Informational constraints on optimal sex allocation in

Jacobus J. Boomsma*†‡, Jannie Nielsen†, Liselotte Sundstro¨ m†§, Neil J. Oldham¶ʈ, 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

Edited by Bert Ho¨lldobler, University of Wu¨rzburg, Wu¨rzburg, Germany, and approved May 14, 2003 (received for review January 16, 2003) Workers of the truncorum specialize in rearing females cuticle, is increasingly considered to be the most generally or males depending on the number of fathers of a colony. These relevant category of recognition cues in societies (8). split sex ratios increase inclusive fitness, but it has remained However, the maintenance of genetic variation for this kind of unknown how workers assess the number of patrilines in their recognition cues is problematic: newly arising genetic cues are colonies and to what extent their reproductive decisions are likely to be selected for initially when they increase the efficiency constrained by lack of information. By analysis of the quantitative of collective discrimination against nonkin from neighboring variation in cuticular hydrocarbon profiles of workers of multiply colonies, but at some point these cues become variable (and thus mated queens, we show that the heritable component of recog- informative) enough to be exploited by emerging nepotistic nition cues is low and that the extent of sex ratio biasing toward phenotypes. The expression of nepotistic conflict will then males is correlated with patriline differences in hydrocarbon pro- prevent further increase of genetic variation for recognition files. Workers are thus able to capitalize on colony-level related- cues, both because of colony-level productivity costs (1, 3), and ness asymmetry, but their inclusive fitness is constrained by un- because the expression of nepotism per se tends to erode the informative recognition cues. These results are consistent with the genetic variation for the cues on which it is based (9, 10). hypothesis that the occasional expression of nepotistic phenotypes The present study uses a population of the ant Formica favoring full-sisters over half-sisters maintains selection against truncorum in the Tva¨rminne archipelago of Southern Finland, informative recognition cues. We evaluate how inclusive fitness where reproductive decision-making has been studied in such theory may be used to predict the number and kind of recognition detail (11–13), that quantitative predictions on extant genetic cues in insect societies of a specific relatedness structure. variation for recognition cues can be tested and the fitness consequences of informational constraints can be evaluated. split sex ratios ͉ inclusive fitness ͉ chemical recognition ͉ cuticular Colonies in the study population have a single queen per nest, hydrocarbons ͉ nepotism but this queen is mated with one or several males. The ensuing variation in relatedness asymmetry of workers toward the re- very kin-structured society is a delicate balance between productive females and males allows the precise prediction of the cooperation for common prosperity and potentially destruc- optimal sex ratio, provided that workers can control investments E and are able to assess what type of colony they belong to: workers tive conflicts between subgroups or individuals. The expression of a singly mated queen should specialize on females, whereas of such conflicts may increase the fitness of some individuals, but workers of a multiply mated queen should specialize on males may entail collective costs when it disrupts social cohesion and (14). Empirical data collected during a period of 10 years have reduces productivity (1). Principles of cooperation and conflict shown that these split sex ratio predictions, based on the inferred in families can generally be understood from Hamilton’s rule (2), ability of workers to assess the number of patrilines, are generally but the key variables in the equation (relatedness and the costs met (11–13). Because it is inconceivable that workers can directly and benefits of helping), may both explain indiscriminate coop- assess the number of ejaculates stored by their mother-queen, eration among relatives of any degree, and nepotistic discrimi- their ability to assess the number of patrilines must be based on nation between relatives of different relatedness. This paradox the expression of heritable variation in cuticular hydrocarbon can only be resolved when recognition efficiency, i.e., the cues among nestmates (8). F. truncorum is therefore one of the availability and perception of information on relatedness, is few social insect model systems where genetic variation for explicitly considered (3). Accurate information allows efficient recognition cues is known to be present. This is because split sex EVOLUTION recognition of individual degrees of kinship promoting the ratios induced by relatedness asymmetry of workers toward the expression of conflict, whereas noisy recognition cues could female and male brood are only possible when half-sister colo- make the cost of nepotism greater than the benefit and thus lead nies express more variation in recognition cues than full-sister to unconditional cooperation among relatives, while still allow- colonies, and when the workers in each colony collectively assess ing discrimination of nonkin. this variation to increase their inclusive fitness (14). The ability to differentiate between kin of different degree has However, the match between the observed sex ratios (11, 12) been amply documented in vertebrate societies (4, 5), but has and the theoretical predictions is not perfect, and it has remained remained difficult to prove in insect societies (ref. 3, but see ref. unclear whether the unexplained variation in sex allocation is 6). Likely reasons for this difference might be that insect colonies caused by environmental noise or by inaccurate genetic infor- are larger than vertebrate societies, whereas the learning capac- mation for the assessment of queen mating status (15). The best ities of are limited compared with vertebrates (4). In way to address these questions is to concentrate on the multiply 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, This paper was submitted directly (Track II) to the PNAS office. which can recognize facial characteristics of nestmates (7). Kin ‡To whom correspondence should be addressed. E-mail: [email protected]. recognition relies thus on chemical cues in most cases. A single ʈPresent address: The Dyson Perrins Laboratory, Oxford University, South Parks Road, class of compounds, the hydrocarbons of the wax layer on the Oxford OX1 3QY, England.

www.pnas.org͞cgi͞doi͞10.1073͞pnas.1430283100 PNAS ͉ July 22, 2003 ͉ vol. 100 ͉ no. 15 ͉ 8799–8804 Downloaded by guest on September 25, 2021 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 colony (far more than used for the cuticular hydrocarbon within colonies headed by a multiply mated queen allows analyses) implied that nondetection and nonsampling errors heritability of worker traits to be estimated with reasonable were negligible, so that all patrilines present can be assumed to accuracy. Any other comparison (e.g., among colonies with have been detected (16) singly mated queens, or among colonies of both mating- Solid pieces of cuticle (refs. 17 and 18; the 1994 samples) or categories) would involve considerable uncontrolled environ- cuticular washes with pentane (the 2000 samples) of the geno- mental variation, inflating heritability estimates and͞or requir- typed ants were analyzed for quantitative variation in cuticular ing unrealistically large sample sizes. Following this approach, hydrocarbon profiles. Details on the gas chromatography and we show that the heritable component of cuticular hydrocarbon mass spectrometry procedures are given in ref. 18. The - recognition cues is low. We further provide correlative evidence specific cuticular hydrocarbon profile of F. truncorum workers that F. truncorum workers are constrained by lack of information consists of relatively few compounds and is dominated by from genetic recognition cues, and that they are thus often saturated (alkane) hydrocarbon peaks (18). The solid sampling unable to maximize their inclusive fitness. Finally, we develop technique applied to the 1994 samples allowed the consistent the hypothesis that these informational constraints are ulti- quantification of five peaks, whereas the cuticular washes of the mately caused by selection against the expression of nepotistic 2000 samples allowed the quantification of nine cuticular hy- traits. We argue that the threat of nepotism is likely to select drocarbons (Fig. 1). Occasionally, small peaks of other hydro- against genetically informative recognition cues in species where carbons were detected as well, but these were excluded from colonies have a single multiply mated queen, but not in species further analysis because they failed to reach the detection limit with multiple-queen colonies or in species where colonies always in most ants. For a compound to be included in the analysis, we have a single once-mated queen. decided that it should be detectable in at least 50% of all individuals sampled and in at least one ant per patriline per Materials and Methods colony, whereas for an individual ant to be included it could Individual profiles of worker cuticular hydrocarbons were ana- maximally have two (1994) or four (2000) missing values. This lyzed in eight colonies that were known to have a multiply mated procedure excluded seven compounds and 29 ants in the 2000 queen (11–13). Four of these colonies were sampled in Septem- data set and two compounds but no ants in the 1994 data set. The ber 1994 (Sundholmen S3, S6 and Mellanska¨r: M5, M11) and final data set analyzed thus consisted of 62 (1994) plus 216 (2000) seven colonies were sampled in June 2000 (Joska¨r: J1, J2; individual ant profiles of known genotype. Bra¨nnska¨r: B5 and B7, while resampling S6, M5, and M11). GC-peak areas were first expressed as percentages to adjust Between 80 and 100 individual worker ants from each of the 11 for differences in absolute quantities of tissue analyzed. To avoid colony samples were genotyped for up to seven codominant biases caused by the compositional nature of the proportional allozyme and microsatellite markers and assigned to one of two data, the hydrocarbon peak areas were further standardized (and or three different patrilines (11–13). The high genetic variation missing values imputed for the 2000 data) by using the procedure at the marker loci and the numbers of workers analyzed per recommended by Aitchison (19). For each single ant the geo-

8800 ͉ www.pnas.org͞cgi͞doi͞10.1073͞pnas.1430283100 Boomsma et al. Downloaded by guest on September 25, 2021 1/D metric mean of peak areas was calculated, g(x) ϭ (⌸xj) , where P ϭ 0.400). Patrilines nested within colonies showed an opposite xj is the peak area of the jth hydrocarbon and D is the number pattern. There were no significant differences for scores at PC1 of hydrocarbons. The standardized peak areas were subse- (1994: F6,52 ϭ 1.13, P ϭ 0.360; 2000: F8,201 ϭ 0.692, P ϭ 0.699), quently obtained by zj ϭ ln[xj͞g(x)] and used in principal whereas patrilines within colonies differed significantly along component analyses. We tested the data for homogeneity of PC2 (1994: F6,52 ϭ 2.55, P ϭ 0.031; 2000: F8,201 ϭ 2.237, P ϭ variances with Levene’s tests, which gave P Ͼ 0.05 in all cases 0.026). In the samples of 2000, PC3 and PC4 explained a further after Bonferroni correction. Principal component analysis 18% and 11% of the variation, respectively, both showing (PCA) was used to visualize the major part of the total variation colony-level differences (P ϭ 0.004 and 0.018, respectively), but in cuticular hydrocarbon profiles of workers. Because of the no patriline within colony differences. The final four PC axes in different extraction methods and peak numbers, the samples of these samples explained together only 13% of the variation and 1994 and 2000 were analyzed separately to evaluate whether were ignored, as were the effects of PC3 and PC4 in the samples colony-level differences and patriline-level differences in cutic- of 1994 (explaining together only 5% of the variation). ular hydrocarbon profiles tended to be consistently explained by The heritability estimates (h2) for the separate peak volumes different axes. Further analyses were done with nested ANOVA, averaged 0.22 (range 0–0.49) in 1994 and 0.031 (range 0–0.100) either on the transformed peak volumes separately, or on the in 2000. For the scores at PC2, representing the only statistically PCA scores at an axis that captured the overall significant significant overall component of patriline-level variation in patriline effect and thus the main genetic component in the cuticular hydrocarbon profiles, estimates of h2 ϭ 0.44 Ϯ 0.41 cuticular hydrocarbon profiles. (1994) and h2 ϭ 0.16 Ϯ 0.15 (2000) were obtained. The 1994 The colony-level variance component reflects both genetic estimate for PC2 has such wide SE’s that no firm conclusions can and environmental differences, but the variance explained at the be drawn, though the range of peak-wise heritability estimates (nested) patriline-level is due only to genetic differences among suggests that values are normally below 0.5. However, the 2000 2 the fathers, because half-sib worker offspring have everything estimate of h ϭ 0.16 for PC2 has an upper 95% confidence limit else (maternal genes, phenotypic maternal effects, nest environ- of 0.45 and is thus significantly lower than 0.5. These estimates ment) in common. The maternal half-sib variance component indicate that neither individual variation at single cuticular thus produces an estimate of the narrow sense heritablity (h2) hydrocarbon compounds, nor the overall variation in the cumu- of the hydrocarbon variation, both for each compound separately lative profiles of these hydrocarbons, provide more than highly and for the scores on any PC-axis that would represent a noisy information about the relatedness structure of colonies. significant overall variance component at the patriline level. Moreover, colonies varied in the extent to which patrilines were 2 Mean heritabilities (h ) were obtained by dividing the proportion segregated along PC2. Fig. 1 illustrates these differences by of variance explained at the nested patriline level (t)by plotting the variation at PC1 (the nonsignificant axis for patriline 2 0.5 (20–22). The SE of h was defined as the square root of differences) and PC2 (the axis showing significant overall sep- the variance of a heritability estimate. This variance was calcu- aration of patrilines within colonies) in some of the opposite lated following Searle et al. (21), taking into account that the extremes in the data set. In colony S6, the patrilines were visibly ANOVA design used to estimate the variance components was segregated along the vertical axis in both years, whereas in unbalanced (22). colony M11 there was a not very distinct separation of patrilines Sex ratio estimates were based on samples of 20–100 individ- in 1994 and essentially no separation in 2000 because the third uals (mostly pupae) per colony, and collection was timed so that patriline was no longer expressed. all larvae had pupated and pupae were differentiated to the We investigated whether differences in distinctness of cutic- extent that they could be sexed. Sex ratio data from the colonies ular hydrocarbon profiles of patrilines were likely to be impor- included in this study were collected almost yearly from 1989 to tant when workers assess the relatedness structure of their 2002 (9–11), but only the 2 years of sex ratio data closest to the colony to make sex allocation decisions. This would imply that moment of worker sampling were used in the analysis. The data colonies with poor separation of patrilines would be less likely to were analyzed with logistic regression taking some missing values be unambiguously assessed as having a multiply mated queen and the actual number of females and males in each sample into and to male-bias the sex ratio. To test this hypothesis, we account (23). Predictor variables entered into the model were regressed the sex ratios produced in 1994, 1995, 1999, and 2000 the period of sampling (categorial), the percentage variance (the clutches closest in time to the two samples of workers explained by patriline (continuous), and the actual relatedness analyzed for cuticular hydrocarbons) against the proportion of asymmetry (paternity skew) (13) (continuous) in each of the cuticular hydrocarbon variance explained by patriline differ- colonies. ences at PC2 (Fig. 2). Only the colonies with distinctly different patrilines produced the extremely male-biased sex ratios pre- Results dicted by split sex ratio theory (14). Other colonies where EVOLUTION In the analyses of the two entire data sets, the first and second multiple fathers accidentally contributed genes that gave their principal component axes (PC1 and PC2) explained 51% and daughter workers similar cuticular hydrocarbon profiles did not 44% of the respective total variation in 1994, whereas these axes or only slightly bias the sex ratio toward males. explained 35% and 23% of the total variation in 2000 (Fig. 1). The overall negative relationship between the realized sex The loadings were: PC1 (1994) C25:0(0.565), C27:0(0.227), ratio and the degree of patriline differences in cuticular hydro- C29:0(0.053), C29:1(Ϫ0.789), C31:0(Ϫ0.056); PC2 (1994) carbons was highly significant (P ϭ 0.0002), and this effect C25:0(Ϫ0.491), C27:0(Ϫ0.089), C29:0(0.279), C29:1(Ϫ0.409), remained significant when the two samples were analyzed sep- C31:0(0.711); PC1 (2000) C20:0(0.448), C25:0(0.217), arately (P ϭ 0.0177 and P Ͻ 0.0001 for 1994͞1995 and 1999͞ C26:0(Ϫ0.337), C27:0(0.250), C27:1(Ϫ0.219), C28:0(Ϫ0.356), 2000, respectively). There was no consistent difference in sex C29:0(0.258), C29:1(Ϫ0.518), C31:0(0.257); PC2 (2000) allocation between the two periods (1994͞1995 versus 1999͞ C20:0(Ϫ0.116), C25:0(Ϫ0.089), C26:0(Ϫ0.073), C27:0(Ϫ0.105), 2000; P ϭ 0.671), but the interaction term between period and C27:1(Ϫ0.693), C28:0(0.459), C29:0(0.010), C29:1(0.095), degree of patriline separation was significant (P ϭ 0.0027) C31:0(0.512). Further ANOVA of the scores at these axes showed indicating that the regression slopes for the two periods were that colonies differed significantly (1994: F3,3.7 ϭ 9.98, P ϭ 0.030; different. This is not surprising, as different techniques (solid 2000: F6,6.0 ϭ 13.66, P ϭ 0.003) in their average cuticular sampling and pentane washes) were used to sample the hydro- hydrocarbon profile at the first principal component axis (PC1), carbons from the ant cuticle. Also, the 1994͞1995 and 1999͞2000 but not at PC2 (1994: F3,4.9 ϭ 0.84, P ϭ 0.527; 2000: F6,7.4 ϭ 1.20, curves would have been less different in slope if we would have

Boomsma et al. PNAS ͉ July 22, 2003 ͉ vol. 100 ͉ no. 15 ͉ 8801 Downloaded by guest on September 25, 2021 trophallaxis has recently also been shown to induce a consider- able mixing of individual cuticular hydrocarbons via the post- pharyngeal gland of workers (8), creating a relatively homoge- neous 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 as- sessment 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. Sundstro¨m, unpublished results) is more female biased than the error-free model predicts (Ϸ10% 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. Fig. 2. Sex ratios (proportion of females) of F. truncorum colonies with a The inference that assessment errors in colonies with a singly multiply mated queen as a function of the proportion of variance in worker mated queen are lower than assessment errors in colonies with cuticular hydrocarbon profiles (PC2) explained by patriline (see Fig. 1). Open a multiply mated queen is based on correlative evidence. In the dots are colonies sampled in 1994. Black dots are colonies sampled in 2000. singly mated class, the sex ratio is positively correlated with Curves are logistic regression lines. See text for further details. colony productivity (P ϭ 0.04), suggesting that small colonies with a singly mated queen have ergonomic reasons for avoiding expressed the x axis scores in Fig. 2 on a relative scale, taking into complete specialization on female reproductives (the more expensive sex) (see also ref. 26), rather than that workers wrongly account that PC2 explained 44% of the total variance in 1994, but only 23% in 2000. The observed relatedness asymmetry (pater- assess the kin structure of their colony. The same correlation is nity skew) per colony was not significant (P Ͼ 0.5) in any of the nonsignificant (with a trend in the opposite direction) in the colonies with a multiply mated queen (P ϭ 0.32), suggesting that analyses, but tended to be in the expected direction (more female assessment errors are the main causal factor here. Both these bias with increasing paternity skew) (13). This implies that the analyses had ‘‘year’’ included, but this factor did not have a number of colonies sampled was probably too small to shed light significant effect (F ϭ 0.37; P ϭ 0.92 and F ϭ 0.60; P ϭ on the possible interaction between the average chemical (cu- 7,84 7,56 0.75, for colonies with a singly and multiply mated queen, ticular hydrocarbon based) separation among patrilines, and the respectively). Workers in colonies headed by a multiply mated relative representation of patrilines (of a specific degree of queen thus realize maximal inclusive fitness by specializing on chemical distinctness) on the assessment of colony kin-structure rearing males, and their inclusive fitness declines monotonously by workers. with increasing female bias. Note that the above evaluation of Discussion the relative magnitude of assessment errors in the two classes of colony is opposite to a tentative appraisal by Ratnieks and Informational Constraints on Worker Inclusive Fitness. Our results Boomsma (15), which was based on a much smaller set of indicate that workers of F. truncorum have the chemical infor- samples (11) and which did not take differential correlations mation on intra-colony relatedness that is required to realize with colony productivity into account. adaptive split sex ratios. However, heritabilities of these recog- When split sex ratio theory was formulated (14), it was stressed ͞ nition assessment cues are low, and the correlation between the that the assessment of queen mating status by workers was likely chemical distinctness of patrilines and male bias in the sex ratio to be based on the same genetically based cues that may allow (Fig. 2) shows that the inclusive fitness of workers in colonies nepotistic discrimination of kin of varying relatedness to self. headed by a multiply mated queen is constrained by lack of However, it was also expected that kin assessment would be information. In other words, workers in colonies with relatively easier to achieve, because it could be based on cumulative nondistinct patrilines are less likely to assess the mating status of information obtained over longer periods, whereas nepotistic their mother correctly and tend to invest too much in female individual recognition would require instantaneous decisions. reproductives. It is important to note that the proximate mech- With the present evidence (Fig. 2) that genetic cues contain anism of liquid food exchange between nestmates (trophallaxis) limited information and apparently constrain kin assessment, it implies that there may be more genetic variation in cuticular seems unlikely that the same cues would normally allow nepo- hydrocarbon profiles of individuals than is actually expressed. tistic discrimination to be effective, despite the 3-fold relatedness Trophallaxis is common in Formica ants (24), where it primarily difference between full sisters (relatedness ϭ 0.75) and half serves to redistribute resources through the colony. However, sisters (relatedness ϭ 0.25). The question remains, however, why

8802 ͉ www.pnas.org͞cgi͞doi͞10.1073͞pnas.1430283100 Boomsma et al. Downloaded by guest on September 25, 2021 cuticular hydrocarbon profiles in F. truncorum do not contain alkanes (33), it seems reasonable to hypothesize that the lack of more information on kinship, so that colony-level kin assessment more complex and thus more informative cuticular hydrocarbon would be more effective and worker inclusive fitness would end profiles may have evolved in response to the expression of up closer to the maximum possible value. nepotistic tendencies in a low and thus difficult to detect frequency (1, 3, 28, 31). The few studies that have demonstrated Inclusive Fitness Predictions of the Number and Chemical Structure of nepotistic tendencies in honey bees are consistent with this Recognition Compounds. In the following paragraphs, we will interpretation. They were all made in colonies where patrilines explore the hypothesis that the rare expression of nepotistic were unnaturally low in number (after artificial insemination) or discrimination is likely to be responsible for the observed lack of unnaturally distinct in their genetic differences (using color genetic variation in recognition cues in F. truncorum. Through- morphs or different races) (reviewed in ref. 34), so that the out, we will assume that there is a colony-level cost to the productivity costs of nepotism and͞or the errors of discrimina- expression of nepotism (1, 3). This implies that colonies that by tion were unusually low. chance have sufficient variation in genetic recognition cues to The hypothesis that the occasional expression of nepotistic allow nepotism to be expressed will leave fewer offspring. The traits would invoke selection against kin-informative recognition erosion of genetic recognition cues due to the expression of compounds can be independently tested in other social insect nepotism therefore has, in all likelihood, both a group-selection genera where colonies have a single, multiply mated queen. Our component and a kin-selection component, which can be treated hypothesis thus predicts simple cuticular hydrocarbon profiles, simultaneously by covariance equations as developed by Price similar to those of F. truncorum and honey bees, to prevail in (see ref. 27). These combined selection forces should work in the Acromyrmex and Atta leafcutter ants (35), Pogonomyrmex har- same direction and are expected to result in both a low number vester ants (36, 37), and Vespula wasps (38). In addition, heri- of recognition compounds and a low degree of information per tabilities of other compounds such as glandular secretions that compound. By implication and all other things being equal, these could secondarily be used for kin-recognition would also be two characteristics should then not only have evolved in F. expected to be low. This matches the recent finding of unex- truncorum but also independently in other social insects that pectedly small patriline differences in the qualitative and quan- have colonies with a single, multiply mated queen. When it could titative profiles of metapleural gland secretions of Acromyrmex further be shown that the pay-off of nepotistic behavior is octospinosus (39). fundamentally different in multiple-queen (polygynous) insect However, polygynous societies may well be characterized by societies, we would have arrived at the first inclusive fitness opposite, highly informative recognition cues. These societies predictions of the adaptive chemical structure of recognition are principally open for additional reproductive females to join, compounds in insect societies of specific kin structure. which implies that queen–queen competition for attention from The typical cuticular hydrocarbon profile of workers in the the workers is probably frequent and that queens advertise their monogynous study population of F. truncorum is indeed poor in fertility by chemical signals (40). Polygyny also incurs a constant peaks (Fig. 1), at least compared with the 30–60 compounds that risk of usurpation or social parasitism by unrelated individuals. are normally found on the cuticle of ants (8). Even neighboring In such contexts, kin-informative recognition cues may thus be polygynous populations of F. truncorum in Finland have many favoured by selection, because the inclusive fitness penalty of more detectable peaks (typically 20–25; N.J.O., unpublished having to raise nonrelatives instead of relatives is much higher data). In addition, the worker profiles in the study population are than the 3-fold reduction of inclusive fitness after the rearing of dominated by straight-chain saturated cuticular hydrocarbons half sisters that could have been full sisters. Even if relatedness (alkanes), which are considered to be relatively ‘‘uninformative,’’ is lower in polygynous colonies (say, 0.3), the ratio 0.3͞0 (when because they lack the functional groups or specific shape that rearing nonrelatives is the alternative) is infinite, and it is this typically stimulate insect chemo-receptors, compared with un- ratio (not the absolute difference in relatedness) that would saturated (alkene) and methyl-branched hydrocarbons (28). A maintain selection for genetically informative cues. The recent straight-chain, alkane-dominated profile may thus be insuffi- finding of nepotism in the polygynous ant Formica fusca (6) is ciently informative for individual recognition, but still adequate therefore not at variance with the absence of informative for patriline assessment based on long-term cumulative evalua- recognition cues in monogynous F. truncorum, provided that tion of genetic diversity among nestmates. Experimental work future work will show that the cuticular hydrocarbon profiles of applying single synthesized hydrocarbons on the cuticle of work- F. fusca workers are significantly more informative than those of ers of Polistes wasps has recently corroborated that alkanes do monogynous F. truncorum workers. A recent study identifying 61 not elicit individual recognition reactions, whereas alkenes and cuticular hydrocarbons in queens of this F. fusca population (40) methyl-branched hydrocarbons do (29). suggests that this may well be the case. Informative cuticular Do we find a similar combination of cuticular hydrocarbon hydrocarbon profiles are also expected in species that invariably EVOLUTION characteristics in other social insects with a single multiply mated have a single once mated queen per colony, because highly queen per colony? The only other well-studied system with heritable recognition cues will likely be adaptive for recognizing full-sister͞half-sister societies is the honey bee, Apis mellifera. nonnestmates, whereas the issue of discrimination between Honey bees have more cuticular hydrocarbon compounds than full-sisters and half-sisters does not apply. F. truncorum (30, 31), but also in this species the typical cuticular hydrocarbon profiles of workers are characterized by relatively Conclusion few straight chain hydrocarbons (28, 30, 31). The cuticular Our present results are consistent with selective fine-tuning of hydrocarbon profiles of honey bee patrilines overlap, but are genetic odor cue diversity to a level preventing selfish exploita- segregated more consistently than in F. truncorum (32). No tion. They indicate that, in closed ant societies consisting of heritability estimates of cuticular hydrocarbon characteristics in full-sisters and half-sisters, selection is more likely to favor and honey bees are available and neither have the reproductive maintain the collective interests of all workers than the selfish consequences of variation in patriline segregation been studied. interests of groups (patrilines) or individuals. Information on However, even moderately high heritabilities would not neces- kinship is sufficient to allow F. truncorum workers to collectively sarily imply that nepotistic discrimination is easier, as the increase and occasionally maximise their inclusive fitness by number of patrilines in a typical honey bee colony is an order of splitting the sex ratio, but is unlikely to be sufficient to let kin magnitude higher than in F. truncorum (32). Although honey bee selection operate at the individual level through the expression workers can discriminate between mixtures of C23 and C25 of nepotism. We cannot exclude the possibility (and find it in fact

Boomsma et al. PNAS ͉ July 22, 2003 ͉ vol. 100 ͉ no. 15 ͉ 8803 Downloaded by guest on September 25, 2021 quite likely) that kin assessment in F. truncorum is based on variation in these recognition cues. If these predictions would be information additional to the cuticular hydrocarbons that we confirmed, they would further add to the notion that social were able to analyze here. However, if there would be such a cohesion of complex insect societies is less surprising than it larger set of relevant recognition cues, e.g., additional volatile seems, even when inclusive fitness theory based on reproductive compounds (41) or glandular secretions (39), there is no reason conflict is the basic organizing principle. to assume that these cues would not be subject to the same selection forces on information content, so that heritability We thank K. Bargum, S. Barker, P. Gertsch, I. Hanski, G. de Jo´ng, L. estimates would likewise be low. The similarities in cuticular Keller, A. P. Møller, D. R. Nash, B. Oldroyd, D. Ortius-Lechner, and F. hydrocarbon characteristics of workers of honey bees and F. Ratnieks for comments and discussion, D. R. Nash and G. Nachman for truncorum ants suggest that multilevel inclusive fitness theory assistance with the final analyses and editing, and three anonymous reviewers for constructive comments. We were supported by the Danish (27) may ultimately have the potential to make rather precise and Natural Science Research Council, the European Union Research- kin-structure-specific predictions about the type of recognition Training Networks ‘‘Social Evolution’’ (FW4) and ‘‘INSECTS’’ (FW5), cues (e.g., alkanes versus alkenes and methyl-branched hydro- the NorFa research network in Conservation Biology, Academy of carbons) and the magnitude of genetic versus environmental Finland Grant 42725, and Tva¨rminne Zoological Station.

1. Ratnieks, F. L. W. & Reeve, H. K. (1992) J. Theor. Biol. 158, 33–65. 22. Becker, W. A. (1992) Manual of Quantitative Genetics (Academic Enterprises, 2. Hamilton, W. D. (1964) J. Theor. Biol. 7, 1–16. Pullman, WA), 5th Ed, pp. 55–66. 3. Keller, L. (1997) Trends Ecol. Evol. 12, 99–103. 23. Boomsma, J. J. & Nachman, G. (2002) in Sex Ratios: Concepts and Research 4. Sherman, P. W., Reeve, H. K. & Pfennig, D. W. (1997) in Behavioural Ecology: Methods, ed. Hardy, I. C. W. (Cambridge Univ. Press, Cambridge, U.K.), pp. An Evolutionary Approach, eds. Krebs, J. R. & Davies, N. B. (Blackwell, 93–111. Oxford), pp. 69–96. 24. Ho¨lldobler, B. & Wilson, E. O. (1990) The Ants (Belknap, Cambridge, MA) 5. Fletcher, D. J. C. & Michener, C. D., eds. (1987) Kin Recognition in 25. Cozier, R. H. & Dix, M. W. (1979) Behav. Ecol. Sociobiol. 4, 217–224. (Wiley, Chichester, U.K.), 465 pp. 26. Sundstro¨m, L. (1995) Am. Nat. 146, 182–201. 6. Hannonen, M. & Sundstro¨m, L. (2003) Nature 421, 910. 27. Hamilton, W. D. (1975) in Biosocial Anthropology, ed. Fox, R. (Malaby, 7. Tibbets, E. A. (2002) Proc. R. Soc. London Ser. B 269, 1423–1428. London), Vol. 4, pp. 133–153. 8. Lenoir, A., Fresneau, D., Errard, C. & Hefetz, A. (1999) in Information 28. Breed, M. D. (1998) in Pheromone Communication in Social Insects, eds. Processing in Social Insects, eds. Detrain, C., Deneubourg, J.-L. & Pasteels, Vander Meer, R. K., Breed, M. D., Espelie, K. E. & Winston, M. L. (Westview, J. M. (Birkha¨user, Basel), pp. 219–237. Boulder, CO), pp. 57–78. 9. Crozier, R. H. (1988) in Invertebrate Historecognition, eds. Grosberg, R. K., 29. Dani, F. R., Jones, G. R., Destri, S., Spencer, S. H. & Turillazzi, S. (2001) Anim. Hedgecock, D. & Nelson, D. (Plenum, New York), pp. 143–156. Behav. 62, 165–171. 10. Ratnieks, F. L. W. (1991) Am. Nat. 137, 202–226. 30. Getz, W. M. & Page, R. E., Jr. (1991) Ethology 87, 298–315. 11. Sundstro¨m, L. (1994) Nature 367, 266–268. 31. Page, R. E., Jr., Metcalf, R. A., Metcalf, R. L., Erickson, E. H., Jr., & Lampman, 12. Sundstro¨m, L. & Ratnieks, F. L. W. (1998) Behav. Ecol. 9, 116–121. R. L. (1991) J. Chem. Ecol. 17, 745–756. 13. Sundstro¨m, L. & Boomsma, J. J. (2000) Proc. R. Soc. London Ser. B 267, 32. Arnold, G., Quenet, B., Cornuet, J.-M., Masson, C., De Schepper, B., Estoup, 1439–1444. A. & Gasqui, P. (1996) Nature 379, 489. 14. Boomsma, J. J. & Grafen, A. (1991) J. Evol. Biol. 4, 383–407. 33. Getz, W. M. & Smith, K. B. (1987) J. Comp. Physiol. A 160, 239–245. 15. Ratnieks, F. L. W. & Boomsma, J. J. (1997) J. Theor. Biol. 185, 423–439. 34. Breed, M. D., Welch, C. K. & Cruz, R. (1994) Behav. Proc. 33, 25–40. 16. Boomsma, J. J. & Ratnieks, F. L. W. (1996) Philos. Trans. R. Soc. London Ser. 35. Villesen, P., Murakami, T., Schultz, T. R. & Boomsma, J. J. (2002) Proc. R. Soc. B 351, 947–975. London Ser. B 269, 1541–1548. 17. Bagne`res, A. G. & Morgan, E. D. (1990) J. Chem. Ecol. 16, 3263–3276. 36. Ho¨lldobler, B. (1976) Behav. Ecol. Sociobiol. 1, 405–423. 18. Nielsen, J., Boomsma, J. J., Oldham, N. J., Petersen, H. C. & Morgan, E. D. 37. Cole, B. J. & Wiernasz, D. C. (1999) Science 285, 891–893. (1999) Insectes Soc. 46, 58–65. 38. Ross, K. G. (1986) Nature 323, 798–800. 19. Aitchison, J. (1986) The Statistical Analysis of Compositional Data (Chapman 39. Ortius-Lechner, D. Maile, R., Morgan, E. D., Petersen, H. C. & Boomsma, J. J. & Hall, London). (2003) Insectes Soc. 50, 113–119. 20. Oldroyd, B. & Moran, C. (1983) Aust. J. Biol. Sci. 36, 323–332. 40. Hannonen, M., Sledge, M. F., Turillazzi, S. & Sundstro¨m, L. (2002) Anim. 21. Searle, S. R., Casella, G. & McCulloch, C. E. (1992) Variance Components Behav. 64, 477–485. (Wiley, New York), pp. 429–431. 41. Getz, W. M., Bru¨ckner, D. & Smith, K. B. (1986) J. Comp. Physiol. A 159, 251–256.

8804 ͉ www.pnas.org͞cgi͞doi͞10.1073͞pnas.1430283100 Boomsma et al. Downloaded by guest on September 25, 2021