Annu. Rev. Entomol. 2001. 46:413–40 Copyright c 2001 by Annual Reviews. All rights reserved MODELS OF DIVISION OF LABOR IN SOCIAL INSECTS Samuel N. Beshers Department of Entomology, University of Illinois, Urbana, Illinois 61801; e-mail: [email protected]Jennifer H. Fewell Department of Biology, Arizona State University, Tempe, Arizona 85287; e-mail: [email protected]Key Words polyethism, task allocation, self-organization, response threshold models, foraging for work, self-reinforcement, social inhibition ■ Abstract Division of labor is one of the most basic and widely studied aspects of colony behavior in social insects. Studies of division of labor are concerned with the integration of individual worker behavior into colony level task organization and with the question of how regulation of division of labor may contribute to colony efficiency. Here we describe and critique the current models concerned with the proximate causes of division of labor in social insects. The models have identified various prox- imate mechanisms to explain division of labor, based on both internal and external factors. On the basis of these factors, we suggest a classification of the models. We first describe the different types of models and then review the empirical evidence supporting them. The models to date may be considered preliminary and exploratory; they have advanced our understanding by suggesting possible mechanisms for division of labor and by revealing how individual and colony-level behavior may be related. They sug- gest specific hypotheses that can be tested by experiment and so may lead to the development of more powerful and integrative explanatory models. CONTENTS INTRODUCTION ................................................ 414 DIVISION OF LABOR AND ITS CAUSES ............................. 415 What is Division of Labor? ........................................ 415 How Do Individual Workers Choose Tasks? ............................ 416 Integrating from the Individual to the Colony ........................... 417 MODEL DESCRIPTIONS .......................................... 417 Response Threshold Models ....................................... 418 Integrated Threshold-Information Transfer Model ........................ 420 0066-4170/01/0101-0413$14.00 413 Annu. Rev. Entomol. 2001.46:413-440. Downloaded from arjournals.annualreviews.org by University of Leiden - Waleus Library on 01/27/10. For personal use only.
Samuel N. BeshersDepartment of Entomology, University of Illinois, Urbana, Illinois 61801;e-mail: [email protected]
Jennifer H. FewellDepartment of Biology, Arizona State University, Tempe, Arizona 85287;e-mail: [email protected]
Key Words polyethism, task allocation, self-organization, response thresholdmodels, foraging for work, self-reinforcement, social inhibition
■ Abstract Division of labor is one of the most basic and widely studied aspectsof colony behavior in social insects. Studies of division of labor are concernedwith the integration of individual worker behavior into colony level task organizationand with the question of how regulation of division of labor may contribute to colonyefficiency.
Here we describe and critique the current models concerned with the proximatecauses of division of labor in social insects. The models have identified various prox-imate mechanisms to explain division of labor, based on both internal and externalfactors. On the basis of these factors, we suggest a classification of the models. Wefirst describe the different types of models and then review the empirical evidencesupporting them.
The models to date may be considered preliminary and exploratory; they haveadvanced our understanding by suggesting possible mechanisms for division of laborand by revealing how individual and colony-level behavior may be related. They sug-gest specific hypotheses that can be tested by experiment and so may lead to thedevelopment of more powerful and integrative explanatory models.
Social insect colonies are groups of individuals that live together and reproduceas a unit. The colony represents a level of organization above the individual or-ganism with its own characteristic morphology, behavior, internal organization,and life history pattern (119, 121). Social insect biologists face the challenge ofintegrating between the individual and colony levels of organization. Recognizingthis challenge, EO Wilson declared that “the reconstruction of mass behavior froma knowledge of the behavior of single colony members is the central problem ofinsect sociology” (121, p. 227).
Classical models of colony organization focused on the adaptive value of socialstructure (73, 120, 125, 126). More recently the focus has shifted to the mecha-nistic processes that generate colony organization and behavior. Recent modelstreat the social insect colony as a self-organized, decentralized system in whichbehavior emerges from the independent actions and decisions of workers. Self-organization models have been used to describe numerous colony processes, in-cluding homeostasis (15, 118), mass action responses (8, 18, 19, 23, 24, 65), andcolony construction (8, 59).
Division of labor, in which different workers specialize on subsets of the tasksperformed by a colony, is one of the most prominent features of social insectcolony behavior (73, 121, 127). Our purpose is to discuss current models of theproximate mechanisms of division of labor and to assess the present and futurecontributions of models to our understanding. In the sections ahead we first providethe necessary background on division of labor and highlight some of the keyissues facing researchers in this area. We describe and critique the current models,proceed to a review of the empirical evidence, and conclude by summarizing thestate of the art and offering our own view of prospects for the future.
Ann
u. R
ev. E
ntom
ol. 2
001.
46:4
13-4
40. D
ownl
oade
d fr
om a
rjou
rnal
s.an
nual
revi
ews.
org
by U
nive
rsity
of
Lei
den
- W
aleu
s L
ibra
ry o
n 01
/27/
10. F
or p
erso
nal u
se o
nly.
P1: UBI
November 6, 2000 13:30 Annual Reviews AR119-14
MODELS OF DIVISION OF LABOR 415
DIVISION OF LABOR AND ITS CAUSES
What is Division of Labor?
Division of labor is fundamentally a stable pattern of variation among workerswithin a colony in the tasks they perform (73). We can characterize it more preciselyby saying that (a) each worker specializes on a subset of the complete repertoire oftasks performed by the colony, and (b) this subset varies across individual workersin the colony.
Two general patterns of division of labor are recognized in social insects: tempo-ral polyethism, or age-correlated patterns of task performance, and morphologicalpolyethism, in which a worker’s size and/or shape is related to its performance oftasks. Temporal polyethism is widespread in social insects and invariably followsthe pattern of young workers performing tasks within the nest and older workersperforming outside tasks such as foraging and defense (reviewed in 89). Morpho-logical polyethism is found in termites and in those ant species with distinguishablesubcastes within the worker caste (73). Patterns of morphological polyethism arevariable; one generalization that appears to hold is that the more extreme sub-castes, in either size or morphology, have more specialized behavior and narrowrepertoires (73). The most common specializations are for defense and foraging.Other roles of morphologically specialized workers include food processing andfood storage (46).
Much early work on division of labor was devoted to discovering correla-tions between behavior and worker age or morphology and to defining behavioral“castes” on the basis of these correlations (73, 98, 125, 126). This approach hasbeen invaluable for studies of the ecology of division of labor, but it is not suf-ficient for studies of mechanisms of division of labor, for at least two reasons.First, much of the variation in task performance among individual workers occursindependently of age or morphological variation (4, 10, 12, 40, 60). Second, un-der normal conditions a worker’s age and/or size may be good predictors of thetasks that worker is likely to perform, but workers often respond to changes inthe social environment by varying task performance independently of specific ageor size constraints. For example, “single-cohort” honey bee colonies, comprisedof workers that are all the same age, differentiate into hive workers and foragers,following the pattern of a normal colony (92). Third, as colony labor demandschange, workers show behavioral flexibility, either performing tasks not previouslyseen in their repertoires or switching from one task to another (41, 89). Thoughthere are examples in the literature of lack of flexibility (e.g. 66, 123), in the ab-sence of more evidence it seems reasonable to assume as a working hypothesis thatthe workers of most social Hymenoptera are totipotent (able to perform all tasks)except for reproduction (64, 121). Thus, models of division of labor must incor-porate both variation in task performance among workers and individual workerflexibility.
Ann
u. R
ev. E
ntom
ol. 2
001.
46:4
13-4
40. D
ownl
oade
d fr
om a
rjou
rnal
s.an
nual
revi
ews.
org
by U
nive
rsity
of
Lei
den
- W
aleu
s L
ibra
ry o
n 01
/27/
10. F
or p
erso
nal u
se o
nly.
P1: UBI
November 6, 2000 13:30 Annual Reviews AR119-14
416 BESHERS¥ FEWELL
Figure 1 Factors known or hypothesized to affect individual workers decisions to perform tasks.The figure is divided into internal and external regions, meaning within and outside the worker.Numbers on the arrows indicate effects that are included in each type of model: 1, responsethresholds; 2, information transfer; 3, self-reinforcement; 4, social inhibition; 5, foraging forwork; 6, network task allocation.
How Do Individual Workers Choose Tasks?
The decision of an individual worker to perform a task lies at the heart of divisionof labor. Figure 1 shows factors that are known or hypothesized to affect indi-vidual task performance. We divide these into internal and external factors, basedon whether they are generated as a result of the internal state of the individualor via interactions with the colony environment. Internal factors include genetic,neural, and hormonal factors and the effects of experience; external factors in-clude the stimuli that elicit task performance and worker-worker interactions thatcommunicate task needs.
Obviously, internal and external conditions interact. For example, interactionswith other workers may affect an individual’s motivational state, and performanceof a task may increase its intrinsic probability of performing that task again. Inturn, performance of a task by a worker affects the stimuli perceived by the rest ofthe colony. Thus, there are more connections and feedback loops between thesefactors than we have drawn.
Ann
u. R
ev. E
ntom
ol. 2
001.
46:4
13-4
40. D
ownl
oade
d fr
om a
rjou
rnal
s.an
nual
revi
ews.
org
by U
nive
rsity
of
Lei
den
- W
aleu
s L
ibra
ry o
n 01
/27/
10. F
or p
erso
nal u
se o
nly.
P1: UBI
November 6, 2000 13:30 Annual Reviews AR119-14
MODELS OF DIVISION OF LABOR 417
Current experimental and theoretical approaches to the problem of individualtask choice are encompassed by three specific questions:
1. What are the decision rules that workers follow in performing tasks? Oneway of understanding division of labor is to find phenomenologicalrules—that is, rules that describe the behavior without reference tomechanisms—that can account for the behavior patterns of individualworkers (e.g. 93, 113).
2. How do workers obtain information about task needs? Task performance isa response by the worker to the environmental information it perceives. Forany given task, workers may gain information from local stimuli within thenest or through interactions with other workers or both (e.g. 42, 48).
3. What internal mechanisms underlie the behavioral rules for taskperformance? Both genetic and hormonal effects have been shown to beimportant (e.g. 57, 68, 92), and studies are beginning to venture into thenervous system (e.g. 96, 97, 117, 128).
Integrating from the Individual to the Colony
Colony organization emerges from the decisions and actions of individual work-ers, and the consequences at the colony level of a particular behavioral rule ormechanism are in general too complex to be guessed intuitively or visualized. Forexample, individual workers can vary in their task repertoires and in how muchthey specialize on one or a few tasks. Colony-level patterns include the sizes oftask groups, the patterns of overlap among task groups (122), and short-termtask allocation (41). The models we discuss attempt to link the patterns of taskperformance at the individual and colony levels.
MODEL DESCRIPTIONS
We describe six classes of models, based on their main hypotheses about thecauses of division of labor: response threshold, integrated information transfer,self-reinforcement, social inhibition, foraging for work (FFW), and network taskallocation models. The response threshold, self-reinforcement, and social inhi-bition models address the question of how workers vary in their responses toinformation about a task. The FFW and network models seek to explain behav-ioral variation in terms of the flow of information about task needs and the localinformation available to each worker. The threshold-information transfer modelexamines how integrated effects of genetic variation in response probability andof information flow affect individual worker behavior.
In this section we describe each class of model and the specific variationsthat have been published, and we indicate how the various models are related. InFigure 1, numbers on the arrows indicate which factors are invoked as causes in
Ann
u. R
ev. E
ntom
ol. 2
001.
46:4
13-4
40. D
ownl
oade
d fr
om a
rjou
rnal
s.an
nual
revi
ews.
org
by U
nive
rsity
of
Lei
den
- W
aleu
s L
ibra
ry o
n 01
/27/
10. F
or p
erso
nal u
se o
nly.
P1: UBI
November 6, 2000 13:30 Annual Reviews AR119-14
418 BESHERS¥ FEWELL
each model, showing that the various models emphasize different subsets of thelist of possible causes.
Response Threshold Models
Response threshold models are framed around the hypothesis that workers haveinternal thresholds for responding to task-specific stimuli and that variation in taskthresholds among workers in a colony generates division of labor (93). Thus, inFigure 1, threshold models relate the internal threshold, the perceived stimulus,and the decision to perform a task. These models postulate that tasks are performedin response to specific stimuli and that a worker performs a task when a stimulusexceeds its internal threshold. The behavioral program of a worker includes aresponse threshold for every task in the repertoire, and it is assumed that the“default state” of a worker is to remain quiescent and not to attempt to performany task (76, 93). Variation among workers in response thresholds could comefrom many sources, both fixed and variable. For example, the model as proposedby Robinson & Page (78, 93) assumes that genotypic differences are responsiblefor most of the variation.
Variation in response thresholds generates a system that combines individualtask specialization and colony task flexibility. The subset of workers with lowestthresholds performs a given task even at low stimulus levels; they become thespecialists for that task (93). However, because all workers have some thresholdfor a task, higher stimulus levels result in the recruitment of additional workersinto a task group.
The response threshold model incorporates a negative feedback loop in whichthe performance of a task by a worker decreases the stimulus level for that task.Thus, if one worker has a lower threshold for a task than another worker, it not onlyis likely to respond sooner to the task, but it reduces the stimulus levels for that taskso that they may never reach the second worker’s threshold (76, 77). As a result,small intrinsic variation in task preferences or responsiveness may be amplifiedinto large differences in task repertoires and in the frequency of task performance(30, 77).
The idea that variation in task performance is driven by differences in responsethresholds is not new; it appeared at least as early as Wilson’s work in 1976(122). However, these models differ from earlier theoretical approaches to colonyorganization in that they identify variation in this worker-environment interactionas a primary driving force in colony social organization. Robinson (88) explicitlyinvoked changes in response thresholds with age to explain temporal polyethismin honey bees. Calabi (11) and Robinson & Page (78, 93) independently presentedverbal response threshold models. The response threshold concept and modelshave been the subject of several recent reviews (3, 5, 7).
Mathematical Treatments of the Response Threshold ModelsThe responsethreshold model, with emphasis on the negative feedback loop, was formally
Ann
u. R
ev. E
ntom
ol. 2
001.
46:4
13-4
40. D
ownl
oade
d fr
om a
rjou
rnal
s.an
nual
revi
ews.
org
by U
nive
rsity
of
Lei
den
- W
aleu
s L
ibra
ry o
n 01
/27/
10. F
or p
erso
nal u
se o
nly.
P1: UBI
November 6, 2000 13:30 Annual Reviews AR119-14
MODELS OF DIVISION OF LABOR 419
modeled by Page & Mitchell (76, 77). They set up a network ofN individualsandK tasks, assuming universal connectivity so that each worker was affectedby the behavior of every other worker. They performed simulation tests of theirmodel in which they exposed workers individually to a stimulus environment foreither one task (76) or two tasks (77). Workers were switched “on” for a task ifthe environment exceeded their intrinsic thresholds. When workers performed atask, stimulus levels were ratcheted down, and if not then stimulus levels wereincreased. The model achieved a stable equilibrium when the number of workersperforming a task matched the stimulus level for the task and when individualworkers maintained constant task performance probabilities. This equilibrium statewas based on, and qualitatively matches, the homeostatic patterns of task allocationto pollen foraging by honey bees (31, 33). When they perturbed away from thisequilibrium, colonies returned to a steady state in which individuals returned totheir prior task state. The equilibrium itself was affected by the mean and varianceof thresholds in the colony (76, 77).
Bonabeau et al (6, 7) built an analytical model using assumptions similar tothose of Page & Mitchell (76). Their model included variation in thresholdsfor performing a task, in which individuali was assigned a probabilityTθ ij ofperforming taskj, given a stimulus levels and a response thresholdθ , so thatTθ ij (sj) = s2
j /(s2j + θ2
ij ). Colony stimulus levels were constantly incremented by afactorδ but were decremented by performance of the task. Because the incrementexceeded task performance effects, the model did not generate an equilibriumstate. However, it did generate expected changes in task performance in relation tochanges in the proportion of individuals with low vs high thresholds that quan-titatively matched empirical data on grooming and social contact in the ant genusPheidole(124).
Bonabeau et al extended their model in several ways, including allowing theperception of a stimulus to depend on a worker’s current task, as would be thecase for spatially localized tasks. Given this assumption, the model yields a weaktemporal polyethism and can also account for variation in behavioral developmentin a colony in which thresholds vary due to genetic effects (7).
Building on Response Threshold ModelsThreshold models focus on a specific,limited part of the process that leads to worker task performance, and they incor-porate numerous simplifying assumptions. For example, most models assume thatthresholds are fixed. There is evidence that response thresholds for some stimulican vary considerably over the lifetime of the worker (81), and this is likely to betrue for task-related stimuli. The self-reinforcement model integrates experience-based variation in task performance into a threshold model (109). FFW and socialinhibition models provide alternative mechanisms for changes in task performance,but which can potentially be integrated with variation in response threshold models(7, 36, 49, 50, 108, 112).
To actually implement a threshold model, one must make assumptions abouthow response thresholds are distributed in the worker population. For example,
Ann
u. R
ev. E
ntom
ol. 2
001.
46:4
13-4
40. D
ownl
oade
d fr
om a
rjou
rnal
s.an
nual
revi
ews.
org
by U
nive
rsity
of
Lei
den
- W
aleu
s L
ibra
ry o
n 01
/27/
10. F
or p
erso
nal u
se o
nly.
P1: UBI
November 6, 2000 13:30 Annual Reviews AR119-14
420 BESHERS¥ FEWELL
Robinson & Page (93) assumed normal distributions. However as Fewell & Bertram(29) (information transfer model, below) demonstrate, the phenotypic distributionfor task performance can strongly affect how variation in individual-worker be-havior integrates into colony-level behavior. Another issue is how the stimuli fora given task are perceived. For most tasks the actual stimulus is not known, andso most threshold models do not address this issue.
Mechanisms for transmitting information about task opportunity and nestmatetask performance must also be considered. Where threshold models make anyassumptions about encounter patterns between workers and tasks, threshold modelsgenerally assume that all workers are equally likely to encounter all tasks (7). Thusthe effects of the spatial distribution of tasks and of worker movements need to beadded. To bridge the gap between individual rules and colony behavior, some orall of these assumptions may need to be made more realistic.
Typical task repertoires of social insect colonies range from 20–40 tasks (73),but no threshold models yet have considered more than three (7, 77). Thus, thecentral question still remains of how much of the organization within a complexsocial group can be explained by variation in response thresholds alone.
Many properties of threshold models remain to be explored. Beshers et al (3)discussed extensions of the response threshold concept, including the organizationof thresholds within an individual and the distribution of thresholds across anentire colony, how the threshold distribution might affect measures of colonyperformance, and how modulation of response thresholds could be used to improvecolony performance.
Integrated Threshold-Information Transfer Model
Fewell & Bertram (29) developed an analytical model which they integratedbetween the questions of (a) how do workers receive information about a task,and (b) how does variation in stimulus perception affect worker task performance.They began with the basic assumption of the threshold model that workers performa task when the stimulus they encounter matches an intrinsic threshold. Both thethreshold distribution for a task and the process by which workers perceived thetask stimulus could be varied in the model. Threshold distributions were geneticallybased. The distribution of thresholds in the colony was based on the number ofloci coding for the threshold and varied from bimodal (a single locus effect withdominance) to normal (multiple loci with purely additive effects). Workers couldperceive task needs either directly, by random encounters with tasks distributednormally in space, or via social information transfer, in which all workers receivedthe information concurrently.
Fewell & Bertram (29) used the information transfer model to predict colony-level response patterns to graded changes in stimulus levels for a given task. For allbut one combination of threshold distribution and stimulus perception mechanism,the number of individuals performing the task increased gradually in response toincreases in task stimulus levels. The model predicted that a normally distributed
Ann
u. R
ev. E
ntom
ol. 2
001.
46:4
13-4
40. D
ownl
oade
d fr
om a
rjou
rnal
s.an
nual
revi
ews.
org
by U
nive
rsity
of
Lei
den
- W
aleu
s L
ibra
ry o
n 01
/27/
10. F
or p
erso
nal u
se o
nly.
P1: UBI
November 6, 2000 13:30 Annual Reviews AR119-14
MODELS OF DIVISION OF LABOR 421
pattern of task thresholds would generate a graded response, independently ofhow individuals received information about a task. However, the combination ofbimodally distributed threshold phenotypes and simultaneous information trans-fer generated a step-wise colony response. The number of workers performing thetask remained low until a set-point was reached, at which the stimulus matched thethresholds of workers with higher thresholds. At this point the colony respondedimmediately by increasing task performance. This response prediction is similarto that of the information center model of recruitment (100) in which social infor-mation transfer is used as the major mechanism for transmitting a task stimulus.However, the information center model assumes no explicit genetic effects on taskperformance.
Self-Reinforcement Models
Another group of models asks whether division of labor can be generated by theeffects of experience. Self-reinforcement is a postulated mechanism in which suc-cessful performance of a task increases the probability of performing that taskagain, while unsuccessful performance or lack of opportunity reduces the proba-bility of performance (24, 84, 109). This is similar to the concept of task fixation(35, 114). It has been suggested (105) that self-reinforcement can potentially ex-plain the occurrence of specialists and generalists in a wide variety of biologicalsystems.
This mechanism can lead to the development of task specialists. Deneubourget al (24) used a self-reinforcement model to show how foraging specializationscould develop that matched food location and quantity. They suggested also thatself-reinforcement, coupled with natural variation in worker age, could accountfor temporal polyethism. Plowright & Plowright (84) used a self-reinforcementsimulation to generate elitism (73), in which some workers perform a particulartask at very high frequencies. In their simulations, the probability of performinga task was affected by both the external stimulus environment (E) and internalreinforcement (I ), according to the equationP = 1−e−IKE (with K as a constant).Workers were given initial random probabilities of encountering tasks, andI wasincremented when tasks were performed. When conditions were set at an interme-diate level of low but positive feedback, the frequency of task performance becamebimodally distributed so that workers were either task specialists or inactive.
Theraulaz et al (109) obtained similar results with a model derived from thefixed-threshold model of Bonabeau et al (6, 7). Individuals were assigned taskperformance probabilities as described by Bonabeau et al (7). However, whena worker performed a task the response thresholdθ was reduced by a “learning”factor61t, where1t is the time period over which the worker performed the task.When the task was not performed within1t, θ was increased by a “forgetting”factorσ1t. From an initial condition in which workers had equal thresholds, theyobtained worker groups that specialized on different tasks and adjusted activitylevels according to the task on which they specialized. The simulated colonies
Ann
u. R
ev. E
ntom
ol. 2
001.
46:4
13-4
40. D
ownl
oade
d fr
om a
rjou
rnal
s.an
nual
revi
ews.
org
by U
nive
rsity
of
Lei
den
- W
aleu
s L
ibra
ry o
n 01
/27/
10. F
or p
erso
nal u
se o
nly.
P1: UBI
November 6, 2000 13:30 Annual Reviews AR119-14
422 BESHERS¥ FEWELL
also showed flexibility; when specialists for one task were removed, other workersbegan to perform that task.
Self-Reinforcement and Social DominanceVariation in task performance isoften correlated with dominance, particularly in social insect colonies in whichreproductive division of labor is not absolute (47, 53, 69, 85, 86, 110, 111). So-cial dominance can lead to division of labor by a mechanism analogous to self-reinforcement. Hogeweg & Hesper (44, 45) and Theraulaz et al (109) found thatsimulated colonies could generate self-organized hierarchical structures that werecorrelated with differences in spatial fidelity and task performance. They both be-gan with populations of identical individuals (with the exception of the bumblebeequeen) and random encounters among workers. The dominance status of workerschanged on the basis of encounters by positive feedback; the dominant individualin an encounter became more dominant and the subordinate became less dominant.The models generated stable hierarchies, in which females that dominated in earlyencounters became and remained the most dominant females.
These patterns of dominance parallel the division of labor generated by self-reinforcement, and they lead to division of labor if dominance is linked to taskperformance. The link can be direct, for example by affecting task responsiveness,or it can be made indirectly through restrictions on access to tasks that are local-ized in the nest (44, 85, 110, 111). However, in a recent paper, Bonabeau et al (9)cautioned that it may be more realistic to assume intrinsic variation in dominanceamong reproductive females and that self-reinforcement may enhance dominancehierarchies, but it need not be invoked to explain their existence.
Foraging for Work
The FFW model shows how a flexible division of labor with temporal polyethismcould emerge from a simple algorithm for individual task performance that is notcausally related to worker age or size (112, 113). FFW has two main components,a behavioral algorithm for task performance and a spatial arrangement of tasks.The algorithm is (a) perform any task for which there is a need, (b) once a task isperformed, continue to perform the same task, (c) if this task no longer needs tobe done, move to another area of the nest and attempt to perform tasks there.
Tasks are arranged spatially in a series of zones (a given task can be performed inonly one zone). In the model the zones are arranged linearly; this is a simplificationof the radial or concentric arrangement of many ant nests, in which the queen,brood, and young workers are found in the center and older workers at the periphery,and the oldest workers may leave the nest to forage.
Tasks are connected functionally in a production line. The input for one taskis the output from another, so that the opportunity to perform a task depends onthe activity of other workers upstream in the sequence. The zone in which an antseeks work depends on how many times she fails to find the task for her currentzone (z), the relative number of ants performing the upstream task (z− 1), and theavailability of the downstream task (z+ 1) (112).
Ann
u. R
ev. E
ntom
ol. 2
001.
46:4
13-4
40. D
ownl
oade
d fr
om a
rjou
rnal
s.an
nual
revi
ews.
org
by U
nive
rsity
of
Lei
den
- W
aleu
s L
ibra
ry o
n 01
/27/
10. F
or p
erso
nal u
se o
nly.
P1: UBI
November 6, 2000 13:30 Annual Reviews AR119-14
MODELS OF DIVISION OF LABOR 423
FFW assumes no intrinsic variation in task preference among workers (112).Heterogeneity in task performance is generated by variation in the spatial encounterprobabilities for tasks [external factors (Figure 1)]. Individual workers in FFW canappear to be specialists that repeat a single task or generalists that move from onezone to another seeking work (112, 113), but as worker repertoires are externallydetermined, any “specialization” or “generalization” is unlikely to persist. Becausechanges in task performance are driven entirely by the opportunity to perform atask, FFW generates flexible task allocation that varies directly in response tochanges in task needs (36, 112).
FFW can theoretically also generate temporal polyethism. New workers emergeon the brood pile and initially seek work in that area. The influx of young workerstends to force older workers to move centrifugally to find tasks, and older workerssuffer mortality, generating a work sink in the areas in which they are located(112). The result is a correlation between distance from the nest center (and thecorresponding task zone) and worker age.
FFW has been controversial (36, 37, 91, 94, 115). Critiques have asserted thatits assumption of no intrinsic effects on task performance is violated by the clearphysiological correlates of temporal polyethism that have been identified (91). Thelevel of temporal polyethism that FFW can actually generate has also been disputed(94). Additionally, the argument has been made that a single behavioral algorithmis unlikely to explain division of labor that has evolved in many different ecologicalcontexts (115). To some extent these criticisms reflect different priorities of theauthors and differing views on the importance of modeling. However, despitethese often legitimate criticisms, foraging for work remains an important model.It sets up a baseline expectation of what level of task organization might occurwithin a social group in the absence of selection effects on intrinsic mechanismsfor worker task performance.
Foraging for Work and Threshold Models The FFW and threshold modelsdiffer in their assumptions and explanatory goals; despite this, they come to similarand important conclusions. Patterns of division of labor can be generated from acombination of simple rule sets and random variation (in either task encounter orperformance threshold). The resulting interactions between individual workers andtheir environment produce global patterns of division of labor. Both models alsoaccomplish the seemingly paradoxical task of generating colony task structuresthat show strong task specialization, but that can also respond quickly and flexiblyto changes in task need.
How then, do the emergent properties of the models vary? Temporal polyethismcan in principle be explained by changes in response thresholds as workers age,but response threshold models do not by themselves explain temporal polyethism.FFW suggests that a component of spatial organization of tasks may be sufficientto generate temporal polyethism even with fixed thresholds and may in any casebe an important component of real systems.
On the other hand, threshold models explain a ubiquitous and often puzzlingcomponent of task structure, the presence of a large number of unproductive
Ann
u. R
ev. E
ntom
ol. 2
001.
46:4
13-4
40. D
ownl
oade
d fr
om a
rjou
rnal
s.an
nual
revi
ews.
org
by U
nive
rsity
of
Lei
den
- W
aleu
s L
ibra
ry o
n 01
/27/
10. F
or p
erso
nal u
se o
nly.
P1: UBI
November 6, 2000 13:30 Annual Reviews AR119-14
424 BESHERS¥ FEWELL
(“lazy”) workers, generated as a byproduct of variation in thresholds (29, 76).According to FFW all workers should either be performing or seeking tasks. Tofts(112) noted that this parameter can be modified to incorporate a subset of lazyworkers which perform only one task (idle). However, this violates the assump-tion of initial homogeneity in task response and must be added as a special case. Asnoted by Tofts (112) the process of task decision making in FFW can be amendedto incorporate learning, strengthening the degree of specialization. It would addi-tionally be possible to merge a spatial FFW model with variation in task thresholds(7). However, although Bonabeau et al (7) argued that FFW is a special case ofthe threshold model, FFW makes specific assumptions about locational effectson task opportunity that have not been incorporated into a threshold model. Atruly integrated model incorporating our current understanding of variation in taskthresholds and spatial effects on task performance has yet to be constructed.
Social Inhibition Models
Social inhibition models explain temporal polyethism as an interaction between anintrinsic process of behavioral development and inhibitory effects of other workers(2, 50). Thus, these models are concerned with the behavioral state of a worker(the internal physiological state as it relates to task performance) and with theexternal factor of worker-worker interactions (Figure 1).
The first social inhibition model, the “activator-inhibitor” model (48, 49), pos-tulated that honey bees have an intrinsic activator that promotes behavioral de-velopment from in-hive tasks to foraging. The activator was initially associatedwith juvenile hormone (JH), for which blood titers correlate strongly with be-havioral ontogeny from in-hive tasks to foraging (50–52). Under the influence ofthe activator alone, all workers would quickly mature into foragers. An inhibitortransferred from foragers to younger workers opposes the effects of the activatorand suppresses their behavioral development. This leads to a stable balance of hiveworkers and foragers. If foragers are lost there will be less inhibition and morehive workers will mature into foragers. If there are too many foragers, inhibitionwill be stronger and delay the maturation of hive workers, even causing foragers torevert to the hive-worker state. The model shows how colony demography can beused to predict the age at which a worker first forages, a key measure of temporalpolyethism, whether this age is within the normal range or abnormally young orold, or whether a forager reverts to within-hive tasks (49).
“Activator-inhibitor” is no longer an appropriate name for this model, becauseit has been shown that JH does not play the activator role assigned to it in theoriginal model (107) (see below, evidence for the social inhibition model). “Socialinhibition” is a more appropriate name, because worker-worker inhibitory effectsare well documented (48–50).
Beshers et al (2) developed a mathematical version of the social inhibitionmodel in which a worker’s physiological state is represented by a single variable,x, that changes from day to day in response to the social environment (the average
Ann
u. R
ev. E
ntom
ol. 2
001.
46:4
13-4
40. D
ownl
oade
d fr
om a
rjou
rnal
s.an
nual
revi
ews.
org
by U
nive
rsity
of
Lei
den
- W
aleu
s L
ibra
ry o
n 01
/27/
10. F
or p
erso
nal u
se o
nly.
P1: UBI
November 6, 2000 13:30 Annual Reviews AR119-14
MODELS OF DIVISION OF LABOR 425
value ofx in the colony). The model consists of rules for mapping a worker’sxat time (t) to x at (t + 1). Essentially, the model describes a set of trajectories forbehavioral development that are intrinsically determined and that are responsiveto the social environment. Simulations show that, for any number of hive workersand foragers in the initial condition, the colony goes to a stable balance of hiveworkers and foragers and that any set of workers, even if they are initially identical,differentiates into hive workers and foragers. This model confirms that the socialinhibition mechanism can explain the generation and maintenance of a division oflabor in which task performance is correlated with age.
A similar model of temporal polyethism in the primitively eusocial waspRopa-lidia marginata has been presented by Naug & Gadagkar (67). The model in-corporates an activator-inhibitor mechanism in which each worker has two poolsof a chemical inhibitor, one that affects its own behavioral state and one that ittransfers to other workers that it encounters. The effect of inhibition on a worker’sbehavioral state is determined by the ratio of its own activator to the quantity ofinhibitor it receives from other workers. Like the Beshers et al (2) model, it gener-ates patterns of individual temporal polyethism coupled with a dynamically stableallocation of workers to different tasks.
The social inhibition models are compatible with a basic colony structure gen-erated by variation in individual task thresholds. However social inhibition modelsdiffer from other division–of-labor models because they specifically address the is-sue of how worker behavioral states are regulated. As currently presented, responsethresholds cannot explain temporal polyethism without additionally consideringthe mechanisms for changing worker thresholds as they age. Social inhibitionprovides such a mechanism. In addition, the organization of the worker force thatresults from social inhibition is not immediately affected by task needs and thismay allow a colony to flexibly respond to changing conditions while maintainingthe integrity of its organization (50, 78).
Network Models of Task Allocation
Gordon et al (42) and Pacala et al (74) used models to explore how task allocationand the dynamics of colony behavior can be explained by simple interactionsamong workers [external effects (Figure 1)]. These models assume no intrinsicdifferences among workers; instead, changes in task allocation result from simpleinteractions among workers that effectively communicate information about thenumber of workers that are active or inactive for a given task.
In the network model presented by Gordon et al (42) workers can perform oneof four tasks, and for each task they can be active or inactive, a total of eight states.Worker interactions are biased toward information transfer with workers in thesame task group and are dependent on whether the worker is currently active orinactive. For example (if one task in the model is identified with midden work), anactive midden worker interacts with all other active ants and with inactive middenworkers, whereas an inactive midden worker interacts with only active and inactive
Ann
u. R
ev. E
ntom
ol. 2
001.
46:4
13-4
40. D
ownl
oade
d fr
om a
rjou
rnal
s.an
nual
revi
ews.
org
by U
nive
rsity
of
Lei
den
- W
aleu
s L
ibra
ry o
n 01
/27/
10. F
or p
erso
nal u
se o
nly.
P1: UBI
November 6, 2000 13:30 Annual Reviews AR119-14
426 BESHERS¥ FEWELL
midden workers. Worker behavior in each iteration is determined by comparing alinear combination of these interactions with a threshold value. Interactions amongworkers move the system toward a stable set-point in which there is a balance ofactive and inactive workers for each task and among the workers allocated to eachof the four tasks.
Pacala et al (74), building on the results with the network model, developed aset of differential-equation models to investigate the effects of social group size ontask allocation. In their models, individuals receive information about task needs(a) directly from the environment, (b) through their success or failure in findingtask opportunities, and (c) through random encounters with other group members.Again, colonies can allocate workers in relation to the need for each task and canadjust the allocation in response to environmental changes. Larger groups trackenvironmental change more efficiently because of a higher rate of informationcollection and frequency of interactions, suggesting that information flow mayfavor the evolution of increased colony size. However, as colony size increases,the rate of interactions grows faster than the rate of information collection fromthe environment. Colonies could keep these two components of information flowin balance by regulating the frequency of interactions among workers (74).
Although the goals of the network and FFW models differ, they share the viewthat division of labor can be generated or maintained via changes in the localinformation encountered by an individual worker. This local information is in turnaffected by availability and performance of other tasks, causing workers to moveinto or out of particular “task spaces.” The network model presents perhaps a moreaccurate view of how task spaces are arranged, as connecting networks rather thanas a linear arrangement. However, it lacks the self-organizational properties of theFFW model. Recent work on social resilience, the ability of colonies to reestablishtheir spatial organization after disturbance or nest moving (102, 103), may providean approach that will allow integration of these perspectives.
MODELS AND EMPIRICAL EVIDENCE
In this section we discuss empirical evidence bearing on the models and evaluatethe contributions and the limitations of each model type. When possible, we addressthe evidence relating to specific hypotheses incorporated into each model.
Response Threshold Model
The response threshold model relies on the fundamental assumption that work-ers vary intrinsically in task preference. Genotypic variation is associated withtask performance in a diversity of social insect taxa (68, 70, 72, 78, 104, 106).These experiments have primarily identified family or subfamily (patriline) dif-ferences in task performance, leading to the criticism that they do not establish acausal relationship between genotype and task preference (10). However, selection
Ann
u. R
ev. E
ntom
ol. 2
001.
46:4
13-4
40. D
ownl
oade
d fr
om a
rjou
rnal
s.an
nual
revi
ews.
org
by U
nive
rsity
of
Lei
den
- W
aleu
s L
ibra
ry o
n 01
/27/
10. F
or p
erso
nal u
se o
nly.
P1: UBI
November 6, 2000 13:30 Annual Reviews AR119-14
MODELS OF DIVISION OF LABOR 427
experiments on pollen and nectar collection in honey bees have demonstrated astrong causal link between genotype and worker foraging-task choice (32, 43, 75).
The response threshold model makes the prediction that under conditions oflow stimulus for a given task a relatively narrow genotypic subset of the colonyshould perform the task (those with the lowest thresholds), but as stimuli increase,a wider genotypic diversity of individuals should perform the task (31, 93). Consis-tent with this, Fewell & Page (31) found that under conditions of high pollen stores(low foraging stimulus) one focal genotypic group dominated the pollen-foragingpopulation. When stores were removed workers from the other focal groups sig-nificantly increased in number. Similar results have been shown in subsequentexperiments (29, 32).
Evidence supporting the prediction that the diversity of the group performing atask should increase with increasing stimuli also comes from caste studies in ants.The organization of prey retrieval in the antPheidole pallidulaappears to be basedin part on differences in response thresholds to recruitment displays between minorand major workers (25). Only minor workers are recruited to patches of small preythat can be carried individually, but more vigorous displays are used to recruit majorworkers to large prey that must be dissected before being retrieved. This suggeststhat major workers have higher thresholds for responding to recruitment than theminors. The same difference between major and minor workers was found in thecontext of responding to recruitment for colony defense (26).
Results from a study of nest emigrations of the antTapinoma erraticumsuggestthat individual variation in response thresholds could explain the division of laborobserved during emigrations (63). For a set of experimental colonies, repeatedtrials showed that a small group of workers in each colony consistently partici-pated in brood transport during emigration, suggesting intrinsic differences amongthe workers. However, when these workers were removed, other workers in thesame colony replaced them as transporters. When the experimental colonies weredivided and reassembled into new colonies containing either all transporters or notransporters, the same proportion of workers transported brood during emigrationsas in the original colonies. This is consistent with the interpretation that all work-ers are potential transporters, and in each colony the “specialists” are the workerswith the lowest thresholds for transporting.
The same patterns of task response have not been reported universally for alltasks. Robinson & Page (90) examined the behavioral responses of workers inhoney bee hives when the subset of workers that undertake (remove dead bees)was eliminated. As predicted, undertakers represented a genetically differentiatedsubgroup in the colony. However, contrary to expectations, other workers did notmove into the undertaker task group, and the dead bees were left in the hive.Why did this occur? For a rare task such as undertaking, most workers may havevery high thresholds and never perform the task. Or removal of the undertakersmay not have caused stimulus levels to increase, as the number of dead bees(the stimulus) did not change. A further interesting possibility is that there wereworkers that could have removed the dead bees but did not because they were
Ann
u. R
ev. E
ntom
ol. 2
001.
46:4
13-4
40. D
ownl
oade
d fr
om a
rjou
rnal
s.an
nual
revi
ews.
org
by U
nive
rsity
of
Lei
den
- W
aleu
s L
ibra
ry o
n 01
/27/
10. F
or p
erso
nal u
se o
nly.
P1: UBI
November 6, 2000 13:30 Annual Reviews AR119-14
428 BESHERS¥ FEWELL
more responsive to stimuli for different tasks and that within each worker responsethresholds are arranged in a hierarchy that reflects colony-level task priorities (90).
If task specialization is an emergent property based on variance in task thresh-olds, as suggested by the response threshold model, then we should expect thatit will appear in any social group with variation in individual task preference.Sakagami & Maeta (95) created artificial social groups in the normally solitarycarpenter beeCeratina flavipes. They were able to establish tolerant pairs of-queens in only five nests, but in each of those nests, the queens showed reproduc-tive division of labor, in which one queen laid eggs and guarded the nest and theother queen foraged.
Fewell & Page (30) created artificial social groups of normally solitary antqueens, which they placed in dirt chambers and allowed to excavate nests. In thesepairs, one queen consistently became the excavator specialist. They could predictwhich one would become the task specialist based on the queen’s behavior whilesolitary or in previous pairs, suggesting that task specialization emerges fromintrinsic differences in behavior. The emergence of task performance in thesegroups involved negative feedback; excavating queens did not vary rates of taskperformance between solitary and social contexts, but the queens that became“non-excavators” significantly decreased excavation rates when paired with anexcavating partner. There was no evidence of self-reinforcement in this system; thebehavior of the task specialist remained constant from a solitary to social context.
Actual measurements of response thresholds are few. There are no measure-ments of individual worker thresholds for responding to stimuli known to elicit theperformance of a task. In theory individual response thresholds could be estimatedfrom individual frequencies of performing a task or from the probability of re-sponding to a stimulus by performing the task. Page et al (81) measured responsethresholds to sucrose solutions of various concentrations in honey bees; individualthresholds were shown to be closely correlated with the forager’s preference fornectar or pollen with pollen foragers having lower thresholds than nectar foragers.Response thresholds in workers that were not yet foraging were shown to correlatewith later resource preferences and selectivity when these same workers began toforage (83). The direct role of the sucrose response threshold in the decision of aworker to forage is not known.
Temporal polyethism can be explained by response thresholds that change overthe lifetime of the worker. Several studies show that responsiveness of honey beeworkers to pheromones can change with age (1, 21, 87). These experiments alltested the responsiveness of groups of workers; in fact, single worker bees do notrespond at all to alarm pheromone, so their responsiveness can only be studied ina social context (87).
Threshold-Information Transfer Model
Fewell & Bertram (29) applied their model to pollen foraging in honey bees. Theyvaried the stimulus for this task by adding or subtracting from the colony’s pollen
Ann
u. R
ev. E
ntom
ol. 2
001.
46:4
13-4
40. D
ownl
oade
d fr
om a
rjou
rnal
s.an
nual
revi
ews.
org
by U
nive
rsity
of
Lei
den
- W
aleu
s L
ibra
ry o
n 01
/27/
10. F
or p
erso
nal u
se o
nly.
P1: UBI
November 6, 2000 13:30 Annual Reviews AR119-14
MODELS OF DIVISION OF LABOR 429
stores (29). Colonies responded to changes in pollen stores in a stepwise manner,rapidly increasing or decreasing pollen foraging as stores moved past a specificset point. These shifts in foraging activity were accompanied by changes in thegenotypic diversity of the pollen-foraging population, as predicted by a thresholdmodel.
These results match the predictions of the model for the combination of a bi-modal threshold distribution and simultaneous information encounter. Resourcepreference in honey bees shows phenotypic dominance for nectar (versus pollen)foraging, resulting in discrete groups of nectar and pollen foragers (32). The mech-anisms for regulating pollen foraging are not yet completely understood, but socialcommunication could be involved (17, but see 27). This suggests that the assump-tion of simultaneous information encounter also holds and that information aboutpollen stores may be propagated very quickly through the colony.
Similar stepwise changes in foraging have been seen in response to gradedchanges in nectar quality (99, 100), suggesting that a similar model may apply tothis task. Forager genotypes were not determined in these studies.
Self-Reinforcement Models
Bonabeau et al (7) and Theraulaz et al (109) emphasize that individual thresholdsdo not have to be rigidly fixed, but instead they may be altered by experience.The effect of task performance on individual thresholds has not to our knowledgebeen directly tested. However there are some suggestive results from responses ofhoney bee foragers to changing stimulus levels. Fewell & Page (32) varied qualityand the need for pollen in honey bee colonies and tracked the resource preferencesof individually marked foragers. As expected, new foragers were recruited intothe nectar- and pollen-foraging populations as need and availability increased, butthose workers already engaged in pollen or nectar foraging did not switch betweentasks after they began collecting one resource and the stimuli for that resourcewere reduced. Instead, they stopped foraging. This fidelity cannot be completelyexplained by genetic variation. If it could, we would expect that workers recruitedinto one task when stimuli for that task were extremely high should move fromthat task group into the other when stimuli were reversed. A more parsimoniousexplanation may be that collection of one resource is self-reinforcing and that aperiod of decay (in which stimuli for that task remain low and performance of thetask is “forgotten”) is required for task switching.
Social Dominance
Social dominance rank has been found to be correlated with task performancein bumble bees (47), in the ponerine antsPachycondyla(53) andOdontomachus(85), and in the wasp generaPolistes(86, 111) andMischocyttarus(69). It is notknown whether dominance is the only cause of division of labor in these speciesor if other mechanisms are also involved.
Ann
u. R
ev. E
ntom
ol. 2
001.
46:4
13-4
40. D
ownl
oade
d fr
om a
rjou
rnal
s.an
nual
revi
ews.
org
by U
nive
rsity
of
Lei
den
- W
aleu
s L
ibra
ry o
n 01
/27/
10. F
or p
erso
nal u
se o
nly.
P1: UBI
November 6, 2000 13:30 Annual Reviews AR119-14
430 BESHERS¥ FEWELL
Foraging for Work
The FFW algorithm has two main parts: workers repeat the same task when pos-sible, and they actively seek work when they are left without a task to perform.There is no direct experimental evidence that social insect workers behave accord-ing to the FFW algorithm, in part because the system set up by the algorithm isnot meant to completely duplicate the complexity of an eusocial insect colony.
An implicit assumption of FFW is that workers are intrinsically identical andthat worker task performance depends on opportunity rather than on intrinsictask preferences (36). However, in those species in which behavioral variationhas been examined there is evidence for intrinsic variation in response to thecolony environment (discussed under response threshold and social inhibitionmodels). Not all workers encountering a task opportunity respond equally. Julian &Cahan (58) placed ant corpses into colonies of individually marked leaf-cutter ants(Acromyrmex versicolor) and found that, although>80% of the workers encoun-tered the corpses, only a small percentage of the workers responded by carryingthem out of the nest. Workers that removed corpses more frequently also com-pleted the task faster, suggesting a link between specialization and efficiency.A similar effect was observed in honey bees: Workers experienced at removingcorpses did so faster than workers removing corpses for the first time, though afterremoval of the first corpse, further experience had little effect on performance(116).
The second main component of FFW is the assumption that tasks are spatiallylocalized within the nest and that there is often a radial-symmetry componentto nest organization. As a consequence of FFW, young workers are expected tobe at the center of the nest near the brood and older workers more towards thenest periphery. Consistent with this, harvester ant nests (genusPogonomyrmex)show spatial stratification in which younger workers are found near the bottom ofnests and older workers are found higher and nearer the entrance (62). Tasks arealso spatially arranged inLeptothorax unifasciatusnests (101). However youngerworkers move extensively through the nest area but avoid the entrance, a detailnot specifically expected from a simple FFW algorithm. Older workers showsocial resilience, in which they returned to specific spatial locations within thenest after displacement (102). Although these data suggest that spatial dynamicsplay a potentially important role in task structure, the basic FFW algorithm doesnot incorporate this level of spatial fidelity. However, the model could certainlybe modified to incorporate social resilience (102, 103) and doing so would likelyincrease the levels of task specialization generated.
Our understanding of the spatial geometry of task information flow within somesocial insect societies argues against a simple task encounter-based model of tem-poral polyethism. Leaf-cutter workers show a pattern of in-nest to outside-nest taskperformance. However foragers deliver plant material to the fungus-brood area,located in the center of the nest (57). Thus, contact between foragers and otherworkers is not spatially limited to the nest entrance. Honey bees exchange infor-mation about nectar and pollen availability from the hive entrance, but they gain
Ann
u. R
ev. E
ntom
ol. 2
001.
46:4
13-4
40. D
ownl
oade
d fr
om a
rjou
rnal
s.an
nual
revi
ews.
org
by U
nive
rsity
of
Lei
den
- W
aleu
s L
ibra
ry o
n 01
/27/
10. F
or p
erso
nal u
se o
nly.
P1: UBI
November 6, 2000 13:30 Annual Reviews AR119-14
MODELS OF DIVISION OF LABOR 431
information about colony need from diverse areas in the hive through contact withbrood, pollen, nurse bees (who process pollen), and nectar receivers (16, 17, 22).Camazine et al (17) found that pollen foragers actually change locations within thehive under conditions of increased pollen need, moving away from the entranceand into the upper chambers. By doing so, they increase contact with nurse beesthat digest and distribute pollen.
In contrast, the structure of information flow for some wasp task sets may fit thelinear dynamics of the FFW model well. Nest construction inPolybia occidentaliswasps involves three tasks, wood-pulp foraging, nest building, and water foraging.Activity levels in each of these three tasks are regulated via interactions withindividuals in the other task group, and information flow is in part linear. Forexample, only builders receive information about nest damage (56).
FFW shows that temporal polyethism need not require age-related differencesin the mechanisms of task choice, and by pushing the limits of a model with nointrinsic variation among workers it has expanded the range of possible explana-tions for division of labor. In addition, although much evidence shows that thereare indeed intrinsic differences among workers, it is not known exactly how in-ternal and external factors interact to yield division of labor, nor is it known howthe physiological correlates of division of labor are expressed behaviorally. Forexample, temporal polyethism could be explained, at least in part, by age-relatedpreferences for location within the nest, which then determine the task stimuli towhich a worker is exposed. Such a hypothesis would combine features of bothFFW and social inhibition models.
Social Inhibition Models
Several lines of evidence have demonstrated how the presence of older foragingbees affects behavioral development, a core argument of the social inhibition model(48–50). The proportion of bees that become foragers is indirectly related to thenumber of older bees present in the colony (49), and removal of foragers decreasesthe age at which younger bees become foragers (61). These data are consistentwith both the social inhibition and FFW models. However when a group of olderforagers is moved into a colony of younger workers, the transition to foraging bythe younger resident bees is delayed relative to that in similar colonies with noolder workers. This occurs even when the transplanted foragers are not allowed toforage (48). Additionally, bees reared initially in social isolation show high ratesof JH biosynthesis and forage precociously when introduced into a colony (48).These data are consistent with a social inhibition model, but they do not fit FFWor network models, in which changes in task distribution (rather than colony agestructure) affect polyethism.
A number of studies have demonstrated a correlation between JH titers andtemporal changes from in-hive tasks to foraging in this system (28, 50, 55, 87, 88).Precocious foragers have high JH titers, whereas foragers that have reverted tonursing have reduced titers (89). Treatment of workers with JH, JH analogs, or JH
Ann
u. R
ev. E
ntom
ol. 2
001.
46:4
13-4
40. D
ownl
oade
d fr
om a
rjou
rnal
s.an
nual
revi
ews.
org
by U
nive
rsity
of
Lei
den
- W
aleu
s L
ibra
ry o
n 01
/27/
10. F
or p
erso
nal u
se o
nly.
P1: UBI
November 6, 2000 13:30 Annual Reviews AR119-14
432 BESHERS¥ FEWELL
mimics accelerates the transition to foraging (28, 50, 54, 55). Methoprene injection(a JH mimic) also accelerates temporal polyethism in wasps (71), although a similareffect has not been found for bumble bees (20).
JH is not an activator, however. It is not required for behavioral maturation;when the corpora allata (the glandular source of JH) of young bees are removedshortly after eclosion, the bees still become foragers, although they do so later thancontrol workers (107). Also, because workers with their corpora allata removedforage sooner when reared in single-cohort colonies than in normal colonies, itis clear that JH does not mediate a worker’s response to change in the socialenvironment (107).
The social inhibition models argue that temporal polyethism is causally me-diated by physiological and specifically hormonal factors. Is there evidence thatage can causally affect task performance? Calderone (13) controlled for environ-mental effects on task ontogeny by incubating honey bee workers separately andintroducing them to a host hive simultaneously with workers of the same geno-type but of younger age. He found significant differences between age cohorts inperformance of multiple tasks, suggesting that task choice is at least in part devel-opmentally driven. In similar studies, Calderone & Page (14) and Page et al (79)found that older workers show higher tendencies to forage, again independentlyof colony environment.
Although the social inhibition model provides a good fit with our empiricalunderstanding of honey bee age polyethism (50, 51), some crucial questions re-main. There is good correlational evidence for physiological effects on temporalpolyethism, but the mechanism for inhibition is less clear. One inhibitory factorhas been identified: Queen mandibular pheromone (QMP) inhibits worker behav-ioral development (82). But it is clear that QMP is not the primary inhibitor in thesocial inhibition mechanism, because worker-worker inhibition does not depend onthe presence of the queen (48).
Network Models of Task Allocation
Like FFW, network models are based around information transfer across and withintask groups, and so the evidence for spatial arrangement of tasks presented inFFW (see above) is relevant also to networks. The network model also predictsthat perturbation of the number of workers in one task group should affect thenumbers of workers in other tasks as well, so that the system “emphasizes” one ormore of the “unperturbed” tasks, as well as the perturbed task before re-stabilizing.Gordon (38–40) found such an interaction between task groups in colonies of theharvester antPogonomyrmex barbatus. When she manipulated the nest surface toincrease the need for nest maintenance, colonies also reduced the number of activeforagers, even though they represented a separate task group (38).
This connection between tasks is not always present. Although pollen and nectarforaging co-vary genetically in honey bees (32, 75), colonies regulate pollen andnectar intake independently. Pollen foraging is regulated homeostatically around
Ann
u. R
ev. E
ntom
ol. 2
001.
46:4
13-4
40. D
ownl
oade
d fr
om a
rjou
rnal
s.an
nual
revi
ews.
org
by U
nive
rsity
of
Lei
den
- W
aleu
s L
ibra
ry o
n 01
/27/
10. F
or p
erso
nal u
se o
nly.
P1: UBI
November 6, 2000 13:30 Annual Reviews AR119-14
MODELS OF DIVISION OF LABOR 433
levels of pollen and brood in the colony (33, 34). However, nectar foraging isrelatively insensitive to hive conditions and instead seems to be positively reg-ulated around nectar availability (99). Furthermore, changes in allocation to onetask do not seem to affect the other. Variation in colony pollen stores produceschanges both in the proportion of foragers collecting pollen and in individualpollen-foraging effort, but it has no effect on allocation to nectar foraging orindividual nectar-foraging rates (33).
CONCLUSIONS
Modeling of division of labor is still in an early stage. These mechanistic and oftenself-organizational models of division of labor have begun to show us the colony-level patterns of division of labor that can result from simple individual behavioralrules. In doing so, they provide us with an understanding of the underlying organi-zational framework on which selection can act. As shown in Figure 1, the modelsthat researchers have developed independently often target different components ofthe processes generating division of labor. Thus, one of our primary goals shouldbe to broaden the integrative scope of these models. As discussed above, thereis considerable opportunity for integration among the approaches currently used.Combining assumptions and comparing between modeling techniques will likelymove our understanding of behavioral complexity forward in a synergistic manner.
Have these models been useful in terms of improving our understanding ofsocial complexity? There is no doubt that this is the case. Models should becomeindispensable in this field, because they allow us to relate causes and patterns fromthe individual to colony level that cannot be generated by intuitive approachesalone. One area that needs to be further developed is relating the behavioral pro-grams of workers not only to patterns of behavior but to colony efficiency andother performance measures. Progress in this area could lead to a revitalization ofergonomic studies that explore how efficiency results from the organization of acolony as well as from the task efficiency of individual workers.
To date, the models have been “exploratory”; they are designed to reveal the con-sequences of specific assumptions about individual task performance. Exploratorymodels can sometimes reveal important general principles. More often, in theprocess of modeling a question arises for which no relevant data exist. Thus, ex-ploratory models can suggest new hypotheses.
Ultimately, the promise of models is that they can be “explanatory”; they cangenerate predictions that can be tested with experimental data. Few models haveyet been tested against quantitative data, although quantitative expectations arepresented in some models (e.g. 6, 29). In part, this is due to the lack of suitable datafor comparisons, and another function of exploratory models will be to show whatkinds of data will be most useful.
Although quantitative explanatory models may be some years away, we suggestthat progress can be made by recognizing that models usually incorporate several
Ann
u. R
ev. E
ntom
ol. 2
001.
46:4
13-4
40. D
ownl
oade
d fr
om a
rjou
rnal
s.an
nual
revi
ews.
org
by U
nive
rsity
of
Lei
den
- W
aleu
s L
ibra
ry o
n 01
/27/
10. F
or p
erso
nal u
se o
nly.
P1: UBI
November 6, 2000 13:30 Annual Reviews AR119-14
434 BESHERS¥ FEWELL
hypotheses. For explanatory power, models should be as broad and integrative aspossible. Hypotheses, in contrast, need to be narrow and focused so that they canbe empirically tested. Development of rigorous and successful explanatory modelswill require that models be statistically compared to the data, that authors be asspecific as possible about their goals in developing and using a model and aboutthe power and limitations of their specific models, and that the assumptions of eachmodel be specifically stated and discussed. For example, several types of modelsexplicitly assume that all workers in a colony are identical without discussing theusefulness or ramifications of such an assumption.
Bonabeau et al (7) have shown how some of the models are related to oneanother and advocated more explicit statements about strengths and weaknessesby the authors of models.
Modeling the Mechanisms and Evolution of Division of Labor
Models of self-organization are especially valuable to studies of the evolution ofsocial complexity. Although selection undoubtedly shapes social organization, itacts on a social unit that already has intrinsic properties. Some of the fundamentalproperties of social organization, including division of labor, are likely present atthe origins of sociality (30, 80). They are not necessarily produced via selection,though they may be subsequently molded by selection. Mechanistic models of so-cial organization allow us to separate out the components of this process, by identi-fying general emergent properties, and assess how selection and self-organizationinteract to produce the complex behavioral properties of social insect colonies.
The mechanisms of division of labor are not yet completely identified, but theefforts of many authors, working independently and in parallel, are beginning toreveal the outlines of a network of causes that includes both internal and externalfactors. The internal factors shown in Figure 1 give us a preliminary picture ofthe “behavioral architecture” of a worker and how it leads to the decision toperform a task. Even such a preliminary picture can be a valuable guide to furtherresearch, and we expect the complexity of this diagram to grow rapidly with moresophistication of behavioral experiments and with the availability of improvedgenetic and neurobiological tools. For example, in honey bees, it is clear thatthe “behavioral state” of a worker involves at least two levels that affect taskperformance: the actual level at which task decisions are made, which may involveresponse thresholds, and the underlying level at which JH acts to modulate aworker’s physiological state and cause it to focus on either foraging or within-hivetasks. Recent data strongly suggest that there is an intervening level of regulationin which the responsiveness of a worker can be modulated by brain levels ofoctopamine (97).
Social insects are proving to be ideal systems for asking questions about thedesign and organization of biological systems. The organizing principles of socialgroups and organismal components are apparently analogous in many ways, but
Ann
u. R
ev. E
ntom
ol. 2
001.
46:4
13-4
40. D
ownl
oade
d fr
om a
rjou
rnal
s.an
nual
revi
ews.
org
by U
nive
rsity
of
Lei
den
- W
aleu
s L
ibra
ry o
n 01
/27/
10. F
or p
erso
nal u
se o
nly.
P1: UBI
November 6, 2000 13:30 Annual Reviews AR119-14
MODELS OF DIVISION OF LABOR 435
social insect colonies are easier to observe and manipulate than cellular or sub-cellular systems and offer greater opportunities for characterizing “individual”behavior. The increasing sophistication of models and techniques for studying thebehavior and physiology of individual workers promises to stimulate a new waveof research and to make the study of social insect behavior broadly relevant withinand beyond the traditional domains of biology.
Visit the Annual Reviews home page at www.AnnualReviews.org
LITERATURE CITED
1. Allan SA, Slessor KN, Winston ML, KingGGS. 1987. The influence of age and taskspecialization on the production and per-ception of honey bee pheromones.J. InsectPhys.33:917–22
2. Beshers S, Huang Z, Oono Y, Robinson G.2000. Social inhibition and the regulation oftemporal polyethism in honey bees.J. Theor.Biol. In press
3. Beshers SN, Robinson GE, Mittenthal JE.1999. Response thresholds and division oflabor in insect colonies. See Ref. 24a, pp.115–39
4. Beshers SN, Traniello JFA. 1996. Poly-ethism and the adaptiveness of worker sizevariation in the attine antTrachymyrmexseptentrionalis. J. Insect Behav.9:61–83
5. Bonabeau E, Theraulaz G. 1999. Role andvariability of response thresholds in the reg-ulation of division of labor in insect soci-eties. See Ref. 24a, pp. 141–63
6. Bonabeau E, Theraulaz G, Deneubourg JL.1996. Quantitative study of the fixed thresh-old model for the regulation of division oflabour in insect societies.Proc. R. Soc. Lon-don B263:1565–69
7. Bonabeau E, Theraulaz G, Deneubourg JL.1998. Fixed response thresholds and the reg-ulation of division of labor in insect soci-eties.Bull. Math. Biol.60:753–807
8. Bonabeau E, Theraulaz G, Deneubourg JL.1998. Group and mass recruitment in antcolonies: the influence of contact rates.J.Theor. Biol.195:157–66
9. Bonabeau E, Theraulaz G, Deneubourg JL.
1999. Dominance orders in animal soci-eties: the self-organization hypothesis re-visited.Bull. Math. Biol.61:727–57
10. Bourke AFG, Franks NR. 1995.SocialEvolution in Ants. Princeton, NJ: Prince-ton Univ. Press. 529 pp.
11. Calabi P. 1988. Behavioral flexibility inHymenoptera: a re-examination of theconcept of caste. InAdvances in Myrme-cology, ed. JC Trager, pp. 237–58. Leiden:Brill
12. Calabi P, Rosengaus R. 1988. Interindi-vidual differences based on behaviortransition probabilities in the antCam-ponotus sericeiventris. In InterindividualBehavioral Variability in Social Insects, ed.RL Jeanne, pp. 61–90. Boulder, CO: West-view
13. Calderone NW. 1995. Temporal divisionof labor in the honey bee,Apis mellif-era: a developmental process or the resultof environmental influences.Can. J. Zool.73:1410–16
25. Detrain C, Pasteels JM. 1991. Caste dif-ferences in behavioral thresholds as a ba-sis for polyethism during food recruit-ment in the ant,Pheidole pallidula(Nyl)(Hymenoptera, Myrmicinae).J. Insect.Behav.4:157–76
26. Detrain C, Pasteels JM. 1992. Castepolyethism and collective defense in theant, Pheidole pallidula: the outcome ofquantitative differences in recruitment.Behav. Ecol. Sociobiol.29:405–12
27. Dreller C, Tarpy DR. 2000. Perception ofthe pollen need by foragers in a honeybeecolony.Anim. Behav.59:91–96
28. Fahrbach S. 1997. Regulation of agepolyethism in bees and wasps by juvenilehormone.Adv. Study Behav.26:285–316
29. Fewell JH, Bertram SM. 1999. Divisionof labor in a dynamic environment: re-sponse by honeybees (Apis mellifera) tograded changes in colony pollen stores.Be-hav. Ecol. Sociobiol.46:171–79
30. Fewell JH, Page RE. 1999. The emergenceof division of labour in forced associationsof normally solitary ant queens.Evol. Ecol.Res.1:537–48
31. Fewell JH, Page REJ. 1993. Genotypicvariation in foraging responses to environ-mental stimuli by honey bees,Apis mellif-era. Experientia49:1106–12
32. Fewell JH, Page REJ. 2000. Colony levelselection effects on individual and colonyforaging task performance in honey bees,Apis melliferaL. Behav. Ecol. Sociobiol.Inpress
33. Fewell JH, Winston ML. 1992. Colonystate and regulation of pollen foraging inthe honey bee,Apis melliferaL. Behav.Ecol. Sociobiol.30:387–93
34. Fewell JH, Winston ML. 1996. Regulationof nectar collection in relation to honeystorage levels by honey bees,Apis mellif-era. Behav. Ecol.7:286–91
35. Forsyth AB. 1978.Studies on the behav-ioral ecology of polygynous social wasps.PhD thesis, Harvard Univ., Cambridge,MA
behavioural development in the honey beeis based on changes in the rate of juve-nile hormone biosynthesis.J. Insect Phys-iol. 37:733–41
53. Ito F, Higashi S. 1991. A linear dominan-ce hierarchy regulating reproduction andpolyethism of the queenless antPachycon-dyla sublaevis. Naturwissenschaften78:80–82
57. Julian GE. 1999.Genetic Variation andTask Organization in the Desert Leaf-Cutter Ant, Acromyrmex versicolor. PhDthesis, Ariz. State Univ., Tempe
58. Julian GE, Cahan S. 1999. Undertak-ing specialization in the desert leaf-cutterant Acromyrmex versicolor.Anim. Behav.58:437–42
59. Karsai I, Theraulaz G. 1995. Nest buildingin a social wasp: postures and constraints.Sociobiology26:83–114
60. Kolmes SA. 1986. Age polyethism inworker honey bees.Ethology71:252–55
61. Lindauer M. 1952. Ein beitrag z¨ur Frageder Arbeitsteilung im Bienenstaat.Z. Vergl.Physiol.34:299–345
62. Mackay WP. 1983. Stratification of work-ers in harvester ant nests (Hymenoptera,Formicidae).J. Kans. Entomol. Soc.56:538–42
63. Meudec M. 1979. Le comportementd’emigration chez la fourmiTapinoma er-raticum (Dolichoderinae): un exemple deregulation sociale.Bull. Biol. Fr. Belg.13:321–74
Ann
u. R
ev. E
ntom
ol. 2
001.
46:4
13-4
40. D
ownl
oade
d fr
om a
rjou
rnal
s.an
nual
revi
ews.
org
by U
nive
rsity
of
Lei
den
- W
aleu
s L
ibra
ry o
n 01
/27/
10. F
or p
erso
nal u
se o
nly.
P1: UBI
November 6, 2000 13:30 Annual Reviews AR119-14
438 BESHERS¥ FEWELL
64. Michener CD. 1974.The Social Behaviorof the Bees. Cambridge, MA: HarvardUniv. Press
65. Millor J, Pham-Delegue M, DeneubourgJL, Camazines S. 1999. Self-organized de-fensive behavior in honeybees.Proc. Natl.Acad. Sci. USA96:12611–15
66. Moglich M, Holldobler B. 1974. Socialcarrying behavior and division of labor dur-ing nest moving in ants.Psyche81(2):219–36
67. Naug D, Gadagkar R. 1999. Flexible di-vision of labor mediated by social inter-actions in an insect colony: a simulationmodel.J. Theor. Biol.97:123–33
68. O’Donnell S. 1996. RAPD markers sug-gest genotypic effects on forager special-ization in a eusocial wasp.Behav. Ecol. So-ciobiol. 38:83–88
69. O’Donnell S. 1998. Dominance and poly-ethism in the eusocial waspMischocyttarusmastigophorus(Hymenoptera: Vespidae).Behav. Ecol. Sociobiol.43:327–31
70. O’Donnell S. 1998. Genetic effects on taskperformance, but not on age polyethism,in a swarm-founding eusocial wasp.Anim.Behav.55:417–26
71. O’Donnell S, Jeanne RL. 1993. Metho-prene accelerates age polyethism in work-ers of a social wasp (Polybia occidentalis).Physiol. Entomol.18:189–94
72. Oldroyd BP, Sylvester HA, Wongsiri S,Rinderer TE. 1994. Task specialization ina wild bee,Apis florea(Hymenoptera, Ap-idae), revealed by RFLP banding.Behav.Ecol. Sociobiol.34:25–30
73. Oster GF, Wilson EO. 1978.Caste andEcology in the Social Insects. Princeton,NJ: Princeton Univ. Press
74. Pacala SW, Gordon D, Godfray HCJ. 1996.Effects of social group size on informa-tion transfer and task allocation.Evol. Ecol.10:127–65
75. Page RE, Fondrk MK. 1995. The effectsof colony-level selection on the social or-ganization of honey bee (Apis melliferaL.) colonies: colony-level components of
pollen hoarding.Behav. Ecol. Sociobiol.36:135–44
76. Page RE, Mitchell SD. 1991. Self organi-zation and adaptation in insect societies. InPSA 1990Vol. 2, ed. A Fine, M Forbes,L Wessels, pp. 289–98. East Lansing, MI:Philos. Sci. Assoc.
77. Page RE, Mitchell SD. 1998. Self-organization and the evolution of divisionof labor.Apidologie29:171–90
78. Page RE, Robinson GE. 1991. The gene-tics of division of labor in honey beecolonies.Adv. Insect Physiol.23:117–69
79. Page RE, Robinson GE, Britton DS, Fon-drk MK. 1992. Genotypic variability forrates of behavioral-development in workerhoneybees (Apis melliferaL.) Behav. Ecol.3:173–80
80. Page REJ. 1997. The evolution of insectsocieties.Endeavor21:114–20
81. Page REJ, Erber J, Fondrk MK. 1998. Theeffect of genotype on response thresholdsto sucrose and foraging behavior of honeybees (Apis melliferaL.) J. Comp. Physiol.A 182:489–500
83. Pankiw T, Page RE. 2000. Responsethresholds to sucrose predict foraging di-vision of labor in honeybees.Behav. Ecol.Sociobiol.47:265–67
84. Plowright RC, Plowright CMS. 1988.Elitism in social insects: a positive feed-back model. InInterindividual BehavioralVariability in Social Insects, ed. RL Jeanne,pp. 419–31. Boulder, CO: Westview
85. Powell S, Tschinkel WR. 1999. Ritualizedconflict in Odontomachus brunneusandthe generation of interaction-based task al-location: a new organizational mechanismin ants.Anim. Behav.58:965–72
86. Reeve HK, Gamboa GJ. 1987. Queen reg-ulation of worker foraging in paper wasps:
Ann
u. R
ev. E
ntom
ol. 2
001.
46:4
13-4
40. D
ownl
oade
d fr
om a
rjou
rnal
s.an
nual
revi
ews.
org
by U
nive
rsity
of
Lei
den
- W
aleu
s L
ibra
ry o
n 01
/27/
10. F
or p
erso
nal u
se o
nly.
P1: UBI
November 6, 2000 13:30 Annual Reviews AR119-14
MODELS OF DIVISION OF LABOR 439
a social feedback control system (Polistesfuscatus, Hymenoptera, Vespidae).Beha-viour 102:147–67
87. Robinson GE. 1987. Modulation of alarmpheromone perception in the honey bee—evidence for division of labor based on hor-monally regulated response thresholds.J.Comp. Physiol. A160:613–19
88. Robinson GE. 1987. Regulation of honey-bee age polyethism by juvenile hormone.Behav. Ecol. Sociobiol.20:329–38
89. Robinson GE. 1992. Regulation of divi-sion of labor in insect societies.Annu. Rev.Entomol.37:637–65
90. Robinson GE, Page RE. 1995. Genotypicconstraints on plasticity for corpse remo-val in honey bee colonies.Anim. Behav.49:867–76
91. Robinson GE, Page RE, Huang ZY. 1994.Temporal polyethism in social insects is adevelopmental process.Anim. Behav.48:467–69
92. Robinson GE, Page RE, Strambi C,Strambi A. 1989. Hormonal and geneticcontrol of behavioral integration in honeybee colonies.Science246:109–12
93. Robinson GE, Page REJ. 1989. Genetic ba-sis for division of labor in an insect society.In The Genetics of Social Evolution, ed.MD Breed, REJ Page, pp. 61–80. Boulder,CO: Westview
94. Robson SK, Beshers SN. 1997. Divisionof labour and ‘foraging for work’: simulat-ing reality versus the reality of simulations.Anim. Behav.53:214–18
95. Sakagami SF, Maeta Y. 1987. Sociality, in-duced and/or natural, in the basically soli-tary small carpenter bees (Ceratina). InAnimal Societies: Theories and Facts, ed.Y Ito, JL Brown, J Kikkawa, pp. 1–16.Tokyo: Jp. Sci. Soc. Press
96. Schulz DJ, Robinson GE. 1999. Biogenicamines and division of labor in honey beecolonies: behaviorally related changes inthe antennal lobes and age-related changesin the mushroom bodies.J. Comp. Physiol.A 184:481–88
98. Seeley TD. 1982. Adaptive significanceof the age polyethism schedule in hon-eybee colonies.Behav. Ecol. Sociobiol.11:287–93
99. Seeley TD. 1995.The Wisdom of the Hive.Cambridge, MA: Harvard Univ. Press.295 pp.
100. Seeley TD, Camazine S, Sneyd J. 1991.Collective decision making in honeybees—how colonies choose among nectarsources.Behav. Ecol. Sociobiol.28:277–90
101. Sendova-Franks A, Franks NR. 1993.Task allocation in ant colonies withinvariable environments (a study of tem-poral polyethism: experimental).Bull.Math. Biol.55:75–96
102. Sendova-Franks AB, Franks NR. 1994.Social resilience in individual worker antsand its role in division of labor.Proc. R.Soc. London, Ser. B256:305–9
103. Sendova-Franks AB, Franks NR. 1995.Spatial relationships within nests of theant Leptothorax unifasciatus(Latr) andtheir implications for the division of la-bor.Anim. Behav.50:121–36
104. Snyder L. 1993. Non-random behaviou-ral interactions among genetic subgroupsin a polygynous ant.Anim. Behav.46:431–39
105. Spencer AJ, Couzin ID, Franks NR. 1998.The dynamics of specialization and gen-eralization within biological populations.Adv. Complex Syst.1:115–27
106. Stuart RJ, Page RE. 1991. Genetic com-ponent to division of labor among work-ers of a leptothoracine ant.Naturwis-senschaften78:375–77
107. Sullivan JP, Jassim O, Fahrbach SE,Robinson GE. 2000. Juvenile hormonepaces behavioral development in the adultworker honey bee.Horm. Behav.37:1–14
Ann
u. R
ev. E
ntom
ol. 2
001.
46:4
13-4
40. D
ownl
oade
d fr
om a
rjou
rnal
s.an
nual
revi
ews.
org
by U
nive
rsity
of
Lei
den
- W
aleu
s L
ibra
ry o
n 01
/27/
10. F
or p
erso
nal u
se o
nly.
P1: UBI
November 6, 2000 13:30 Annual Reviews AR119-14
440 BESHERS¥ FEWELL
108. Theraulaz G, Bonabeau E. 1995. Mod-elling the collective building of complexarchitectures in social insects with latticeswarms.J. Theor. Biol.177:381–400
109. Theraulaz G, Bonabeau E, DeneubourgJL. 1998. Response threshold reinforce-ment and division of labour in insect so-cieties.Proc. R. Soc. Lond. B265:327–32
110. Theraulaz G, Gervet J, Thon B, PrattM, Semenoff-Tian-Chanski S. 1992. Thedynamics of colony organization in theprimitively eusocial wasp Polistes domin-ulus Christ.Ethology91:177–202
111. Theraulaz G, Gervet J, Semenoff-Tian-Chanski SS. 1991. Social regulation offoraging activities inPolistes dominulusChrist: a systematic approach to behav-ioral organization.Behaviour116:292–320
112. Tofts C. 1993. Algorithms for task allo-cation in ants (a study of temporal poly-ethism: theory).Bull. Math. Biol. 55:891–918
113. Tofts C, Franks NR. 1992. Doing the rightthing—ants, honeybees and naked mole-rats.Trends Ecol. Evol.7:346–49
114. Traniello JFA. 1978. Caste in a primitiveant: absence of age polyethism inAmbly-opone. Science202:770–72
115. Traniello JFA, Rosengaus RB. 1997.Ecology, evolution and division of labourin social insects.Anim. Behav.53:209–13
116. Trumbo ST, Robinson GE. 1997. Learn-ing and task interference by corpse-removal specialists in honey bee colonies.Ethology103:966–75
117. Wagener-Hulme C, Kuehn JC, Schulz DJ,Robinson GE. 1999. Biogenic aminesand division of labor in honey bee
colonies.J. Comp. Physiol. A184:471–79
118. Watmough J, Camazine S. 1995. Self-organized thermoregulation of honeybeeclusters.J. Theor. Biol.176:391–402
119. Wheeler WM. 1928.The Social Insects:Their Origin and Evolution. London:Kegan Paul, Trench, Trubner and Co. Ltd.378 pp.
120. Wilson EO. 1968. The ergonomics ofcaste in the social insects.Am. Nat.102:41–66
121. Wilson EO. 1971.The Insect Societies.Cambridge, MA: Harvard Univ. Press
122. Wilson EO. 1976. Behavioral discretiza-tion and number of castes in an antspecies.Behav. Ecol. Sociobiol.1:141–54
123. Wilson EO. 1983. Caste and division oflabor in leaf-cutter ants (Hymenoptera,Formicidae, Atta) III: Ergonomic re-siliency in foraging byA. cephalotes. Be-hav. Ecol. Sociobiol.14:47–54
124. Wilson EO. 1984. The relation betweencaste ratios and division of labor in the antgenusPheidole(Hymenoptera: Formici-dae).Behav. Ecol. Sociobiol.16:89–98
125. Wilson EO. 1985. The principles ofcaste evolution. InExperimental Behav-ioral Ecology and Sociobiology, ed. BHolldobler, M Lindauer, pp. 307–24.Stuttgart: Gustav Fischer Verlag
126. Wilson EO. 1985. The sociogenesis of in-sect colonies.Science228:1489–95
127. Wilson EO. 1987. Causes of ecologicalsuccess—the case of the ants: the 6thTansley lecture.J. Anim. Ecol.56:1–9
128. Withers GS, Fahrbach SE, Robinson GE.1993. Selective neuroanatomical plastic-ity and division of labour in the honeybee.Nature364:238–40
Ann
u. R
ev. E
ntom
ol. 2
001.
46:4
13-4
40. D
ownl
oade
d fr
om a
rjou
rnal
s.an
nual
revi
ews.
org
by U
nive
rsity
of
Lei
den
- W
aleu
s L
ibra
ry o
n 01
/27/
10. F
or p
erso
nal u
se o
nly.
Annual Review of Entomology Volume 46, 2001
CONTENTS
BIOGEOGRAPHY AND COMMUNITY STRUCTURE OF NORTH AMERICAN SEED-HARVESTER ANTS, Robert A. Johnson 1
MATING BEHAVIOR AND CHEMICAL COMMUNICATION IN THE ORDER HYMENOPTERA, M. Ayasse, R. J. Paxton, J. Tengö 31
INSECT BIODEMOGRAPHY, James R. Carey 79
PREDICTING ST. LOUIS ENCEPHALITIS VIRUS EPIDEMICS: Lessons from Recent, and Not So Recent, Outbreaks, Jonathan F. Day 111
EVOLUTION OF EXCLUSIVE PATERNAL CARE IN ARTHOPODS, Douglas W. Tallamy 139MATING STRATEGIES AND SPERMIOGENESIS IN IXODID TICKS, Anthony E. Kiszewski, Franz-Rainer Matuschka, Andrew Spielman 167
GENETIC AND PHYSICAL MAPPING IN MOSQUITOES: Molecular Approaches, David W. Severson, Susan E. Brown, Dennis L. Knudson 183
INSECT ACID-BASE PHYSIOLOGY, Jon F. Harrison 221EVOLUTION AND BEHAVIORAL ECOLOGY OF HETERONOMOUS APHELINID PARASITOIDS, Martha S. Hunter, James B. Woolley 251
SPECIES TRAITS AND ENVIRONMENTAL CONSTRAINTS: Entomological Research and the History of Ecological Theory, Bernhard Statzner, Alan G. Hildrew, Vincent H. Resh 291
Genetic Transformation Systems in Insects, Peter W. Atkinson, Alexandra C. Pinkerton, David A. O'Brochta 317
TESTS OF REPRODUCTIVE-SKEW MODELS IN SOCIAL INSECTS, H. Kern Reeve, Laurent Keller 347
BIOLOGY AND MANAGEMENT OF GRAPE PHYLLOXERA, Jeffrey Granett, M. Andrew Walker, Laszlo Kocsis, Amir D. Omer 387
MODELS OF DIVISION OF LABOR IN SOCIAL INSECTS, Samuel N. Beshers, Jennifer H. Fewell 413
POPULATION GENOMICS: Genome-Wide Sampling of Insect Populations, William C. Black IV, Charles F. Baer, Michael F. Antolin, Nancy M. DuTeau 441
THE EVOLUTION OF COLOR VISION IN INSECTS, Adriana D. Briscoe, Lars Chittka 471
METHODS FOR MARKING INSECTS: Current Techniques and Future Prospects, James R. Hagler, Charles G. Jackson 511
RESISTANCE OF DROSOPHILA TO TOXINS, Thomas G. Wilson 545
Ann
u. R
ev. E
ntom
ol. 2
001.
46:4
13-4
40. D
ownl
oade
d fr
om a
rjou
rnal
s.an
nual
revi
ews.
org
by U
nive
rsity
of
Lei
den
- W
aleu
s L
ibra
ry o
n 01
/27/
10. F
or p
erso
nal u
se o
nly.
CHEMICAL ECOLOGY AND SOCIAL PARASITISM IN ANTS, A. Lenoir, P. D'Ettorre, C. Errard, A. Hefetz 573
COLONY DISPERSAL AND THE EVOLUTION OF QUEEN MORPHOLOGY IN SOCIAL HYMENOPTERA, Christian Peeters, Fuminori Ito 601
JOINING AND AVOIDANCE BEHAVIOR IN NONSOCIAL INSECTS, Ronald J. Prokopy, Bernard D. Roitberg 631
BIOLOGICAL CONTROL OF LOCUSTS AND GRASSHOPPERS, C. J. Lomer, R. P. Bateman, D. L. Johnson, J. Langewald, M. Thomas 667
NEURAL LIMITATIONS IN PHYTOPHAGOUS INSECTS: Implications for Diet Breadth and Evolution of Host Affiliation, E. A. Bernays 703
FOOD WEBS IN PHYTOTELMATA: ""Bottom-Up"" and ""Top-Down"" Explanations for Community Structure, R. L. Kitching 729
