• Kazuki Tsuji,

    1. Department of Agro-Environmental Sciences, Faculty of Agriculture, University of the Ryukyus (The United Graduate School of Agricultural Sciences, Kagoshima University), Okinawa 903–0213, Japan
    2. Department of Biology, Faculty of Science, Toyama University, Toyama 930–8555, Japan
    3. E-mail:
    Search for more papers by this author
  • Noritsugu Kikuta,

    1. Department of Biology, Faculty of Science, Toyama University, Toyama 930–8555, Japan
    Search for more papers by this author
  • Tomonori Kikuchi

    1. Department of Agro-Environmental Sciences, Faculty of Agriculture, University of the Ryukyus (The United Graduate School of Agricultural Sciences, Kagoshima University), Okinawa 903–0213, Japan
    2. Marine Biosystems Research Center, Chiba University, Tokawa 1, Choshi, Chiba 288–0014, Japan
    Search for more papers by this author


Workers of social Hymenoptera can usually produce male offspring, but rarely do so in the presence of a queen despite the potential individual fitness benefit. Various mechanisms have been hypothesized to regulate worker reproduction, including avoiding the colony-level cost of worker reproduction. However, firm quantitative evidence is lacking to support that hypothesis. Here, we accurately quantified this cost by studying an ant species (Diacamma sp.) in which worker reproduction is rare in the presence of the gamergate (the functional queen). A series of experiments to manipulate worker–gamergate contact revealed that short-term brood-production efficiency is not changed by the presence of worker reproduction. However, when workers reproduce, their average life span is reduced to between 74% and 88% of that in the absence of reproduction, indicating a long-term cost to the colony. In theory, this cost can explain the policing of worker reproduction under a queen-single mating system, but the cost does not appear to be high enough to stop worker reproduction. When contact with the gamergate is lost, it is only the nonreproductive workers whose life span was reduced; the reproductive workers lived as long as nonorphaned workers. We suggest that an increased workload can account for the reduction in life span better than a trade-off between reproduction and longevity.

Biological systems are characterized by hierarchies, for example, ensembles of genes forming genomes, cells combined to become multicellular organisms, and groups of individuals organized into societies. A fundamental question is how conflicts among the lower level units is resolved to form the higher units (the major evolutionary transitions; Maynard Smith & Szathmáry 1995; Bourke 2011). Policing is a mechanism that enables the formation of group harmony by preventing selfish behavior of the lower units (Ratnieks 1988; Frank 1995, 1996; El Mouden et al. 2010).

Well-organized cooperative colonies of social insects provide a good opportunity to study policing (Wenseleers and Ratnieks 2006), because the presence of various conflicts due to differential interests among colony members is well established by inclusive fitness theory (Bourke and Franks 1995; Crozier and Pamilo 1996). Among those conflicts that over male parentage in hymenopteran societies is most frequently discussed in the context of policing. In the majority of ants, some bees, and some wasps, the workers have lost their ability to mate but can still lay haploid eggs that will become male offspring. Because workers are more closely related to their male offspring than to their brothers and nephews, they should compete with the mother queen and with sister workers over male parentage (Bourke and Franks 1995; Crozier and Pamilo 1996). However, although recent studies revealed that direct male production by workers actually occurs at various frequencies depending on species or populations (Hammond and Keller 2004; Wenseleers and Ratnieks 2006), it is also true that even in colonies with worker reproduction the majority of workers usually remain sterile in the presence of the queen (Choe 1988; Bourke and Franks 1995; Wenseleers and Ratnieks 2006). Three behavioral mechanisms that could inhibit worker reproduction have been proposed (Cole 1986; Ratnieks 1988): (1) policing by the queen, (2) policing by workers, and (3) self-restraint by workers. The first mechanism should be effective only in small colonies, and chemical control by the queen is likely to be unstable against invasion by pheromone-resistant workers (Ratnieks 1988; Keller and Nonacs 1993). Therefore, the two latter hypotheses are most likely to be important in species with large colonies, such as many ants and honeybees. It is also known that workers’ attempt at self-reproduction seems to be more widespread than effective worker reproduction resulting in actual male production (Wenseleers et al. 2005; Wenseleers and Ratnieks 2006; Helanterä and Sundström 2007), demonstrating the effectiveness of policing in determining male parentage. Note that these behavioral mechanisms are not independent. For example, self-restraint can evolve secondarily in response to effective worker policing (e.g., Wenseleers et al. 2004; Ohtsuki and Tsuji 2009).

As for the ultimate factor of worker policing, there are two alternative hypotheses; the relatedness hypothesis and the group-productivity-cost hypothesis. Relatedness hypothesis predicts that worker policing can evolve when workers are on average more genetically related to queen's sons than other workers’ sons (Ratnieks 1988). Such a shared interest of workers is expected when the colony has a multiply mated queen (Woyciechowski and Lomnicki 1987; Ratnieks 1988; Wenseleers et al. 2004; Ohtsuki and Tsuji 2009) or multiple related queens (Nonacs 1993).

Cole (1986) first addressed the importance of the colony-level cost of worker reproduction. Actual conflicts caused by worker reproduction, such as aggression among colony members, can waste colony resources. Furthermore, workers are usually physically or physiologically specialized to some extent to fulfill nonreproductive tasks (Bourke and Franks 1995; Crozier and Pamilo 1996; Khila and Abouheif 2008), so that selfish oviposition by workers would reduce the efficiency of the entire colony by disrupting the division of labor even in the absence of aggression. Theory predicts that a small cost leads to the evolution of worker policing and a large cost results in total self-restraint of worker reproduction (Cole 1986; Ratnieks 1988; Wenseleers et al. 2004; Ohtsuki and Tsuji 2009).

Comparative evidence largely supports the relatedness hypothesis but simultaneously reveals huge variation in the occurrence of worker policing and that of worker reproduction (Wenseleers and Ratnieks 2006; see also Hammond and Keller 2004), which encourages a direct test of the cost hypothesis. In fact, many researchers have tried to measure or detect the cost of worker reproduction and that of the resultant intracolonial conflicts. This has been done in terms of the reduced time spent on brood care in Temnothorax (previously Leptothorax) ant workers (Cole 1986), the metabolic rate in Pachycondyla ant workers (Gobin et al. 2003), the reduced fungus garden volume in two fungus-culturing ants (Dijkstra and Boomsma 2007), the decreased immunocompetence of Diacamma ant workers (Bocher et al. 2007), the failure of brood raising due to overproduction of eggs and overnumerous egg layers in two parthenogenetic ants (Tsuji 1994; Hartmann et al. 2003), the short-term reduction in worker production by a slave-maker ant (Bourke et al. 1988), and the production of sexual offspring in a bumble bee (Lopez-Vaamonde et al. 2003). Some of these results detected the cost (Cole 1986; Tsuji 1994; Gobin et al. 2003; Hartmann et al. 2003; Bocher et al. 2007), but the others did not (Bourke et al. 1988; Lopez-Vaamonde et al. 2003; Dijkstra and Boomsma 2007). However, detection of a cost itself is not a sufficient empirical support for the cost hypothesis, because as already mentioned, theory predicts differential evolutionary outcomes depending on the size of the cost. Furthermore, whether such estimated “costs” are translated into actual costs is unclear. Thus far, except Lopez-Vaamonde et al. (2003), no one has examined the cost by measuring an aspect of colony performance that relates directly to colony fitness. In social insects, lifetime reproductive success is often difficult to measure even in the queens of annual species (but see, e.g., Queller and Strassmann 1988; Lopez-Vaamonde et al. 2003, 2009). The underground life of perennial ants makes the lifetime reproductive success of colonies or queens extremely difficult to measure. We thus focused on two parameters in Diacamma sp. from Japan, which founds colonies by means of colony fission: brood production per unit time and worker life span. In fission-reproducing social insects, colony growth rate is closely associated with colony fitness: that is, the faster the colony's growth, the greater the probability of fission, a hypothesis that was empirically supported by research on an army ant (Franks 1985). Worker life span determines the upper limit of how long the insect can work, and therefore decreased life span as a result of reproduction is a direct estimate of the cost to the colony.

Focusing on life span is also significant from the perspective of testing evolutionary theories of aging (Keller and Genoud 1997; Schrempf et al. 2005; Bourke 2007; Heinze and Schrempf 2008; Münch et al. 2008). The mated reproductive queens usually live far longer than sterile virgin workers, challenging the hypothesized trade-off between reproduction and longevity (e.g., Franks 1995; Keller and Genoud 1997), which has been a basic assumption in evolutionary life-history theory (Stearns 1989; Roff 2002). The protection from external mortality factors that is permitted by social life has been proposed as the ultimate explanation for the long life of queens (Keller and Genoud 1997; Bourke 2007; Heinze and Schrempf 2008). However, in most eusocial insects, the queens and workers differ physically, so a simple proximate explanation that physically superior queens (large and often totipotent) live longer than workers (small and nontotipotent) is also a possible explanation. Thus, empirical studies should try to separate these possibilities.

Diacamma sp. from Japan is a useful species for such studies because it lacks a physical caste; that is, there is no morphological queen caste. A mated worker, called a “gamergate,” reproduces in each colony (Fukumoto et al. 1989). As in many other ants, oviposition by virgin workers is rare in the presence of the gamergate (a functional queen), but is frequent in orphan colonies that lack a gamergate (Peeters and Tsuji 1993; Fukumoto et al. 1989). The proximate regulatory mechanism for reproduction by virgin workers is self-restraint in response to the gamergate's presence (Tsuji et al. 1999). When workers lose contact with the gamergate, their physiology changes within a few hours and they begin to lay male eggs within 7–10 days (Peeters and Tsuji 1993; Tsuji et al. 1999; Kikuchi et al. 2008, 2010). Although gamergates are singly mated, both worker- and queen-policing occurs; therefore, the cost of worker reproduction has been proposed as the ultimate factor responsible for worker policing in this species (Kikuta and Tsuji 1999; Kikuchi et al. 2008). A previous study (Tsuji et al. 1996) also showed that gamergates live on average ca. two to three times as long as nonreproductive workers, providing evidence against the reproduction versus longevity trade-off. However, although gamergates and workers are physically similar, their mating status rather than reproduction could explain this difference in life span (Schrempf et al. 2005). In the present study, we therefore examined only virgin workers. Diacamma workers have small variation in body size (Fukumoto et al. 1989), so they provide a good model for testing the effect of reproduction on life span without physical differences affecting the results.

In this article, we quantitatively examined the colony-level cost of worker reproduction in Diacamma sp. by manipulating workers’ contact to the gamergate thereby inducing worker reproduction. We tested if the short-term colony performance (brood care efficiency per unit time) and the mean worker life span were changed by the induced worker reproduction. Furthermore, we tested whether the observed change in worker life span was caused by a trade-off between reproduction and life span or by a trade-off between workload and life span.



Diacamma species lack a morphological queen. Instead, mated workers called “gamergates” reproduce. In Diacamma sp., the only species of this genus found in Japan, all young workers can potentially become gamergates. Social manipulation soon after adult eclosion leads to a reproductive division of labor with only one gamergate per colony. All workers emerge with a pair of small thoracic appendages (gemmae). Callow workers whose gemmae are removed by other colony members, usually by the gamergate, lose their mating ability and become sterile helpers. When the colony loses its gamergate, a newly eclosed worker (usually the first that emerges) becomes dominant and retains her gemmae. She eventually mates and becomes the new gamergate (Fukumoto et al. 1989). In orphan colonies that lack a gamergate, no new workers emerge, but workers that lack gemmae begin to lay male eggs (Fukumoto et al. 1989; Peeters and Tsuji 1993).

We used 12 colonies that were collected in Nakijin Village, on Okinawa's main island, from 1997 to 1999 and from 2007 to 2008. Each colony was kept in the laboratory at 25 ± 2°C under a 16L:8D photoperiod, and was housed in a plastic cage, as described by Kikuta and Tsuji (1999). Ants received mealworms ad libitum and honey water once a week. All adults were individually marked with enamel paint before being used in the experiments.


Each colony (N= 5) was divided into two randomly assigned, equally sized subcolonies, one with the gamergate and the other without. Each subcolony was kept in a plastic cage (Kikuta and Tsuji 1999). We expected the egg supply to differ between the two subcolonies: the orphan subcolony produced only male eggs, whereas the gamergate laid female eggs. To equalize background conditions for the paired subcolonies, we further controlled their characteristics: We classified all workers into foragers and nest workers by 7–12 h of observations of each colony immediately before the experiment. Then workers of each category were randomly assigned into the paired subcolonies. Brood was also divided into two groups that contained the same number and the same growth-stage distribution (eggs, small, middle and large larvae, and pupae). The paired subcolonies therefore had equal initial numbers of nest workers, foragers, and brood. We checked the brood composition every three weeks, and exchanged half of the brood between the paired subcolonies also every three weeks, as we did at the start of the experiment. Simultaneously, we equalized the worker populations of the two subcolonies by removing random orphan workers or by transferring random workers from the gamergate subcolony to the orphan subcolony. As the developmental time from egg to adult is ca. two months (Nakata and Tsuji 1996), this equalized the brood composition and let us compare the workers’ brood care efficiency. To our knowledge, no previous study controlled the backgrounds of paired colonies this carefully, because all previous studies focused only on short-term performance (see introductory paragraphs). We recorded the incidences of pupation and adult emergence daily for up to 30 weeks and used these parameters as an indicator of colony productivity. We summed brood production (pupae and adults separately) every three weeks and compared the results using a general linear mixed model (GLMM) with treatment (a fixed factor based on whether or not the colony had a gamergate) and colony (random factors: with N= 5 colonies before division into subcolonies, and the colonies × treatment interaction).

The number or the proportion of reproductive workers may influence the size of the cost, that is, the more egg-laying workers, the larger colony-level cost. In Diacamma sp., it is known that after orphaning many workers start to develop ovaries and begin to oviposit. However, eventually, dominance interactions among orphan workers lead to the reduction in the number of egg layers. Finally, in two months or so, the alpha worker monopolizes the haploid-egg production (Peeters and Tsuji 1993; K. Tsuji, unpubl. data). This suggests the hypothesis that the size of cost of worker reproduction might depend on time after orphaning. To test this hypothesis, we analyzed the above-mentioned per three-week brood production data with another general linear model in which we chose three fixed factors: treatment, time, and their interaction. As for “time,” the value 1 was assigned to the first data obtained at three weeks after the orphaning, and the value 2 was assigned to the second data obtained at six weeks after orphaning, and so on. A statistically significant treatment × time interaction should suggest a change of the cost size over time. Alternatively, it is also possible that brood production per time does not change over time, because in orphan colonies of Diacamma sp., frequent aggressive dominance interactions continue to occur among workers even after monopolization of reproduction by the alpha (Peeters and Tsuji 1993; Tsuji et al. 1998; K. Tsuji, unpubl. data). When this hypothesis is right, the treatment × time interaction will not be statistically significant.

Newly emerged workers were also marked and randomly assigned to the paired subcolonies. New workers that emerged in orphan subcolonies were moved to the counterpart subcolony with a gamergate for one day to allow removal of their gemmae. We recorded the adult life span, that is, from the date of adult eclosion to that of death (hereafter, we simply call life span), of newly emerged workers (whose exact age was therefore known). The worker life span was compared between the subcolonies of each pair using the log-rank test of survival rates for each colony, and later combined the results using Stouffer's z-score method. The gamergates of two colonies died after 30 weeks, so we used data only from the other three colonies for the life span comparison. We stopped monitoring worker life span of the paired subcolonies when the gamergate died. We excluded data of worker cohorts (groups of workers emerged on the same day or later) in which any individual belonging to the cohorts was still alive when the gamergate died from the analysis.

We also observed each colony for 3–6 h every week to detect aggressive dominance and oviposition behaviors. These observations allowed us to classify all virgin workers as either dominant (i.e., they exhibited dominance behavior toward another worker at least once during the observation period) or subordinate (they never initiated aggressive behavior). Only dominant workers are known to lay male eggs in orphan colonies (Peeters and Tsuji 1993; Nakata and Tsuji 1996; Kikuta and Tsuji 1999). We also compared life span between these subcategories of workers, as described above.


Orphan colonies may differ from colonies with a gamergate in ways other than the presence of worker reproduction. Thus, we carried out another experiment in which we induced worker reproduction without using orphan colonies. Each colony (N= 2) was placed in a nest cage with two chambers (Tsuji et al. 1999). We used quick-drying glue to attach a small piece of fishing string (0.205 mm in diameter, 7 mm long) to the dorsal alitrunk of the gamergates and to that of a random two-thirds of workers sideways to the body's long axis. These adults were thereby confined to one of the nest's two chambers. Half of the impeded workers were randomly assigned to the gamergate chamber (“A workers”; Tsuji et al. 1999), and the rest to the chamber with no gamergate (“B workers”). The remaining one-third of the workers could move freely between chambers (“C workers”). We know that the entire cage functions as a single colony, but only B workers who lost direct contact with the gamergate begin to lay eggs in this setting (Tsuji et al. 1999). Newly emerged workers were individually marked and randomly assigned to treatments A to C. We compared worker life span in the three categories in the same way as in the orphan colony experiment. We observed the colonies for 3–6 h a week to record dominance behavior and oviposition.


Using colonies (N= 5) different from those used for the above cost measurement, we studied behavioral change associated with orphaning. We used a scanning technique (Altmann 1974) to observe the behavior of all individually marked adults (n= 35–77). The interval between scans was at least 10 min, and each individual was scanned at least 100 times during four days to allow estimation of the time budget for each behavioral act. The gamergates were then removed and the scanning was repeated for another four days. We compared the frequencies of resting, selfish behavior, and social behavior before and after removal of the gamergates. The selfish behavior consisted of dominance behavior and self-grooming. The social behavior included brood care, allo-grooming (donating), food handling, nest construction, and outside-nest activities. We regarded walking within the nest as nonspecific behavior, because its function was difficult to identify. We treated motionless ants as resting, even when they were being groomed by other workers. We examined the difference in the frequency of each of the four major behavioral categories (resting, selfish behavior, social behavior, and the other nonspecific behavior) between before and after orphaning. The change in the frequency of each behavioral category was examined by the Wilcoxon matched pair signed rank test using individual workers as blocks for each colony. Then we combined the results of colonies using unweighted Stouffer's z-score method (Whitlock 2005). We repeated these analyses focusing on dominant and subordinate workers, which were distinguished by observation under the orphaned condition, as defined previously.

All statistical analyses were performed using JMP 7.1J software (SAS Inc., Cary, NC).



To test the colony-level cost hypothesis, we first compared the brood production (pupae and adults) between colonies with and without worker reproduction under carefully controlled conditions. Neither the production of pupae nor that of adults per unit time differed statistically significantly between the pairs (the factor “treatment” in the production of pupae P= 0.59, that in the production of adults P= 0.39: Table 1). Furthermore, the treatment (orphan vs. nonorphan) was statistically insignificant also in the other models in which the effect of time after orphaning was taken into account (Table 2). Although the number of egg-laying workers is known to change over time as already mentioned, the statistically nonsignificant time × treatment interaction (Table 2) suggests that the temporal change in the number of egg-laying workers did not seriously affect the short-term brood production rate. The effect of time on adult eclosion was significant (Table 2), which means that the paired subcolonies, regardless of orphan or nonorphan, similarly changed the brood production rate over time. Such a temporal fluctuation in brood production rate is known in captive colonies of this species (Nakata 1996).

Table 1.  Results of the GLMM analysis of colony productivity (pupae and adults produced per unit time). The number of individuals (A) that pupated and (B) that eclosed as adults during each three-week period is the dependent variable. SE is the standard error of the colony mean. The “treatment” factor compares colonies with a gamergate present and absent (orphan subcolonies).
 Gamergate present (mean±SE) N Orphan subcolony (mean±SE) N  
(A) Pupation per three weeks     
 FactorsSum of squaresdfMean square F P
(B) Adult eclosion per three weeks     
 FactorsSum of squaresdfMean square F P
Table 2.  Results of the general linear model analysis of colony productivity taking into account time after the beginning of the experiment. The number of individuals (A) that pupated and (B) that eclosed as adults during each three-week period is the dependent variable. For the meaning of the factor “treatment,” see Table 1.
FactorsSum of squaresdfMean square F P
(A) Pupation per three weeks
(B) Adult eclosion per three weeks

The average number of individuals that have directly observed to oviposit was 4.6 ± 2.51 (mean ± SD) per colony, and their proportion to all workers was 10.1%± 6.09%. These vales must underestimate the population of egg layers, because our observation time was limited. Therefore, it is informative to note the number of dominant workers, most of which are known to oviposit one to three weeks after orphaning (Peeters and Tsuji 1993). The average number of dominant workers was 15.4 ± 6.39 (mean ± SD) and their proportion to all workers was 33.4%± 14.3%.


The above result does not disprove the cost hypothesis, because we found a long-term effect on worker life span. The average life span of the workers decreased statistically significantly, by 12%± 9.5% (mean ± SE, P= 0.02), in the orphan subcolonies in comparison to that in nonorphan subcolonies (Fig. 1A).

Figure 1.

Comparison of the mean life span of workers. Life span was measured from the day of adult eclosion to that of death. Values represent colony means ± SD. (A) Orphan workers versus nonorphan workers (z=−2.323, P= 0.02). Sample sizes for orphan and nonorphan workers were 140 (41–55 per colony) and 166 (40–71 per colony). (B) Workers without a gamergate present (B workers), workers with gamergate present (A workers), and workers that were free to move between chambers (C workers) (A vs. B workers, z=−5.368, P < 0.001; B vs. C, z=−3.683, P < 0.001; A vs. C, z= 0.3814, NS). The sample sizes for A, B, and C workers were 90 (40 and 50 per colony), 92 (41 and 51), and 65 (29 and 36), respectively. The sample size differed among categories owing to missing marked individuals. (C) Dominant versus subordinate workers (z=−4.293, P < 0.001), dominant versus control workers (z=−1.303, P= 0.192), subordinate versus control workers (z=−6.178, P < 0.001). The sample sizes of dominant and subordinate workers were 77 (8–23 per colony) and 157 (21–46), respectively. All statistical analyses were first performed separately for each colony using the log-rank test, and then the probabilities were combined using Stouffer's z-score method. All values are two-tailed probabilities.

As in the orphan colony experiment, in the impeded worker experiment the B workers had significantly shorter (P < 0.001) life span than the A and C workers (Fig. 1B). By treating the A workers as the control, we estimated the life span cost of worker reproduction as 26%± 7.9% (mean ± SE). When we combined the data from the two experiments, no statistically significant difference was detected between the two cost estimates, and the estimated life span cost was 18%± 6.8%.


Next, we addressed the proximate mechanism for shortened worker life span as a result of selfish oviposition. We formulated two alternative hypotheses: (1) that there is a trade-off between longevity and reproduction (Stearns 1989; Roff 2002), and (2) that an excess workload is created by the less-efficient division of labor that results from selfish reproduction. We can test the trade-off hypothesis, because only dominant workers lay male eggs when they have lost contact with the gamergate, whereas subordinates remain sterile; thus, both the orphan workers and the B workers have egg-laying and nonegg-laying workers (Peeters and Tsuji 1993; Nakata and Tsuji 1996; Tsuji et al. 1999). The reproductive trade-off hypothesis predicts that egg-laying dominant workers will have shorter life spans than the other workers within the same treatment. However, the opposite was the case. Dominant workers lived longer than nonlaying orphan or B workers (Fig. 1C). Furthermore, dominant workers lived as long as the nonorphan control workers (Fig. 1C).

On the other hand, our behavioral observations using the other five colonies support the excess-workload hypothesis. The behavioral responses of workers to orphaning differed depending on an individual's social status. The orphaned dominant workers spent significantly more time on selfish activities such as dominance behavior and self-grooming and less time (although nonsignificantly) on brood care (Fig. 2B). In contrast, subordinates increased social behaviors (Fig. 2C). Both dominant and subordinate workers increased nonspecific behavior (walking) and reduced their resting time under orphan conditions, suggesting an increase in the general activity level (Fig. 2B, C). The colony as a whole decreased social behavior significantly compared with the nonorphan conditions but only to a small extent (Fig. 2A).

Figure 2.

The average time budgets for eight behavioral categories spent by (A) all workers combined (total), (B) dominant workers, and (C) subordinate workers under gamergate and orphan conditions. Selfish behavior, social behavior, resting, and nonspecific behavior are represented in red, blue, black, and white, respectively. We compared the frequencies in these categories between the gamergate and orphan conditions. First, we used Wilcoxon matched pair sign rank test separately for each colony, and then we combined the probabilities using the z-score method. (A) Total: selfish behavior, z=−3.65, P < 0.001; social behavior, z=−4.31, P= < 0.001; nonspecific behavior, z=−6.79, P < 0.001; resting, z=−5.91, P < 0.001. (B) Dominants: selfish behavior, z=−5.03, P < 0.001; social behavior, z=−1.37, NS; nonspecific behavior, z=−5.45, P < 0.001; resting, z=−3.49, P < 0.001. (C) Subordinates: selfish behavior, z=−1.56, NS; social behavior, z=−5.35, P < 0.001; nonspecific behavior, z=−5.62, P < 0.001; resting, z=−6.23, P < 0.001.



Although workers spent more time on nonproductive acts such as dominance behavior in orphan subcolonies, as was reported in Temnothorax (Cole 1986), the production of pupae and adults per unit time did not differ significantly between the pairs (Table 1). This indicates that worker reproduction did not seriously alter the colony's short-term productivity. Although aggressive dominance behavior in animals often appears to be costly and is motivated by selfish interests to increase personal direct fitness (Monnin and Ratnieks 1999; Tsuji and Tsuji 2005), in theory it can sometimes improve group productivity (Úbeda and Duénẽ z-Guzmán 2010). Dominance behavior in social insects often leads to form a linear hierarchy, through a decentralized control (Tsuchida and Suzuki 2006), where only one or a few top rankers can reproduce (Monnin and Ratnieks 1999; Tsuji and Tsuji 2005). Such hierarchy-mediated reproductive division of labor might compensate for the energetic costs of aggression and could diminish the short-term group productivity cost.

However, the short-term productivity unaffected by worker reproduction in Diacamma sp. does not disprove the cost hypothesis, because we found a long-term effect on worker life span. The estimated life span costs of 12% and 18% fall in the theoretically predicted range (ca. 4– 22%) in which conflicts among workers can lead to the evolution of worker policing (Cole 1986; Ratnieks 1988; Wenseleers et al. 2004; Ohtsuki and Tsuji 2009). The higher estimate of 26% (Fig. 1B) seems to exceed the theoretical lower threshold of 17%, above which workers totally forgo reproduction as a result of self-restraint (Cole 1986). However, the estimate of 26% did not differ significantly from 17%, owing to the large standard error (t= 1.17, df = 1, one-tailed P= 0.23). Because Diacamma sp. workers sometimes (rarely) attempt selfish oviposition in the presence of a gamergate (Nakata and Tsuji 1996), the real costs may fall within the range at which conflicts exist, and workers are selected to raise their own male offspring while not favoring the production of other workers’ male offspring. Furthermore, none of the above estimates differed significantly from the predicted colony-level cost of 14.3% (Ohtsuki and Tsuji 2009), that colonies should bear when both worker policing and worker reproduction simultaneously occur under queen single-mating and monogyny (12%, t= 0.243, df = 2, one-tailed P= 0.42; 18%, one-tailed t= 0.482, df = 4, P= 0.33; 26%, t= 1.52, df = 1, one-tailed P= 0.18). Overall, the estimated cost can explain the presence of worker policing in Diacamma sp. but can also allow worker reproduction to occur. Future studies should focus on male parentage in this ant species to test this prediction.

The behavioral responses that we observed are likely to be the consequence of an adaptation to natural orphaning events. When an orphan colony fails to recruit a new gamergate, as can happen in nature when no female brood is available, natural selection should favor male production by the workers. Rapid production might be important in this situation, because the worker population will be declining as a result of mortality and a lack of replacement of dead workers. Small ant colonies are generally less competitive and thus more likely to die off, so orphaned workers should work harder even at the expense of their own longevity. However, such a colony-level consensus on common inclusive fitness does not account for all of the observed phenomena; for example, it does not explain the prevalence of aggressive dominance in the orphaned colonies. The aggression reflects reproductive competition among colony members, and such competition should be an inevitable consequence of worker reproduction, otherwise the reproductive output of individual workers would be diluted as each becomes “one of many” (Bourke 1999), and because one or a few workers could actually monopolize reproduction in orphaned colonies in this species (Peeters and Tsuji 1993; Nakata and Tsuji 1996). Losers of this conflict (the subordinate workers) would then work harder to enhance their indirect fitness. The above logic applies in the case of orphaned colonies; therefore, one may assume that our experiment might not apply to the cost of worker reproduction in the presence of a gamergate.

We believe, however, that our experiment does apply to the presence of the cost of worker reproduction under the gamergate-right condition for the following reason. Suppose that some workers started reproducing directly in the presence of the gamergate. Those laying workers should abandon nonreproductive tasks as documented in this and other ants. If subordinate workers did not work harder to compensate for the diminished colony efficiency caused by the loss of work performed by egg-laying workers, the average worker life span might be unchanged (as we assume a trade-off between workload and longevity, see later), but the colony productivity per unit time would likely decrease. Because no worker compensates for the labor abandoned by laying workers, colonies will lack the workforce required for brood care and other activities except egg laying. As a result the cost should manifest itself in terms of short-term brood production. This situation has been empirically shown in some other ant species, in which the ratio of egg (or brood) supply to the number of workers was manipulated (Tsuji 1994; Hartmann et al. 2003). We consider, therefore, that our result is not an artifact of the workers’ natural response to orphaning (i.e., diminished life span of subordinate workers due to an enhancement of their working activity) but empirical evidence of the cost of worker reproduction.

Single mating of gamergates has been suggested in Diacamma sp. (Fukumoto et al. 1989; Nakata et al. 1998) and in other Diacamma species (André et al. 2001). Although a frequent turnover of gamergates can create a low degree of relatedness among workers that can lead to the evolution of worker policing (Nonacs 1993), our experiments indicate that the cost of worker reproduction alone is substantial enough to cause the worker policing that is known to exist in this species (Kikuta and Tsuji 1999; Kikuch et al. 2008). Worker reproduction, which is common only under orphan conditions and is associated with seemingly cost-bearing aggressive dominance, is a common phenomenon in some ants, wasps, and bees (Bourke and Franks 1995; Crozier and Pamilo 1996). We therefore hypothesize that in social insects, the cost of worker reproduction is an important ultimate factor that causes worker policing, and that worker policing would be widespread among social insects irrespective of the queen mating frequency and the number of queens. In support of this view, comparative data suggest that relatedness structure can explain the occurrence pattern of worker policing but there is also large unexplained variation (Wenseleers and Ratnieks 2006).

Nevertheless, given the presence of worker policing and the cost of worker reproduction, a recent theory predicts that some workers will reproduce under monandry and monogyny when the colonies enter the reproductive stage (Ohtsuki and Tsuji 2009). Supporting this prediction in Diacamma sp., the attempts by workers to reproduce themselves become more prominent in large colonies (Nakata and Tsuji 1996; Kikuchi et al. 2008). We therefore need further studies particularly on male parentage in this ant; in particular, the proportion of worker-produced males and that of gamergate-produced males should be estimated using genetic markers. Such empirical data will improve our understanding of the evolution of insect societies with potential conflicts. Furthermore, male parentage data will also be useful to test the general hypotheses in social evolution: if social enforcement preventing selfish behavior of group members evolves, is it because it enhances the average fitness of group members or because it increases genetic output of group members (Ratnieks 1988; Frank 1995, 1996; El Mouden et al. 2010)?


The fact that reproductively dominant workers lived longer than nonreproductive subordinate workers contradicts the reproduction-longevity trade-off hypothesis. Our behavioral observation suggests that when dominant workers act more selfishly, the other workers compensate by working harder, implying that subordinate workers were likely to have been overworked, which reduced their life span. This compensation by subordinates for the selfish behavior of dominants can also explain why the short-term colony productivity (pupae and adults per unit time) did not differ significantly between the orphan and nonorphan subcolonies (Tables 1 and 2). In fact, although the time spent on dominance behavior after orphaning increased by about six times (Fig. 2B), brood care by the colony as a whole did not dramatically decrease (Fig. 2A). Instead, both dominant and subordinate workers reduced their resting time under orphan conditions (Fig. 2B, C). Note that this compensation is imperfect in the long term, because the average worker life significantly decreased in the presence of worker reproduction (Fig. 1). Together with empirical information from other species (Keller and Genoud 1997; Bourke 2007; Heinze and Schrempf 2008; Münch et al. 2008), our results suggest that in social insects, the cost of reproduction is not the primary determinant of individual life span. We set out a hypothesis that the life span of social insect workers is influenced by another important trade off, that is, the trade off between workload and life span.

However, underlying physiological mechanisms of these individual-level changes in behavior and life span in Diacamma sp. are still to be studied. Recently, some researchers have started to study the physiological and molecular mechanisms of reproductive differentiation in this ant (Okada et al. 2010a,b), as reproductive differentiation is experimentally inducible in this system. Future studies should extend such approaches to investigate the mechanism determining adult life span.

Associate Editor: S. West


We thank A. Bourke, Ed Vargo, S. West, and an anonymous referee for comments on earlier drafts of this article. This study was supported in part by KAKENHI grants from the JSPS (11740428, 17657029, 18047017, 18370012, 20033015 and 21247006) to KT.