ROBERT E. PAGE, JR., AND ERIC H. ERICKSON, JR.
Bee Research Unit, U.S. Department of Agriculture, Agricultural Research Service, Department of Entomology, University of Wisconsin, Madison, Wisconsin 53706
FORUM: Ann. Entomol. Soc. Am. 78: 149-158 (1985)
We investigate via computer simulation the effectiveness of regulating the process of Africanization of commercial honey bee populations in North America following the establishment of an Africanized feral population. Our results show that regulatory programs designed with the specific objective of identification and certification of non-Africanized, commercial colonies within an Africanized environment are not effective except under a narrow set of unlikely conditions. Under the conditions of our model, even low levels of immigration of Africanized drones into non-Africanized areas results in a high percentage of Africanization and a high probability of error of classification. We discuss the limitations of such programs, the kinds of data necessary to assess the likelihood of their success, and recommend a more effective direction for future research.
The spread of Africanized honey bees throughout South and Central America has resulted in much speculation regarding the bees’ potential impact on North American beekeeping and considerable interest in methods of identification and control. The United States Department of Agriculture Animal and Plant Health Inspection Service (APHIS) has recently published an action plan outlining the procedures to be used for the control and eradication of Africanized bees in the event of an infestation or establishment of a feral population in the United States (USDA 1983, Stibick 1984). APHIS intends to regulate queen rearing, package bee production, hobby beekeeping, and migratory beekeeping in and out of Africanized areas with the objective of preventing the spread of undesirable Africanized behavioral characteristics in the North American honey bee population. The central requirement of their regulatory plan is the ability to identify Africanized bees irrespective of the behavioral attributes in question and certify colonies of honey bees as non-Africanized (including introgressive hybrids) by making an accurate determination of the genotypes of queens and workers in colonies.
The current morphometric method used by APHIS to identify Africanized bees (Daly and Balling 1978) is slow, costly, and probably not acceptable for a large-scale program (Rinderer and Sylvester 1981). The most commonly proposed alternative method is identification based on electrophoretic phenotypes of discrete allozymes (Daly and Balling 1978, Nunamaker and Wilson 1981a, Rinderer and Sylvester 1981, Sylvester 1982). The rationale for using this method is based on the work of Ayala and Powell (1972) for classifying individuals belonging to different noninterbreeding sibling species of Drosophila. However, the effectiveness of this method for distinguishing between Africanized and non-Africanized colonies in a common environment has not been considered, is of paramount importance, and is the subject of this paper.
We present the results of a computer model that simulates the population genetics of a possible identification and certification program based on electrophoretic determination of individual honey bee genotypes. We demonstrate that even low levels of gene flow from an Africanized feral population into a commercial population can result in a high percentage of commercial colonies becoming Africanized and a large error in classification. We then discuss the efficacy of such a program in the context of a commercial beekeeping environment and make specific recommendations for further research.
Methods and Materials
The Model. For this study, we assume (although it is by no means certain) that some undesirable changes will occur in North American feral honey bee populations as a consequence of gene flow from South America via Central America and Mexico, or by an invasion population or successful point-source introduction. After this occurs, we assume the feral population is always Africanized and undesirable. Individual workers are considered Africanized if they are descendents of any individual from the feral population after Africanization has occurred. Colonies are Africanized if they have one or more Africanized subfamilies, the result of a queen mating with one or more Africanized drones.
The model simulates conditions that might exist in the United States with migratory beekeeping following Africanization of areas of the southern United States. A commercial beekeeper moves his colonies from a northern, non-Africanized location each fall or spring into a southern, Africanized environment to overwinter, to sell his pollination services, or to take advantage of early nectar flows. We assume that all feral colonies in this area are Africanized and are considered as such even with progressive gene introgression from the commercial population. While located in this area, his colonies supersede their queens, or swarm, and the new virgin queens that are produced mate with drones in the area, both Africanized (feral) and non-Africanized (commercial). (We will hence-forth refer to the North American population as European since it originated primarily from imported European races.) Before transporting these colonies back to their northern apiary locations, they are inspected and certified based on a determination of Africanized and European phenotypes in random samples of workers from colonies. Exact genotypic determination is made from electrophoretic phenotypes for one, two, or three independent gene loci. Queens of Africanized colonies are replaced (or the whole colony is replaced) by European queens raised in closed breeding populations or other non-Africanized environments. We assume that there is no systematic, annual requeening program for the commercial colonies. We also assume that there is no progressive Africanization of the commercial population that serves as the source of replacement queens. Violation of each of these assumptions changes the expectations we present in an opposite direction.
The model simulates the conditions of:
(1) haplodiploid population genetics;
(2) multiple, random mating of queens;
(3) differential population size and mating advantage of European and Africanized drones;
(4) random sampling of workers from commercial colonies;
(5) identification of Africanized colonies within commercial apiaries based on one, two, or three unlinked loci, two alleles per locus;
(6) replacement or destruction of queens or colonies identified as Africanized;
(7) progressive Europeanization of an Africanized, feral population;
(8) progressive Africanization of a regulated commercial population.
The sequence of events for the first generation is:
(1) Select the initial 100 commercial virgin queens from a large European base population in Hardy-Weinberg equilibrium for one, two, or three loci used for classification;
(2) mate each of these queens to 10 different males selected at random from this same large population; each male contributes an equal number of equally viable spermatozoa; spermatozoa are randomly mixed within the spermatheca;
(3) sample workers of each colony by random selection of an egg and a sperm from the spermatheca from each queen and make a determination whether the colony is Africanized based upon predetermined criteria;
(4) calculate all statistics;
(5) enter genes from the commercial population into the feral population in predetermined proportions, so newly established gene frequencies in the feral population are those expected from the binomial expansion of each locus;
(6) replace all queens heading colonies determined to be Africanized with new queens selected from the initial base population.
The sequence for subsequent generations is:
(1) Supersede all commercial queens by randomly selecting an egg and a stored sperm from each queen to represent the genotype of the new queen;
(2) mate all commercial, supersedure queens to 10 males each, each male selected at random from an infinitely large pool of drones composed of drones of commercial and feral colonies in proportions based on the predetermined levels of gene flow; steps 3 through 5 are identical to the initial generation above.
Each generation, the simulation calculates the following:
(1) The number of commercial colonies that have Africanized workers present;
(2) the number of commercial colonies determined to be Africanized by the classification of randomly selected workers;
(3) the number of European colonies classified as Africanized based on worker samples;
(4) the number of Africanized colonies classified European based on worker samples;
(5) the number of European commercial colonies correctly classified;
(6) the number of Africanized colonies correctly classified;
(7) the cumulative proportion of Africanized queens within commercial colonies.
Classification Criteria. We classify the allele at each locus that is in highest frequency in the feral population as Africanized. Our starting conditions for all simulations assume that the most frequent allele at each locus in the Africanized population occurs at a frequency of at least 0.95 and occurs in the commercial population at a frequency not greater than 0.05. This results in an initial probability of correct determination of a single gene of Africanized origin at this locus of at least 0.95, a 0.05 probability of calling an Africanized gene European, and a 0.05 probability of classifying a European gene in the commercial population as Africanized. The probability of misclassifying randomly selected, individual workers in a population where Africanized and European workers are equally frequent is 0.05 (Ayala and Powell 1972, Sylvester 1982); however, this error is dependent on the frequencies of Africanized and European individuals in the sample population and can range from 0.0025 to 0.0975.
For the case of two classifying loci, the probability of misclassification of an individual worker selected at random from a population where Africanized and European workers are equally frequent is reduced to 0.0025 if the classification of an individual as Africanized is made on the basis of “any Africanized allele at any locus.” The probability of error is further reduced to 0.000125 for the three-loci case. These probabilities are gene-frequency-dependent and change dramatically with gene flow between these populations.
In practice, colonies need to be classified on the basis of analyses of individual workers sampled from those colonies. Colony samples involve three separate sampling events: sampling of the drone population by the queens for mates; sampling of the gametes, those of the queen that become eggs, and those of the males from the queen spermathecae; selection of workers for analysis. Each of these events affect the outcome of the classification procedure. For these simulations, we eliminate the third sampling event by selecting worker samples directly from the gametes. This simplifies the procedure and yields a good approximation of the genotypic array of workers in large colonies.
The classification of colonies based on evaluation of worker samples can be complex. We sampled 2, 6, or 10 workers from each colony for identification. Classification of a colony as Africanized was determined by one of three methods that varied among simulations: (1) any worker sampled has an Africanized allele at any locus; (2) a sample of workers from a colony has an Africanized allele represented at each locus, but not necessarily in a single sampled worker; (3) any one of the workers sampled has an Africanized allele at all loci considered. We used methods 1 and 3 for most simulations because they represent the extreme classification criteria.
We chose six workers per colony as our sample size for most of our simulations because it results in a >0.95 probability of sampling both genes at a given locus of the queen. An equal probability of sampling each of the mates of each queen requires much larger sample sizes and would be too costly in application.
We tested different models by varying the number of loci used for determination, the gene frequencies at these loci, the number of workers sampled from each colony for classification, the method of classification, and the amount of gene flow from the Africanized (feral) population into the commercial population and from the commercial population into the feral population. We simulated 10 iterations of 20 generations for each model. We calculated directly from the results the mean and standard deviation for each characteristic measured.(1)
Results and Discussion
Migration. Low levels of gene migration from commercial populations to feral populations are necessary to maintain the gene frequency dispersion between populations that is needed to classify individual members. As European genes increase in frequency in the feral population, the probability of classifying an Africanized worker as European increases, resulting in an increase in frequency of Africanized colonies and queens in the commerical apiary (Tables 1-4; all tables are located in the Appendix). It should be noted, however, that increased migration of the more desirable European genes into the feral population should result in a more desirable, less Africanized feral gene pool.
High success rates of Africanized drones mating with commercial supersedure queens enhance colony classification because they increase the proportion of Africanized worker genotypes to be sampled and thereby increase the chance that one of them will be sampled. Even low levels of immigration result in high levels of Africanization because of the multiple mating of queens. Immigration rates of 0.50 result in virtually 100% Africanization of colonies headed by queens that mate in that environment.
A particularly notable deficiency in all of the classification methods we tested is their inefficiency in detecting Africanization under low levels of gene flow from the feral to the commercial populations. Rinderer and Sylvester (1981) state that detection of Africanized bees in the field is not at all difficult on the basis of colony behavior. If this is the case, we need a method for detecting low levels of gene introgression. In order to do this, large sample sizes are needed from each colony, a condition that greatly increases the misclassification of European colonies. Based on these simulations, Africanization will be widespread before levels of gene flow are sufficient for efficient classification and control.
These results point out the need to define clearly acceptable levels of Africanization in commercial and feral environments. A single subfamily of workers having an Africanized father may alter colony behavior sufficiently to merit rejection of the whole colony. There is also the associated probability that one or more of the members of this subfamily may become queens. We, by necessity, assume that all feral populations are Africanized, undesirable, and always remain so. In reality the composition of the feral population is expected to change with gene flow from the commercial populations. The effect of this gene flow upon the behavioral attributes of the feral population will determine the necessity to continue an identification and certification program. If behavioral attributes remain adverse even with significant change in the genetic composition of the population, then the result will be a diminished ability to detect Africanization and a progressively undesirable commercial population. If, however, the attributes of the feral population become less adverse with gene flow from the commercial population, then the ability to detect Africanization will still diminish but with less significant consequences.
(1) Authors will provide copies of the simulation program on request.
Classification Criteria. The method of classification affects both the probability of correctly classifying Africanized colonies and the probability of misclassifying those that are not. Method 1 is most efficient for detecting Africanized colonies, but it results in a large percentage of misclassification of European colonies when there is an overlap of allelic frequencies at classifying loci (Tables 1 and 2). Method 1 also results in the slowest Africanized gene introgression into the commercial population as well as the slowest accumulation of Africanized queens.
Method 3 results in fewer misclassifications of European colonies but is less effective for determining Africanized colonies (Tables 3 and 4).
Method 3 works very well under conditions of high mating activity or numerical advantage of Africanized drones but poorly under conditions of high rates of introgression from commercial to feral populations. Method 2 yields results intermediate between those of methods 1 and 3 and is not presented separately.
Gene Frequencies. Misclassification of European colonies is completely eliminated when the commercial base population does not have any Africanized alleles at the loci considered (Tables 2 and 4). However, when this is the case, the ability to identify correctly Africanized colonies in the commercial population decreases with classification method 1 under conditions of gene flow from commercial populations into Africanized feral populations. This decrease is a consequence of the reduced immigration of European alleles of the Africanized type into the Africanized population. The presence of alleles of Africanized type in the commercial population reduces the detectable rate of change of allele frequencies in the feral population with continuing introgression, and results in a higher rate of commercial colony rejection with queen supersedure. It is the progressive increase of European alleles in the Africanized, feral population that reduces the efficiency of the classification system. Initial gene frequencies over the range we tested have little effect on classification efficiency for classification method 3 and when migration only occurs from the feral into the commercial population.
Sample Size. Classification of Africanized colonies improves with increasing worker sample sizes (Table 5). Costs of analyses also increase because more workers must be processed and because more European colonies are rejected without cause. Economic analyses incorporating risk assessment are needed to determine how much misclassification is tolerable.
Numher of Loci. An increase in numbers of loci used for classification results in an increasing probability of correctly classifying individual workers (Table 6). However, the effect of increasing numbers of loci on the overall identification and certification program is complex.
Increasing numbers of loci result in lower classification precision under the conditions of classification method 3. This counterintuitive result is a consequence of gene introgression and increasing frequencies of Africanized workers of overlap genotypes that are not classified Africanized. When method 1 is used for classifying colonies, increasing numbers of loci do result in increasing precision for identifying Africanized colonies because overlap genotypes are classified as Africanized. However, as a consequence, this method results in greater misclassification of European commercial colonies under the initial conditions of allele frequency overlap.
After numerous efforts directed toward finding electrophoretic variability in and among honey bee populations, there are only five potentially usable marker loci available for this type of program. Three of these show possible gene frequency differentiation between South American and North American bee populations; two of these are expressed in adult honey bees (Mestriner 1969, Mestriner and Contel 1972, Bruckner 1974, Sylvester 1976, Contel et al. 1977, Martins et al. 1977, Pamilo et al. 1978, Gartside 1980, Nunamaker and Wilson 1981a,b, 1982, Sheppard and Berlocher 1984). Even if more usable markers are detected, they will not necessarily result in more precision in classification and can in fact lead to less precision within cross mating populations.
The results of these simulations clearly demonstrate that regulatory programs directed toward the identification and certification of Africanized colonies in a commercial context are effective only under a strict set of conditions: extreme numerical or mating advantage of Africanized drones; low levels of gene flow from the commercial populations into the feral populations; and extreme initial gene frequency differences between commercial and feral populations at the loci used for classification.
It is unlikely that the restrictive conditions necessary for the success of an identification and certification program as outlined by APHIS will be realized in commercial beekeeping areas of the United States for the following reasons: (1) Cross matings between European drones and Africanized queens do occur with apparently little constraint (Kerr and Bueno 1970). There is no published evidence for sufficient mating advantage of Africanized drones to meet the stated conditions (see also Kerr et al. 1980). (2) Extensive hybridization appears to have occurred between African and European populations throughout Brazil. This is suggested by the large number of sex alleles segregating in some of these populations (Adams et al. 1977). Africanized bees show considerable variability in behavior and appearance in Venezuelan populations, suggesting a broad genetic background (R. Page, personal observations). (3) Estimated frequencies of proposed allozyme markers in North and South America show considerably more overlap than those we assume for our simulations (Sylvester 1976, 1982).
New APHIS research is currently directed toward finding and identifying specific hydrocarbons of taxonomic significance (Carlson and Bolton 1984, McDaniel et al. 1984). This method, as well as morphometric analysis, has the same inherent difficulty as electrophoretic methods: cross breeding between commercial and Africanized populations will change the expected genotypic and phenotypic distributions of both populations toward those intermediate between the two, resulting in more phenotypic overlap and the inability to classify the intermediate populations. Introgressive breeding throughout Central America and Mexico, even if minimal, can greatly alter the reliability of the classification system.
It may still be possible to construct usable programs for the purpose of slowing the spread of undesirable effects of gene flow from South America, but these programs must be based on information regarding the genetic mechanisms behind the behavioral attributes, gene frequency distributions for marker loci, population densities, reproductive dynamics, mating behavior, and the costs of analyses. We believe that for the present time the most promising area of research directed toward the goal of certification of European bees is the development of closed-population breeding programs and faster, more efficient methods of queen marking and finding.
Adams, J., E. D. Rothman, W. E. Kerr, and Z. L. Paulino. 1977. Estimation of the number of sex alleles and queen matings from diploid male frequencies in a population of Apis mellifera. Genetics 86: 583-596.
Ayala, F. J., and J. R. Powell. 1972. Allozymes as diagnostic characters of sibling species of Drosophila. Proc. Natl. Acad. Sci. U.S.A. 69: 1094-1096.
Bruckner, D. 1974. Reduction of biochemical polymorphism in honeybees (Apis mellifica). Experientia 30: 618-619.
Carlson, D. A., and A. B. Bolton. 1984. Identification of Africanized and European bees using extracted hydrocarbons. Bull. Entomol. Soc. Am. 30: 32-35.
Contel, L. P. B., M. A. Mestriner, and E. Martins. 1977. Genetic control and Developmental expression of malate dehydrogenase in Apis mellifera. Biochem. Genetics 15: 859-876.
Daly, H. V., and S. S. Balling. 1978. Identification of Africanized honeybees in the Western Hemisphere by discriminant analysis. J. Kans. Entomol. Soc. 51: 857-869.
Gartside, D. F. 1980. Similar allozyme polymorphism in honeybees (Apis mellifera) from different continents. Experientia 36: 649-650.
Kerr, W. E., and D. Bueno. 1970. Natural crossing between Apis mellifera adansonii and Apis mellifera ligustica. Evolution 24: 145-155.
Kerr, W. E., M. R. Martinho, and L. S. Goncalves. 1980. Short communication: kinship selection in bees. Rev. Brasil. Genet. 3: 339-344.
Martins, E., M. A. Mestriner, and E. P. B. Contel. 1977. Alcohol dehydrogenase polymorphism in Apis mellifera. Biochem. Genetics 15: 357-366.
McDaniel, C. A., R. W. Howard, G. J. Blomquist, and A. M. Collins. 1984. Hydrocarbons of the cuticle, sting apparatus, and sting shaft of Apis mellifera L. Identification and preliminary evaluation as chemotaxonomic characters. Sociobiol. 8: 287-298.
Mestriner, M. A. 1969. Biochemical polymorphism in bees (Apis mellifera ligustica). Nature (London) 223: 188-189.
Mestriner, M. A., and E. P. B. Contel. 1972. The P-3 and EST loci in the honeybee Apis mellifera. Genetics 72: 733-738.
Nunamaker, R. A., and W. T. Wilson. 1981a. Comparison of MDH allozyme patterns in the African honey bee (Apis mellifera adansonii L.) and the Africanized populations of Brazil. J. Kans. Entomol. Soc. 54: 704-710.
1981b. Malate dehydrogenase and nonspecific esterase isoenzymes of eggs of the honey bee (Apis mellifera L.). Comp. Biochem. Physiol. 70B: 607-609.
1982. Isozyme changes in the honeybee, Apis mellifera L., during larval morphogenesis. Insect Biochem. 12: 99-104.
Pamilo, P., S.-L. Varvio-Aho, and A. Pekkarinen. 1978. Low enzyme variability in Hymenoptera as a consequence of haplodiploidy. Hereditas 88: 93-99.
Rinderer, T. E., and H. A. Sylvester. 1981. Identification of Africanized bees. Am. Bee J. 121: 512-516.
Sheppard, W. S., and S. H. Berlocher. 1984. Enzyme polymorphism in Apis mellifera from Norway. J. Apic. Res. 23: 64-69.
Stibick, J. N. L. 1984. Animal and Plant Health Inspection Service strategy and the African honey bee. Bull. Entomol. Soc. Am. 30: 22-26.
Sylvester, H. A. 1976. Allozyme variation in honeybees (Apis mellifera L.). Ph.D. dissertation, University of California, Davis.
1982. Electrophoretic identification of Africanized honeybees. J. Apic. Res. 21: 93-97.
U. S. Department of Agriculture. 1983. African honey bee action plan. Animal and Plant Health Inspection Service, Plant Protection and Quarantine, Emergency Programs. Government Printing Office, Washington, D.C.
Received for publication 18 July 1984; accepted 26 November 1984.
This appendix contains the tabulated results of the computer simulations discussed in the text. For each table, migration rates are presented for feral to commercial populations (F -> C) and commercial to feral (C -> D). Means and standard deviations are presented based on 10 iterations of each simulation. Row values are as follows:
I. The number of colonies in the commercial population (based on 100 colonies) that have Africanized workers;
II. the number of Africanized colonies correctly identified based on the worker samples;
III. the number of colonies classified Africanized based on worker samples;
IV. the number of colonies with Africanized queens;
V. the number of colonies having Africanized queens that are correctly identified based upon the workers sampled.
Row III for generation 1 shows the continual error associated with misclassification of European queens mated outside of the Africanized area. The difference between I and II shows the efficiency for determination of Africanized colonies based on worker samples. The difference between II and III shows the number of European supersedure queens incorrectly classified. The number of European colonies correctly classified is given by the formula 100 – row I – (row III – row II). The difference between rows IV and V demonstrates the efficiency of detecting Africanized queens.
The relevant simulation parameters for individual tables are:
Table 1. Results of computer simulations for the following conditions: (1) two classifying loci; (2) initial frequency of the designated Africanized allele is 0.95 for each locus in the feral population; (3) initial frequency of the Africanized allele is 0.05 for each locus in the commercial population; (4) six workers are randomly selected from each colony for identification; (5) method 1 is used for classification of Africanization.
Table 2. The initial conditions for the models generating the data presented here are identical to those for Table 1 except for the following: (1) the frequency of the Africanized allele is 1.00 at both determining loci in the feral population; (2) the frequency of the Africanized allele is 0.00 at both loci in the commercial populations.
Table 3. The initial conditions for the models generating the data presented here are identical to those for Table 1 except that method 3 was used for classifying worker genotypes.
Table 4. The initial conditions for the models generating the data presented here are identical to those for Table 1 except for the following: (1) the frequency of the Africanized allele is 1.00 at both determining loci in the feral population; (2) the frequency of the Africanized allele is 0.00 at both loci in the commercial population; and (3) method 3 was used for classifying worker genotypes.
Table 5. Initial conditions for the models generating these results are as those presented for Tables 1 and 3 except for differences in the number of workers sampled from each colony, each generation, as indicated.
Table 6. These data show the effects of increasing numbers of loci used for classifying workers. Sets A, B, and C assume maximum gene frequency dispersion between feral and commercial populations with one, two, and three loci, respectively, using classification method 3 and samples of six workers per colony. Set D has starting conditions identical to Table 3 except three loci are used. Set E has starting conditions identical to Table 1 except three loci are used.