides. Additionally, the daily diet consumption for larval
stages was not measured; therefore, whether the increased mortality with larval age is due to
pesticide accumulation with chronic exposure is not clear. The antifeedant or other sensory
effects on consumption, or an age-dependent sensitivity to pesticides cannot be excluded from
consideration. Meanwhile, because neither food intake nor exact doses of pesticides consumed by
each larva were measured during oral feeding; our larval rearing method does not allow exact
quantification of the level of interaction but makes only an initial qualitative assessment of
synergism or antagonism. Further studies to examine the distribution and accumulation of
pesticides and their metabolites in honey bees through different developmental stages are needed.
In this study, we used the predicted adult toxicity estimated from the adult acute topical LD50 data
converted to whole-bee LC50 values using the inverse probit model. This is the first quantitative
evidence to show significant differences in adult and larval sensitivity to the same pesticide
exposure; however, given the lack of chronic toxicity data on adult bees, using the compilation of
acute data from different sources may complicate the accurate estimation of the adult toxicity
because of the heterogeneity introduced by differences among the studies. Chronic laboratory-
based toxicity testing has not been routinely required on individual bees in the US; therefore,
139
further studies need to be developed to evaluate the chronic toxicity of pesticide exposures at
realistic levels on adult bees for more accurate evaluation of the sensitivity differences between
larvae and adults. Furthermore, for a comprehensive understanding of pesticide impacts, future
risk assessment should include the long‐term whole hive effects following chronic exposure to a
treated crop conducted under semi-field or full field conditions.
Given our current state of knowledge and colony-level data, our model was developed
initially based on simplified assumptions including the mechanisms regulating foraging behavior
and division of labor. For instance, for the nectar foraging module, our model assumes an
unchanging, predator and competition free environment with constant forager round-trip times
and single flower source containing nectar of constant quality. Thus, how colonies adapt to
dynamic changes of nectar resources in the environment should be elucidated further which will
allow us to better the simulation of the dynamics of honey stores inside the colony. Currently, the
plasticity during honey bee behavioral development was simulated by assigning specific
probabilities for accelerated, delayed, or reverted transition within each phenotype in response to
changes in internal and external colony stressors. A further extension would be to integrate the
dynamic feedback of the behavioral development at a finer timescale, suggested by the future
research on neurochemical regulatory networks in regulating the honey bee social organization,
into the stage-structured honey bee population model. The addition of a honey bee disease
dynamic module will also be needed in order to further clarify the colony-level impacts of bee
diseases and their interactions with other hive factors. The integration of dynamic feedback
systems of nectar foraging demonstrated that this population model can be justified for the
eventual extrapolation to more generic conditions.
The simulation results indicate that the ratio between brood, foragers and hive bees in the
late field season, which is typically the preparing period for the coming winter season, can serve
as the direct measure of colony health. Although whether the change in task-based ratio may be a
140
consequence or a contributing cause of the colony collapse is not clear, this brood/nurse ratio can
be used as a simplified index to predict the colony fate. To further validate this conclusion, more
extensive colony-level studies are needed.
The role of pesticides particularly fungicides and the potential to mitigate these by
increased nutrition are both long-range goals that we have begun to address by the combinational
modeling and laboratory approach. The possibility of supplemental pollen feeding (the diversity
and duration of pollen feeding) in the late field season in helping restoration of the colony social
organization, thereby moderating negative impacts from pesticide exposures and other stressors,
will continue to be the focus of the future field research efforts.
Development of A Decision Support System for Colony Management
Rationales and Objectives
Due to the complexity of the honey bee colony system, the lack of quantitative colony
data, and the uncertainty in characterizing colony dynamics under multiple disturbances, it is
extremely difficult to explore the mechanisms by which a colony either succumbs to or
overcomes combinational effects of potential threats. Therefore, timely and reliable diagnoses of
the potential stressors and identification of colonies that are vulnerable to collapse is urgently
needed so that appropriate counter-measures can be formulated and implemented by beekeepers.
Such questions cannot usually be answered by reductionist experimental approaches alone, but
require the help of simulation models. As an initial step to explore the processes leading to colony
failure, we developed a stage-structured matrix population model of honey bee colony dynamics
and validated its capability in capturing the key adaptive feedback mechanisms in the population
and resource dynamics of a healthy honey bee colony. However, the complex and large amount
of technical information in the model presents a difficult hurdle for adequately communicating
141
model outputs to beekeepers and the public. Additionally, extension meetings and scientific
publications are a slow route of education. Therefore, we propose to develop a decision support
system (DSS) model (also called knowledge-based system model, expert system) (Stone et al.,
1986, Saunders et al., 2005) as a new and innovative method of transferring timely, up-to-date
management decisions to farmers, growers and beekeepers.
Knowledge-based systems are applied artificial intelligence (AI) tools designed for
providing intelligent decisions with justification, and were developed by the AI community in the
mid-1960s (Dhaliwal and Benbasat, 1996). The major framework is that the vast task-specific
knowledge can be transferred from an expert to a computer using various knowledge
representation techniques rules, then stored in the computer and users can call upon the computer
for specific advice with regard to their questions or problems (Shu-Hsien, 2005). The basic
advantages offered by such systems are documentation of knowledge, intelligent decision
support, self-learning, and symbolic reasoning and explanation capabilities. Acting as a human
consultant, the knowledge-based systems acquire information related with problems, make
inferences, arrive at a specific conclusion, give advices and explain the reasoning logic behind the
decision as needed (Cyran et al., 2009). This system facilitates the integration of quantitative and
qualitative reasoning methods, as well as the integration of knowledge and experiences of
different specialties; therefore, it can provide powerful and flexible solutions to a variety of
problems in a timely manner that cannot be achieved by traditional methods. Nowadays, the
knowledge-based systems have been successfully proliferating in various fields, where their
applications have been proving to be critical in the process of decision support and problem
solving (Dhaliwal and Benbasat, 1996, Shu-Hsien, 2005).
Although expert systems have been used in environmental sciences and management for
years (Stone et al., 1986), their application in apiculture is rare to date (Cohen and Shoshany,
2002, Shu-Hsien, 2005). BEE AWARE (McClure et al., 1993) developed at Penn State, and one
142
diagnostic expert system for honey bee pest management (Mahaman et al., 2002) are the only two
widely known expert system applications for honey bee colony management. BEE AWARE was
designed for the control and management of diseases, pests, predators and parasites of honey bees
(McClure et al., 1993).The more recent diagnostic expert system is a rule-based expert system to
diagnose pests based on the description of the external appearance or behavior of the affected
honey bee colony, and to suggest the appropriate colony treatments. The system can be used as a
diagnostic tool for beekeepers and as for educational and extension purposes in bee pathology
(Mahaman et al., 2002).
However, with the increasing complexity of honey bee colony problems such as Colony
Collapse Disorder and the incidents of colony losses on the rise, there is pressing need to make
up-to-date information. Therefore, by integrating with our current mathematical model which
depicts the most refined aspects of colony population and resource dynamics, our decision
support systems will be unique in determining the honey bee colony health in relation to the
threshold conditions suggested by the model, as well as identifying gaps in knowledge and
potential threats of honey bee colony dynamics (chapter 4). We will use an object oriented
software application developed at Penn State University, which implements a dependency
network approach to knowledge-based system development. This software, called NetWeaver™,
runs under the Microsoft Windows operating system (Saunders et al., 2005). NetWeaver™
provides the software tools to construct dependency networks between data and conclusions
within a fully editable graphic representation. By following paths leading from data through the
logic connectors between data, the heuristic values of a domain expert during the knowledge
elicitation process can be provided in a graphic format (Saunders et al., 2005).
As a further step to extend our current modeling efforts, we propose to use the
NetWeaverTM knowledge base development tool to: 1) initially, assist beekeepers as a diagnostic
tool to determine the condition of colony health; 2) secondly, help them to identify stressors
143
impacting colony health; 3) thirdly, assist them as a prediction tool to determine the potential of
colony collapse; 4) further, provide them with cost-effective decision support needed to manage
bees and help disrupted/weakened colonies to recover; 5) develop criteria for determining the
success of restoration treatments while assuring the honey harvesting 6) evaluate the capability of
our existing model in predicting colony dynamics; and 7) use as a valuable educational tool to
assist the traditional educational methodologies. Our overall goal is to develop a honey bee DSS
model that integrates colony field data and information to provide a comprehensive narrative and
graphic assessment of honey bee colony health and offer decision support for colony management.
Ultimately, we hope to build a web-based knowledge base thereby making it available to
practitioners and managers for whom access to scarce and specialized expertise is problematic. It
will encourage greater participation of beekeepers and other stakeholders in the design of
comprehensive expert system for honey bee colony management.
Proposed Methods
First, we will develop four task modules of the DSS model including: 1) test seasonal
colony health, 2) diagnose the potential stressors in the weak colony, 3) predict the survival time
of the weak colony, 4) inform the threshold of colony indicators to insure overwintering survival,
and 5) suggest the specific colony management actions to be taken to restore the weak colony.
The development of the DSS model using NetWeaverTM software has been depicted in the
Appendix A.
Secondly, we will evaluate this system by: initially testing the code, accomplished by
other programmers experienced with current development practices; further testing the
knowledge base and user interface, accomplished by apiculturists and extension specialists at the
research and production level. The close collaboration with Professor Michael Saunders, who is a
co-developer of NetWeaverTM software, will facilitate the development and evaluation of this
decision support system.
144
Lastly, we will implement this DSS on the internet, and deliver through the extension
website of the Department of Entomology at Pennsylvania State University. It will be available to
practitioners and managers for timely and effective colony management.
Preliminary Results
Healthy honey bee colony
Figure 5-1 has been developed to illustrate the application of this DSS model in
evaluating the colony health and any management actions as needed. This figure contains two
task Group windows from the model, which were initialized with the data presented in Table 5-1.
From this acquired colony status at March, the data suggest that the colony is healthy during the
spring season and has a high chance to survival continually. There is no evidence that this hive
may experience any potential stressors listed, considering the high level of FALSEness of this
colony’s status in each of the various stressor categories. This colony has enough pollen storage,
a good laying queen, suitable hive temperature, a high quality of nectar resource and early nectar
flow. Therefore, this colony does not need any colony management for restoration.
Table 5-1. The summary of simple data values (user-input initial data), calculated data through arguments and functions, and the level of Trueness for each dependency networks in the honey bee colony knowledgebase.
Simple Data (User Input) Name Value Critical Time of the Season March Days from the beginning of nectar flow or honey harvesting 20 Direct measurement of egg laying rate 600 Hive Temperature 36 Nectar Flow Time 2 Nectar Source 3 Pollen Storage during field season 2 Queen Age 3
Calculated Data Name Value Accurate estimate of in-hive adult bees Amount of honey for harvesting Honey Store (in kg) Optimal egg laying rate of a healthy queen in a good bee hive 45.84
145
Optimal honey storage in a healthy hive during fall season 350.21 Ratio of inadequate egg production 13.09
Dependency Networks Name Trueness Value A healthy honey bee colony 0.90 Fall -1 Food Quality and Quantity 1 Food quality is a concern-potential pesticide contamination 0 Honey Harvesting 0 Honey quantity is a concern -1 Pollen quantity is a concern 0 Queen is a concern -0.17 Queen Reproduction 0.75 Remove pesticide contamination 0 Requeen the colony -0.17 Single threat 1 Spring 0.90 Summer -1 Supplemental Feeding with Honey before winter -1 Supplemental Feeding with Pollen 0 The high frequency of Varroa mite in the fall is a threat -1 The insufficient egg production of the queen bee is a threat 1 The insufficient foraging force is a threat -1 Threshold of pollen before winter 1 Winter -1 A.
146
B.
Figure 5-1. Two Group windows for the honey bee knowledge base given the data depicted in Table 5-1. This first window A shows the evaluation results of a honey bee colony health during the spring season and the second window B shows the level of TRUEness of this colony’s health in each of the various stressor categories. The color bar represents the risk evaluation results: the green means the argument of the dependency network is firmly true; however, the red means the goal is completely wrong. For this colony, it shows stable growth during the spring and there is no evidence that the colony is under any disturbances/impacts form all the stressors listed above. Yellow means the goal is under precautious condition. The left bottom section shows data that are not used for the colony evaluation during the spring. The right section shows all the data required for the spring evaluation.
Weak honey bee colony
Figure 5-2 has been developed to illustrate the application of this DSS model in
evaluating the colony health during the fall and determining if and what management actions are
needed. This figure contains three task group windows from the model, which were initialized
with the data presented in Table 5-2. From this acquired colony status at October, the data
suggest that the colony is considerably disrupted in October and has a slim chance for survival
over the winter. There are two strong indicators that this hive may experience two potential
stressors- insufficient nursing and Varroa mite infection, considering the high level of Trueness of
this colony’s status in the stressor categories. Although this colony has enough pollen and honey
147
storage, a good laying queen, abundant brood and forager population, the major problems are
unbalanced brood and nurse ratios during the fall. Moreover, this scenario is worsened by the
high frequency of mite infection, resulting in a high potential of sudden collapse. Therefore, there
is a pressing need to save the colony by the suggested management actions with a high priority,
including treating the colony with mite control and improving the brood care behaviors.
Table 5-2. The summary of simple data values (user-input initial data), calcuated data through arguments and functions, and the level of Trueness for each dependency networks in the honey bee colony knowledgebase.
Simple Data (User Input) Name Value Colony Weight (kg) 80.0 Critical Time of the Season October Days from the beginning of nectar flow or honey harvesting 200.0 Direct measurement of egg laying rate 100.0 Foraging Frequency. 50.0 Hive Temperature 32.0 Number of brood frames in a hive 4.0 Number of frames covered by adult bees (double-sided) 2.0 Pollen Storage during field season 2.0 Queen Age 1.0 The size of a frame Medium (4.9 mm) Time of foraging during one day (in hrs) 2.0 Varroa mite Number per bee sampling in October 0.5
Calculated Data Name Trueness Value Amount of honey for harvesting 57.4 Brood nurse ratio during the fall season 7.4 Brood+Forager vs. Nurse Ratio 9.4 Honey Store (in kg) 90.2 Indirect Estimate of Forager Population 6000.0 Number of adult bees in a bee hive 3000.0 Number of brood in a bee hive 22176.0 Number of brood on one full frame (double-sided) 5544.0 Number of forager bees in one day 6000.0 Optimal brood nurse ratio during the fall season 0.4 Optimal egg laying rate of a healthy queen in a good bee hive 324.4 Optimal honey storage in a healthy hive during fall season 21.9 Ratio of inadequate egg production 0.3 Ratio of inadequate honey storage during the fall season 4.1 Ratio of inadequate nursing in the hive 20.0
148
Rough estimation of number of adult bees in the hive 3000.0 Rough estimation of number of brood in the hive 22176.0 Time (days) of a hive preparing for over wintering since the end of summer season
50.0
Dependency Networks Name Value A healthy honey bee colony -1.0 Fall -1.0 Honey quantity is a concern -1.0 Nursing is not a concern -1.0 Queen Reproduction -1.0 Spring -1.0 Summer -1.0 Supplemental Feeding with Honey before winter -1.0 Winter -1.0 Queen is a concern -0.3 Re-queen the colony -0.3 Food quality is a concern-potential pesticide contamination 0.0 Foraging Quality 0.0 Honey Harvesting 0.0 Number of foragers per day 0.0 Pollen quantity is a concern 0.0 Remove pesticide contamination 0.0 Supplemental Feeding with Pollen 0.0 Food Quality and Quantity 0.2 Provide diverse pollen diet 0.5 Varroa mite is a concern 0.9 Varroa Mite Treatment 0.9 Combine weak colony with strong colony before winter 1.0 Honey in the hive ready for Harvesting 1.0 Improve hive temperature 1.0 Improve Nursing Efficiency 1.0 Nursing is a concern 1.0 Over-winter Collapse 1.0 Use Brood pheromone 1.0
149
A.
B.
150
C.
Figure 5-2. Three Group windows for the honey bee knowledge base given the data depicted in Table 5-2. The first window A shows the evaluation results of a honey bee colony health in October; the second window B shows the level of TRUEness of this colony’s health in each of the various stressor categories; the last window C shows the suggestions of colony management to help restore the colony health. The color bar represents the risk evaluation results: the green means the argument of the dependency network is firmly true; however, the red means the goal is completely wrong. Yellow means the goal is under a precautious condition. For this colony, there is strong evidence that this colony has been experiencing serious nursing and mite problems, resulting in the considerable disturbance in the colony survival and health. Therefore, appropriate management actions are needed to improve the nursing efficiency and control the Varroa mite infection in the hive. The left bottom section shows data that are not used for the colony evaluation during the fall. The right section shows all the data required for the fall evaluation.
Implications and further extensions
This honey bee knowledge-based models are designed to provide a comprehensive
narrative and graphic assessment of honey bee colony health and offer decision support for
colony management. It captures the reasoning processes of domain experts and gives advice on
the identification and management of the common honey bee problems including queen health,
nursing behavior, foraging efficiency, Varroa mite, diseases, and nutrition factors such as the
quality and quantity of food sources both inside and outside of the hive. This system, as
151
illustrated in two colony examples, is divided into four modules: Information Acquisition, Colony
Assessment, Stressor Diagnosis, and Management Suggestions. Through a series of questions on
the important colony predictors, answered by the beekeeper, the computer program narrows down
the possibilities until it comes up with a suitable diagnosis of stressors impacting colony
dynamics most significantly, eventually prioritize suggestions for restoration of colony health.
Even with conflicting or inconsistent information provided, this expert system can still make a
reliable diagnosis in a timely manner by integrating both qualitative and quantitative reasoning
processes. This integrative approach offers advantages over a conventional reference or textbook
or a single domain expert. The graphical and executable rendering of knowledge bases within
NetWeaver™ also greatly facilitates the knowledge engineering process of colony management.
It represents the real honey bee system in a form that corresponds closely to the way beekeepers
perceive it. This expert system is unique in depicting the honey bee colony health in relation to
the threshold conditions suggested by the quantitative model, thereby diagnosing the potential
stressors affecting colony health, and prioritizing management plans based on relevance to colony
health. In particular, the inclusion of fuzzy arguments, against which data values can be
compared, provides a robust description of the complexity of honey bee society, of which the
traditional quantitative modeling approaches are often complicated by the combinational factors
within various uncertainties inherent in the complex systems. Thus, the model is easily
understandable, even by a non-professional audience, and each parameter has a readily
perceivable meaning. It eliminates the artificial distance between domain experts and beekeepers
and other stakeholders by effectively communicating the substantial and complex technical
information in the mathematical model to the general public.
This honey bee knowledge base is built on top of the recent honey bee stage population
model, which has demonstrated its capability to provide the most refined and comprehensive
information of the population and resource dynamics in a honey bee colony. Mathematical
152
models incorporate substantial amounts of knowledge (expertise) about systems’ behavior;
therefore, this close incorporation lays a solid foundation for the bee expert system in assessing
the colony performance and providing sound decision support for colony management.
Meanwhile, the mathematical modeling can benefit substantially from this qualitative modeling
approach. Implemented as a fuzzy rule-based system (Saunders et al., 2005), the qualitative
model provides an expandable and accessible platform for testing, verifying and refining the
mathematical models. Further, the expert system permits the inclusion of the domain expert(s) in
the knowledge representation process (Saunders et al., 2005) and encourages their greater
participation in the effective development and improvement of both quantitative and qualitative
models for honey bee systems. Therefore, this DSS model can be a powerful decision-support
system for colony management. In addition, it can be a valuable educational tool to assist the
traditional educational methodologies.
Using NetWeaver™ based model, it is a simple task to link to spatially referenced
databases. Therefore, it is plausible that the current honey bee NetWeaver™ models can be linked
to the geographic information systems to classify multiple hives at different locations. The
coordination of landscape data sets, knowledge-based reasoning, and geographic information
systems will provide a powerful tool to characterize the landscape impacts on colony health,
thereby stimulating further development of comprehensive knowledge-based systems for honey
bee colony management.
Acknowledgements
The ongoing work of developing honey bee expert system cannot be done without the
contribution of Bruce Miller and Professors Mike Saunders, who have provided constructive help
153
and suggestions with the model development and evaluation. This work is funded by 2013
research grants approved by the National Honey Board.
REFERENCES
Aizen MA, Harder LD (2009) The global stock of domesticated honey bees is growing slower
than agricultural demand for pollination. Curr Biol 19:915-918.
Amdam GV (2011) Social context, stress, and plasticity of aging. Aging Cell 10:18-27.
Amdam GV, Aase AL, Seehuus SC, Kim Fondrk M, Norberg K, Hartfelder K (2005) Social
reversal of immunosenescence in honey bee workers. Exp Gerontol 40:939-947.
Amdam GV, Rueppell O, Fondrk MK, Page RE, Nelson CM (2009) The nurse's load: early-life
exposure to brood-rearing affects behavior and lifespan in honey bees (Apis mellifera).
Exp Gerontol 44:467-471.
Amdam GV, Simoes ZL, Hagen A, Norberg K, Schroder K, Mikkelsen O, Kirkwood TB, Omholt
SW (2004) Hormonal control of the yolk precursor vitellogenin regulates immune
function and longevity in honeybees. Exp Gerontol 39:767-773.
Ament SA, Wang Y, Robinson GE (2010) Nutritional regulation of division of labor in honey
bees: toward a systems biology perspective. Wiley Interdiscip Rev Syst Biol Med 2:566-
576.
Anderson C, Ratnieks FL (1999) Worker allocation in insect societies: coordination of nectar
foragers and nectar receivers in honey bee (Apis mellifera) colonies. Behav Ecol
Sociobiol 46:73-81.
Atkins E, Kellum D, Atkins K (1981) Reducing pesticide hazards to honey bees: mortality
prediction techniques and integrated management strategies. Leafl. 2883 Division of
Agriculture, University of California, Riverside.
Atkins EL, Kellum D (1986) Comparative morphogenic and toxicity studies on the effect of
pesticides on honeybee brood. J Apicult Res 25:242-255.
Aupinel P, Fortini D, Michaud B, Marolleau F, Tasei JN, Odoux JF (2007) Toxicity of
dimethoate and fenoxycarb to honey bee brood (Apis mellifera), using a new in vitro
standardized feeding method. Pest Manag Sci 63:1090-1094.
Babendreier D, Joller D, Romeis J, Bigler F, Widmer F (2007) Bacterial community structures in
honeybee intestines and their response to two insecticidal proteins. FEMS Microbiol Ecol
59:600-610.
155
Bauer DM, Sue Wing I (2010) Economic consequences of pollinator declines: A synthesis. Agr
Resource Econ Rev 39:368-383.
Becher MA, Hildenbrandt H, Hemelrijk CK, Moritz RFA (2010) Brood temperature, task
division and colony survival in honeybees: A model. Ecol Model 221:769-776.
Bendahou N, Fleche C, Bounias M (1999) Biological and biochemical effects of chronic
exposure to very low levels of dietary cypermethrin (cymbush) on honeybee colonies
(Hymenoptera: Apidae). Ecotoxicol Environ Saf 44:147-153.
Beshers SN, Huang ZY, Oono Y, Robinson GE (2001) Social inhibition and the regulation of
temporal polyethism in honey bees. J Theor Biol 213:461-479.
Bodenheimer FS (1937) Studies in animal populations. II. seasonal population-trends of the
honey-Bee. Q Rev Biol 12:406-425.
Bodenheimer FS, Ben-Nerya A (1937) One-year studies on the biology of the honey-bee in
Palestine. Ann Appl Biol 24:385-403.
Bogdanov S (2006) Contaminants of bee products. Apidologie 37:1-18.
Bonzini S, Tremolada P, Bernardinelli I, Colombo M, Vighi M (2011) Predicting pesticide fate in
the hive (part 1): experimentally determined tau-fluvalinate residues in bees, honey and
wax. Apidologie 42:378-390.
Bouvier JC, Boivin T, Beslay D, Sauphanor B (2002) Age-dependent response to insecticides and
enzymatic variation in susceptible and resistant codling moth larvae. Arch Insect
Biochem 51:55-66.
Bretschko J (1985) Naturgemäße Bienenzucht: Entscheidungshilfen für eine erfolgreiche,
erwerbsorientierte Imkerei (in verschiedenen Magazinstocksystemen). Graz ; Stuttgart: L.
Stocker.
Brodschneider R, Crailsheim K (2010) Nutrition and health in honey bees. Apidologie 41:278-
294.
Bühlmann G (1985) Assessing population dynamics in a honeybee colony. Mitteilungen der
Deutschen Gesellschaft fuer Allgemeine und Angewandte Entomologie 4.
California Department of Food and Agriculture (CDFA) (1988) Myclobutanil Risk Assessment.
(Division of Pest Management, E. P., and Worker Safety, California, ed).
Camazine S, Franks NR, Sneyd J, Bonabeau E, Deneubourg J-L, Theraula G (2001) Self-
Organization in Biological Systems: Princeton University Press.
Camazine S, Sneyd J (1991) A model of collective nectar source selection by honey bees: Self-
organization through simple rules. J Theor Biol 149:547-571.
156
Caux PY, Kent RA, Fan GT, Stephenson GL (1996) Environmental fate and effects of
chlorothalonil: A Canadian perspective. Crit Rev Environ Sci Technol 26:45-93.
Cedergreen N, Streibig JC (2005) The toxicity of herbicides to non-target aquatic plants and
algae: assessment of predictive factors and hazard. Pest Manag Sci 61:1152-1160.
Chan QW, Mutti NS, Foster LJ, Kocher SD, Amdam GV, Wolschin F (2011) The worker
honeybee fat body proteome is extensively remodeled preceding a major life-history
transition. PLoS One 6:e24794.
Chauzat M-P, Martel A-C, Cougoule N, Porta P, Lachaize J, Zeggane S, Aubert M, Carpentier P,
Faucon J-P (2011) An assessment of honeybee colony matrices, Apis mellifera
(Hymenoptera: Apidae) to monitor pesticide presence in continental France. Environ
Toxicol Chem 30:103-111.
Chauzat MP, Faucon JP (2007) Pesticide residues in beeswax samples collected from honey bee
colonies (Apis mellifera L.) in France. Pest Manag Sci 63:1100-1106.
Ciarlo TJ, Mullin CA, Frazier JL, Schmehl DR (2012) Learning impairment in honey bees caused
by agricultural spray adjuvants. PLoS One 7:e40848.
Claudianos C, Ranson H, Johnson RM, Biswas S, Schuler MA, Berenbaum MR, Feyereisen R,
Oakeshott JG (2006) A deficit of detoxification enzymes: pesticide sensitivity and
environmental response in the honeybee. Insect Mol Biol 15:615-636.
Cohen Y, Shoshany M (2002) A national knowledge-based crop recognition in Mediterranean
environment. Int J Appl Earth Obs Geoinf 4:75-87.
Collins A, Pettis J (2012) Correlation of queen size and spermathecal contents and effects of
miticide exposure during development. Apidologie 1-6.
Corona M, Velarde RA, Remolina S, Moran-Lauter A, Wang Y, Hughes KA, Robinson GE
(2007) Vitellogenin, juvenile hormone, insulin signaling, and queen honey bee longevity.
Proc Natl Acad Sci USA 104:7128-7133.
Cowles RS, Cowles EA, McDermott AM, Ramoutar D (2000) "Inert" formulation ingredients
with activity: toxicity of trisiloxane surfactant solutions to two spotted spider mites
(Acari: Tetranychidae). J Econ Entomol 93:180-188.
Cox C, Surgan M (2006) Unidentified inert ingredients in pesticides: Implications for human and
environmental health. Environ Health Persp 114:1803-1806.
Cox DR (1972) Regression Models and Life-Tables. J Roy Stat Soc B 34:187-190.
Cox-Foster DL, Conlan S, Holmes EC, Palacios G, Evans JD, Moran NA, Quan PL, Briese T,
Hornig M, Geiser DM, Martinson V, vanEngelsdorp D, Kalkstein AL, Drysdale A, Hui J,
157
Zhai J, Cui L, Hutchison SK, Simons JF, Egholm M, Pettis JS, Lipkin WI (2007) A
metagenomic survey of microbes in honey bee colony collapse disorder. Science
318:283-287.
Crailsheim K (1992) The flow of jelly within a honeybee colony. J Comp Physiol B 162:681-689.
Crailsheim K, Hrassnigg N (1998) Adaptation of hypopharyngeal gland development to the brood
status of honeybee (Apis mellifera L.) colonies. J Insect Physiol 44:929-939.
Cyran KA, Kozielski S, Peters JF, Stanczyk U, Wakulicz-Deja AE (2009) Man-Machine
Interactions. Springer: Advances in Intelligent and Soft Computing.
Davies PE (1987) Physiological, anatomic and behavioral-Changes in the respiratory system of
Salmo-Gairdneri Rich on acute and chronic exposure to chlorothalonil. Comp Biochem
Phys C 88:113-119.
Davis A (1989) The study of insecticide poisoning of honey bee brood. Bee World 70:163-174.
Davis AR, Solomon KR, Shuel RW (1988) Laboratory studies of honeybee larval growth and
development as affected by systemic insecticides at adult-sublethal levels. J Apicult Res
27:146-161.
Decourtye A, Devillers J, Cluzeau S, Charreton M, Pham-Delegue MH (2004) Effects of
imidacloprid and deltamethrin on associative learning in honeybees under semi-field and
laboratory conditions. Ecotox Environ Safe 57:410-419.
DeGrandi-Hoffman G, Roth SA, Loper GL, Erickson Jr EH (1989) BEEPOP: A honeybee
population dynamics simulation model. Ecol Model 45:133-150.
DeGrandi-Hoffman G, Sammataro D, Alarcon R (2009) The importance of microbes in nutrition
and health of honey bee colonies Part II of three parts. Am Bee J 149:667-669.
DeLorenzo ME, Fulton MH (2012) Comparative risk assessment of permethrin, chlorothalonil,
and diuron to coastal aquatic species. Mar Pollut Bull 64:1291-1299.
DeLorenzo ME, Serrano L (2003) Individual and mixture toxicity of three pesticides: atrazine,
chlorpyrifos, and chlorothalonil to the marine phytoplankton species Dunaliella
tertiolecta. J Environ Sci Health, Part B: Pestic, Food Contam, Agric Wastes 38:529-538.
Desneux N, Decourtye A, Delpuech JM (2007) The sublethal effects of pesticides on beneficial
arthropods. Annu Rev Entomol 52:81-106.
Dhaliwal JS, Benbasat I (1996) The use and effects of knowledge-based system explanations:
Theoretical foundations and a framework for empirical evaluation. Inform Syst Res
7:342-362.
158
Dolejsky W, Schley P (1980) A mathematical simulation model to estimate the population
development in a bee (Apis mellifera L.) colony following infestation with the mite
Varroa jacobsoni Oud. J Vet Med B 27:798-805.
Donovan Y (2006) Reregistration Eligibility Decision for Chlorpyrifos. U.S. Environmental
Protection Agency, Office of Pesticide Programs.
Dyer FC (2002) The biology of the dance language. Annu Rev Entomol 47:917-949.
Ebert GV (1922) Zur Massenentwicklung der Bienenvoelker. Arch f Bienenkde 4:37-38.
Edwards JR, Myerscough MR (2011) Intelligent decisions from the hive mind: Foragers and
nectar receivers of Apis mellifera collaborate to optimise active forager numbers. J Theor
Biol 271:64-77.
EPA US (2007) Mycobutanil. Ecological Risk Assessment on New Uses for Tropical Fruit,
Fruiting Vegetables and Artichokes [D336613] (10/12/07). U.S. Environmental
Protection Agency, Office of Pesticide Programs, Environmental Fate and Effects
Division and Health Effects Division. Arlington, VA.
European Commission (2009) Council Directive 91/414/EEC (Plant Protection Products) as
repealed under Regulation (EC) No 1107/2009. (EC, ed), pp 1-50: Official Journal of the
European Union.
Everich R, Schiller C, Whitehead J, Beavers M, Barrett K (2009) Effects of captan on Apis
mellifera brood development under field conditions in California almond orchards. J
Econ Entomol 102:20-29.
Fewell JH, Winston ML (1996) Regulation of nectar collection in relation to honey storage levels
by honey bees, Apis mellifera. Behav Ecol 7:286-291.
Fiedler L (1987) Assessment of chronic toxicity of selected insecticides to honeybees. J Apicult
Res 26:115-122.
Frazier M, Mullin C, Frazier J, Ashcraft S (2008) What have pesticides got to do with it? Am Bee
J 148:521-523.
Fukuda H (1971) Improvement of Bodenheimer’s method for estimating individual number in
honeybee colonies. J Fac Sci Hokkaido Univ Ser VI 18:128–143.
Fukuda H (1983) The relationship between work efficiency and population size in a honeybee
colony. Res Popul Ecol 25:249-263.
Gallagher EP, Cattley RC, Digiulio RT (1992) The acute toxicity and sublethal effects of
chlorothalonil in channel catfish (Ictalurus-Punctatus). Chemosphere 24:3-10.
Gisi U (1996) Synergistic interaction of fungicides in mixtures. Phytopathology 86:1273-1279.
159
Gregorc A, Ellis JD (2011) Cell death localization in situ in laboratory reared honey bee (Apis
mellifera L.) larvae treated with pesticides. Pestic Biochem Phys 99:200-207.
Haarmann T, Spivak M, Weaver D, Weaver B, Glenn T (2002) Effects of fluvalinate and
coumaphos on queen honey bees (Hymenoptera: Apidae) in two commercial queen
rearing operations. J Econ Entomol 95:28-35.
Hardstone MC, Scott JG (2010) Is Apis mellifera more sensitive to insecticides than other
insects? Pest Manag Sci 66:1171-1180.
Harris JL (1985) A model of honeybee colony population dynamics. J Apicult Res 24:228-236.
Health and Safety Executive (1997) N-Methyl-2-pyrrolidone (Risk assessment document) EH
72/10. London, UK.
Hendriksma HP, Hartel S, Steffan-Dewenter I (2011) Honey bee risk assessment: new approaches
for in vitro larvae rearing and data analyses. Methods Ecol Evol 2:509-517.
Henry M, Beguin M, Requier F, Rollin O, Odoux JF, Aupinel P, Aptel J, Tchamitchian S,
Decourtye A (2012) A common pesticide decreases foraging success and survival in
honey bees. Science 336:348-350.
Herb BR, Wolschin F, Hansen KD, Aryee MJ, Langmead B, Irizarry R, Amdam GV, Feinberg
AP (2012) Reversible switching between epigenetic states in honeybee behavioral
subcastes. Nat Neurosci 15:1371-1373.
Hertzberg RC, MacDonell MM (2002) Synergy and other ineffective mixture risk definitions. Sci
Total Environ 288:31-42.
Heylen K, Gobin B, Arckens L, Huybrechts R, Billen J (2011) The effects of four crop protection
products on the morphology and ultrastructure of the hypopharyngeal gland of the
European honeybee, Apis mellifera. Apidologie 42:103-116.
Jeffree EP, Allen DM (1956) The annual cycle of pollen storage by honey bees. J Econ Entomol
50:211-212.
Jeroen D, Johan B (2005) Age-dependent morphology and ultrastructure of the hypopharyngeal
gland of Apis mellifera workers (Hymenoptera, Apidae). Apidologie 36:49-57.
Johansen C, Kleinschmidt M (1972) Insecticide formulations and their toxicity to honeybees. J
Apic Res 11:59-62.
Johnson BR (2010) Division of labor in honeybees: form, function, and proximate mechanisms.
Behav Ecol Sociobiol 64:305-316.
Johnson RM (2011) Managed pollinator CAP Coordinated Agricultural Project miticide and
fungicide interactions. Am Bee J 151:975-977.
160
Johnson RM, Ellis MD, Mullin CA, Frazier M (2010) Pesticides and honey bee toxicity - USA.
Apidologie 41:312-331.
Johnson RM, Pollock HS, Berenbaum MR (2009) Synergistic interactions between in-hive
miticides in Apis mellifera. J Econ Entomol 102:474-479.
Johnson RM, Wen Z, Schuler MA, Berenbaum MR (2006) Mediation of pyrethroid insecticide
toxicity to honey bees (Hymenoptera: Apidae) by cytochrome P450 monooxygenases. J
Econ Entomol 99:1046-1050.
Kalbfleisch J, Prentice R (1980) The Statistical Analysis of Failure Time Data. John Wiley, New
York, NY, USA.
Kamakura M (2011) Royalactin induces queen differentiation in honeybees. Nature 473:478-483.
Kennaugh L, Pearce D, Daly JC, Hobbs AA (1993) A piperonyl butoxide synergizable resistance
to permethrin in Helicoverpa Armigera which is not due to increased detoxification by
Cytochrome-P450. Pestic Biochem Phys 45:234-241.
Kitulagodage M, Astheimer LB, Buttemer WA (2008) Diacetone alcohol, a dispersant solvent,
contributes to acute toxicity of a fipronil-based insecticide in a passerine bird. Ecotox
Environ Safe 71:597-600.
Krogh KA, Halling-Sorensen B, Mogensen BB, Vejrup KV (2003) Environmental properties and
effects of nonionic surfactant adjuvants in pesticides: a review. Chemosphere 50:871-
901.
Kudsk P, Mathiassen SK (2004) Joint action of amino acid biosynthesis-inhibiting herbicides.
Weed Res 44:313-322.
Kunert K, Crailsheim K (1988) Seasonal changes in carbohydrate, lipid and protein content in
emerging worker honeybees and their mortality. J Apicult Res 27:13-21.
Ladurner E, Bosch J, Kemp WP, Maini S (2005) Assessing delayed and acute toxicity of five
formulated fungicides to Osmia lignaria Say and Apis mellifera. Apidologie 36:449-460.
Lan CH, Peng CY, Lin TS (2004) Acute aquatic toxicity of N-Methyl-2-Pyrrolidinone to
Daphnia magna. Bull Environ Contam Toxicol 73:392-397.
Lee GH, Ellersieck MR, Mayer FL, Krause GF (1995) Predicting chronic lethality of chemicals
to fishes from acute toxicity test data - Multifactor probit analysis. Environ Toxicol Chem
14:345-349.
Leoncini I, Le Conte Y, Costagliola G, Plettner E, Toth AL, Wang M, Huang Z, Becard JM,
Crauser D, Slessor KN, Robinson GE (2004) Regulation of behavioral maturation by a
161
primer pheromone produced by adult worker honey bees. Proc Natl Acad Sci USA
101:17559-17564.
Mahaman BD, Harizanis P, Filis I, Antonopoulou E, Yialouris CP, Sideridis AB (2002) A
diagnostic expert system for honeybee pests. Comput Electron Agr 36:17-31.
Maisonnasse A, Alaux C, Beslay D, Crauser D, Gines C, Plettner E, Le Conte Y (2010) New
insights into honey bee (Apis mellifera) pheromone communication. Is the queen
mandibular pheromone alone in colony regulation? Front Zool 7:18.
Maisonnasse A, Lenoir JC, Beslay D, Crauser D, Le Conte Y (2010) E-beta-ocimene, a volatile
brood pheromone involved in social regulation in the honey bee colony (Apis mellifera).
PLoS One 5:e13531.
Makela ME, Rowell GA, Sames Iv WJ, Wilson LT (1993) An object-oriented intracolonial and
population level model of honey bees based on behaviors of European and Africanized
subspecies. Ecol Model 67:259-284.
Mao W, Schuler MA, Berenbaum MR (2011) CYP9Q-mediated detoxification of acaricides in
the honey bee (Apis mellifera). Proc Natl Acad Sci USA 108:12657-12662.
Marino S, Hogue IB, Ray CJ, Kirschner DE (2008) A methodology for performing global
uncertainty and sensitivity analysis in systems biology. J Theor Biol 254:178-196.
Mayer DF, Lunden JD (1986) Toxicity of fungicides and an acaricide to honey-bees
(Hymenoptera, Apidae) and their effects on bee foraging behavior and pollen viability on
blooming apples and pears. Environ Entomol 15:1047-1049.
Mayer FL, Krause GF, Buckler DR, Ellersieck MR, Lee GH (1994) Predicting chronic lethality
of chemicals to fishes from acute toxicity test data - Concepts and linear-regression
analysis. Environ Toxicol Chem 13:671-678.
McClure JE, Tomasko M, Collison CH (1993) BEE AWARE, an expert system for honey bee
diseases, parasites, pests and predators. Comput Electron Agr 9:111-122.
McLellan AR, Rowland CM, Fawcett RH (1980) A monogynous eusocial insect worker
population model with particular reference to honeybees. Ins Soc 27:305-311.
McMahon TA, Halstead NT, Johnson S, Raffel TR, Romansic JM, Crumrine PW, Boughton RK,
Martin LB, Rohr JR (2011) The fungicide chlorothalonil is nonlinearly associated with
corticosterone levels, immunity, and mortality in Amphibians. Environ Health Persp
119:1098-1103.
162
Meled M, Thrasyvoulou A, Belzunces LP (1998) Seasonal variations in susceptibility of Apis
mellifera to the synergistic action of prochloraz and deltamethrin. Environ Toxicol Chem
17:2517-2520.
Milum VG (1956) An analysis of twenty years of honey bee colony weight changes. J Econ
Entomol 49:735-738.
Morris MD (1991) Factorial sampling plans for preliminary computational experiments.
Technometrics 33:161-174.
Morse RA, Calderone NW (2000) The value of honey bees as pollinators of US crops in 2000.
Bee Culture 128:1-15.
Mott C, Breed M (2012) Insulin modifies honeybee worker behavior. Insects 3:1084-1092.
Mullin C, Ciarlo T, Chen J, Zhu W, Frazier M, Frazier J (2013) Pesticide formulation adjuvants
and their health impacts on pollinators and other non-target species. Pest Manag Sci in
review.
Mullin CA, Frazier M, Frazier JL, Ashcraft S, Simonds R, VanEngelsdorp D, Pettis JS (2010)
High levels of miticides and agrochemicals in North American apiaries: Implications for
honey bee health. PLoS One 5:e9754.
Mutti NS, Dolezal AG, Wolschin F, Mutti JS, Gill KS, Amdam GV (2011) IRS and TOR
nutrient-signaling pathways act via juvenile hormone to influence honey bee caste fate. J
Exp Biol 214:3977-3984.
National Academy of Sciences (2007) Status of pollinators in North America. Washington, DC:
National Academy Press.
Nelson CM, Ihle KE, Fondrk MK, Page RE, Amdam GV (2007) The gene vitellogenin has
multiple coordinating effects on social organization. PLoS Biol 5:e62.
Nolan W (1932) The development of package-bee colonies. USDA Technical Bulletin No. 309.
Oakes DJ, Pollak JK (2000) The in vitro evaluation of the toxicities of three related herbicide
formulations containing ester derivatives of 2,4,5-T and 2,4-D using sub-mitochondrial
particles. Toxicology 151:1-9.
Ohashi K, Natori S, Kubo T (1997) Change in the mode of gene expression of the
hypopharyngeal gland cells with an age-dependent role change of the worker honeybee
Apis mellifera L. Eur J Biochem 249:797-802.
Oldroyd BP, Thompson GJ (2006) Behavioural genetics of the honey bee Apis mellifera. In:
Advances in Insect Physiology, vol. 33 (Simpson, S. J., ed), pp 1-49: Academic Press.
163
Omholt SW (1986) A model for intracolonial population dynamics of the honeybee in temperate
zones. J Apic Res 25:9-21.
Pariseau J, Saint-Louis R, Delaporte M, El Khair MA, McKenna P, Tremblay R, Davidson TJ,
Pelletier E, Berthe FCJ (2009) Potential link between exposure to fungicides
chlorothalonil and mancozeb and haemic neoplasia development in the soft-shell clam
Mya arenaria: A laboratory experiment. Mar Pollut Bull 58:503-514.
Patel A, Fondrk MK, Kaftanoglu O, Emore C, Hunt G, Frederick K, Amdam GV (2007) The
making of a queen: TOR pathway is a key player in diphenic caste development. PLoS
One 2:e509.
Peng Y-SC, Mussen E, Fong A, Montague MA, Tyler T (1992) Effects of chlortetracycline of
honey bee worker larvae reared in vitro. J Invertebr Pathol 60:127-133.
Pereira JL, Antunes SC, Castro BB, Marques CR, Goncalves AMM, Goncalves F, Pereira R
(2009) Toxicity evaluation of three pesticides on non-target aquatic and soil organisms:
commercial formulation versus active ingredient. Ecotoxicology 18:455-463.
Pettis JS, Collins AM, Wilbanks R, Feldlaufer MF (2004) Effects of coumaphos on queen rearing
in the honey bee, Apis mellifera. Apidologie 35:605-610.
Pettis JS, vanEngelsdorp D, Johnson J, Dively G (2012) Pesticide exposure in honey bees results
in increased levels of the gut pathogen Nosema. Naturwissenschaften 99:153-158.
Pilling ED, Bromleychallenor KAC, Walker CH, Jepson PC (1995) Mechanism of synergism
between the pyrethroid insecticide Lambda-Cyhalothrin and the imidazole fungicide
prochloraz, in the honeybee (Apis mellifera L). Pestic Biochem Phys 51:1-11.
Pilling ED, Jepson PC (1993) Synergism between Ebi fungicides and a pyrethroid insecticide in
the honeybee (Apis mellifera). Pestic Sci 39:293-297.
Pinto LZ, Bitondi MM, Simoes ZL (2000) Inhibition of vitellogenin synthesis in Apis mellifera
workers by a juvenile hormone analogue, pyriproxyfen. J Insect Physiol 46:153-160.
Potts SG, Biesmeijer JC, Kremen C, Neumann P, Schweiger O, Kunin WE (2010) Global
pollinator declines: trends, impacts and drivers. Trends Ecol Evol 25:345-353.
Ricarda S, Robert EP, Joachim E (2004) Sucrose responsiveness and behavioral plasticity in
honey bees (Apis mellifera). Apidologie 35:133-142.
Robinson GE (1987) Regulation of honey bee age polyethism by juvenile hormone. Behav Ecol
Sociobiol 20:329-338.
Robinson GE (1992) Regulation of division of labor in insect societies. Annu Rev Entomol
37:637-665.
164
Rortais A, Arnold G, Halm M-P, Touffet-Briens F (2005) Modes of honeybees exposure to
systemic insecticides: estimated amounts of contaminated pollen and nectar consumed by
different categories of bees. Apidologie 36:71-83.
Rueppell O, Bachelier C, Fondrk MK, Page RE, Jr. (2007) Regulation of life history determines
lifespan of worker honey bees (Apis mellifera L.). Exp Gerontol 42:1020-1032.
Sagili R, Breece C (2012) Effects of brood pheromone (SuperBoost) on consumption of protein
supplement and growth of honey bee (Hymenoptera: Apidae) colonies during fall in a
northern temperate climate. J Econ Entomol 105:1134-1138.
Sagili R, Pankiw T (2007) Effects of protein-constrained brood food on honey bee (Apis mellifera
L.) pollen foraging and colony growth. Behav Ecol Sociobiol 61:1471-1478.
Sagili R, Pankiw T, Metz B (2011) Division of labor associated with brood rearing in the honey
bee: how does it translate to colony fitness? PLoS One 6:e16785.
Saillenfait AM, Gallissot F, Sabaté JP (2007) Developmental toxic effects of N-ethyl-2-
pyrrolidone administered orally to rats. J Appl Toxicol 27:491-497.
Sakagami Si, Fukuda H (1968) Life tables for worker honeybees. Res Popul Ecol 10:127-139.
Saunders MC, Sullivan TJ, Nash BL, Tonnessen KA, Miller BJ (2005) A knowledge-based
approach for classifying lake water chemistry. Knowl-Based Syst 18:47-54.
Schmickl T, Crailsheim K (2004) Inner nest homeostasis in a changing environment with special
emphasis on honey bee brood nursing and pollen supply. Apidologie 35:249-263.
Schmickl T, Crailsheim K (2007) HoPoMo: A model of honeybee intracolonial population
dynamics and resource management. Ecol Model 204:219-245.
Schmidt JO, Johnson BE (1984) Pollen feeding preference of Apis mellifera (Hymenoptera,
Apidae) a polylectic bee. Southwest Entomol 9:41-47.
Schmuck R (2004) Effects of a chronic dietary exposure of the honeybee Apis mellifera
(Hymenoptera: Apidae) to imidacloprid. Arch Environ Contam Toxicol 47:471-478.
Schmuck R, Pfuger W, Grau R, Holihn U, Fischer R (1994) Comparison of short-term aquatic
toxicity - formulation vs active ingredients of pesticides. Arch Environ Contam Toxicol
26:240-250.
Schmuck R, Stadler T, Schmidt HW (2003) Field relevance of a synergistic effect observed in the
laboratory between an EBI fungicide and a chloronicotinyl insecticide in the honeybee
(Apis mellifera L, Hymenoptera). Pest Manag Sci 59:279-286.
165
Schneider CW, Tautz J, Grunewald B, Fuchs S (2012) RFID tracking of sublethal effects of two
neonicotinoid insecticides on the foraging behavior of Apis mellifera. PLoS One
7:e30023.
Seeley TD (1985) Honeybee ecology: a study of adaptation in social life: Princeton University
Press.
Seeley TD (1989) Social foraging in honey bees: how nectar foragers assess their colony's
nutritional status. Behav Ecol Sociobiol 24:181-199.
Seeley TD (1995) The wisdom of the hive: the social physiology of honey bee colonies.
Cambridge, Mass.: Harvard University Press.
Sheppard WS, Meixner MD (2003) Apis mellifera pomonella, a new honey bee subspecies from
Central Asia. Apidologie 34:367-375.
Sherrard RM, Murray-Gulde CL, Rodgers JH, Shah YT (2003) Comparative toxicity of
chlorothalonil: Ceriodaphnia dubia and Pimephales promelas. Ecotox Environ Safe
56:327-333.
Shu-Hsien L (2005) Expert system methodologies and applications- a decade review from 1995
to 2004. Expert Syst Appl 28:93-103.
Silva de Moraes RLM, Bowen ID (2000) Modes of cell death in the hypopharyngeal gland of the
honey bee (Apis mellifera L). Cell Biol Int 24:737-743.
Smith CR, Toth AL, Suarez AV, Robinson GE (2008) Genetic and genomic analyses of the
division of labour in insect societies. Nat Rev Genet 9:735-748.
Solomon KR, Thompson DG (2003) Ecological risk assessment for aquatic organisms from over-
water uses of glyphosate. J Toxicol Env Heal B 6:289-324.
Stone ND, Coulson RN, Frisbie RE, Loh DK (1986) Expert systems in entomology: three
approaches to problem solving. Bull Entomol Soc Am 32:161-166.
Suchail S, Guez D, Belzunces LP (2001) Discrepancy between acute and chronic toxicity induced
by imidacloprid and its metabolites in Apis mellifera. Environ Toxicol Chem 20:2482-
2486.
Sullivan JP, Fahrbach SE, Harrison JF, Capaldi EA, Fewell JH, Robinson GE (2003) Juvenile
hormone and division of labor in honey bee colonies: effects of allatectomy on flight
behavior and metabolism. J Exp Biol 206:2287-2296.
Sullivan JP, Fahrbach SE, Robinson GE (2000) Juvenile hormone paces behavioral development
in the adult worker honey bee. Horm Behav 37:1-14.
166
Surgan M, Condon M, Cox C (2010) Pesticide risk indicators: unidentified inert ingredients
compromise their integrity and utility. Environ Manage 45:834-841.
Tasei J-Nl (2001) Effects of insect growth regulators on honey bees and non-Apis bees. A review.
Apidologie 32:527-545.
Tautz J, Sandeman DC, Heilmann HR (2008) The buzz about bees: Biology of a superorganism.
Berlin: Springer.
Thompson HM (2003) Behavioural effects of pesticides in bees--their potential for use in risk
assessment. Ecotoxicology 12:317-330.
Thompson HM (2010) Risk assessment for honey bees and pesticides - recent developments and
'new issues'. Pest Manag Sci 66:1157-1162.
Thompson HM, Wilkins S (2003) Assessment of the synergy and repellency of
pyrethroid/fungicide mixtures. B Insectol 56:131-134.
Tipping C, Bikoba V, Chander GJ, Mitcham EJ (2003) Efficacy of Silwet L-77 against several
arthropod pests of table grape. J Econ Entomol 96:246-250.
Toth AL, Kantarovich S, Meisel AF, Robinson GE (2005) Nutritional status influences socially
regulated foraging ontogeny in honey bees. J Exp Biol 208:4641-4649.
Toth AL, Robinson GE (2005) Worker nutrition and division of labour in honeybees. Anim
Behav 69:427-435.
Vanburen NWM, Marien AGH, Oudejans RCHM, Velthuis HHW (1992) Perizin, an acaricide to
combat the mite Varroa Jacobsoni - its distribution in and influence on the honeybee
Apis mellifera. Physiol Entomol 17:288-296.
Vandame R, Belzunces LP (1998) Joint actions of deltamethrin and azole fungicides on honey
bee thermoregulation. Neurosci Lett 255:61-61.
Vandenberg LN, Colborn T, Hayes TB, Heindel JJ, Jacobs DR, Jr., Lee DH, Shioda T, Soto AM,
vom Saal FS, Welshons WV, Zoeller RT, Myers JP (2012) Hormones and endocrine-
disrupting chemicals: low-dose effects and nonmonotonic dose responses. Endocr Rev
33:378-455.
VanEngelsdorp D, Evans JD, Donovall L, Mullin C, Frazier M, Frazier J, Tarpy DR, Hayes J,
Pettis JS (2009) "Entombed Pollen": A new condition in honey bee colonies associated
with increased risk of colony mortality. J Invertebr Pathol 101:147-149.
VanEngelsdorp D, Evans JD, Saegerman C, Mullin C, Haubruge E, Nguyen BK, Frazier M,
Frazier J, Cox-Foster D, Chen Y, Underwood R, Tarpy DR, Pettis JS (2009) Colony
collapse disorder: a descriptive study. PLoS One 4:e6481.
167
VanEngelsdorp D, Meixner MD (2010) A historical review of managed honey bee populations in
Europe and the United States and the factors that may affect them. J Invertebr Pathol 103
Suppl 1:S80-95.
VanEngelsdorp D, Speybroeck N, Evans JD, Nguyen BK, Mullin C, Frazier M, Frazier J, Cox-
Foster D, Chen Y, Tarpy DR, Haubruge E, Pettis JS, Saegerman C (2010) Weighing risk
factors associated with bee colony collapse disorder by classification and regression tree
analysis. J Econ Entomol 103:1517-1523.
Wallner K (1999) Varroacides and their residues in bee products. Apidologie 30:235-248.
Warne MS, Hawker DW (1995) The number of components in a mixture determines whether
synergistic and antagonistic or additive toxicity predominate: the funnel hypothesis.
Ecotox Environ Safe 31:23-28.
Wilson E (1975) Sociobiology-The New Synthesis: Belknap Press of Harvard University Press.
Winston M, Dropkin J, Taylor O (1981) Demography and life history characteristics of two
honey bee races (Apis mellifera). Oecologia 48:407-413.
Winston ML (1987) The biology of the honey bee. Cambridge, Mass.: Harvard University Press.
Wolschin F, Mutti NS, Amdam GV (2011) Insulin receptor substrate influences female caste
development in honeybees. Biol Lett 7:112-115.
Wu JY, Anelli CM, Sheppard WS (2011) Sub-lethal effects of pesticide residues in brood comb
on worker honey bee (Apis mellifera) development and longevity. PLoS One 6:e14720.
Yang EC, Chang HC, Wu WY, Chen YW (2012) Impaired olfactory associative behavior of
honeybee workers due to contamination of imidacloprid in the larval stage. PLoS One
7:e49472.
Yang EC, Chuang YC, Chen YL, Chang LH (2008) Abnormal foraging behavior induced by
sublethal dosage of imidacloprid in the honey bee (Hymenoptera: Apidae). J Econ
Entomol 101:1743-1748.
Yu SJ (2008) The toxicology and biochemistry of insecticides. Boca Raton: CRC Press/Taylor &
Francis.
Zhu WY, Schmehl DS, Mullin CA, Frazier JL (2013) Four common pesticides and their mixtures
in the hive environment have high oral toxicity to honey bee larvae.
Appendix A
The Development of Honey Bee Expert System
First, knowledge acquisition: determine the subset of colony components (indicators) to
represent the overall condition of colony health. This proposed research builds upon recent
modeling work that determined, described and quantified the major indicators of the colony
dynamics including: 1) reproduction quality of the queen, 2) brood-caring quality of nurse bees,
3) foraging quality of foragers, 4) quality of the nectar and pollen resource in the environment, 5)
quantity of the honey and pollen storage in the hive, and 6) disease levels of the hive (the levels
of Varroa mite infection). We have already collected data and information for this set of
indicators based on existing peer-reviewed literature and model simulations; therefore, this will
enable us to determine the reference condition (measures of indicator quality in the absence of
biotic and abiotic disturbances) and thresholds for selected colony indicators (the point/boundary
at which there is an abrupt change in hive attributes which may produce large and irreversible
responses in colony dynamics). Data and information related to these indicators and their
measures will be screened, cleaned, checked for accuracy and consolidated into an organized
format within the DSS model. Models in NetWeaver™ are based on dependency networks, which
are graphical depictions of rules (Saunders et al., 2005) (Figure 6-1, 2). At the bottom of a
dependency network are data links, which are used to hold, fetch, or modify raw data. There are
two types of data links; simple (Table 6-1) and calculated (Table 6-2). Simple data links fetch and
hold data from various sources (databases, GIS map layers, flat files, direct input, environmental
variables, etc.) (Saunders et al., 2005). Calculated data links modify data through networks of
calculation nodes chosen from a toolbox of arithmetic, trigonometric, selection, summation, etc
(Saunders et al., 2005).The scope of this program is refined to determine potential stressors
169
among biotic (pest, disease) and abiotic (temperature, pesticide contamination) stressors in a hive,
and to provide management suggestions such as nutrition supplemental feeding, mite treatment,
honey harvesting, and winter maintenance.
Table 6-1. The simple data requirements (user input) identified by the domain expert panel as sufficient for characterization of honey bee colony health
Data Description Colony Weight The unit is kg. Critical Time of the Season The month of evaluating the colony health Days since nectar flow Days since the start date of nectarflow in the same year
Direct estimate of forager population in one day
This is a nondestructive trapping system to capture returning honey bees, using weight to estimate the number of forager bees in one day.
Direct measurement of egg laying rate
Direct measurement of egg laying rate of queen within 24 hr using caging queen experiment
Does the environment have enough floral source for bees
To determine if the colony has low foraging force is caused by lacking of food source in the environment
Foraging Frequency No. of foragers leave the colony every three minutes during the peak flight time. Sunny, outside temperature is above 70F
Hive Temperature The temperature in the central area (brood area) of the hive. If the temperature is too high (>36c) or too low (<32c), the brood survival will be significantly affected.
Nectar Source The average sucrose concentration (mg/L) Number of adult bees per frame
This is a rough estimation. Use an average number of bees covered on one full frame (double sided)
Number of brood frames Including open and capped brood frames in a hive Number of frames covered by adult bees
This is a rough estimation. Use an average number of frames covered by adult bees (double-sided).
Pollen Storage during field season
The total weight of pollen stored (kg) in the hive. The sensitive season of pollen storage is spring and fall. It is a two-tailed boundary. Too much pollen indicates less food consumption and less brood population; too few pollen indicates the insufficient pollen foraging and less nursing.
Queen Age The queen egg laying is largely determined by the age of the queen and the starting time of nectar flow
Temperature in the hive The egg laying of queen is affected by the hive temperature. The size of a frame Choice: Deep/Medium frame, each cell size: 4.9/5.4mm2 The size of one cell in the frame
Use the method provided in the website to measure the size of one cell and the size of brood area
Time of foraging during one day
During the summer and fall season, the foraging time (hrs) usually varies from 1 to 6 hrs, depending on weather
Total area of adult bees The unit is square centimeter.
170
Total size of brood area The unit is square centimeter.
Varroa mite Number per bee sampling in October
Overwinter survival is highly correlated with the mite number in the hive. Now that the direct correlation has not been established, so here using a fuzzy logic to show the threshold of mite number per bee in a hive that is critical to colony survival. For a colony doomed to collapse within one year, we use a crisp argument to show the threshold of mite infection before winter
Table 6-2. The summary of calculated data (based on the stage model calculation) sufficient for characterization of honey bee colony health
Calculated Data Description/Function
Accurate estimate of brood number
This is estimated from the total size of brood area (open and capped) in the hive compared to the size of one cell in the frame
Accurate estimate of in-hive adult bees
The percentage of coverage of adult bees in the hive compared to the size of one deep frame, and multiplied with the linear coefficient.
Amount of honey for harvesting
It is estimated by subtracting the current honey stores by the 1.5 times optimal honey stores required for healthy colony growth and survival.
Brood nurse ratio (Fall) Winter bees which required intensive nursing from in-hive adult bees
(Brood+Forager) vs. Nurse Ratio
It is a simple representation of the ratio of nutrition distribution among three critical divisions of labor in the hive: whether the food consumer, vs. food provider can achieve the balance at the end of field season
Honey Store (in kg) Colony Weight as an index of honey production and nectar flow: Honey (g)= -1416+0.7604CW-57.142*Days+0.487*(Days)^2+0.00142*CW*Days
Indirect Estimate of Forager Population
Use the foraging frequency and the time of foraging during each day to estimate the number of foragers per day.
Number of adult bees in a bee hive
There are two methods to estimate: the preferred way is accurate estimate; if the preferred is missing, another way is rough estimation
Number of brood in a bee hive
There are two ways to estimate: one preferred way is accurate estimation by counting brood area; if the preferred method is not available, choose rough estimation.
Number of brood on one full frame
This depends on different size of frames used in the managed beehive. Choose the closet size of frame in the four options list here
Number of forager bees in one day
There are two ways to estimate: one preferred way is indirect estimation by counting the forager frequency; if the preferred is not available, the other option is direct estimation.
171
Optimal brood nurse ratio during the fall season
Quadratic function fitted from the stage model of a healthy colony during the fall season. The formula is y=2e-005x^2-0.0042x+0.53, x is the time of colony entering into preparation stage.
Optimal egg laying rate of a healthy queen
Relativedate = mod(date,360); maxProduction= (0.0000434)*(relativedate)^4.98293*exp(-0.05287*relativedate);
Optimal honey storage in a healthy hive
in Kg. From the stage model during the fall season: y=-2.9e 005Days^3+0.008 Days ^2-0.81 Days +46
Ratio of inadequate egg production
It is derived from comparing the actual egg production with the optimal egg laying rate. It has two tail boundaries.
Ratio of inadequate honey storage
It is derived from comparing the actual honey storage with the optimal honey stores during the fall season. It has two tail boundaries.
Ratio of inadequate nursing in the hive
This is estimated by comparing the brood nurse ratio by user input with the optimal brood nurse ratio during the fall season.
Rough estimation of number of adult bees in the hive
This is estimated by the number of adult bee frames and the estimated number of bees covered on one frame, assuming the average number of bees covered fully on one frame is 1500
Rough estimation of number of brood in the hive
Including both capped and open brood frames. Same method as estimation of adult bees, using the number of brood frames multiplied with the number of brood per frame
Second, decision support system (DSS) model: The development of DSS model includes
design, testing, evaluation, delivery and maintenance. This DSS model for honey bee colony
management has been developed and evaluated using the knowledge-based development software
NetWeaverTM. It is composed of a fuzzy-logic-based inference engine and a graphical user
interface for knowledge base developers (Saunders et al., 2005). It provides tools to construct a
hierarchy of dependency networks (graphical depictions of critical rules of colony dynamics)
within a fully editable graphic representation, and runs under the Microsoft Windows operating
system (Saunders et al., 2005). Using the six major colony indicators and associated reference
conditions and thresholds, the DSS model can be developed to compare the current condition
(user input) with reference conditions or threshold values, based on a range of arguments and
logical relationships, which depict how the key factors characterize the colony dynamics and is
172
defined by our existing model and domain experts. Figure 6-1 gives an example of how
beekeepers can determine whether one colony indicator (queen) is healthy in order to maintain
colony survival.
Figure 6-1. A dependency network for evaluating queen health as represented in NetWeaver™. In this dependency network, there are three items of data represented by the squares at the bottom of the figure: critical time of the year, the age of the queen, the ratio of egg laying rate of queen compared to that of healthy queen. Each of the data items is evaluated relative to the degree to which it satisfies its arguments. This network can be read as a rule as follows: “IF the time of year satisfies the specified argument. AND queen age does not satisfy the specified argument. OR Ratio of queen egg production compared to the optimal egg laying rate of a healthy queen (derived from the stage model) does not satisfy the argument. THEN the assertion of “queen is a concern” is true”. The degree to which the assertion is true is a function of the degree(s) to which the individual data satisfy their arguments and the types and arrangements of the logical nodes used within the network.
173
Figure 6-2. A dependency network for evaluating food quality in the hive as represented in NetWeaver™. In this dependency network, there are four items of data represented by the squares at the bottom of the figure: critical time of the year, the foraging frequency, the foraging time per day when outside temperature is suitable for foraging trips, and ratio of inadequate egg production compared to that of a healthy queen. Each of the data items is evaluated relative to the degree to which it satisfies its arguments. This network can be read as a rule as follows: “IF the time of year satisfies the specified argument, AND SOR node satisfies the three specified arguments related with foraging frequency, foraging time, and queen health. THEN the assertion of “food quality is a concern” is true”. The degree to which the assertion is true is a function of the degree(s) to which the individual data satisfy their arguments and the types and arrangements of the logical nodes used within the network. The SOR node is used to pick among these three competing evaluation methods where the methods are arranged in descending order of preference. Often the first choice method using foraging frequency of determining some property is not available, but another, less desirable method of foraging time during the day is available. The third method is using queen egg production, which is time-consuming evaluation. The SOR node provides a way of switching between these competing methods based on how well each method is currently functioning based on the sufficiency of the data driving the method.
We will develop four task modules of the DSS model including: 1) test seasonal colony
health (Figure 6-3), 2) diagnose the potential stressors in the weak colony (Figure 6-4), 3) predict
the survival time of the weak colony (Figure 6-5), 4) inform the threshold of colony indicators to
174
insure overwintering survival, and 5) suggest the specific colony management actions to be taken
to restore the weak colony (Figure 6-6).
Figure 6-3. The first task module-testing the seasonal colony health. It examines the colony health at each season: Spring colony health is determined by the queen reproduction (dependent on the queen’s age, the hive temperature, and the queen egg laying rate compared with optimal rate) and food quality and quantity; Summer colony health is determined by the foraging frequency, pollen quantity; Fall colony health is determined by the nursing quality represented by brood vs. nurse ratio and hive temperature; Winter colony health is determined by the adult bees number, the frequency of Varroa Mite infection, and amount of honey stores.
175
Figure 6-4. The second task module-diagnosing the potential stressors in a disrupted hive. There are six candidate stressors that can all weaken the colony fitness. This task is designed to check whether the colony has any problems of queen, pollen stores, honey stores, food quality, nursing or Varroa mite infection.
176
Figure 6-5. The third task module-predicting the survival of a disrupted colony. There are three major threats affecting colony survival: single threat caused by single stressors listed above; combinational threat caused by the synergistic or additive effects of multiple stressors; the overwinter collapse caused by insufficient adult population, or inadequate honey storage, mite infection and pollen stores before winter. This task is designed to evaluate whether the weakened colony is in an emergency situation and should be prioritized for restoration.
177
Figure 6-6. The last task module-making suggestion of colony management for restoring the weakened colony. There are eight groups of actions: if the queen is a concern, then requeen the colony; if nursing is a concern, then improve hive temperature/ supply brood pheromone/ provide diverse fresh pollen to stimulate egg production and rearing, or combine colonies with strong ones before winter; if the pollen/honey quantity is a concern, then feed the colony with supplemented pollen/honey; if the mite is a concern, then treat the colony; if the colony has extra honey, then determine the amount of honey ready for harvesting.
178
Appendix B
The Mixture Toxicity of Fungicide Formulations
Figure 7-1. Additive effects of the binary mixture of common fungicide formulations. Panel A shows the Kaplan-Meier survival plots for honey bee larvae reared on 24ppm Bravo/ 30ppm Pristine fungicide mixture and each component (ppm: mg/L). Panel B shows the Kaplan-Meier survival plots for honey bee larvae reared on 21ppm Nova/ 30ppm Pristine fungicide mixture and each component.
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
Pro
po
rtio
n S
urv
ivin
g
1 2 3 4 5 6
Time
24 ppm Bravo
30 ppm Pristine
30 ppm Pristine+ 24 ppm Bravo
Fungicide
A.
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
Pro
po
rtio
n S
urv
ivin
g
1 2 3 4 5 6
Time
21 ppm Nova
21 ppm Nova+30 ppm Pristine
30 ppm Pristine
Fungicide
B.
VITA
Wanyi Zhu
EDUCATION
• Ph.D. in Entomology and Operations Research, The Pennsylvania State University 08/2008~05/2013
• B.S. in Life Science and Technology, Northeast Forestry University, China 09/2003~09/2007
RESEARCH EXPERIENCE
Center for Pollinator Research, Department of Entomology, The Pennsylvania State University Research Assistant 01/2009 ~ present • Designed and conducted experimental and modeling study of chronic toxicity of common
pesticides to honey bee health • Designed and conducted field study of the effects of supplemental pollen feeding on
pesticide-disrupted colony health • Developed and analyzed population model of honey bee intracolonial dynamics using
differential and difference equations model • Develop and evaluate a knowledge-based decision support system for honey bee colony
management using NetWeaver TM software
RESEARCH GRANTS FUNDED
• Principal investigator for Pennsylvania Pollinators Research Grant funded by PA State Beekeepers Association, The Center for Pollinator Research 04/2011 ~ 08/2012 The effects of fungicides on honey bee colony health. (Grant: $7,000)
• Co-principal investigator for National Honey Board Grant 01/2011 ~ 09/2012 From Subtle to Substantial: a Stage-Structured Matrix Population Model for predicting combined roles of nutrition and pesticides on honey bee colony health. (Grant: $17,000)
• Co-principal investigator for National Honey Board Grant 11/2012 ~ 03/2012 A decision support system for honey bee colony management (Grant: $11,000)
AWARDS
• Graduate Student Travel Awards, The College of Agricultural Science, The Pennsylvania State University 09/2012
• William Yendol Memorial Research Fund, The Pennsylvania State University 11/2011 • Scholarship of the Foundation for the Preservation of Honey Bees for 2010-2011 • Graham Endowed Fellowship Award, The Pennsylvania State University 2008 ~ 2009 • College Scholarships for four consecutive years (GPA Top 5%), Northeast Forestry University,
China 2003 ~ 2007