The Ants Chapter 8

CHAPTER 8. CASTE AND DIVISION OF LABOR

It is possible to distinguish two very general problems in the study of social insects, to which all other topics and theoretical reflection play a tributary role. The first is the origin of social behavior itself. Biologists are especially interested in the advanced state of eusociality, which characterizes all of the ants and a small percentage of aculeate wasps and bees. This evolutionary grade is defined as the combination of three features: care of the young by adults, overlap in generations, and a division of labor into reproductive and nonreproductive castes.

The last of the eusocial traits, the existence of a subordinate or even completely sterile worker caste, is the rarest of the three. It is also by far the most significant with reference to the further evolutionary potential of social life, for when individuals can be turned into specialized working machines, an intricate division of labor can be achieved and a complicated social organization becomes attainable even with a relatively simple repertory of individual behavior. It does not matter, to use a rough metaphor, if the separate pieces of a clock are simple in construction, so long as they can be shaped to serve particular roles and then fitted together into a working whole. You cannot construct a proper clock if every little piece must also keep some sort of time on its own. By the same token natural selection cannot readily assemble a complex colony if every member is designed to complete a reproductive life cycle of its own in the vertebrate manner.

This brings us to the second major problem in the study of social insects, which can be called strategic design, or, more fully, the strategic design of colonies. Once a sterile worker is in place, that is, once the equivalent of the less-than-independent levers and wheels of the clock can be manufactured, the important consideration becomes the best arrangement of castes and division of labor for the functioning and reproduction of the colony as a whole. The colony can be most effectively analyzed if it is treated as a factory within a fortress. Natural selection operates so as to favor colonies that contribute the largest number of mature colonies in the next generation, so that the number of workers per colony is only of incidental importance. In other words, the key measure is not how big, or strong, or aggressive a colony of a particular genotype can become but how many successful new colonies it generates. Hence the functioning of the workers in gathering energy and converting it into virgin queens and males is vital. This part of colony functioning constitutes the factory. But the colony is simultaneously a tempting target for predators. The brood and food stores are veritable treasure houses of protein, fat, and carbohydrates. As a consequence colonies must have an adequate defense system, which often takes the form of stings and poisonous secretions and even specialized soldier castes. This set of adaptations constitutes the fortress.

The ultimate currency in the equations of colony fitness is energy. The workers appear programmed to sweep the nest environs in such a way as to gain the maximum net energetic yield. Their size, diel rhythms, foraging geometry, recruitment techniques, and methods of food retrieval are the qualities most likely to be shaped by natural selection. But even as energy is being collected and distributed to the queen and brood, the colony is subject to predation and competition. A certain number of individuals must be sacrificed in periodic defenses of the colony and during the riskier steps of the daily foraging expeditions. The loss of energy required to replace them is entered on the debit side of the fitness ledger.

Colony design then becomes effectively a problem in economics. The more precise expression is ergonomics, to acknowledge that work and energy are the sole elements of calculation, and also that nothing resembling human transactions with credit and money is involved (Wilson, 1968; Oster and Wilson, 1978).

Caste, task, role
The core of any ergonomic (as well as any human economic) analysis is the prescription of a complete balance sheet, and to this end it is necessary to define as precisely as possible the agents entailed. The fundamental distinctions among individual workers in a colony are based on the actual tasks they perform and the various roles they play in accomplishing the functions of the colony as a whole. An example of the behavioral repertories of two physical castes is given in Table 8-1.

Lists of distinct behavioral acts are made from protracted observations of individuals, selected sets of individuals, or, the entire colony. If distinct physical and temporal groups can be distinguished, separate repertories for each are compiled. In the unlikely event that some of these arbitrarily defined categories prove to have identical repertories, they cannot be distinguished as separate castes --and the data concerning their repertories can be subsequently combined. The next step is to compute the relative frequencies of the acts and to fit them to standard distributions such as the lognormal Poisson or negative binomial. This method has been used with success by Fagen and Goldman (1977) to make estimates of total repertory size even before a complete catalog is actually compiled. An observer thus can obtain a measure of the adequacy of his sample size. For example, after 1,222 separate behavioral observations had been made on the minor worker caste of a colony of Pheidole dentata, 26 kinds of behavioral acts could be recognized; the latter constituted the known behavioral repertory of this caste. By fitting the frequency data (see Table 8-1) to a lognormal Poisson distribution, the actual number of kinds of behavioral acts was estimated to be 27, with a 95 percent confidence interval of (26, 28). The major workers were observed performing 8 kinds of behavioral acts; the true number was estimated to be 9, with a confidence interval of (8, 10).

As analyzed by Wilson and Fagen (1974) there are notable strengths and weaknesses in this method of tabulating behavioral repertories, which has gained wide usage in the study of behavior. One obvious advantage of the technique is that it improves unaided intuition without forcing any new, unsupportable assumptions on the analysis. Another is that the tabulation allows a more precise judgment of the point of diminishing return during ethological studies. That point comes surprisingly early in the case of the ants. A typical example is the study of worker behavior in Leptothorax curvispinosus conducted by Wilson and Fagen. After only 51 hours of observation, during which 1,962 separate acts were recorded, the mode of the frequency curve emerged and the estimated sample coverage attained was 99.95 percent. Thus the effort required to secure a nearly complete repertory is at least a full order of magnitude less than in the case of vertebrates, which typically consumes months or years of arduous field research. The result is that comparative behavioral studies have proceeded much more rapidly in ants and other insects.

The tractability of ant studies has a rationale of considerable biological interest. This is the small number of rare behavioral acts that exist relative to common acts (see Figure 8-1). In other words, whatever ants do they do rather frequently in comparison with vertebrates; few if any rare behaviors exist to surprise the investigator in the late stages of the study. It is possible that the small size of the ant brain precludes the storage of acts that are not used commonly or at least are not of crucial importance to the colony. This interpretation is consistent with the principle of parsimony in the evolution of ant communication systems, which we documented in Chapter 7. In briefest form, the concept of parsimony holds that a characteristic of ant behavior is the repeated use of the same communicative signals and responses in different contexts to achieve multiple purposes.

There are two disadvantages of the method which we do not regard as particularly serious. The first is the probability that the repertories and especially the frequency distributions change in different contexts. It remains for the biologist to define those contexts and to repeat the analysis within them. In the case of ants distinguishable contexts are not only finite but also probably quite limited in number. By far the greatest part of an ant's life is conducted in the homeostatic environment of the nest interior. Thus the lifetime sample coverage in the Pheidole dentata study, defined as the cumulative probability of all behavior for all contexts, was probably very high in spite of the fact that it was limited to one environment. We suggest that the following list might exhaust the remaining contexts for the worker caste: extended foraging periods and territorial patrolling and defense; major disturbances of the nest, including invasion by alien colonies, flooding, and overheating; emigration to a new nest site; and assisting during the initiation of nuptial flights on the part of the reproductive forms.

The second difficulty in repertory estimation is the arbitrariness of the definition of the behavioral act. One observer might see three distinct neuromuscular patterns where another sees only one. Thus “foraging” as defined in Table 8-1 could easily be broken down into several acts. This is essentially a problem of language, and different observers can solve it by a straightforward mapping procedure. One observer's acts a, b, and c will be recognized as comprising the second observer's act a; the first observer's act h will be seen as representing the second observer's acts m and n; and so forth. No great difficulty should occur when the same species is considered or closely related species are compared. Serious conceptual problems might exist, however, when an attempt is made to compare the size and frequency characteristics across radically different species.

The next step beyond the behavioral repertory is the construction of an ethogram, which incorporates not only the repertory of a caste but also the transition probabilities connecting individual acts and the time distributions spent in each act. When the ethogram also takes into account the interactions of parents and offspring, dominant and subordinate individuals, and other members of the society, the description is referred to as a sociogram. Ethograms can cover all of the repertory or certain well-defined portions of it. Such quantitative studies have been conducted on ants, honeybees, hermit crabs, mantis shrimps, and rhesus monkeys, with promising results (see reviews by Dingle, 1972; Wilson, 1975b: 194-200).

Figure 8-2 illustrates an imaginary example of an ethogram with role and castes delimited. For any individual, certain of the behavioral acts (ai) will be linked together by relatively high transition probabilities: in Pheidole dentata, for example, pupal licking is associated with high frequency to pupa carrying, egg licking, and egg transport; nest construction gives way with high probability to foraging outside the nest; and so on. A set of closely linked behavioral acts can be defined as a role, even if the acts are otherwise quite different. It is generally true, for example, that the act of grooming the queen is closely linked with the very different acts of regurgitating to the queen and removing freshly oviposited eggs. All of these responses can be considered part of the single role of queen care. We can define a caste as a group that specializes to some extent on one or more roles. Broadly characterized, a caste is any set of a particular morphological type, age group, or physiological state (such as inseminated versus barren) that performs specialized labor in the colony. A natural classification follows from this definition. A physical caste is distinguished not only by behavior but also by distinctive anatomical traits. A temporal caste, in contrast, is distinguished by age. A physiological caste is distinguished by a principal physiological state that is frequently but not necessarily linked to anatomy or age, such as insemination. Sometimes all of the workers together are referred to as a caste, as distinct from the queen, and subsets of workers are called subcastes.

The term task is used to denote a particular sequence of acts that accomplishes a specific purpose, such as foraging or nest repair. Ordinarily a task is identical to a role or is composed of the subset of acts within a role, but it might conceivably consist of acts distributed across two or more roles. Finally, the division of labor by the allocation of tasks among various castes is often referred to as polyethism, a term apparently first employed by Weir (1958a,b). We further speak of physical caste polyethism as opposed to temporal caste polyethism.

Few studies have probed to the depth necessary to yield quantitative measures of castes and tasks. Wilson (1976c) recognized four worker subcastes of Pheidole dentata, comprising a unified age category in the major worker and 3 age categories in the minor worker. Together the castes perform a known total of 26 acts in undisturbed colonies, of which 23 can be classified as true social tasks. Similarly, a total of 29 tasks have been identified in the leafcutter ant Atta sexdens (Wilson, 1980a). These are performed by an estimated four physical castes, of which three are further subdivided into temporal castes to make a total of at least seven castes overall. The most thorough studies of labor division have been conducted on Lasius niger (Lenoir, 1979a,c) and on species of Leptothorax (Herbers, 1983; Herbers and Cunningham, 1983). These authors combined matrix analysis with methods of multivariate analysis to cluster the tasks into roles. Two or three roles, or “components” as they were called by Lenoir, could be recognized in Lasius. In the Leptothorax, the workers comprised three behavioral castes filling a total of four roles.

It will be useful at this point to introduce a short glossary of terms applied to castes that have gained at least moderate employment by myrmecologists. This is a difficult procedure, for two reasons. First, an extraordinary number of technical terms has been proposed. Many of them were suggested by William Morton Wheeler, who was not only a great entomologist but a classical scholar fluent in Greek with an inordinate fondness for neologisms. In 1907 he proposed a caste system based on anatomy with no fewer than 30 categories. He recognized pathological forms such as the phthisaner, a pupal male with appendage growth caused by an Orasema wasp larva; the pseudogyne, a worker-like form with swollen mesothorax in the genus Formica caused by the presence of lomechusine staphylinid beetles in the nest; and the pterergate, a worker ant with vestiges of wings. During his lifetime and especially in his last work on the subject in 1937, Wheeler believed he had found good reasons for multiplying and naming every qualitatively distinguishable form. On the one hand he considered the parasitogenic forms to be different enough and rare enough to fit comfortably into the system. More important, however, he believed (erroneously) that most nonparasitogenic castes arise by genetic mutations. He saw no fundamental difference between normal functional castes and true anomalies. All except the queen and typical males were basically anomalous forms to Wheeler, and he referred to his categories alternatively as castes, phases, and anomalies. In a reevaluation of Wheeler's system, Wilson (1953b) concluded that some of the names are superfluous, some are virtually synonymous with others, and some are no more than stages in an allometric progression. Over the years a substantial literature of teratology has accumulated on gynandromorphs and other developmental anomalies, some of which possess traits intermediate between normal castes, but there has been no inclination to recognize any of these forms as castes; descriptions and reviews have been provided variously by Wheeler (1937b), Novak (1948), Kusnezov (1951a), Buschinger and Stoewesand (1971), Wilson and Fagen (1974), Sokolowski and Wisniewski (1975), Peru (1984), and Taber and Francke (1986).

A second cause of ambiguity in classification is the dispute among leading students of the subject over whether castes should be defined primarily by role or by anatomy. In particular, Buschinger (1987b) has proposed that “queen” and “worker” be used to designate females with reproductive and nonreproductive roles, regardless of whether they differ anatomically in a particular species, while various other terms such as gynomorph and intermorph be used to denote the anatomical variants within these two principal categories. Peeters (1987a,b), in opposition, prefers a morphological definition of the castes “queen” and “worker," with the term “gamergate” being used for the relatively uncommon condition in which mated, completely worker-like forms have replaced the queen caste in “queenless” societies. The two positions are conceptually not far apart and can be semantically resolved.  Our opinion is that the classification should be kept somewhat loose, incorporating either anatomy or roles in a manner that maximizes convenience, precision, and clarity of expression.  It must be recognized that anatomy and reproductive role are related across the various ant species not on a one-to-one linkage but by a many-to-many linkage.  That is, anatomical workers often join the anatomical queens in reproduction although usually to a limited extent.  In such species the reproductive role can be said to be assumed by both anatomical castes. And queens in their turn assume many roles other than reproduction. Such many-to-many relationships are nearly universal among the anatomical worker subcastes and labor roles. It is also true that anatomically distinct (i.e., physical) castes are almost always biased toward certain roles, while individual roles are almost always filled preferentially by one anatomical caste or another in the case of polymorphic species. With these qualifications in mind, the following annotated glossary is the most efficient and for the most part reflects current usage (see also Figure 8-3).

1. Male. In the great majority of ant species, males (designated by the symbol ) fill no role in the colony that generated them, being content to receive food from their sisters while awaiting the nuptial flight that will end their lives. In such cases it is misleading to refer to them as a caste. In some species of Camponotus, the males are long-lived and serve as donors in food exchange within the colony. Hence they qualify marginally as castes. To take another, very different phenomenon, worker-like or “ergatomorphic” males occur in several genera (Hypoponera, Cardiocondyla, Formicoxenus, Technomyrmex). However, so far as known they do not contribute to colony labor.

2. Queen. Often designated by the symbol  (contrast with symbols for worker, to follow). In the broad sense, the principal female reproductive type, in other words any form anatomically distinguishable from the worker caste and responsible to a disproportionate extent for reproduction. In the narrow sense, employed in more technical literature on caste biology, the fully developed reproductive female, possessing a generalized hymenopterous thorax and functional but deciduous wings. The queen is sometimes referred to loosely as the “female” of the colony. The term gyne was used synonymously with queen by Wheeler (1907a). Brian (1957b) employed it to denote more specifically “a sexual female that is not socially a functional reproductive.” In recent years the word has gained wider usage in conformity with Wheeler's original meaning. It is employed most consistently as a neutral expression, to cover any reproductive female whether virgin or inseminated, laying or infertile. In these writings, “queen” then denotes a functioning gyne. In a few species such as the primitive Australian Nothomyrmecia macrops, the wings are reduced in size and probably nonfunctional (Figure 8-4). Such individuals are designated micropterous queens (or micropterous females) in the literature. Occasionally, queens are exceptionally small, even smaller than the workers in some dimensions. Such microgynes, as they are called, are usually social parasites. Examples are found in the South American myrmicine Pheidole microgyna (Wilson, 1984c), the parasitic and microgynous form of the European Myrmica rubra complex (Pearson, 1981), and members of the North American Formica microgyna group (Wheeler, 1910a; Creighton, 1950). Microgynes coexist in the same colonies with normal, “macrogynous” queens, in the arboreal Pseudomyrmex veneficus (Janzen, 1973b) and an arboreal Australian species of Polyrhachis subgenus Cyrtomyrma (Bellas and Hölldobler, 1985; see Figure 8-5).

3. Worker. Often represented by the symbols  or. The ordinarily sterile female, possessing reduced ovarioles (or none at all) and a greatly simplified thorax, the nota of which typically consist of no more than a single sclerite each (Tulloch, 1935). Even when ovarioles are present, the workers usually lack spermathecae in which sperm can be stored. Hence when they become reproductive they cannot produce daughters except by thelytokous parthenogenesis--an uncommon event in ants generally (Crozier, 1975; Buschinger and Winter, 1978; Choe, 1988; Bourke, 1988). The disparity between the reproductive apparatus of workers and queens varies considerably among ant species. In Nothomyrmecia macrops, a very primitive ant, workers possess four to seven ovarioles and queens eight to ten (Hölldobler and Taylor, 1983). In Leptothorax nylanderi only two ovarioles are found in workers and eight in queens (Plateaux, 1970; Figure 8-6), while in Formica polyctena workers have four to six ovarioles as against 90 and 270 in queens (Otto, 1958, 1962; Figure 8-7). On the other hand in some species of Rhytidoponera whose mated workers have replaced the queen as the functional egglayers, there is no significant difference between the ovaries of reproductives and those of non-reproductives. However, only inseminated workers appear to lay eggs (Peeters, 1987b). The queen of Eciton hamatum has approximately 4600 ovarioles (Hagan, 1954), and that of Eciton burchellii has at least twice as many (Whelden, 1963; Schneirla, 1971). In some other species, such as the members of Pheidole and Solenopsis, workers possess no ovarioles at all (Wilson, 1971). The worker caste is often subdivided into additional castes, or subcastes, as follows:

4. Minor, media, major. Castes in a size-variable worker series are designated according to ascending size as minor, media, or major workers. In some species the media subcaste mostly or entirely drops out, and the minor and majors are easily distinguished as two frequency distributions with distinct modes. In this case, the symbol  designates the major, and the symbol   usually designates the minor. When their function is partly or wholly fighting on behalf of the colony, the majors are often referred to as soldiers.

5. Ergatogyne. A reproductive form anatomically intermediate between the worker and the queen. A variety of relatively informal terms have been used to denote the intermediates. These terms range from “apterous females” (Bolton, 1986b), and “gynomorph” or “gynomorphic workers,” “ergatomorph” or “ergatomorphic queens,” and “intermorphs” (Buschinger and Winter, 1976). We are inclined to favor a bipartite classification suggested to us by Christian Peeters (personal communication). It makes a distinction between intercastes, which are forms that are intermediate in anatomy between workers and queens, commonly lack a spermatheca (and hence are incapable of mating), and occur in conjunction with typical workers and queens; and ergatoid queens, which are also intermediate in anatomy but possess spermathecae and typically replace the ordinary queen caste as the female reproductive.

Intercastes as just defined occur most commonly in social parasites of the leptothoracine genera Formicoxenus, Harpagoxenus, and Protomognathus (see Figure 8-53), but they also occur in some free-living species of Leptothorax and Monomorium. They give the impression of arising from a failure of endocrine regulation in the differentiation of queens and workers. In the most detailed study to date, Plateaux (1970) showed that there is a continuous anatomical progression in intercastes of the free-living Leptothorax nylanderi, manifested in body size, ocelli, segmentation of the thorax, number of ovarioles, the spermatheca, and elsewhere. While the character states are strongly concordant in the workers and queens, they vary independently of one another in the intercastes. For example, the thorax and gaster do not increase in size in a well coordinated manner, and some individuals occur that combine large ovaries with worker-like thoraces. Intercastes of most of the species, such as Harpagoxenus canadensis, Leptothorax nylanderi, and Protomognathus americanus, lack spermathecae. On the other hand, the intercastes of Formicoxenus species and Harpagoxenus sublaevis do have spermathecae and hence often mate and lay eggs.

Ergatoid queens, on the other hand, replace the typical queen entirely and are not connected to the worker caste by a graded series. This second, truly functional type of ergatogyne is especially common in the primitive genus Myrmecia and in several genera of the subfamily Ponerinae. Haskins and Haskins (1955) have pointed out that the tendency is partly correlated with the habit of queens of primitive ants to forage for food outside the nest during colony founding. In higher ants claustral founding is the rule, and queens usually must rear their first brood entirely on the reserves contained in their own fat bodies and degenerating flight muscles. Hence fully differentiated queens are a necessity when colonies are founded claustrally. Ergatoid queens nevertheless do occur in some free-living higher ants. They are the rule, for example, in the aberrant dolichoderine genus Leptomyrmex (Wheeler, 1934a) and in the legionary cerapachyines (a tribe of the Ponerinae) and dorylines (Wilson, 1958e). They occur in Aphaenogaster phalangium of Central America and Blepharidatta brasiliensis (Wilson, unpublished). They are common in species of the Old World Monomorium salomonis group, where a graded series on the queen-worker gradient can be found among various species but not within species (Bolton, 1986b; see Figure 8-8). Ergatoid queens are also the sole form of reproductive in a high percentage of the endemic species of New Caledonia, belonging to such phylogenetically advanced genera as Monomorium (= Chelaner), Lordomyrma, Prodicroaspis, and Promeranoplus (Wilson, 1971). New Caledonia is an old, very isolated island, and ergatogyny in its ants corresponds to the flightless state found so commonly among the endemic species of birds and insects on oceanic islands. Where ergatoid and micropterous queens have replaced true queens in higher species, it seems likely that either workers accompany the reproductives during colony founding, or else the reproductives revert to foraging on their own during colony founding. Supporting evidence of the former alteration has been adduced in the case of Monomorium (= Chelaner) (Briese, 1983). Many ergatoid queens show a divergent trend away from the queen-worker gradient in that the development of the gaster and postpetiole outpaces the development of the thorax and head. Such variants possess a gaster that approaches (or surpasses) in size that of the typical queen, while the thorax and head are more typically worker-like.

6. Gamergates. A growing number of ponerine ants has been discovered in which the queen caste is replaced by reproductives that are anatomically indistinguishable from the worker caste. The species in this category include Diacamma rugosum (Wheeler and Chapman, 1922; Fukumoto, 1983; Moffett, 1986h), Dinoponera grandis (Haskins and Zahl, 1971), Leptogenys schwabi (Peeters, 1987a), Ophthalmopone berthoudi (Peeters and Crewe, 1984, 1985), Pachycondyla krugeri (Peeters and Crewe, 1986b), Platythyrea schultzei (Peeters, 1987a), and several species of Rhytidoponera (Haskins and Whelden, 1965; Ward, 1981a,b, 1983a,b; Peeters, 1987b). The only examples known from the Myrmicinae are apparently a few species belonging to the South American Megalomyrmex leoninus group (Brandão, 1987). Peeters has suggested the term gamergate for such inseminated workers, which seems a useful and clarifying step. It is the one case known to us of a true physiological caste, in which a particular physiological state (insemination and enhanced oogenesis) rather than anatomy or age distinguish a labor group. On the other hand, it may be pushing the term too far to include the reproductive workers of Pristomyrmex pungens, which are not fertilized and produce females by parthenogenesis. In this extreme, aberrant case the colony can be said to lack a physical caste system (Itow et al., 1984; Tsuji and Itô, 1986). On the other hand, Pristomyrmex pungens have a typical age caste system: young workers work inside the nest and lay eggs, while older workers have reduced ovaries and tend to work outside the nest (K. Tsuji, personal communication).

7. Dichthadiiform ergatogyne. The dichthadiiform female is the extreme stage of the phylogenetic trend toward enlargement of the gaster in the ergatogyne, and it is usually recognized as a distinct category. In fact, it is just an aberrant queen and is never accompanied by ordinary queens. The total size of the female is greatly increased, the gaster is huge, and the postpetiole is expanded to the extent that it has come secondarily to resemble the first gastric segment (see Figure 8-9). In addition, the head is broadened and rounded, the mandibles are often falcate, and the petiole is commonly bilaterally cornulate. Dichthadiiform females are limited to ants with a legionary mode of life. The extreme development is seen in the subfamilies Dorylinae, Ecitoninae, and Leptanillinae and the ponerine genus Onychomyrmex (Wilson, 1958e; see Figure 8-10). An intermediate stage in the evolution of this form occurs in the ergatogynes of the ponerine Acanthostichus quadratus and species of the ponerine genus Simopelta (Gotwald and Brown, 1966). All of these phyletic lines, both intermediate and advanced, evidently evolved their dichthadiiform form independently.

Temporal castes. We now take up a literally different dimension in caste formation. A temporal caste is a set of individuals distinguished both by behavior and by age. It is thus wholly different from physical castes, which are marked by some anatomical feature, and physiological castes, which may in a few cases be distinguished by insemination or some other principal physiological feature alone, in a manner unrelated to anatomy and age. In Pheidole dentata, for example, one temporal caste consists of the youngest workers, from the moment of emergence from the pupa to about two days into adult life. These individuals care preferentially for the queen, eggs, and young larvae. The second temporal caste, which tends to care for medium-sized larvae, extends to about the sixteenth day of adult life. The third temporal caste, which prefers nest work and foraging to a greater extent, consists of still older workers (Wilson, 1976c).

Adaptive demography
A key principle of caste evolution is the adaptive nature of colony demography (Wilson, 1968; Oster and Wilson, 1978). Ordinary demography, of the kind found in nonsocial organisms, in social vertebrates, and in more primitively social insects such as subsocial wasps and bees, is a function of the individual parameters of growth, reproduction, and death. A large amount of documentation from free-living and laboratory populations supports the general belief that growth and natality schedules are direct adaptations shaped by natural selection at the level of the individual (Krebs and Davies, 1981). On the other hand, the size and age structure of the population as a whole are epiphenomena, in the sense that they reflect the individual-level adaptations but do not constitute adaptations in their own right. Thus a sharply tapered age distribution in any species of fishes and birds results from a high birth rate and a high mortality schedule throughout the life span, but in itself does not contribute to the survival of the population or the individual members. The exact reverse is the case of the eusocial insect colony defined as a population. The demography, not its causal parameters, is directly adaptive. The workers are for the most part sterile; their birth and death schedules have meaning only with reference to the survival and reproduction of the queen. Hence the unit of selection is the colony as a whole. The traits of the colony are the larger features of demography, the age and size-frequency distributions. What matters are such higher-level traits as the number of very large adults available to serve as soldiers, the number of small young adults functioning as nurses, and so on through the entire caste roster and behavioral repertory. Each species has a characteristic age-size frequency distribution, and the evidence is strong that this colony-wide trait is not an epiphenomenon, but has been shaped by natural selection, constituting a direct adaptation (see Figure 8-11).

A striking example of adaptive demography is provided by developmental changes in the caste system of the leafcutter ant Atta cephalotes (Wilson, 1983b). Beginning colonies, started by a single queen from her own body reserves, have a nearly uniform size-frequency distribution across a relatively narrow head-width range of 0.8-1.6 mm. The key to the arrangement is that workers in the span 0.8-1.0 mm are required as gardeners of the symbiotic fungus on which the colony depends, while workers with head widths of 1.6 mm are the smallest that can cut vegetation of average toughness. This range also embraces the worker size groups most involved in brood care. Thus, remarkably, the queen produces about the maximum number of individuals which together can perform all of the essential colony tasks. As the colony continues growing, the worker size variation broadens in both directions, to head width 0.7 mm or slightly less at the lower end to over 5.0 mm at the upper end, and the frequency distribution becomes more sharply peaked and strongly skewed to the larger size classes (see Figures 8-12 and 8-13).

A more physiological question immediately arises from the observed sociogenesis (colony ontogeny) of Atta: which is more important in determining the size-frequency distribution, the size of the colony or its age? In order to learn the answer, Wilson selected four colonies 3-4 years old and with about 10,000 workers and reduced the population of each to 236 workers, giving them a size-frequency distribution characteristic of natural young colonies of the same size collected in Costa Rica. The worker pupae produced at the end of the first brood cycle possessed a size-frequency distribution like that of small, young colonies rather than larger, older ones. Thus colony size and indirectly the amount of food produced are more important than age.

The Atta example is just an extreme case of what appears at least on the surface to be programmed colony demography among ant species. However, the physiological control mechanisms remain almost entirely unexplored. The “rejuvenation” effect in Atta cephalotes indicates that a feedback loop of some kind is involved, as opposed to an irreversible maturation of the size-frequency distribution; but its nature has not been investigated.

The ontogeny of the physical caste system has also been traced in the red imported fire ant Solenopsis invicta, as illustrated in Figure 8-14. In this species the first adult generation of workers consists entirely of “nanitics” or “minims,” with head widths about 0.50 mm. These individuals, as Porter and Tschinkel (1986) have demonstrated experimentally, are more effective as a group in rearing the second generation of adults, but less efficient energetically on an individual basis than the slightly larger “minor” workers. When the population grows by the addition of the second and later broods, it is to be expected on the basis of ergonomic selection at the colony level that the minims will be quickly replaced by minors and still larger workers. Just such a transition actually occurs, as shown in Figure 8-14. Subsequent field studies by Tschinkel (1988a) revealed that as the fire ant colonies grew, they changed from a monomorphic to a polymorphic worker force. The production of larger, major workers causes the overall size-frequency curve to become skewed in the manner just illustrated (Figure 8-9) for Atta. The skewed distributions are actually bimodal. They consist of two slightly overlapping normal distributions, a narrow one defined as the minor workers and a much broader one defined as the major workers. The “media” workers have no clear developmental definition and are simply individuals in the zone of overlap between the minor and major frequency curves. In full-sized colonies, the majors make up about 35 percent of the total worker force and 68 percent of its biomass.

It is evident that each ant species possesses a particular programmed ontogeny in the ratios of its subcastes (size-frequency distribution). In other words, the ratios are species-specific and hence genetically prescribed. Even differences encountered among colonies of the same species may have a partial genetic basis, as suggested by laboratory studies conducted on the North American myrmicine ant Pheidole dentata by Johnston and Wilson (1985). The worker force of colonies of Pheidole dentata, like those of other species of Pheidole, is divided into two discrete castes: small-headed minor workers, which conduct most of the quotidian tasks of the colony, and large-headed major workers, which are specialized for defense (see Table 8-1 and Figure 8-20). In natural populations of Pheidole dentata in northern Florida, majors make up between 5 and more than 50 percent of the worker force. The higher figures are rare and almost certainly pathological, and the great majority of colonies in the field and laboratory maintain a representation of majors between 5 and 15 percent. When Johnston and Wilson altered laboratory colonies displaying this latter range of variation to a uniform percentage of 5 percent and then allowed them to grow freely, the colonies changed to the original rank ordering and in most cases not far from their original subcaste percentages. Wheeler and Nijhout (1984) have shown that excess numbers of majors in Pheidole bicarinata ( vinelandica) colonies inhibit further production of majors in a manner that restores the species-characteristic ratio. The inhibition is evidently due to a pheromone that affects the endocrine system of presumptive soldiers. It seems likely that a similar mechanism operates in Pheidole dentata, and that the variation in ratios among colonies arises from genetic differences in major pheromone production, sensitivity to the pheromone, or both. The result is potentially important for an understanding of the evolution of caste ratios, since genetic variation in any trait is a prerequisite for its modification through time.

To summarize, there appears to be a species average or “norm” in caste ratios around which individual colonies vary due to both genetic and environmental differences. The environmental factors most important to the variance have only begun to be explored. For example, it is a common observation that majors of ant species live longer than minors, so that when the queen dies or ceases oviposition for long periods of time the major/minor ratio rises. It also increases above ordinary levels when the colony is starved or desiccated, because larger ants withstand such stress longer than their nestmates. The differential is commonly observed in the laboratory and has been documented explicitly in Camponotus and Formica by Kondoh (1977). In Veromessor pergandei, a seed-harvesting myrmicine living in the deserts of the southwestern United States, there is an annual cycle in average worker size. Smaller workers appear in the foraging force as a result of the wintertime “triple crunch” caused by a reduced seed crop, shorter suitable foraging time, and the added expense of production of reproductive forms (Rissing, 1987). In a reverse trend, major workers of the fire ant Solenopsis invicta become relatively most common in the late winter and early spring, probably due to their lower mortality under adverse conditions (Markin and Dillier, 1971). Such unusual fluctuations may well be just epiphenomena, that is, “noise” in the system of no adaptive significance to the colonies.

The evolution of physical castes
In ants the diversification of the female castes is based mostly on allometry. During larval development, the imaginal discs (patches of undifferentiated tissue destined to be transformed into adult organs at the pupal state) grow at different rates, a process that swiftly accelerates during pupal development (Brian, 1957a, 1979a; Schneirla et al., 1968). The principal result of differential growth rate in the imaginal discs is that various organs end up with different sizes relative to one another according to how large the individual is at the termination of the larval period. That is, the final adult size determines how much overall growth the various organs have attained. Thus, if the disc destined to transform into part of the head is growing faster than the disc destined to transform into part of the thorax, it will finish proportionately larger in an ant that attains a larger overall size. In short, big ants will have proportionally even bigger heads. If each disc grows exponentially, and if the disc growth rates do not change much in the course of development, the sizes of two parts will be related by a simple power law: log y = log b + a log x, or, equivalently, y = bxa where y and x are linear measures of the two body parts and a and b are fitted constants the values of which depend on the nature of the measurement taken. This simple relation is referred to as allometry or heterogonic growth. Its study in ants and other organisms was pioneered by Julian Huxley (1932) and has been reviewed extensively by Gould (1966). On a double logarithmic plot the curve is a straight line. Its slope a is determined by the rate of divergence of the two body parts with increasing total size and can be referred to as the allometric constant. If a is equal to unity, no divergence takes place with an overall increase in size; the growth is then referred to as isometric. The greater the departure of a from unity, the more striking the differential growth. Skellam, Brian, and Proctor (1959) found that in Myrmica rubra the imaginal discs of the wings and legs actually grow in this elementary fashion during development of adult queens and males. The organs predictably conform to the basic allometry equation in their final adult form.

It is possible to produce a wide array of deviations from elementary allometric growth by simply speeding or slowing the growth rates of different discs according to different time schedules. Further complexity can be introduced by making the rates dependent on the total size of the larva reached by certain ages. This last effect is crucial in the determination of the queen and worker castes in Myrmica, as documented by M. V. Brian. It also occurs during the differentiation of the worker caste of the Eciton army ants, as discovered by Tafuri (1955). These modifications are crucial for the discretization of physical castes within the adult instar of ants, a process that will now be examined in some detail.

Wilson (1953b) demonstrated that the allometry equation, or relatively simple modifications of it, can be applied almost universally to continuous variation in the hard parts of ants. The comparative study of allometry has proved fruitful in tracing the evolution of castes. Polymorphism, as this research has led us to understand it, embraces three variable characters in the adult females within colonies of any species: size variation, shape variation due to allometric growth at the time of adult formation, and the frequency distribution of different sized workers. A physical caste system (or polymorphism as it is also frequently called) is defined as growth occurring over a sufficient range to produce individuals of distinctly different sizes, body proportions, or both. Where polymorphism exists, it has always been found to be closely linked to division of labor.

By comparing a large fraction of the more than 10,000 living ant species, it has been possible to infer five major steps in the evolution of physical castes (Wilson, 1953b):

(1) Monomorphism. The workers of the normal mature colony display isometry (with a log-log curve slope of approximately 1.0) and very limited size variability. A plot of their size-frequency distribution is symmetrical and has only a single mode. In other words, the properties of variation are not basically different from those in a typical random collection for nonsocial insects. The worker castes of most ant genera and species are monomorphic. Also, within the majority of genera and higher taxonomic groups monomorphism is evidently the primitive state.

(2) Monophasic allometry. The relative growth is nonisometric, meaning that the allometric constant a is greater or less than unity to a subjectively noticeable degree. In the most elementary form of monophasic allometry, and hence of worker polymorphism generally, feeble nonisometric variation is displayed over a short span of size variation; this variation in turn is grouped around a single mode with possible skewing in the direction of the major caste. A more advanced stage, involving an increased variation in size together with an apparent bimodal size-frequency distribution, is exemplified by the fire ant Solenopsis invicta (Figure 8-15). The two modal groups are still overlapping, and hence the majors and minors are connected by intermediate-sized workers, or “medias.”

(3) Diphasic allometry. The allometric regression line, when plotted on a double logarithmic scale “breaks” and consists of two segments of different slopes that meet at an intermediate point. In the several species known to possess this condition, including the leafcutter ant Atta texana (Wilson, 1953), African driver ant [[Dorylus nigricans (Hollingsworth, 1960), and carpenter ant Camponotus sericeiventris'' (Busher et al., 1985), the size-frequency curve is bimodal. Also, the saddle between the two frequency modes falls just above the bend in the allometry curve.  Diphasic allometry permits the stabilization of the body form in the small caste while providing for the production of a markedly divergent major caste by means of a relatively small increase in size.  The lower segment of the relative growth curve is nearly isometric, so that individuals falling within a large segment of the size range are nearly uniform in structure; but the upper segment leading to the major caste is strongly nonisometric, with the result that a modest increase in size yields a new morphological type.

(4) Triphasic and tetraphasic allometry. The allometric line breaks at two points and consists of three straight segments. The two terminal segments, representing the minor and major castes respectively, deviate only slightly from isometry while the middle segment, encompassing the media caste, has a very steep slope. The effect of triphasic allometry is the stabilization of body proportions in the minor and major castes. An example from the weaver ant genus Oecophylla is presented in Figures 8-16 and 8-17. Baroni Urbani (1976b) has reported a case of tetraphasic allometry in the antennal length of the West African ant Camponotus maculatus. The curve resembles that of triphasic species except that in the largest size classes a high degree of allometry is resumed. The biological significance of this extreme pattern is wholly unknown.

(5) Complete dimorphism. Two morphologically very distinct size groups exist, separated by a gap in which no intermediates occur. Each class is either isometric or non-isometric, but the allometric regression curves are not aligned, a condition suggesting that complete dimorphism can arise directly from triphasic allometry. Examples include most queen-worker differences in ants. They also include minor-major divisions in no fewer than eight phylogenetically scattered genera: the myrmicines Acanthomyrmex, Oligomyrmex, Pheidole, and Zacryptocerus; the aneuretine Aneuretus; the dolichoderine Zatapinoma; and the formicines Camponotus (subgenera Colobopsis and Tanaemyrmex) and Pseudolasius (Plate 7).

By arranging species of ants along a gradient from what appear to be the simplest to the most advanced systems, worker castes can be found that display virtually every conceivable step in a transition from perfect monomorphism to a complete dimorphism (see Figure 8-18). Certain large taxonomic groups, such as the subfamilies Myrmicinae and Formicinae, embrace the entire evolutionary sequence within themselves. The evolution has thus occurred repeatedly within multiple phyletic lines in the ants and produced a remarkable degree of convergence between the lines.

Of the three principal qualities of polymorphism (size variation, allometry, and size frequency), the size-frequency distribution has been subject to the stricter and more notable convergence. When individual colonies of a given species show a slight increase in size variance, the frequency curve is almost always skewed toward the larger size classes. When the intracolonial size variance is still greater, the frequency curve is bimodal. In at least one species of army ant with extreme size variation, data published by Topoff (1971) and da Silva (1972) show the existence of three size modes. Three modes may occur in the extremely polymorphic myrmicine Pheidologeton diversus (Moffett, 1987b). Otherwise enlarged intracolonial size variation is typically associated with bimodality, with the large workers being less common. In weaver ants of the genus Oecophylla, the reverse is true: the majors are more common than the minors, and they fill most of the labor roles of the colony. This unusual form of polymorphism has existed since at least as far back as Miocene times, as shown by Wilson and Taylor (1964). These authors describe the extinct species Oecophylla leakeyi from a colony fragment that had been preserved intact, including even pupae and packets of larvae. With this extraordinary material, the only fossil insect colony found to date, it was possible to measure allometry and size frequency and compare these traits with those of the modern species.

The ways by which ant species create castes out of the adult instar are few. It is reasonable to ask how a colony limited to simple skewing and a maximum of two or three modes is able to produce castes in proportions that match the frequencies of numerous environmental contingencies. The answer appears to be that the evolutionary sequence unfolds within narrow physiological constraints. Species can only improve their situation to the extent of increasing the intracolonial size range and arranging something close to the optimum numbers of majors, minors, and medias. Physical caste systems, in a word, are coarse-tuned rather than fine-tuned (Oster and Wilson, 1978).

Temporal castes
The adult workers of almost all kinds of social insects change roles as they grow older, ordinarily progressing from nurse to forager. Each species has its own distinctive pattern of temporal polyethism, and in many the behavioral changes are accompanied by patterned shifts in the activity of exocrine glands. For example, as honeybee workers shift during a two-week period from a preoccupation with brood care and nest work to an emphasis on foraging, the activity of the hypopharyngeal and wax glands declines somewhat while that of the labial glands increases. For the purposes of ergonomic analysis it is useful to consider different age groups as constituting distinct temporal castes (Wilson, 1968). Just as a species may manipulate its own developmental biology in the course of evolution to produce optimum ratios of physical castes, it can adjust the program of role change during adult life to approach optimum ratios of temporal castes.

Two extreme alternatives are open to a species in the process of evolving temporal castes. These are represented in Figure 8-19. The aging period depicted is that which occurs from the moment of the worker's eclosion from the pupal skin to the moment of its death by senescence. The division of the life span into six periods in this imaginary example is arbitrary. The worker undergoes physiological change with age such that its responsiveness to various environmental stimuli changes. For example, suppose that T1 is the responsiveness to a misplaced egg: the curve indicates that when the worker is very young (age I) it is likely both to be in the vicinity of the egg and to react by picking the egg up and putting it on an egg pile. Its location and/or behavioral responsiveness change as it ages in such a way that its probability of response to the contingency drops off rapidly after ages I or II. Such age-dependent responsiveness has been abundantly documented in ants (Topoff et al., 1972; Cammaerts-Tricot, 1974b; Jaisson, 1975; Topoff and Mirenda, 1978; MacKay, 1983b).

Let us now consider the possibilities. In the upper diagram of Figure 8-20, labeled Model 1, the response curves to four contingencies (T1 through T4) are all out of phase. The curve of response to T1 (misplaced egg) is different from the curve of response to T2 (say, a hungry larva), and so on. As a result, the ensemble of age groups, represented on the right-hand side by the frequency distribution of workers in different age groups that attend to task T1, is different from that attending to T2, and so on. As the number of contingencies is increased, and their response curves are all made discordant, there will be one age-group ensemble for each task. Let us now define an age-group ensemble as a temporal caste. In the extreme case represented in Model 1 there is a caste for each task. However, the distinction between age-group ensembles will be blurred as more tasks are added. The overlap in the age-group frequency curves is so extensive that after ten or so contingencies are added, the system becomes effectively continuous. For this reason we suggest that such an arrangement be called a continuous caste system.

The approach to continuity in Model 1 is marked by complexity and subtlety. The evolving ant species can easily adjust the programming to individual worker responsiveness to attain discordance that in turn yields one caste specialized for each task. Thus, only a relatively elementary alteration in physiology is needed to produce a complex caste system.

Next consider the alternative option, depicted in the lower half of Figure 8-20. Here various of the response curves are concordant, or at least approximately so. As a result the same statistical ensemble of workers attends to more than one task. As the number of tasks increases, the number of castes does not keep pace; conceivably it could remain low, say corresponding to as few as 2 or 3 distinct ensembles. Thus, the species has chosen to operate with a discrete caste system--comprising relatively few, easily recognized age-group ensembles. The evolutionary process leading to such a system can be called “behavioral discretization” (Wilson, 1976c). It can operate through physiological alterations as potentially simple as those that yield continuous caste systems.

Few studies of temporal polyethism have been designed in a way that permits a consideration of the hypothesis of discretization. An analysis of the ant Pheidole dentata by Wilson (1976c) showed that the system has been discretized. Virtually all of the 26 behavioral acts recorded in the minor worker caste can be divided according to three age periods in which they are performed approximately in concert (see Figures 8-20 and 8-21). Data presented by Higashi (1974) on the Japanese wood ant Formica yessensis indicate that temporal castes have also been discretized in this species, but the behavioral categories followed through time were too few to be certain.

The observed discretization in Pheidole dentata appears to represent an adaptation that increases spatial efficiency. It is obviously more efficient for a particular ant grooming a larva to regurgitate to it as well. Similarly, a worker standing “guard” at the nest entrance can be expected to be especially prone to excavate soil when the entrance is buried. The other juxtapositions in the Pheidole dentata repertory make equal sense when the spatial arrangement of the colony as a whole is considered. The queen, eggs, first instar larvae (microlarvae), and pupae are typically clustered together and apart from the older larvae, although the positions are being constantly shifted, and pupae are often segregated for varying periods of time well away from other immature stages. Thus, the A-ensemble of workers can nicely care for all of these groups, moving from egg to pupa to queen with a minimum of travel. The mean free path of a patrolling worker is minimized by such versatility so that the least amounts of time and energy are consumed. It makes equal sense for A workers to assist the eclosion of adults from pupae, since pupae are already under their care.

It will be of interest to learn to what extent and by which patterns the temporal polyethism of other species of social insects has been discretized. Calabi et al. (1983) found the phenomenon lacking in the Asian Pheidole hortensis; in other words, the caste system is continuous. While it is too early to put the matter on a sound empirical basis, we suspect that the recognizable age ensembles--that is, temporal castes--will prove to be related functionally to nest architecture. Species with complex nest structure, which provide a wider array of housekeeping tasks, and the opportunity for a more precise distribution of brood stages into chambers with differing microclimates, can be expected to have more temporal castes than those with relatively simple structures. For example, leafcutter ants of the genus Atta, which construct elaborate chambers for the gardening of symbiotic fungi, are expected to possess a correspondingly complex system of temporal polyethism. The patterns of temporal polyethism may be related to the dietary specialization of the species and the external environment of its nests. An ant species that forages widely outside the nest for small particles of food, while simultaneously defending its nest entrances from frequent attacks by predators, is likely to have an unusually discrete polyethic division at the end of worker life. The C age ensemble of Pheidole dentata represents just such a case. Finally, the more dispersed the tasks for which temporal castes are specialized, the more likely is the labor to be discretized.

To the extent that temporal castes are discretized, all of the castes of a species can be counted. Pheidole dentata, for example, has five adult castes: three temporal stages of the minor worker, a major worker (which has only one temporal stage), and the queen. The male does not occupy any known labor role. The larva might provide gluconeogenesis or some other metabolic service and hence constitute a sixth caste, but this possibility remains uninvestigated in Pheidole. There is every reason to suspect that the modest temporal caste system of Pheidole dentata is typical for the majority of ant species.

More generally, the total characterization of an insect society appears more feasible than it did only a few years ago. It is likely that the number of physical and temporal castes will not exceed 10 in ants and 20 in termites. The categories of behavior recorded in individual physical castes of ants have so far ranged between 20 and 42, with a broad overlap of categories among castes and a total species repertory probably not much exceeding 50 (Oster and Wilson, 1978). Moreover, the smaller the size of the worker characterizing a species, the smaller the behavioral repertory of the species. The largest ants have a repertory size somewhat more than 1.5 times that of the smallest ants (Cole, 1985). The number of categories of signals used in communication, mostly chemical, is likely to fall between 10 and 20 (Wilson, 1971).

The study of the relation of age and behavior in ants has proceeded slowly because it is a technically difficult task. A few authors, including Otto (1958) and Lenoir (1979c), have simply marked and followed individual ants through portions of their lives. This brute-force procedure, while unassailably accurate, is also very tedious. Other investigators have sought short-cuts that allow the rapid estimation of the ages of masses of workers engaged in particular tasks. In other words, instead of piecing together many individual trajectories and from those data inferring the age bias among workers generally with reference to particular tasks, the reverse is accomplished with approximately the same results: the age bias among the bulk of the workers is ascertained, and from this information “typical” individual repertories are inferred. An age marker used most commonly is the darkening color of workers as they age. In Pheidole dentata, for example, newly eclosed adult minor workers are uniformly clear yellow; after 2 days the gaster darkens to a contrasting yellowish brown; when the ant reaches about a week in age the head also darkens to contrast with the middle of the body; and so on until after about 16 days the body is nearly uniformly dark brown (Wilson, 1976c). A closely similar series has been effectively employed for Myrmica by Weir (1958a,b) and by Cammaerts (1974b). In the Neotropical cryptobiotic ant Basiceros manni we recognized four stages in coloration and cumulative body deposits that were correlated with age and hence useful in the study of temporal division of labor (Wilson and Hölldobler, 1986).

Another kind of marker altogether, employed by Higashi (1974) in his studies of Formica yessensis, is the degree of mandible wear. The three apical teeth of the masticatory border are needle-sharp in newly eclosed workers and gradually wear down through the course of adult life.

A third widely used marker is the condition of the ovaries, which typically reach maximal development early in the adult stage and gradually atrophy as the ants age and undertake work away from the brood chambers (Figure 8-22). In Formica sanguinea, for example, yolk deposition commences about the tenth day of adult life and reaches a maximum between the 26th and 35th days (Billen, 1982a). Oogenesis by the young workers either produces trophic eggs that are fed to the queen and larvae, or viable eggs that almost always develop into males. The progression of ovarian development has been used in other studies of the division of labor in Formica by Otto (1958), Kneitz (1964b, 1969), Hohorst (1972), and Möglich and Hölldobler (1974), and in Oecophylla by Hölldobler (1983). These authors have all noted the close correspondence between the position of the ants in the nest and the state of their ovaries.

Each of these methods of age estimation is subject to considerable error due to individual differences among the workers and fluctuations in the colony environment. But if large samples of workers are taken, at least the main course of age polyethism can be charted. Its reliability increases when the colonies are maintained under controlled laboratory conditions and the marker calibrated by following marked workers through the early part of their lives.

The technical advance most needed in studies of temporal division of labor is a physiological or biochemical “clock,” one that changes steadily as the ant grows older and is subject to minimal error due to environmental covariance. One candidate is the density of particles in the cytoplasm of oenocytes (Buschinger, 1967; Ehrhardt, 1970). In Pheidole dentata it increases at a seemingly steady rate (Wilson, unpublished). Another is the accretion of daily rings in the endocuticle, a process that works well in the honeybee for up to 11 days and less reliably thereafter (Menzel et al., 1969). Endocuticular rings occur in the harvester ant Pogonomyrmex badius but have not been calibrated (Wilson, unpublished).

Case histories
Temporal polyethism, in other words division of labor by age, is far more prevalent in ants than caste polyethism (the more easily recognizable division of labor correlated with anatomical differences). Only a single species, Amblyopone pallipes, is known that lacks temporal polyethism altogether. All of the remaining studies to date display some variation on “typical” age polyethism, in which young workers devote themselves to brood care and other forms of labor inside the nest, while older workers tend to travel outside the nest for nest construction, defense, and foraging. Ants, as the German investigators have succinctly put it, pass from Innendienst (inside work) to Aussendienst (outside work). In contrast, only a minority of genera of ants contain one or more polymorphic species, in other words species in which the workers are divided into two or more physical subcastes. Of the 297 living genera recognized during the preparation of this book (Table 2-2), only 46 or 15 percent contain at least some polymorphic species. These taxa are listed in Table 8-2, along with the small number of genera whose caste systems are unknown. One of the remarkable features of the phylogenetic distribution revealed by this tabulation is the scarcity of physical worker subcastes within the large and relatively primitive subfamily Ponerinae. Only a single genus (Megaponera), out of 58 known living ponerine genera, or 2 percent of the total, is known to be polymorphic. According to Breed and Harrison (1988) another case might be the giant Neotropical ponerine Paraponera clavata, where worker variation appears to be mildly allometric and the size-frequency distribution of mature colonies is unimodal and skewed to the right. While the sizes of workers performing different tasks overlap widely, there is a statistical association between size and task performance. Large workers function more often as guards and foragers. One other ponerine genus, Aenictogiton, is of unknown status with respect to caste.

Most of the studies of caste and division of labor conducted to the present time are listed in Table 8-3. It is informative to compare this compilation with Table 2-2, which contains all of the living genera, and Table 8-2, which lists the genera known to display polymorphism; and then to note that almost all ant species probably have some form of division of labor, even if just based on age. It will then be obvious that while a large amount of information is now at our disposal, it applies to only a small fraction of genera and species. The exploration of caste and division of labor in ants is in an early stage, and many surprises surely lie ahead.

The best way to encompass existing knowledge of division of labor is to compare several grades of complexity across the entire range of the living ants. A well-studied example of weak polymorphism correlated with polyethism has been provided by Brandão (1978) for Formica perpilosa. This American montane species displays weak polymorphism of a kind widespread in ants. There is a modest size variation within the worker caste of individual colonies, accompanied by a slight but distinct allometry in which the bodies grow more robust and the heads broader relative to length with an overall increase in size. Medium-sized workers have the broadest repertories, covering all of the quotidian tasks of brood and queen care, nest work, and foraging. The smallest workers are very similar in overall behavior while performing slightly fewer acts. The largest workers, in contrast, constitute a true “major” caste specialized for liquid food storage. They engage in few other social acts. Overall, Formica perpilosa closely parallels caste and division of labor in the species of Proformica, a formicine genus found in dry habitats in Europe and Asia that also possesses weak polymorphism and a major subcaste specialized for food storage (Figure 8-23). The similarity is almost certainly due to convergent evolution.

In almost all species for which differences in size have been found to be correlated with division of labor, the size variation is also distinctly allometric. In addition to this first rule of caste evolution we encounter a second rule: monomorphic ant species rarely divide labor according to size. More precisely, the greater the combined amounts of size variation and allometry, the more pronounced the division of labor. Considerable size variation often occurs, but unless it is accompanied by allometry strong enough to be clear-cut by casual inspection, it is not correlated with marked role differences (Wilson, 1971; Herbers et al., 1985). No deviation is known from the first rule relating polymorphism to division of labor. However, one exception has been discovered to the second rule. Leptothorax longispinosus is a typically monomorphic species, yet the large workers forage more than the smaller ones, a foreshadowing of the division of labor characterizing the most simply allometric ant species (Herbers and Cunningham, 1983). In contrast, the monomorphic Leptothorax ambiguus, when subjected to the same form of analysis, evinced no sign of size bias in behavior. Herbers (1983) suggested that the difference between the two species might be due to the greater size variation that occurs within colonies of Leptothorax longispinosus. Another exception is the formicine honey ant Myrmecocystus mimicus, where Hölldobler found that larger workers are significantly more engaged in tournament behavior than smaller workers.

Successive grades in the early evolution of polymorphism and division of labor is nicely demonstrated among species of fire ants. The more primitive pattern of Solenopsis invicta is typical of almost all of the fire ant species, which comprise the subgenus Solenopsis of Solenopsis (as opposed to the thief ants, which comprise the subgenus Diplorhoptrum). The allometry of Solenopsis invicta is very slight (see Figures 8-15 and 8-24). The majors have heads that are only somewhat shorter, broader, and more quadrate with reference to the remainder of the body. The behavioral differences are also modest, since the majors perform the same tasks as the minors with the exception of caring for the eggs and young larvae. Workers in these larger size classes simply handle pieces of brood, prey, and soil particles that are correspondingly larger in size. Hence transport and nest excavation appear to be enhanced by the polymorphism, although that subjective impression has not yet been put to a quantitative test. The more important point is that Solenopsis geminata has added something entirely new. Allometry is steep, so that the larger medias and majors are grotesque creatures possessing massive heads and blunt, toothless mandibles. These two features together contribute to a substantial increase in the crushing power of the ants. Wilson (1978) was able to show that the largest workers were in fact specialized for milling seeds, the majors so much so that they have become restricted to the smallest behavioral repertory thus far recorded in the social insects--milling and self-grooming. Solenopsis geminata has undergone this striking revision in its caste system and division of labor as part of its trend toward granivory, which is the most extensive known among the fire ant species.

Fire ants illustrate a further principle in the division of labor, which can be summarized under the expression alloethism: the regular and disproportionate change in behavior in a particular category of behavior as a function of size of the workers (Wilson, 1978). Alloethism parallels allometry, or the regular and disproportionate change in one anatomical dimension relative to other dimensions as a function of worker size. In both species in which sufficiently detailed measurements have been taken, the fire ant Solenopsis geminata and the leafcutter Atta sexdens, the curves were found to possess steeper slopes than the corresponding allometric curves of the anatomical structures employed in the behavior (Wilson, 1978, 1980a). Put succinctly, alloethism amplifies allometry. Part of the fire ant pattern is illustrated in Figure 8-25.

Passing to the opposite extreme, let us next consider one of the most advanced grades of caste and division of labor discovered so far in ants, that of Pheidologeton diversus of southeast Asia. Colonies of this “marauder ant” contain hundreds of thousands of workers and conduct wide-ranging raids to collect insects and a variety of vegetable materials. The workers display only a moderate amount of allometry, which is largely confined to the proportions of the head. But their size variation is the greatest ever recorded within single ant colonies: a full tenfold range in head width and a 500-fold range in dry weight (see Figure 8-26). The size-frequency distribution of the workers is trimodal. Superimposed on this pattern are the following four size-related roles.

Minors (head widths less than 0.8 mm) conduct most of the tasks of the colony, including brood care, much of the foraging, construction of the nest and trail arcades, and defense.

Medias (head widths 1.0-2.6 mm) assist the minors in foraging and construction.

Majors (head widths greater than 2.6 mm) hoist obstructions and smooth the surface of the trails.

Repletes are medias and majors that remain in the nest and acquire food until their abdomens are swollen; they share this stored food with other members of the colony.

The medias and majors act as the colony bulldozers. They move obstructions on the trails too large for their small nestmates to handle, pack the soil with their massive heads, and chew through the tougher fibers of fruit and the more massive appendages of captured prey. In addition to the physical castes, Moffett has distinguished two temporal castes in the small workers that conform to the usual pattern in ants. That is, the youngest workers associate more with the brood, while older workers presumably expand their repertory to fill the entire range of tasks. In addition, exceptionally dark-colored, hence old or wounded workers are disproportionately represented among the workers guarding the trails. The caste system of Pheidologeton diversus, when combined with its large colony size and ability to field a large army of workers, has enabled it to utilize an extraordinary wide variety of foods.

Precisely the opposite dietary adaptation has been achieved by the equally complex system of the leafcutter ants in the genus Atta, illustrated in Figure 8-27. Like Pheidologeton, Atta generates a broad array of physical types by combining extreme size variation with moderate allometry. In Atta sexdens, the head width varies eightfold and the dry weight 200-fold from the smallest minors to the huge majors; the polymorphism thus runs a close second to Pheidologeton diversus. However, the Atta caste diversity is not directed at broadening the diet of the ants, as in Pheidologeton, but at narrowing it severely. Colonies of Atta subsist entirely on the sap of plants and a symbiotic fungus they grow on fragments of vegetation. Atta and its sister genus Acromyrmex are unique in the Animal Kingdom in their ability to utilize freshly cut leaves and other vegetation for the rearing of fungus. Their unusual caste system has made the adaptation possible.

The fungus-growing ants of the ant tribe Attini, to which Atta belongs, are of unusual interest in biology because (to cite a familiar metaphor) they alone among the ants have achieved the transition from a hunter-gatherer to an agricultural existence. But this major shift did not require an elaborate caste system. The great majority of attine genera and species are monomorphic, including the presumably most primitive forms belonging to Cyphomyrmex. The complex caste system and division of labor of Atta represent a much narrower, more idiosyncratic adaptation to the collecting of fresh vegetation as a novel form of fungal substrate. Most of the monomorphic attines such as Cyphomyrmex utilize decaying vegetation, insect remains, or insect excrement, all of which are materials ready-made for fungal growth. Fresh leaves and petals, in contrast, require a whole series of special operations before they can be converted into substrate. They must first be cut down, then chopped into fine pieces, next chewed and treated with enzymes, and finally incorporated into the garden comb. Beyond the harvesting process, the fungus must also be provided constant care after it sprouts on the substratum.

The Atta workers organize the gardening operation in the form of an assembly line, which is illustrated in Figures 8-28 and 8-29 (see also Plates 9 to 11). Tough vegetation can be cut only by workers with head widths of 1.6 mm or greater. The most frequent size group among foragers consists of workers with head widths 2.0 to 2.2 mm. At the opposite end of the line, the care of the delicate fungal hyphae requires very small workers, and this task is filled within the nests by workers with head widths predominantly 0.8 mm. The intervening steps in gardening are conducted by workers of graded intermediate size. After the foraging medias drop the pieces of vegetation onto the floor of a nest chamber, they are picked up by workers of a slightly smaller size, who clip them into fragments about one to two millimeters across. Within minutes, still smaller ants take over, crush and mold the fragments into moist pellets and carefully insert them into a mass of similar material. The resulting “comb” ranges in size from a clenched fist to a human head and is riddled with channels. Resembling a gray cleaning sponge, it is light and fluffy and crumbles under slight pressure. On its surface the fungus spreads out like a frost, sinking its hyphae into the leaf paste to digest the abundant cellulose and proteins held there in partial solution.

The remainder of the gardening cycle proceeds. Worker ants even smaller than those just described pluck loose strands of the fungus from places of dense growth and plant them onto the newly constructed surfaces. Finally, the very smallest--and most abundant--workers patrol the beds of fungal strands, delicately probing them with their antennae, licking their surfaces clean and plucking out the spores and hyphae of alien species of mold. These colony dwarfs are able to travel through the narrowest channels deep within the garden masses. From time to time they pull loose tufts of fungal strands resembling miniature stalked cabbage heads and carry them out to feed their larger nestmates.

Although the assembly line of fungal cultivation is the core of caste and division of labor in the leafcutter colony, it is far from the entire story. The defense of the colony is also organized to some extent according to size. All of the size groups attack intruders, but in addition there is a true soldier caste. Among the scurrying ants can be seen a few extremely large majors. Their sharp mandibles are powered by massive adductor muscles that fill the swollen, five-millimeter-wide head capsules. Working like miniature wire clippers, these specialists chop enemy insects into pieces and easily slice through the skin of vertebrate intruders. The giants are especially adept at repelling large enemies. When entomologists digging into a nest grow careless, their hands become nicked all over as if pulled through a thorn bush.

When brood care, nest construction, and other tasks are added, the ants perform a total of 22 such social functions. The workers of Atta sexdens fall into four size groups defined by the clusters of these tasks on which they specialize, together with size. These four physical castes can be characterized respectively as gardeners-nurses, within-nest generalists, foragers-excavators, and defenders. As shown in Figure 8-30, this analysis starts with the drawing of polyethism curves, or size-frequency curves of the workers engaged in each of the 22 tasks in turn. The polyethism curves are then clustered to define the roles. Three of the four physical castes recognized in this fashion have also been shown to pass through changes of behavior with aging, allowing the discrimination of two temporal castes in each. The total number of castes recognizable in Atta sexdens, physical plus temporal, is therefore seven (Wilson, 1980a).

The absolute commitment of the Atta worker caste to the social life of the species is reflected by its intricate polyethism. It is also underscored by the distinctive patterns of size-related variation in the exocrine glands and body spines. In every instance where the function of an organ is known, the organ proves to be maximally developed in the size classes that specialize on the tasks served by the organ (see Figure 8-31). Hence the pronotal spines, taken as representative of armament over the entire body, are proportionately longest in the size classes (with head widths 1.6-2.6 mm) that spend the most time foraging outside the nest and hence are exposed most frequently to danger from predators and competitor ants. The same size classes do most of the trail-laying, and sure enough, they have the proportionately largest poison glands, the source of the trail pheromone. The postpharyngeal gland is the source of larval food in species of ants that have been studied in this respect, and it turns out that in Atta sexdens the postpharyngeal gland is largest in the smallest size class, with head widths 0.6 to 1.2 mm; these workers are the ones that feed larvae by regurgitation. The mandibular gland is known to produce the alarm substances citral and 5-methyl-3-heptanone in species of Atta, including, at least in the case of the methylheptanone, Atta sexdens (Blum et al., 1968a). As expected, this paired organ is proportionately most massive in the largest size classes of Atta sexdens, which are specialized for colony defense. The consistency of correlation suggests that proportionate size alone might be used in the future as a first clue concerning the roles of previously unstudied or experimentally less tractable organs. For example, it can be predicted that secretions of the labial and hypopharyngeal glands in Atta sexdens function in brood care or fungus gardening, while those of the Dufour's gland are used either in defense or alarm communication. The technique can also be used to infer the roles of sensory organs. Jaisson (1972c), for example, reports that the sensilla ampulacea and sensilla coeloconica of the antennae vary according to worker size in Atta laevigata. This important aspect of sensory physiology awaits closer study.

Although caste and division of labor in Atta are very complex in comparison with other ant systems, they are still derived from surprisingly elementary processes of increased size variation, allometry, and alloethism. This important point is made graphic by Figure 8-32, in which the properties of the true Atta sexdens system are compared with those of an imaginary, more complex arrangement that could be generated by only slightly more elaborate rules. Ant species in general and Atta sexdens in particular have thus been remarkably restrained in the elaboration of their castes. They have relied on a single rule of deformation to create physical castes, which translates into a single allometric curve for any pair of specified dimensions such as head width versus pronotal width or, as illustrated in Figure 8-31, exocrine gland size versus pronotal width. The Atta could have created a more complex and precise array of castes by programming an early divergence of developing larval lines along with differing allometry among those lines at pupation, producing the effect illustrated in the upper diagram of Figure 8-32. But neither the Atta nor to our knowledge any of the many kinds of polymorphic ants have ever done so.

Behavior follows similarly elementary rules. The polyethism curves presented in Figure 8-30 are all of relatively simple form. They are unimodal and show only limited amounts of skewing. These properties suggest the existence of underlying alloethism functions of fundamental simplicity: for each behavior, a monotonic rise of responsiveness to a peak along the size range is followed by a monotonic decline.

To summarize, even though Atta sexdens possesses one of the most complicated caste systems found in the ants, it has not evolved anywhere close to the conceivable (and we believe evolutionarily attainable) limit. There are far more tasks than castes: by the first crude estimate seven castes cover a total of between 20 and 30 tasks. Furthermore, one can discern another important phenomenon in Atta sexdens that constrains the elaboration of castes: polyethism has evolved further than polymorphism. As we showed in the case of the fire ant Solenopsis geminata, the alloethic curves rise and descend more steeply than the size-frequency distributions and they are generally steeper than the allometric curves drawn from any selected pair of physical measurements. Consequently, ensembles specialized on particular tasks are more differentiated by behavior than by age or anatomy. In the course of evolution, Atta created its division of labor primarily by greatly expanding the size variation of the workers, while adding a moderate amount of allometry and a relatively much greater amount of alloethism.

Specialization by majors: defense
It is remarkable that the anatomically more deviant major workers of various ant species have arisen in evolution solely as specialists for one or the other of three primary tasks: defense, milling (chewing and comminution) of seeds, and food storage (by the extreme distension of the crop to create repletes). In cases of advanced polymorphism, especially complete dimorphism where intermediates have dropped out and the majors and minor workers have begun to evolve allometric patterns of their own, the greatest modifications in the major caste are found in the head and mandibles. The majors often look like members of an entirely different species.

The dominant role of majors, first noted by Westwood (1838) and since repeatedly confirmed, is defense of the nest and food. The majors, or “soldiers” as they are properly designated, are adapted to one or the other four basic fighting techniques: shearing, piercing, blocking and bouncing. Each will now be briefly characterized.

(1) Shearing. The mandibles are large but otherwise typical, the head is massive and cordate, and the soldiers are adept at cutting the integument and clipping off the appendages of enemy arthropods. Examples are found in Pheidole (Figure 8-3), Pheidologeton (Figure 8-26), Atta (Figure 8-28), Oligomyrmex, Aneuretus, Zatapinoma, Camponotus (Figure 8-18), and a few other genera of diverse relationships. Wheeler, in his essay “The physiognomy of insects” (1927), pointed out that the peculiar head shape of this form of soldier is simply due to an enlargement of the adductor muscles of the mandibles, which imparts greater cutting or crushing power to the mandibles.

(2) Piercing. The mandibles are pointed and sickle-shaped or hook-shaped. Soldiers of the most highly evolved army ants (Dorylus and Eciton) often line up along the flanks of the moving columns, the heads facing outward and mandibles gaping. An identical guard posture is assumed around nest entrances by the saber-jawed soldiers of the formicine Cataglyphis bombycina in the Sahara Desert (Délye, 1957). These formidable looking individuals rush at any moving object when the nest is disturbed. They seldom engage in other tasks. Contrary to an earlier suggestion by Felix Santschi, Délye found that the Cataglyphis soldiers are not well suited for carting particles during the excavation of nests.

Closely similar blocking maneuvers occur in the Neotropical Zacryptocerus varians (Wilson, 1976a; see Figure 8-35). But this is only part of the total strategy of defense. In fact, both minor and major workers are very active. The minor workers respond at a lower threshold. Thus they form the “early warning system” of the colony and are able to dispose of less formidable intruders without help. Soldiers respond less readily, but once activated they are individually more effective.

The specialization of the Zacryptocerus soldiers is augmented by a unique encrusting layer of filamentous material that covers the surface of their heads. This odd material, which resembles a mass of fungal mycelia, gives a grimy appearance to the part exposed to the outside during entrance guarding. W. M. Wheeler (1942) suggested that it serves as camouflage for the nest entrance. He commented that the cephalic disc of older soldiers and the queen were often coated with “dirt and extraneous particles so that it closely resembled the bark of the plant.” Clean, shiny head surfaces might be more easily spotted by visual predators, such as birds and lizards, which could break open twigs for the rich reward of ant brood. The resemblance of head to bark would effectively conceal the location of the nest entrance. Recent research by D. Wheeler and Hölldobler (1985) has disclosed that the filaments are secreted through a large number of secretory pores only 1-3 m in diameter and scattered over the disc surface.

A case of abdominal phragmosis has been reported in the ponerine Proceratium melinum by Poldi (1963). The second gastric segment is strongly convex, so much so that the succeeding gastric segments point obliquely forward. The posterior surface of the second segment is therefore the posteriormost part of the body, and it is used to block the nest galleries against intruders. The trait is possessed by all of the workers, rather than by a specialized major caste.

(3) Blocking. The behavior of soldiers that block nest entrances is the most specialized of all. The members of Colobopsis can be taken as typical of this category. Colobopsis truncatus of Europe has been studied in detail by Forel (1874) and later authors. Wheeler (1910a) described the American Colobopsis etiolatus under natural conditions, and Wilson (1974c) made observations on colonies of the American Colobopsis fraxinicola housed in glass tubes. The soldiers seldom leave the nests, which consist of narrow cavities in the dead wood of standing trees and shrubs. One or more of them stand guard at the nest entrances, where they serve literally as living doors. When minor workers approach them from either end and give the right signal (presumably a combination of simple touch and colony-recognition scent, although the matter has never been experimentally investigated) the soldiers pull back into the nest to allow their nestmates free passage. The nest entrances are cut into wood or plastered with carton so as just to accommodate the head of a soldier. It is a remarkable fact that it is the minors, not the soldiers, who engineer this fit. In those instances when the entrance is larger, several soldiers join to plug it with the combined mass of their heads. Both arrangements are shown in the accompanying illustration of Colobopsis truncatus by Szabó-Patay (Figure 8-33).

The specialized role of the Colobopsis soldiers was demonstrated still more convincingly in experiments conducted by Wilson (1974c). Undecane, stored in the Dufour's gland of the abdomen, is a general formicine alarm pheromone. When small quantities of this substance were allowed to evaporate near the nest entrance, all members of the Colobopsis fraxinicola colony were thrown into the typical excited running movements of the fraxinicola alarm response. But some of the soldiers moved to the nest entrances, filling even holes that had been unattended prior to the alarm reaction. On the other hand, the soldiers were not especially adept in combat. When twigs containing fraxinicola colonies were first broken open, both minor workers and soldiers rushed out. Many attack any accessible alien objects, such as the observer's hand or a bit of cloth offered to them, biting it and spraying it with formic acid. The same response is obtained in the laboratory by permitting fire ant workers (Solenopsis invicta) to invade the nests. Individuals of both castes are about equally aggressive and effective in repelling these invaders. The total population of minor workers, by virtue of its greater size, was more effective than that of the soldiers. In the Australia species Camponotus ephippium the soldiers appear to function solely in nest defense. They have never been observed foraging. They are fed by minors (Figure 8-17) but rarely if ever regurgitate food to nestmates. The nest entrance in the soil, which is slightly larger than the head of a soldier, is always blocked by the head of one member of this caste. When a stick is touched to the nest entrance the guarding soldier attacks it with its powerful mandibles. When the stick is then removed the soldier usually does not release its grip and is pulled out. Immediately another soldier takes its place at the entrance (Hölldobler, unpublished; Figure 8-34). Camponotus ephippium often nests within the territories of the dominant species Iridomyrmex purpureus. One of the major functions of the soldiers of Camponotus ephippium appears to be to protect the colony against raids by these aggressive “meat ants.”

A different form of blocking behavior is exhibited by the North American cephalotine Zacryptocerus texanus. The entrance hole to the arboreal nest is somewhat larger than the head of the soldier and is blocked by the combined mass of the head and expanded prothorax, the latter structure being heavily armored and pitted like the head. The head is held obliquely, rather like the animated blade of a miniature bulldozer. This posture, combined with the thrust and pull of the short, powerful legs, allows the soldier to press the intruders right out of the nest--a kind of nonviolent defense. When a minor worker returns to the entrance hole, the following sequence unfolds:

The returning minor may or may not touch the antennae of the guard, although it usually does so. Thereafter the guard crouches down. This brings the anterior rim of the head below the level of the floor of the passage or, if the guard stands completely inside the passage, the front of the head is raised as the guard crouches. The dorsum of the guard's thorax is now no longer close to the roof of the passage and the minor can, if it is sufficiently active, wriggle between the dorsum of the thorax and the roof of the passage. . . . If the passageways are made large enough to accommodate two majors simultaneously, they ordinarily assume a position where they are back to back. Under such circumstances the two opposed cephalic discs form a V-shaped area. The bottom of this V is open but the space behind it is closed by the closely approximated thoracic dorsi of the two guards. When minors are admitted to the nest both majors crouch and the entering worker struggles through the narrow space between the thoraces of the guards. It seems scarcely necessary to state that there is no part of this behavior which at all resembles that of the Colobopsis major, which must back away from the nest entrance to admit the returning minor (Creighton and Gregg, 1954).

(4) Bouncing. A wholly different mode of defense occurs in the Australian dacetine ant Orectognathus versicolor (Carlin, 1981). As depicted in Figure 8-36, the mandibles of the largest individuals, in other words the soldiers, are broadened and flattened from bottom to top. The apical teeth, which the minor and media workers use to capture prey, are shortened and blunted. When an alien ant enters the Orectognathus nest, the soldiers spread their peculiar mandibles about 120 degrees apart. At the moment the intruder's body comes within range of a soldier's gape, she snaps her mandibles shut, pinching the intruder with such force as to shoot it away--like a slippery seed pressed hard between the fingers. Only the largest workers, with their broad, nearly toothless mandibles and powerful adductor muscles, are equipped to perform the bouncing maneuver. They are very effective in this specialized role, successfully propelling enemy ants up to 10 cm through the air. The workers of Odontomachus species use a similar bouncing technique to repel invaders from their nest entrances.

Specialization by majors: milling and food storage
Turning to the next principal specialization, the addition of a major caste for milling has arisen in evolution only when seeds constitute a substantial (but less than exclusive) supplement to the diet. The phenomenon occurs in Solenopsis geminata (Figure 8-25), Pogonomyrmex badius, the species of Acanthomyrmex (Figure 8-37) and granivorous members of the large, cosmopolitan genus Pheidole (Wilson, 1978, 1984b; Moffett, 1985b). The majors employ their massive, blunt-edged mandibles to strip away the coat and break apart the endosperm of the seed. They are demonstrably more efficient at this task than the much smaller and weaker minor workers. Paradoxically, where seeds are the principal or sole dietary items, the species tend to be monomorphic or weakly polymorphic. In other words, all of the workers are adapted to milling, yet remain generalists in other tasks and hence are more “normal” in appearance. Examples of the second category are believed to include species of Monomorium (“subgenus Holcomyrmex”) and the north American members of the genus Messor, in other words the species formerly placed in the separate genus Veromessor.

The storage of liquid food in the crop has been carried to great heights by the “repletes” of certain ant species, individuals whose abdomens are so distended they have difficulty moving and are forced to remain permanently in the nests as “living honey casks” (Plate 8). The extreme examples are ground-dwelling species that live in arid habitats: species of Myrmecocystus, a genus confined to the western United States and Mexico; Camponotus inflatus, Melophorus bagoti, and M. cowlei of the deserts of Australia; some species of Leptomyrmex in Australia, New Guinea, and New Caledonia; and Tetramorium trimeni of Natal (McCook, 1882; Wheeler, 1910; Creighton, 1950). Leptomyrmex is a dolichoderine, and the remainder belong to the Formicinae. Australian aborigines dig up and eat repletes of Camponotus and Melophorus as a kind of candy. Intermediate stages of repletion are seen in such diverse genera as Erebomyrma, Pheidologeton, Prenolepis, Proformica, and Oligomyrmex. Repletes are usually drawn from the ranks of the largest workers, who apparently enter their servile role as callows while their abdomens are still soft and elastic (McCook, 1882; Wheeler, 1908a, 1910a; Rissing, 1984). No fewer than 1,500 such individuals were recovered from the nest of a colony of Myrmecocystus melliger by Creighton and Crandall (1954). The complete population of a mature Myrmecocystus colony typically comprises a single queen and about 15,000 workers (Hölldobler, 1984d). Crandall, with the aid of professional gravediggers, followed the nest galleries through 16 ft 3 in of Arizona desert soil until he recovered the nest queen from a small chamber at the very bottom. Thus the earlier conjecture by Wheeler (1908a) that the Myrmecocystus nests are shallow and the population of replete workers is small has been thoroughly refuted. Rissing (1984) found that the repletes belong to the largest workers in colonies of Myrmecocystus mexicanus. This appears to be generally true for Myrmecocystus, because we have confirmed it in numerous laboratory colonies of several other Myrmecocystus species, including Myrmecocystus depilis, Myrmecocystus mimicus, Myrmecocystus navajo, and Myrmecocystus placodops. The repletes of Myrmecocystus often vary in color from clear yellowish-brown to dark amber. Conway (1977) analyzed the crop contents of the repletes of Myrmecocystus mexicanus and found that the dark fluid “contained more dissolved solids and that they were mainly glucose and fructose. In the more dilute clear sample, sucrose made up the bulk of the solids and there were only traces of glucose and fructose. Clear repletes may function primarily as water-storage vesicles, an adaptation well-suited to their semi-arid habitats.”

The pseudomutant technique has the great advantage of permitting the same colony to be used over and over again, with the experimenter modifying it in a precise manner as though it were a mutant in a certain trait, but with little or no alteration of the remainder of its traits. In other words, there is no genetic noise or pleiotropy to confuse the analysis. The procedure used in the Atta experiments is roughly comparable to an imaginary study of the efficiency of various conceivable forms of the human hand in the employment of a tool. In the morning we painlessly pull off a finger and measure performance of the four-fingered hand, then restore the missing finger at the end of the day; the next morning we cut off the terminal digits (again painlessly) and measure the performance of a stubby-fingered hand, restoring the entire fingers after the experiment; and so on through a wide range of variations and combinations of hand form. Finally, we are able to decide whether the natural hand form is near the optimum.

The data from the pseudomutant scanning revealed that the size-frequency distribution of the leafcutter caste in Atta sexdens conforms closely to the optimum predicted by the energetic efficiency criterion for harder varieties of vegetation, such as thick, coriaceous leaves. The distribution is optimum with reference to both construction and maintenance costs (see Figure 8-38). It does not conform to other criteria conceived prior to the start of the experiments.

Wilson next constructed a model in which the attraction of the ants to vegetation and their initiative in cutting were allowed to “evolve” genetically to uniform maximum levels. The theoretical maximum efficiency levels obtained by this means were found to reside in the head-width 2.6-2.8 mm size class, or 8 percent from the actual maximally efficient class. In the activity of leaf cutting, Atta sexdens can therefore be said not only to be at an adaptive optimum but also, within at most a relatively narrow margin of error, to have been optimized in the course of evolution. In other words, there appears to be no nearby adaptive peak that is both higher and attainable by gradual macroevolution.

Not all such studies reveal optimal responses or even adaptive mechanisms of the sort intuitively expected. For example, what would be the physiological response of an Atta colony, as opposed to the purely behavioral reaction described earlier, if a large part of its foraging specialists were suddenly removed by some catastrophe outside the nest? One prudent response on the part of a colony would be to manufacture a higher proportion of workers in the 1.8-2.2 mm size class during the next brood cycle to make up the deficit. However, when over 90 percent of this group in experimental colonies were removed, no differential increase in the production of 1.8-2.2 mm workers could be detected in comparison with sham-treated control colonies. As a consequence, this group remained under-represented in the foraging arenas by about 50 percent at the end of the first brood cycle, a period of eight weeks (Wilson, 1983a).

Another striking example of a feedback loop that did not evolve is stress-induced change in the production of major workers in the ant Pheidole dentata. What happens when a colony is attacked repeatedly over a long period by fire ants, the enemy to which its alarm-defense system is specially tuned? An obvious adaptation would be to raise the production of majors (increase defense expenditures, to put it in more familiar human terms) until the stress is relieved. But this response could not be induced by Johnston and Wilson (1985) in laboratory colonies. They stressed Pheidole dentata colonies heavily through four brood cycles by regularly forcing them to fight and destroy Solenopsis invicta workers. Contrary to their expectation, the proportion of majors did not change significantly from that in colonies stressed with another species of ant (Tetramorium caespitum) or from the proportion that prevailed prior to the experiments, when the colonies were free of any pressures through many brood cycles.

The optimization theory pursued in the case of caste ratios assumes selection at the colony level. In fact, colony selection in the advanced social insects does appear to be the one example of group selection that can be accepted unequivocally so long as we are careful to bear in mind that the group in this case is the colony and not the population of colonies. To be sure, it is the queen, the mother of all other workers and second generation reproductives in the colony that transmits the gametes and is the ultimate focus of selection. In this special sense colony selection differs from group selection in the hypothetical Wynne-Edwards sense, where most or all of the mature individuals of the populations are involved in reproduction (Wynne-Edwards, 1962, 1986; G. C. Williams, 1966). But it remains true that the colony is selected as a whole, and its members contribute to colony fitness rather than to individual fitness. It therefore seems a sound procedure to accept colony selection as a mechanism and to press on in search of optimization theory based on the assumption that the mechanism operates generally. For, if selection is mostly at the colony level, workers can be altruistic with respect to the remainder of the colony, and their numbers and behavior can be regulated in evolution to achieve maximum colony fitness. What is required is a theory of group behavior, a way of abstracting our empirical knowledge of caste and colony ergonomics into a form that can be used to analyze optimality. The term ergonomics was borrowed from human sociology (e.g., Murrell, 1965) to identify the quantitative study of the distribution of work, performance, and efficiency in insect societies (Wilson, 1968). Wilson attempted a first formulation by means of the techniques of linear programming and obtained some surprising but still largely theoretical results. The essential arguments and results are presented here in a simplified form.

First, consider the concept of cost in colony reproduction. As colonies grow, their caste ratios change. Very young colonies founded by single queens typically consist only of the queens and minor workers. As they approach maturity, these same colonies may add medias and soldiers. Finally, they produce males and new, virgin queens. Here we will consider ergonomics and cost in the mature colony only. A mature colony is defined as a colony large enough to produce new, virgin queens. Also, for convenience, the category “caste” will include both physical castes, such as minor workers and soldiers, and temporal castes, the various periods of labor specialization that most ants pass through in the course of their lives. What determines the efficiency of the mature colony is the number of workers in each temporal caste at any given moment. This conception is spelled out in the example given in Figure 8-39.

In the mature colony, depending on the species, the adult force may contain anywhere from a few tens of workers to several millions. The number is a species characteristic. It has been evolved as an adaptation to ultimate limiting factors in the environment. An ultimate limit may be imposed by the constraints of a peculiar kind of nest site to which the species is adapted, by the restricted productivity of some prey species on which the species specializes, or, conversely, by a prey species or competitor so physically formidable as to require a larger worker force as a minimum for survival. The mature colony, on reaching its predetermined size, can be expected to contain caste ratios that approximate the optimal mix. This mix is simply the ratio of castes that can achieve the maximum rate of production of virgin queens and males while the colony is at or near its maximum size.

It is helpful to think of a colony of social insects as operating somewhat like a factory constructed inside a fortress. Entrenched in the nest site and harassed by enemies and capricious changes in the physical environment, the colony must send foragers out to gather food while converting the secured food inside the nest into virgin queens and males as rapidly and as efficiently as possible. The rate of production of the sexual forms is an important, but not an exclusive, component of colony fitness. Suppose we are comparing two genotypes belonging to the same species. The relative fitness of the genotypes could be calculated if we had the following complete information: the survival rates of queens and males belonging to the two genotypes from the moment they leave the nest on the nuptial flights; their mating success; the survival rate of the fecundated queens; and the growth rates and survivorship of the colonies founded by the queens. Such complete data would, of course, be extremely difficult to obtain. In order to develop an initial theory of ergonomics, however, it is possible to get away with restricting the comparisons to the mature colonies. In order to do this and still retain precision, it would be necessary to take the difference in survivorship between the two genotypes outside the period of colony maturity and reduce it to a single weighting factor. But we can sacrifice precision without losing the potential for general qualitative results by taking the difference as zero. Now we are concerned only with the mature colony, and the production of sexual forms becomes (keeping in mind the artificiality of our convention) the exact measure of colony fitness. The role of colony-level selection in shaping population characteristics within the colony can now be clearly visualized. If, for example, colonies belonging to one genotype contain on the average 1,000 sterile workers and produce 10 new, virgin queens in their entire life span, and colonies belonging to the second genotype contain, on the average, only 100 workers but produce 20 new, virgin queens in their life span, the second genotype has twice the fitness of the first, despite its smaller colony size. As a result, selection would reduce colony size. The lower fitness of the first could be due to a lower survival rate of mature colonies, or to a smaller average production of several forms for each surviving mature colony, or to both. The important point is that the rate of production in this case is the measure of fitness, and evolution can be expected to shape mature colony size and organization to maximize this rate.

The production of sexual forms is determined in large part by the number of “mistakes” made by the mature colony as a whole in the course of its fortress-factory operations. A mistake is made when some potentially harmful contingency is not met--a predator successfully invades the nest interior, a breach in the nest wall is tolerated long enough to desiccate a brood chamber, a hungry larva is left unattended, and so forth. The cost of the mistakes for a given category of contingencies is the product of the number of times a mistake is made times the reduction in queen production per mistake. With this formal definition, it is possible to derive in a straightforward way a set of basic theorems on caste. In the special model, the average output of queens is viewed as the difference between the ideal number made possible by the productivity of the foraging area of the colony and the number lost by failure to meet some of the contingencies. (The model can be modified to incorporate other components of fitness without altering the results.) The evolutionary problem postulated to have been faced by social insects can be solved as follows:  the colony produces the mixture of castes that maximizes the output of queens. In order to describe the solution in terms of simple linear programming, it is necessary to restate the solution in terms of the dual of the first statement: the colony evolves the mixture of castes that allows it to produce a given number of queens with a minimum quantity of workers. In other words, the “objective” is to minimize the energy cost. Meeting the objective confers higher genetic fitness to the colony members, both collectively and individually.

This form of ergonomic theory leads to at least two results that appear at first to be counterintuitive but in fact are straightforward consequences of selection at the colony level. One, illustrated in Figure 8-40, involves the relation between the efficiency and the numerical representation of a given caste. If in the course of evolution one caste increases in efficiency while others do not, the proportionate total weight of the improving caste will decrease. In other words, the expected result of colony-level selection is precisely the opposite of that of individual selection, which would be an increase in the more efficient form.

This prediction was tested in the following way. Increased efficiency at a given task implies increased specialization on that task, a relation that has been documented abundantly and virtually without exception throughout the ants. The theoretical prediction can therefore be translated to the simple statement that the less members of a caste do, the fewer there are. Pheidole is an excellent group with which to test such relationships. With the exception of several parasitic species, the members of this myrmicine genus are consistently dimorphic, with distinctive, large-headed major workers that are specialized for defense, milling, food storage, or some combination of these three roles. Yet there is variation: the majors differ from species to species in size and degree of anatomical and behavioral specialization, as well as numerical representation within the colony. Pheidole is also the largest of all ant genera, with hundreds of species available for sampling and comparison. Finally, most species of Pheidole have proved relatively easy to culture in the laboratory. Wilson (1984b) obtained the behavioral repertories and major/minor ratios of ten species, representing an equal number of phylogenetic distinct species groups from various localities around the world. The proportionate representation of the majors was found to decrease significantly as their behavioral repertory decreased among the species, in accordance with ergonomic theory. This effect is shown in Figure 8-41.

A second counterintuitive result of ergonomic theory is the predicted generalization that species with initially unspecialized castes will have on the average fewer castes and more variable caste ratios, and this effect will be enhanced in fluctuating environments. In other words, generalized castes are prone to extinction. The more specialized the castes become in evolution, the more entrenched they become, in the sense that they are more likely to be represented in the optimal mix regardless of long-term fluctuations in the environment (Wilson, 1968, 1971). In classical evolutionary theory, which entails individual selection, it is the generalized genotypes and species, and not the specialized ones, that are more likely to survive in the face of long-term fluctuation in the environments. This result will be hard to test in any definitive manner. Well-developed major castes have persisted with little change in Camponotus, Oecophylla, and Pheidole since at least Miocene times, in other words about 15 million years, but data are lacking on the longevity of less differentiated caste systems.

The ergonomic models lead to the results that in a perfectly constant environment and in the absence of developmental constraints of any kind, it is of advantage to colonies to evolve so that there is one caste specialized to respond to each kind of contingency. In other words, one caste should come into being that perfects the appropriate response, even at the expense of losing proficiency in other tasks. But this is very far from what prevails in nature. Behavioral repertories prepared for a variety of species in Atta, Camponotus, Cephalotes, Leptothorax, Pheidole, Solenopsis, and other genera (references in Table 8-3) indicate that as a rule workers of each colony perform between 20 and 50 distinct tasks, the precise number varying according to species. Yet most ant species have only a single worker physical subcaste plus three or four discernible temporal subcastes. Even species possessing the most complex polymorphism and division of labor, such as the members of Atta and Pheidologeton, appear to have no more than seven combined physical and temporal castes.

Oster and Wilson (1978) considered at some length the possible existence of constraints on the proliferation of physical and temporal castes. Through a combination of deduction and inference based on empirical evidence they suggested a restrictive role for the following properties:

1. Holometabolism. In ants, bees, and wasps adult size is fixed permanently at the end of larval growth, while most of the labor is performed later, following the attainment of the adult instar. As a consequence the size-frequency curve of the existing adult population cannot be altered to meet new contingencies imposed by the environment. There exists a time lag between the onset of the new conditions and the attainment of a more efficient size-frequency distribution that minimally equals the period between the moment of cessation of larval growth and the eclosion of new adults, in other words roughly the duration of the pupal stage. The coarse nature of this adjustment to the environment reduces the number of castes to be expected in the optimal mix below that predicted by the elementary linear programming models.

2. Allometry. In the ants, which are the only social hymenopterans that display a significant amount of worker polymorphism, the physical differentiation is based on allometry--the regular disproportionate increase of body parts relative to one another. Thus there is usually only a single rule of deformation, or (in the case of diphasic and triphasic allometry) two or three, which substantially diminishes the number of castes that can be generated over a given amount of total size variation.

3. Fidelity costs during development. In order to regulate not only allometry but also the characteristic size-frequency distributions that fix the proportions of physical castes, feedback mechanisms must be employed that correct repeatedly for the dispersion in growth of each age cohort. As precision is increased, the energetic costs of the regulatory mechanisms are likely to grow in a greater than linear relation. Above some size the costs must become prohibitive, and species that reach this limit are expected to compromise accordingly in the degree of complexity of their caste systems. Diana Wheeler (1986a) has argued further that species relying on larval nutrition as the principal trigger for queen determination will find it more difficult to add worker physical subcastes on top of the prior and requisite differentiation of queens and workers. The physiological systems needed to maintain both levels at the same time would be difficult to operate with any precision. On the other hand, systems of queen determination based on direct pheromonal or hormonal intervention will pose less of a barrier to the evolution of worker polymorphism. Pheromonal and hormonal intervention is in fact the rule in ants, as we shall see shortly.

4. Environmental variance. Prey captured by the colony foragers and soil particles moved by the excavators have a substantial size variance. It should be of advantage for the worker force of colonies to have a built-in size variance to compensate for this environmental uncertainty. If true, the result will be a decrease in the number of discrete castes that can be maintained.

5. Task overlap. Tasks that are very different from one another may, nevertheless, require anatomical features that are closely similar. For example, prey items and the pupal stages of the ants' own nestmates often resemble each other in size and shape, so that the same worker caste could be employed to handle both categories of objects.

6. Behavioral plasticity. Workers specialized for one task are usually required to perform other rare but essential tasks. When ant nests are broken open, to take one familiar example, most workers stop what they are doing and either attack the intruder or carry larvae and other immature forms into deeper chambers. This minimal flexibility limits the degree of specialization of individual castes and perhaps also the number that can be differentiated within a single colony population.

7. Ergonomic costs. Soldiers, millers, and other exceptionally large castes are energetically expensive to manufacture and maintain. Although one of these forms might be “ideal” in their capacity to function, they would never evolve because the energy they would expend is greater than the energy their presence would gain for the colony. Put another way, it is necessary for majors to perform some special service of exceptional importance to make their creation profitable to the colony. It is not surprising, then, to find that majors of strongly polymorphic species are characteristically very specialized in behavior.