Informational constraints on optimal sex allocation in ants Jacobus J. Boomsma*†‡, Jannie Nielsen†, Liselotte Sundström†§, Neil J. Oldham¶储, Jutta Tentschert¶, Hans Christian Petersen**, and E. David Morgan¶ *Department of Population Ecology, Zoological Institute, University of Copenhagen, Universitetsparken 15, 2100 Copenhagen, Denmark; †Department of Ecology and Genetics, Institute of Biological Sciences, University of Aarhus, 8000 Aarhus C, Denmark; §Department of Ecology and Systematics, Helsinki University, POB 17, FIN-00014 Helsinki, Finland; ¶Department of Chemistry, Keele University, Keele, Staffordshire ST5 5BG, England; and **Department of Statistics, University of Southern Denmark, Sdr. Boulevard 23A, 5000 Odense C, Denmark Workers of the ant Formica truncorum specialize in rearing females or males depending on the number of fathers of a colony. These split sex ratios increase inclusive fitness, but it has remained unknown how workers assess the number of patrilines in their colonies and to what extent their reproductive decisions are constrained by lack of information. By analysis of the quantitative variation in cuticular hydrocarbon profiles of workers of multiply mated queens, we show that the heritable component of recognition cues is low and that the extent of sex ratio biasing toward males is correlated with patriline differences in hydrocarbon profiles. Workers are thus able to capitalize on colony-level relatedness asymmetry, but their inclusive fitness is constrained by uninformative recognition cues. These results are consistent with the hypothesis that the occasional expression of nepotistic phenotypes favoring full-sisters over half-sisters maintains selection against informative recognition cues. We evaluate how inclusive fitness theory may be used to predict the number and kind of recognition cues in insect societies of a specific relatedness structure. split sex ratios 兩 inclusive fitness 兩 chemical recognition 兩 cuticular hydrocarbons 兩 nepotism E very kin-structured society is a delicate balance between cooperation for common prosperity and potentially destructive conflicts between subgroups or individuals. The expression of such conflicts may increase the fitness of some individuals, but may entail collective costs when it disrupts social cohesion and reduces productivity (1). Principles of cooperation and conflict in families can generally be understood from Hamilton’s rule (2), but the key variables in the equation (relatedness and the costs and benefits of helping), may both explain indiscriminate cooperation among relatives of any degree, and nepotistic discrimination between relatives of different relatedness. This paradox can only be resolved when recognition efficiency, i.e., the availability and perception of information on relatedness, is explicitly considered (3). Accurate information allows efficient recognition of individual degrees of kinship promoting the expression of conflict, whereas noisy recognition cues could make the cost of nepotism greater than the benefit and thus lead to unconditional cooperation among relatives, while still allowing discrimination of nonkin. The ability to differentiate between kin of different degree has been amply documented in vertebrate societies (4, 5), but has remained difficult to prove in insect societies (ref. 3, but see ref. 6). Likely reasons for this difference might be that insect colonies are larger than vertebrate societies, whereas the learning capacities of insects are limited compared with vertebrates (4). In addition, the social life of insects happens mostly in the darkness of nest cavities. Visual cues are therefore of limited use, the only known exception being paper wasps with daylight-exposed nests, which can recognize facial characteristics of nestmates (7). Kin recognition relies thus on chemical cues in most cases. A single class of compounds, the hydrocarbons of the wax layer on the www.pnas.org兾cgi兾doi兾10.1073兾pnas.1430283100 cuticle, is increasingly considered to be the most generally relevant category of recognition cues in insect societies (8). However, the maintenance of genetic variation for this kind of recognition cues is problematic: newly arising genetic cues are likely to be selected for initially when they increase the efficiency of collective discrimination against nonkin from neighboring colonies, but at some point these cues become variable (and thus informative) enough to be exploited by emerging nepotistic phenotypes. The expression of nepotistic conflict will then prevent further increase of genetic variation for recognition cues, both because of colony-level productivity costs (1, 3), and because the expression of nepotism per se tends to erode the genetic variation for the cues on which it is based (9, 10). The present study uses a population of the ant Formica truncorum in the Tvärminne archipelago of Southern Finland, where reproductive decision-making has been studied in such detail (11–13), that quantitative predictions on extant genetic variation for recognition cues can be tested and the fitness consequences of informational constraints can be evaluated. Colonies in the study population have a single queen per nest, but this queen is mated with one or several males. The ensuing variation in relatedness asymmetry of workers toward the reproductive females and males allows the precise prediction of the optimal sex ratio, provided that workers can control investments and are able to assess what type of colony they belong to: workers of a singly mated queen should specialize on females, whereas workers of a multiply mated queen should specialize on males (14). Empirical data collected during a period of 10 years have shown that these split sex ratio predictions, based on the inferred ability of workers to assess the number of patrilines, are generally met (11–13). Because it is inconceivable that workers can directly assess the number of ejaculates stored by their mother-queen, their ability to assess the number of patrilines must be based on the expression of heritable variation in cuticular hydrocarbon cues among nestmates (8). F. truncorum is therefore one of the few social insect model systems where genetic variation for recognition cues is known to be present. This is because split sex ratios induced by relatedness asymmetry of workers toward the female and male brood are only possible when half-sister colonies express more variation in recognition cues than full-sister colonies, and when the workers in each colony collectively assess this variation to increase their inclusive fitness (14). However, the match between the observed sex ratios (11, 12) and the theoretical predictions is not perfect, and it has remained unclear whether the unexplained variation in sex allocation is caused by environmental noise or by inaccurate genetic information for the assessment of queen mating status (15). The best way to address these questions is to concentrate on the multiply This paper was submitted directly (Track II) to the PNAS office. ‡To whom correspondence should be addressed. E-mail: jjboomsma@zi.ku.dk. 储Present address: The Dyson Perrins Laboratory, Oxford University, South Parks Road, Oxford OX1 3QY, England. PNAS 兩 July 22, 2003 兩 vol. 100 兩 no. 15 兩 8799 – 8804 EVOLUTION Edited by Bert Hölldobler, University of Würzburg, Würzburg, Germany, and approved May 14, 2003 (received for review January 16, 2003) Fig. 1. Typical cuticular hydrocarbon profiles of a single F. truncorum worker (Right) obtained after solid sampling (1994) and washing with pentane (2000) and plots of the first two principal components (PC1 and PC2) evaluating the quantitative variation in these profiles for a representative colony with relatively distinct patriline differences at PC2 (S6) and little (1994) or no (2000) patriline differences at PC2 (M11). The two or three patrilines are represented by different symbols. Compounds according to standard notation are: C20:0 (1), C25:0 (2), C26:0 (3), C27:0 (4), C28:0 (5), C29:0 (6), C31:0 (7) (all alkanes), and C27:1 (8) and C29:1 (9) (both alkenes). Some additional small peaks that could only be quantified in a small fraction of the ant workers have been omitted from further analysis. [Upper Right has been reproduced with permission from ref. 18 (copyright 1999, Birkhauser Publishers Ltd., Basel).] mated class of colonies. Only the comparison of patrilines nested within colonies headed by a multiply mated queen allows heritability of worker traits to be estimated with reasonable accuracy. Any other comparison (e.g., among colonies with singly mated queens, or among colonies of both matingcategories) would involve considerable uncontrolled environmental variation, inflating heritability estimates and兾or requiring unrealistically large sample sizes. Following this approach, we show that the heritable component of cuticular hydrocarbon recognition cues is low. We further provide correlative evidence that F. truncorum workers are constrained by lack of information from genetic recognition cues, and that they are thus often unable to maximize their inclusive fitness. Finally, we develop the hypothesis that these informational constraints are ultimately caused by selection against the expression of nepotistic traits. We argue that the threat of nepotism is likely to select against genetically informative recognition cues in species where colonies have a single multiply mated queen, but not in species with multiple-queen colonies or in species where colonies always have a single once-mated queen. Materials and Methods Individual profiles of worker cuticular hydrocarbons were analyzed in eight colonies that were known to have a multiply mated queen (11–13). Four of these colonies were sampled in September 1994 (Sundholmen S3, S6 and Mellanskär: M5, M11) and seven colonies were sampled in June 2000 (Joskär: J1, J2; Brännskär: B5 and B7, while resampling S6, M5, and M11). Between 80 and 100 individual worker ants from each of the 11 colony samples were genotyped for up to seven codominant allozyme and microsatellite markers and assigned to one of two or three different patrilines (11–13). The high genetic variation at the marker loci and the numbers of workers analyzed per 8800 兩 www.pnas.org兾cgi兾doi兾10.1073兾pnas.1430283100 colony (far more than used for the cuticular hydrocarbon analyses) implied that nondetection and nonsampling errors were negligible, so that all patrilines present can be assumed to have been detected (16) Solid pieces of cuticle (refs. 17 and 18; the 1994 samples) or cuticular washes with pentane (the 2000 samples) of the genotyped ants were analyzed for quantitative variation in cuticular hydrocarbon profiles. Details on the gas chromatography and mass spectrometry procedures are given in ref. 18. The speciesspecific cuticular hydrocarbon profile of F. truncorum workers consists of relatively few compounds and is dominated by saturated (alkane) hydrocarbon peaks (18). The solid sampling technique applied to the 1994 samples allowed the consistent quantification of five peaks, whereas the cuticular washes of the 2000 samples allowed the quantification of nine cuticular hydrocarbons (Fig. 1). Occasionally, small peaks of other hydrocarbons were detected as well, but these were excluded from further analysis because they failed to reach the detection limit in most ants. For a compound to be included in the analysis, we decided that it should be detectable in at least 50% of all individuals sampled and in at least one ant per patriline per colony, whereas for an individual ant to be included it could maximally have two (1994) or four (2000) missing values. This procedure excluded seven compounds and 29 ants in the 2000 data set and two compounds but no ants in the 1994 data set. The final data set analyzed thus consisted of 62 (1994) plus 216 (2000) individual ant profiles of known genotype. GC-peak areas were first expressed as percentages to adjust for differences in absolute quantities of tissue analyzed. To avoid biases caused by the compositional nature of the proportional data, the hydrocarbon peak areas were further standardized (and missing values imputed for the 2000 data) by using the procedure recommended by Aitchison (19). For each single ant the geoBoomsma et al. Results In the analyses of the two entire data sets, the first and second principal component axes (PC1 and PC2) explained 51% and 44% of the respective total variation in 1994, whereas these axes explained 35% and 23% of the total variation in 2000 (Fig. 1). The loadings were: PC1 (1994) C25:0(0.565), C27:0(0.227), C 29:0 (0.053), C 29:1 (⫺0.789), C 31:0 (⫺0.056); PC 2 (1994) C 25:0 (⫺0.491), C 27:0 (⫺0.089), C 29:0 (0.279), C 29:1 (⫺0.409), C 31:0 (0.711); PC 1 (2000) C 20:0 (0.448), C 25:0 (0.217), C 26:0 (⫺0.337), C 27:0 (0.250), C 27:1 (⫺0.219), C 28:0 (⫺0.356), C 29:0 (0.258), C 29:1 (⫺0.518), C 31:0 (0.257); PC 2 (2000) C20:0(⫺0.116), C25:0(⫺0.089), C26:0(⫺0.073), C27:0(⫺0.105), C 27:1 (⫺0.693), C 28:0 (0.459), C 29:0 (0.010), C 29:1 (0.095), C31:0(0.512). Further ANOVA of the scores at these axes showed that colonies differed significantly (1994: F3,3.7 ⫽ 9.98, P ⫽ 0.030; 2000: F6,6.0 ⫽ 13.66, P ⫽ 0.003) in their average cuticular hydrocarbon profile at the first principal component axis (PC1), but not at PC2 (1994: F3,4.9 ⫽ 0.84, P ⫽ 0.527; 2000: F6,7.4 ⫽ 1.20, Boomsma et al. P ⫽ 0.400). Patrilines nested within colonies showed an opposite pattern. There were no significant differences for scores at PC1 (1994: F6,52 ⫽ 1.13, P ⫽ 0.360; 2000: F8,201 ⫽ 0.692, P ⫽ 0.699), whereas patrilines within colonies differed significantly along PC2 (1994: F6,52 ⫽ 2.55, P ⫽ 0.031; 2000: F8,201 ⫽ 2.237, P ⫽ 0.026). In the samples of 2000, PC3 and PC4 explained a further 18% and 11% of the variation, respectively, both showing colony-level differences (P ⫽ 0.004 and 0.018, respectively), but no patriline within colony differences. The final four PC axes in these samples explained together only 13% of the variation and were ignored, as were the effects of PC3 and PC4 in the samples of 1994 (explaining together only 5% of the variation). The heritability estimates (h2) for the separate peak volumes averaged 0.22 (range 0–0.49) in 1994 and 0.031 (range 0–0.100) in 2000. For the scores at PC2, representing the only statistically significant overall component of patriline-level variation in cuticular hydrocarbon profiles, estimates of h2 ⫽ 0.44 ⫾ 0.41 (1994) and h2 ⫽ 0.16 ⫾ 0.15 (2000) were obtained. The 1994 estimate for PC2 has such wide SE’s that no firm conclusions can be drawn, though the range of peak-wise heritability estimates suggests that values are normally below 0.5. However, the 2000 estimate of h2 ⫽ 0.16 for PC2 has an upper 95% confidence limit of 0.45 and is thus significantly lower than 0.5. These estimates indicate that neither individual variation at single cuticular hydrocarbon compounds, nor the overall variation in the cumulative profiles of these hydrocarbons, provide more than highly noisy information about the relatedness structure of colonies. Moreover, colonies varied in the extent to which patrilines were segregated along PC2. Fig. 1 illustrates these differences by plotting the variation at PC1 (the nonsignificant axis for patriline differences) and PC2 (the axis showing significant overall separation of patrilines within colonies) in some of the opposite extremes in the data set. In colony S6, the patrilines were visibly segregated along the vertical axis in both years, whereas in colony M11 there was a not very distinct separation of patrilines in 1994 and essentially no separation in 2000 because the third patriline was no longer expressed. We investigated whether differences in distinctness of cuticular hydrocarbon profiles of patrilines were likely to be important when workers assess the relatedness structure of their colony to make sex allocation decisions. This would imply that colonies with poor separation of patrilines would be less likely to be unambiguously assessed as having a multiply mated queen and to male-bias the sex ratio. To test this hypothesis, we regressed the sex ratios produced in 1994, 1995, 1999, and 2000 (the clutches closest in time to the two samples of workers analyzed for cuticular hydrocarbons) against the proportion of cuticular hydrocarbon variance explained by patriline differences at PC2 (Fig. 2). Only the colonies with distinctly different patrilines produced the extremely male-biased sex ratios predicted by split sex ratio theory (14). Other colonies where multiple fathers accidentally contributed genes that gave their daughter workers similar cuticular hydrocarbon profiles did not or only slightly bias the sex ratio toward males. The overall negative relationship between the realized sex ratio and the degree of patriline differences in cuticular hydrocarbons was highly significant (P ⫽ 0.0002), and this effect remained significant when the two samples were analyzed separately (P ⫽ 0.0177 and P ⬍ 0.0001 for 1994兾1995 and 1999兾 2000, respectively). There was no consistent difference in sex allocation between the two periods (1994兾1995 versus 1999兾 2000; P ⫽ 0.671), but the interaction term between period and degree of patriline separation was significant (P ⫽ 0.0027) indicating that the regression slopes for the two periods were different. This is not surprising, as different techniques (solid sampling and pentane washes) were used to sample the hydrocarbons from the ant cuticle. Also, the 1994兾1995 and 1999兾2000 curves would have been less different in slope if we would have PNAS 兩 July 22, 2003 兩 vol. 100 兩 no. 15 兩 8801 EVOLUTION metric mean of peak areas was calculated, g(x) ⫽ (⌸xj)1/D, where xj is the peak area of the jth hydrocarbon and D is the number of hydrocarbons. The standardized peak areas were subsequently obtained by zj ⫽ ln[xj兾g(x)] and used in principal component analyses. We tested the data for homogeneity of variances with Levene’s tests, which gave P ⬎ 0.05 in all cases after Bonferroni correction. Principal component analysis (PCA) was used to visualize the major part of the total variation in cuticular hydrocarbon profiles of workers. Because of the different extraction methods and peak numbers, the samples of 1994 and 2000 were analyzed separately to evaluate whether colony-level differences and patriline-level differences in cuticular hydrocarbon profiles tended to be consistently explained by different axes. Further analyses were done with nested ANOVA, either on the transformed peak volumes separately, or on the PCA scores at an axis that captured the overall significant patriline effect and thus the main genetic component in the cuticular hydrocarbon profiles. The colony-level variance component reflects both genetic and environmental differences, but the variance explained at the (nested) patriline-level is due only to genetic differences among the fathers, because half-sib worker offspring have everything else (maternal genes, phenotypic maternal effects, nest environment) in common. The maternal half-sib variance component thus produces an estimate of the narrow sense heritablity (h2) of the hydrocarbon variation, both for each compound separately and for the scores on any PC-axis that would represent a significant overall variance component at the patriline level. Mean heritabilities (h2) were obtained by dividing the proportion of variance explained at the nested patriline level (t) by 0.5 (20–22). The SE of h2 was defined as the square root of the variance of a heritability estimate. This variance was calculated following Searle et al. (21), taking into account that the ANOVA design used to estimate the variance components was unbalanced (22). Sex ratio estimates were based on samples of 20–100 individuals (mostly pupae) per colony, and collection was timed so that all larvae had pupated and pupae were differentiated to the extent that they could be sexed. Sex ratio data from the colonies included in this study were collected almost yearly from 1989 to 2002 (9–11), but only the 2 years of sex ratio data closest to the moment of worker sampling were used in the analysis. The data were analyzed with logistic regression taking some missing values and the actual number of females and males in each sample into account (23). Predictor variables entered into the model were the period of sampling (categorial), the percentage variance explained by patriline (continuous), and the actual relatedness asymmetry (paternity skew) (13) (continuous) in each of the colonies. Fig. 2. Sex ratios (proportion of females) of F. truncorum colonies with a multiply mated queen as a function of the proportion of variance in worker cuticular hydrocarbon profiles (PC2) explained by patriline (see Fig. 1). Open dots are colonies sampled in 1994. Black dots are colonies sampled in 2000. Curves are logistic regression lines. See text for further details. expressed the x axis scores in Fig. 2 on a relative scale, taking into account that PC2 explained 44% of the total variance in 1994, but only 23% in 2000. The observed relatedness asymmetry (paternity skew) per colony was not significant (P ⬎ 0.5) in any of the analyses, but tended to be in the expected direction (more female bias with increasing paternity skew) (13). This implies that the number of colonies sampled was probably too small to shed light on the possible interaction between the average chemical (cuticular hydrocarbon based) separation among patrilines, and the relative representation of patrilines (of a specific degree of chemical distinctness) on the assessment of colony kin-structure by workers. Discussion Informational Constraints on Worker Inclusive Fitness. Our results indicate that workers of F. truncorum have the chemical information on intra-colony relatedness that is required to realize adaptive split sex ratios. However, heritabilities of these recognition兾assessment cues are low, and the correlation between the chemical distinctness of patrilines and male bias in the sex ratio (Fig. 2) shows that the inclusive fitness of workers in colonies headed by a multiply mated queen is constrained by lack of information. In other words, workers in colonies with relatively nondistinct patrilines are less likely to assess the mating status of their mother correctly and tend to invest too much in female reproductives. It is important to note that the proximate mechanism of liquid food exchange between nestmates (trophallaxis) implies that there may be more genetic variation in cuticular hydrocarbon profiles of individuals than is actually expressed. Trophallaxis is common in Formica ants (24), where it primarily serves to redistribute resources through the colony. However, 8802 兩 www.pnas.org兾cgi兾doi兾10.1073兾pnas.1430283100 trophallaxis has recently also been shown to induce a considerable mixing of individual cuticular hydrocarbons via the postpharyngeal gland of workers (8), creating a relatively homogeneous colony ‘‘gestalt’’ (25). This implies that the expression of heritable variation in cuticular hydrocarbon profiles is likely to be a direct consequence of trophallactic activity and may thus vary with general metabolic activity of ants across the seasons. The fitness penalty of producing too many females in the multiply mated class of colonies can only be estimated when the frequency of multiple queen mating is constant over time and when the overall sex allocation ratio of the entire population approaches equilibrium (14). To a reasonable approximation this is indeed the case (L.S., unpublished data). The proportion of colonies with singly and multiply mated queens has been fairly stable since 1989 around an average of 39% multiply mated queens. In addition, the average population-wide sex allocation (proportional investment in females; see ref. 11 for methods) has been 0.67 (2:1), which is exactly the predicted optimum value assuming no assessment errors (14) and which is statistically indistinguishable from the predicted optimum value when assessment errors in colonies with a singly mated queen are low compared with those in colonies with a multiply mated queen) (see upper right panel of figure 2 in ref. 15). The observed range of yearly sex allocation totals for the multiply mated class of colonies (0.19–0.34) (L. Sundström, unpublished results) is more female biased than the error-free model predicts (⬇10% investment in females, see refs. 14 and 15), and adding assessment errors for the multiply mated class shift this predicted optimum to 100% investment in males (15), increasing the mismatch between prediction and observation. The inference that assessment errors in colonies with a singly mated queen are lower than assessment errors in colonies with a multiply mated queen is based on correlative evidence. In the singly mated class, the sex ratio is positively correlated with colony productivity (P ⫽ 0.04), suggesting that small colonies with a singly mated queen have ergonomic reasons for avoiding complete specialization on female reproductives (the more expensive sex) (see also ref. 26), rather than that workers wrongly assess the kin structure of their colony. The same correlation is nonsignificant (with a trend in the opposite direction) in the colonies with a multiply mated queen (P ⫽ 0.32), suggesting that assessment errors are the main causal factor here. Both these analyses had ‘‘year’’ included, but this factor did not have a significant effect (F7,84 ⫽ 0.37; P ⫽ 0.92 and F7,56 ⫽ 0.60; P ⫽ 0.75, for colonies with a singly and multiply mated queen, respectively). Workers in colonies headed by a multiply mated queen thus realize maximal inclusive fitness by specializing on rearing males, and their inclusive fitness declines monotonously with increasing female bias. Note that the above evaluation of the relative magnitude of assessment errors in the two classes of colony is opposite to a tentative appraisal by Ratnieks and Boomsma (15), which was based on a much smaller set of samples (11) and which did not take differential correlations with colony productivity into account. When split sex ratio theory was formulated (14), it was stressed that the assessment of queen mating status by workers was likely to be based on the same genetically based cues that may allow nepotistic discrimination of kin of varying relatedness to self. However, it was also expected that kin assessment would be easier to achieve, because it could be based on cumulative information obtained over longer periods, whereas nepotistic individual recognition would require instantaneous decisions. With the present evidence (Fig. 2) that genetic cues contain limited information and apparently constrain kin assessment, it seems unlikely that the same cues would normally allow nepotistic discrimination to be effective, despite the 3-fold relatedness difference between full sisters (relatedness ⫽ 0.75) and half sisters (relatedness ⫽ 0.25). The question remains, however, why Boomsma et al. Inclusive Fitness Predictions of the Number and Chemical Structure of Recognition Compounds. In the following paragraphs, we will explore the hypothesis that the rare expression of nepotistic discrimination is likely to be responsible for the observed lack of genetic variation in recognition cues in F. truncorum. Throughout, we will assume that there is a colony-level cost to the expression of nepotism (1, 3). This implies that colonies that by chance have sufficient variation in genetic recognition cues to allow nepotism to be expressed will leave fewer offspring. The erosion of genetic recognition cues due to the expression of nepotism therefore has, in all likelihood, both a group-selection component and a kin-selection component, which can be treated simultaneously by covariance equations as developed by Price (see ref. 27). These combined selection forces should work in the same direction and are expected to result in both a low number of recognition compounds and a low degree of information per compound. By implication and all other things being equal, these two characteristics should then not only have evolved in F. truncorum but also independently in other social insects that have colonies with a single, multiply mated queen. When it could further be shown that the pay-off of nepotistic behavior is fundamentally different in multiple-queen (polygynous) insect societies, we would have arrived at the first inclusive fitness predictions of the adaptive chemical structure of recognition compounds in insect societies of specific kin structure. The typical cuticular hydrocarbon profile of workers in the monogynous study population of F. truncorum is indeed poor in peaks (Fig. 1), at least compared with the 30–60 compounds that are normally found on the cuticle of ants (8). Even neighboring polygynous populations of F. truncorum in Finland have many more detectable peaks (typically 20–25; N.J.O., unpublished data). In addition, the worker profiles in the study population are dominated by straight-chain saturated cuticular hydrocarbons (alkanes), which are considered to be relatively ‘‘uninformative,’’ because they lack the functional groups or specific shape that typically stimulate insect chemo-receptors, compared with unsaturated (alkene) and methyl-branched hydrocarbons (28). A straight-chain, alkane-dominated profile may thus be insufficiently informative for individual recognition, but still adequate for patriline assessment based on long-term cumulative evaluation of genetic diversity among nestmates. Experimental work applying single synthesized hydrocarbons on the cuticle of workers of Polistes wasps has recently corroborated that alkanes do not elicit individual recognition reactions, whereas alkenes and methyl-branched hydrocarbons do (29). Do we find a similar combination of cuticular hydrocarbon characteristics in other social insects with a single multiply mated queen per colony? The only other well-studied system with full-sister兾half-sister societies is the honey bee, Apis mellifera. Honey bees have more cuticular hydrocarbon compounds than F. truncorum (30, 31), but also in this species the typical cuticular hydrocarbon profiles of workers are characterized by relatively few straight chain hydrocarbons (28, 30, 31). The cuticular hydrocarbon profiles of honey bee patrilines overlap, but are segregated more consistently than in F. truncorum (32). No heritability estimates of cuticular hydrocarbon characteristics in honey bees are available and neither have the reproductive consequences of variation in patriline segregation been studied. However, even moderately high heritabilities would not necessarily imply that nepotistic discrimination is easier, as the number of patrilines in a typical honey bee colony is an order of magnitude higher than in F. truncorum (32). Although honey bee workers can discriminate between mixtures of C23 and C25 Boomsma et al. alkanes (33), it seems reasonable to hypothesize that the lack of more complex and thus more informative cuticular hydrocarbon profiles may have evolved in response to the expression of nepotistic tendencies in a low and thus difficult to detect frequency (1, 3, 28, 31). The few studies that have demonstrated nepotistic tendencies in honey bees are consistent with this interpretation. They were all made in colonies where patrilines were unnaturally low in number (after artificial insemination) or unnaturally distinct in their genetic differences (using color morphs or different races) (reviewed in ref. 34), so that the productivity costs of nepotism and兾or the errors of discrimination were unusually low. The hypothesis that the occasional expression of nepotistic traits would invoke selection against kin-informative recognition compounds can be independently tested in other social insect genera where colonies have a single, multiply mated queen. Our hypothesis thus predicts simple cuticular hydrocarbon profiles, similar to those of F. truncorum and honey bees, to prevail in Acromyrmex and Atta leafcutter ants (35), Pogonomyrmex harvester ants (36, 37), and Vespula wasps (38). In addition, heritabilities of other compounds such as glandular secretions that could secondarily be used for kin-recognition would also be expected to be low. This matches the recent finding of unexpectedly small patriline differences in the qualitative and quantitative profiles of metapleural gland secretions of Acromyrmex octospinosus (39). However, polygynous societies may well be characterized by opposite, highly informative recognition cues. These societies are principally open for additional reproductive females to join, which implies that queen–queen competition for attention from the workers is probably frequent and that queens advertise their fertility by chemical signals (40). Polygyny also incurs a constant risk of usurpation or social parasitism by unrelated individuals. In such contexts, kin-informative recognition cues may thus be favoured by selection, because the inclusive fitness penalty of having to raise nonrelatives instead of relatives is much higher than the 3-fold reduction of inclusive fitness after the rearing of half sisters that could have been full sisters. Even if relatedness is lower in polygynous colonies (say, 0.3), the ratio 0.3兾0 (when rearing nonrelatives is the alternative) is infinite, and it is this ratio (not the absolute difference in relatedness) that would maintain selection for genetically informative cues. The recent finding of nepotism in the polygynous ant Formica fusca (6) is therefore not at variance with the absence of informative recognition cues in monogynous F. truncorum, provided that future work will show that the cuticular hydrocarbon profiles of F. fusca workers are significantly more informative than those of monogynous F. truncorum workers. A recent study identifying 61 cuticular hydrocarbons in queens of this F. fusca population (40) suggests that this may well be the case. Informative cuticular hydrocarbon profiles are also expected in species that invariably have a single once mated queen per colony, because highly heritable recognition cues will likely be adaptive for recognizing nonnestmates, whereas the issue of discrimination between full-sisters and half-sisters does not apply. Conclusion Our present results are consistent with selective fine-tuning of genetic odor cue diversity to a level preventing selfish exploitation. They indicate that, in closed ant societies consisting of full-sisters and half-sisters, selection is more likely to favor and maintain the collective interests of all workers than the selfish interests of groups (patrilines) or individuals. Information on kinship is sufficient to allow F. truncorum workers to collectively increase and occasionally maximise their inclusive fitness by splitting the sex ratio, but is unlikely to be sufficient to let kin selection operate at the individual level through the expression of nepotism. We cannot exclude the possibility (and find it in fact PNAS 兩 July 22, 2003 兩 vol. 100 兩 no. 15 兩 8803 EVOLUTION cuticular hydrocarbon profiles in F. truncorum do not contain more information on kinship, so that colony-level kin assessment would be more effective and worker inclusive fitness would end up closer to the maximum possible value. quite likely) that kin assessment in F. truncorum is based on information additional to the cuticular hydrocarbons that we were able to analyze here. However, if there would be such a larger set of relevant recognition cues, e.g., additional volatile compounds (41) or glandular secretions (39), there is no reason to assume that these cues would not be subject to the same selection forces on information content, so that heritability estimates would likewise be low. The similarities in cuticular hydrocarbon characteristics of workers of honey bees and F. truncorum ants suggest that multilevel inclusive fitness theory (27) may ultimately have the potential to make rather precise and kin-structure-specific predictions about the type of recognition cues (e.g., alkanes versus alkenes and methyl-branched hydrocarbons) and the magnitude of genetic versus environmental We thank K. Bargum, S. Barker, P. Gertsch, I. Hanski, G. de Jóng, L. Keller, A. P. Møller, D. R. Nash, B. Oldroyd, D. Ortius-Lechner, and F. Ratnieks for comments and discussion, D. R. Nash and G. Nachman for assistance with the final analyses and editing, and three anonymous reviewers for constructive comments. We were supported by the Danish Natural Science Research Council, the European Union ResearchTraining Networks ‘‘Social Evolution’’ (FW4) and ‘‘INSECTS’’ (FW5), the NorFa research network in Conservation Biology, Academy of Finland Grant 42725, and Tvärminne Zoological Station. Ratnieks, F. L. W. & Reeve, H. K. (1992) J. Theor. Biol. 158, 33–65. Hamilton, W. D. (1964) J. Theor. Biol. 7, 1–16. Keller, L. (1997) Trends Ecol. Evol. 12, 99–103. Sherman, P. W., Reeve, H. K. & Pfennig, D. W. (1997) in Behavioural Ecology: An Evolutionary Approach, eds. Krebs, J. R. & Davies, N. B. (Blackwell, Oxford), pp. 69–96. Fletcher, D. J. C. & Michener, C. D., eds. (1987) Kin Recognition in Animals (Wiley, Chichester, U.K.), 465 pp. Hannonen, M. & Sundström, L. (2003) Nature 421, 910. Tibbets, E. A. (2002) Proc. R. Soc. London Ser. B 269, 1423–1428. Lenoir, A., Fresneau, D., Errard, C. & Hefetz, A. (1999) in Information Processing in Social Insects, eds. Detrain, C., Deneubourg, J.-L. & Pasteels, J. M. (Birkhäuser, Basel), pp. 219–237. Crozier, R. H. (1988) in Invertebrate Historecognition, eds. Grosberg, R. K., Hedgecock, D. & Nelson, D. (Plenum, New York), pp. 143–156. Ratnieks, F. L. W. (1991) Am. Nat. 137, 202–226. Sundström, L. (1994) Nature 367, 266–268. Sundström, L. & Ratnieks, F. L. W. (1998) Behav. Ecol. 9, 116–121. Sundström, L. & Boomsma, J. J. (2000) Proc. R. Soc. London Ser. B 267, 1439–1444. Boomsma, J. J. & Grafen, A. (1991) J. Evol. Biol. 4, 383–407. Ratnieks, F. L. W. & Boomsma, J. J. (1997) J. Theor. Biol. 185, 423–439. Boomsma, J. J. & Ratnieks, F. L. W. (1996) Philos. Trans. R. Soc. London Ser. B 351, 947–975. Bagnères, A. G. & Morgan, E. D. (1990) J. Chem. Ecol. 16, 3263–3276. Nielsen, J., Boomsma, J. J., Oldham, N. J., Petersen, H. C. & Morgan, E. D. (1999) Insectes Soc. 46, 58–65. Aitchison, J. (1986) The Statistical Analysis of Compositional Data (Chapman & Hall, London). Oldroyd, B. & Moran, C. (1983) Aust. J. Biol. Sci. 36, 323–332. Searle, S. R., Casella, G. & McCulloch, C. E. (1992) Variance Components (Wiley, New York), pp. 429–431. 22. Becker, W. A. (1992) Manual of Quantitative Genetics (Academic Enterprises, Pullman, WA), 5th Ed, pp. 55–66. 23. Boomsma, J. J. & Nachman, G. (2002) in Sex Ratios: Concepts and Research Methods, ed. Hardy, I. C. W. (Cambridge Univ. Press, Cambridge, U.K.), pp. 93–111. 24. Hölldobler, B. & Wilson, E. O. (1990) The Ants (Belknap, Cambridge, MA) 25. Cozier, R. H. & Dix, M. W. (1979) Behav. Ecol. Sociobiol. 4, 217–224. 26. Sundström, L. (1995) Am. Nat. 146, 182–201. 27. Hamilton, W. D. (1975) in Biosocial Anthropology, ed. Fox, R. (Malaby, London), Vol. 4, pp. 133–153. 28. Breed, M. D. (1998) in Pheromone Communication in Social Insects, eds. Vander Meer, R. K., Breed, M. D., Espelie, K. E. & Winston, M. L. (Westview, Boulder, CO), pp. 57–78. 29. Dani, F. R., Jones, G. R., Destri, S., Spencer, S. H. & Turillazzi, S. (2001) Anim. Behav. 62, 165–171. 30. Getz, W. M. & Page, R. E., Jr. (1991) Ethology 87, 298–315. 31. Page, R. E., Jr., Metcalf, R. A., Metcalf, R. L., Erickson, E. H., Jr., & Lampman, R. L. (1991) J. Chem. Ecol. 17, 745–756. 32. Arnold, G., Quenet, B., Cornuet, J.-M., Masson, C., De Schepper, B., Estoup, A. & Gasqui, P. (1996) Nature 379, 489. 33. Getz, W. M. & Smith, K. B. (1987) J. Comp. Physiol. A 160, 239–245. 34. Breed, M. D., Welch, C. K. & Cruz, R. (1994) Behav. Proc. 33, 25–40. 35. Villesen, P., Murakami, T., Schultz, T. R. & Boomsma, J. J. (2002) Proc. R. Soc. London Ser. B 269, 1541–1548. 36. Hölldobler, B. (1976) Behav. Ecol. Sociobiol. 1, 405–423. 37. Cole, B. J. & Wiernasz, D. C. (1999) Science 285, 891–893. 38. Ross, K. G. (1986) Nature 323, 798–800. 39. Ortius-Lechner, D. Maile, R., Morgan, E. D., Petersen, H. C. & Boomsma, J. J. (2003) Insectes Soc. 50, 113–119. 40. Hannonen, M., Sledge, M. F., Turillazzi, S. & Sundström, L. (2002) Anim. Behav. 64, 477–485. 41. Getz, W. M., Brückner, D. & Smith, K. B. (1986) J. Comp. Physiol. A 159, 251–256. 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 8804 兩 www.pnas.org兾cgi兾doi兾10.1073兾pnas.1430283100 View publication stats variation in these recognition cues. If these predictions would be confirmed, they would further add to the notion that social cohesion of complex insect societies is less surprising than it seems, even when inclusive fitness theory based on reproductive conflict is the basic organizing principle. Boomsma et al.