Mutually beneficial host exploitation and ultra-biased sex ratios in quasisocial parasitoids

Selfish interests usually preclude resource sharing, but under some conditions collective actions enhance per capita gains. Such Allee effects underlay early explanations of social evolution but current understanding focusses on kin selection (inclusive fitness). We find an Allee effect that explains unusual quasisociality (cooperative brood care) among parasitoid wasps without invoking or precluding kin selection effects. In Sclerodermus harmandi, individual females produce most offspring when exploiting small hosts alone. However, larger hosts are more successfully exploited by larger groups of females, with the per-female benefits outweighing the costs of host sharing. Further, the extremely biased sex ratios (97% female) are better explained by mutually beneficial female–female interactions that increase the reproductive value of daughters (local resource enhancement), rather than by the usually invoked local mate competition between males. Thus, atypical quasisocial behaviour in a parasitoid wasp directly enhances reproductive success and selects for very extremely female-biased sex ratios.

T he selfish interests of individuals usually preclude them from opting to share resources, but under some conditions collective actions enhance per capita gains: such Allee effects on reproduction can explain mutually beneficial interactions and formed the basis of early explanations of social behaviour 1,2 . Most current studies of social evolution utilize kin selection theory based on genetic relatedness (inclusive fitness) [3][4][5] . Eusocial insects, with cooperative brood care, reproductive division of labour and overlapping generations provide the major empirical focus for these studies 4,6,7 and also key tests of inclusive fitness-based sex allocation theory 6,8,9 , but there are other social insects that lack some eusocial life-history characteristics 2,10 .
Some parasitoid wasps exhibit aggressive brood defence by guarding mothers 11 , classing them as sub-social 10 , but few exhibit more complex social biologies and most, lacking parental care, are socially solitary 10,12 . In contrast, the quasisocial bethylid wasp Sclerodermus harmandi and some of its congeners are among the most socially complex parasitoids known [13][14][15] : not only do females typically exhibit maternal care 16,17 , but multiple foundress females may tend a group of offspring developing on a single host 13,18,19 (Fig. 1) and mothers do not exclusively care for their own offspring 16 .
Adult female S. harmandi are highly active in searching for host larvae in the field [20][21][22][23] . Once a host is found, females first inspect it with their antennae and mandibles, paralyse it by stinging with the ovipositor, clean the host's body surface, then feed on the host haemolymph before laying eggs if the host is sufficiently large [23][24][25][26][27] . Although larger hosts are preferred for oviposition, they require more time to paralyse, and parasitism rates and offspring survival are both lower than when eggs are laid on smaller hosts 28 . Field evidence shows that multiple females may attack a single long-horned beetle host (with, for instance, up to four females observed attacking single Saperda populnea larvae (ca.15 mm in body length) 29 . Under multifemale attack, less time is required for paralysis and parasitism rates can be higher than under single-female attack 30,31 . After laying eggs, females tend the brood until the progeny mature 16,17,25,32 : as with host attack, brood tending can be performed by single or multiple females 16,17,25 with broods on large hosts tended by multiple females producing more progeny than those tended by single females 31 .
Here we provide experimental evidence that mutual host exploitation by S. harmandi maximizes the average reproductive success (direct fitness) of individual females, thus explaining quasisociality in this species without recourse to kin selection. We also show that the very extremely female-biased sex ratios observed are better explained by local resource enhancement (LRE) 9,33,34 , due to mutually beneficial foundress-foundress interactions contributing to the value of female offspring, than by the much more widely applied explanation of local mate competition (LMC) between males 9,33,35,36 .

Results
Oviposition and development. The percentage of hosts that were oviposited on was 66.3% overall. The probability of oviposition decreased as host size increased (logistic analysis of covariance (ANCOVA): G 1 ¼ 14.43, Po0.001, %deviance explained ¼ 5.1, n ¼ 220) and as foundress number increased (logistic ANCOVA: G 4 ¼ 19.18, Po0.001, %dev ¼ 27.3, n ¼ 220); there was no significant interaction between these effects (logistic ANCOVA: G 4 ¼ 0.28, P ¼ 8.93, n ¼ 220). The time taken for females to begin ovipositing on a host ranged from 3 to 20 days, with longer periods for larger hosts (ANCOVA: F 1,140 ¼ 6.14, P ¼ 0.014, %dev ¼ 2.  (Fig. 2). Exploring the clutch size data in terms of the mean number of eggs laid per foundress showed that this average increased with host weight (log-linear ANCOVA: F 1,140 ¼ 9.96, P ¼ 0.002, n ¼ 146) and decreased with foundress number (log-linear ANCOVA: F 4,140 ¼ 24.12, Po0.001, n ¼ 146), without significant interaction (log-linear ANCOVA: F 4,136 ¼ 0.19, P ¼ 0.941, n ¼ 146). So, provided that eggs were laid, more eggs were laid on larger hosts and by larger groups of females, but the average contribution of eggs by individual females was smaller when foundress groups were larger.   Total developmental time from egg to adult emergence ranged between 23 and 31 days and was shorter on smaller hosts (ANCOVA: F 1,109 ¼ 8.11, Po0.005, r 2 ¼ 0.05, n ¼ 112) and shorter in broods developing from larger numbers of eggs (ANCOVA: F 1,109 ¼ 32.48, Po0.001, r 2 ¼ 0.22, n ¼ 112); the interaction between these main effects was marginally nonsignificant (ANCOVA: F 1,108 ¼ 3.62, P ¼ 0.060, n ¼ 112). The developmental time was also shorter when offspring were produced by more foundresses (analysed separately to avoid multicollinearity with numbers of eggs laid, ANCOVA: F 4,107 ¼ 6.04, Po0.001, r 2 ¼ 0.18, n ¼ 112). Mortality of offspring during development was common, with 67.17% (s.e.m. þ 2.13, À 2.21) of eggs dying before adulthood. Mortality was highly overdispersed (heterogeneity factor ¼ 37.0) indicating a tendency for offspring on a given host to survive or die collectively. Developmental mortality was unrelated to host weight (logistic ANCOVA: F 1,139 ¼ 0.78, P ¼ 0.378, %dev ¼ 0.5, n ¼ 146), number of eggs laid (logistic ANCOVA: F 1,140 ¼ 1.98, P ¼ 0.162, %dev ¼ 1.3, n ¼ 146) or foundress number (logistic ANCOVA: F 4,141 ¼ 1.96, P ¼ 0.104, % dev ¼ 5.2, n ¼ 146). The size (mean individual weight) of adult female offspring increased with increasing host weight (ANCOVA: F 1,109 ¼ 6.88, P ¼ 0.010, r 2 ¼ 0.05, n ¼ 111) but was not significantly affected by the numbers of adults produced per host (an index of resource availability, ANCOVA: F 1,108 ¼ 2.52, P ¼ 0.116, r 2 ¼ 0.02, n ¼ 111) or by the interaction between host weight and adult offspring number (ANCOVA: Offspring production. The probability of producing adult offspring from a given host was enhanced when foundress groups were larger (logistic ANCOVA: Fig. 3a). Enhanced offspring production was due to higher probability of oviposition (logistic ANCOVA: G 4 ¼ 19.18, Po0.001, %dev ¼ 27.3, n ¼ 220) and greater number of eggs being laid (log-linear ANCOVA: F 4,140 ¼ 50.64, Po0.001, %dev ¼ 56.5, n ¼ 220) when foundress groups were larger and not due to developmental mortality of offspring, which was unrelated to foundress number (logistic ANCOVA on grouped binary data: Fitness gains per foundress were contingent on an interaction between foundress number and host weight (log-linear ANCOVA: Individual foundresses were most productive when attacking small hosts (r ca. 0.2 g) alone. For intermediate-sized hosts (ca. 0.2-0.33 g), foundresses benefited most by being in groups of two or four foundresses, whereas for the largest hosts (40.33 g), females benefited most by being in groups of eight foundresses (Fig. 3b). Differences in per foundress offspring production within each host size category (Fig. 4), defined according to where the uppermost regression lines on Fig. 3b cross, were explored using three separate Kruskal-Wallis tests. For small hosts, differences in mean ranking were not significant due to limited sample size (n ¼ 17; H ¼ 1.229) and for medium hosts production did not differ between foundress number categories (n ¼ 105, H ¼ 2.676, P ¼ 0.613). For large hosts, production differed significantly (n ¼ 98, H ¼ 31.47, Po0.001) and post hoc multiple comparison testing (with critical values calculated for the 10 possible comparisons using: a ¼ 0.05, k ¼ 5 groups, z ¼ 2.807) showed that this difference was due to groups of eight foundresses being significantly more productive than single foundresses (difference in mean ranking ¼ 41.11, Po0.05) and two-foundress groups (difference in mean ranking ¼ 31.09, Po0.05), and also to groups of four foundresses being more productive than single foundresses (difference in mean ranking ¼ 25.95, Po0.05).
Sex ratios. The sex ratios of groups of maturing adults were very strongly female biased (mean proportion males ¼ 0.032, The probability of adult offspring being produced declines as host size increases and is additively higher for larger foundress groups (logistic ANCOVA: host weight,  Fig. 5) but for multiple foundress cases sex ratios were far more biased than would be predicted by classical LMC theory. Figure 5 contrasts observed sex ratios with predicted evolutionarily stable sex ratio responses to foundress number in haplodiploid species with single-generation mating groups 9,35 . While these models assume foundresses are unrelated, some extended LMC models predict increased female bias when foundresses are related, but the differences from classical predictions are not large 9,37 . Mechanistically, the relationship between sex ratio and foundress number appears to be due to individual foundresses producing a constant number of males (mean ¼ 0.797, s.e.m. ¼ þ 0.073, À 0.067) irrespective of foundress number (log-linear regression: F 1,108 ¼ 3.04, P ¼ 0.084, %dev ¼ 2.7), but fewer females when in larger foundress groups (log-linear regression: F 1,108 ¼ 36.82, Po0.001, %dev ¼ 26.5, n ¼ 110), tallying with Mamaev's 15 observation that 1-2 males are included among the first eggs laid in Sclerodermus clutches.

Discussion
Multiple foundress oviposition is generally disadvantageous to individual parasitoid mothers, due to subsequent resource competition among developing offspring, and is usually adaptive only when single-foundress opportunities are rare 12,[38][39][40] . Where present, Allee effects are usually reported in relation to single (foundress) females laying multiple eggs to suppress host immune defences 41 or in terms of large offspring groups being required to consume the host and complete development 42 . In contrast, in S. harmandi an Allee effect operates via a foundress numberdependent probability of parasitizing (ovipositing on and  to groups of eight foundresses being more productive than one-foundress and two-foundress groups and also to four-foundress groups being more productive than single foundresses (Po0.05 in all three post hoc multiple comparison tests). Overall, the size of the foundress group that maximizes the median per capita production of offspring increases as host size increases. obtaining offspring from) larger hosts. Our data thus constitute an experimental demonstration of mutually beneficial host exploitation among adult parasitoid wasps, and an independent evolution of cooperative reproductive behaviour within the Hymenoptera likely promoted by phylogenetic constraints on body size and clutch size 43 . Our results also accord with prior reports that larger hosts, once suppressed, enhance reproductive success for single Sclerodermus females but also take longer to handle and are more likely to kill the attacking parasitoid than are smaller hosts 13,26,28,30,31,44 . S. harmandi sex ratios were very strongly female biased, as reported in other studies of this species and its congeners 13,15,16,18,28,32,44,45 . Female-biased sex ratios are often explained by the reproductive value of nondispersing male offspring being reduced (by male-male competition for mates, LMC), compared with dispersing daughters 9,35,46 . LMC models can be formulated in several equivalent ways, with an inclusive fitness approach considered the most flexible 9,12,33 . Further, under LMC, sex ratios are predicted, and often observed, to become less biased as foundress numbers increase due to sons being progressively able to mate with the daughters of other foundresses 9,12,35 . Although the observed increase in S. harmandi sex ratio with increasing foundress number is in qualitative agreement with expectations from classical LMC theory, and also with anecdotally reported trends for Sclerodermus pupariae 44 , the quantitatively small effect of foundress number, coupled with the greater extent of the sex ratio bias than would be predicted (Fig. 5), casts doubt on whether LMC can fully explain these results. LMC could best explain the observed sex ratios of S. harmandi if only one dominant foundress produced offspring on each host irrespective of the number of other females present, which could be determined using molecular genetic markers 36 . However, the facts that the total numbers of eggs laid on a host were greater when more foundress were present (Fig. 2), and that females that have not laid their own eggs do not tend broods 16,17 strongly suggest that such extreme reproductive skew is unlikely.
Male-male combat interactions can intensify LMC and lead to greater female bias in some parasitoids 47,48 , but combat is not observed in S. harmandi or its congeners. Furthermore, S. harmandi females are typically wingless (Fig. 1) and males winged 32 , which is expected to promote more intrasex interactions between female than male siblings, due to a greater dispersal of males 9,33,35 . If females competed for hosts, local resource competition (the generalized form of LMC) 9,33 would be expected to select for decreased investment in females 49 , but given that we have shown that the reproductive value of females is enhanced by mutualistic exploitation of large hosts, LRE, when the production of one sex increases the fitness of relatives more than does production of the other 9,33,34,50 , offers explanation for the extreme female bias. Our results thus indicate that femalebiased sex ratios are favoured because they increase the probability that a daughter's future environment will contain more females with which she can more successfully exploit hosts. The extent of any selected bias is likely to be influenced by factors such as the mating capacity of males and the number and relatedness of the foundress females.
LRE is more often documented in vertebrates and is typically based on intergenerational interactions, such as offspring helping at the parental nest 9 . Among invertebrates, LRE is a strong candidate explanation for female bias in some solitary and facultatively quasisocial bees, with the fitness enhancement operating via multiple-foundresses reducing the probability of post-oviposition parasitism or predation of offspring 21,51 . We know of no prior reports of LRE operating in parasitoid wasps. While our results indicate LRE, they contrast with offspring production patterns under LRE in allodapine bees in which production per female follows a dome-shaped relationship with the number of cooperating females 34 . In S. harmandi offspring production per foundress declines with foundress number (log-linear regression: F 1,109 ¼ 35.91, Po0.001, %dev ¼ 24.8, n ¼ 111) for cases in which at least some offspring survived, for comparison with analysis in Schwarz 34 . Mechanistically, this difference may be due to the number of failed nests in allodapine bees being unknown 34 , while our data on offspring production comprised both successful and unsuccessful attempts at host suppression. Biologically, the differences may be due to the allodapine bee mutualisms deriving from reduced predation of offspring 34,51,52 and being relatively weakly coupled to offspring food supply, while in S. harmandi the mutualism is facultatively based on the acquisition of a subsequently limited and shared food resource.
After parasitizing a host, foundress S. harmandi females remain in close association with each other and each other's offspring for extended periods, tending and moving eggs and larvae 13,14,16,17,25,31 . Such foundress associations are predicted, by iterated public goods game expansion of LMC theory 48 , to select for strong female bias, due to cooperative sex allocation (whereby all foundresses benefit by the production of only the lowest number of males required for insemination, thus utilizing the maximum host resource for female production). Due to decreasing likelihood of long-term cooperation by all females in larger groups, the type of gradual response to foundress number that we have observed (Fig. 5) is also predicted 48 . While this theory provides a candidate alternative explanation for S. harmandi sex ratios, for it to operate foundresses would have to recognize the sex ratio of eggs on the host and to respond by adjusting their own sex allocation decisions (a form of policing against cheating by some foundresses laying higher proportions of male eggs 48 ); hymenopteran parasitoids generally do not have this ability 53 and the egg stage of S. harmandi clutches is short relative to the prolonged period of brood tending through the larval stages 16,17 . Although the sex of developing larvae may be more readily detectable by tending foundresses 53 , offspring mortality was unrelated to foundress number in S. harmandi so it does not seem that infanticide is common (see also Hu et al. 16 ) (in contrast to solitary and subsocial bethylids 11,54 ), making postovipositional sex ratio adjustment of S. harmandi brood sex ratios a less likely explanation for sex ratio bias than LRE, although both selective mechanisms could potentially operate simultaneously.
While S. harmandi foundresses care for each other's offspring, females that have not laid their own eggs do not exhibit brood care 16,17 . The conditional occurrence of brood care by adult females suggests that (largely selfish) behaviours, such as reproductive dominance and skew, may operate during the phase of (apparently) cooperative brood care 7,52,55,56 . However, these may be tempered by mutual policing and/or by foundressfoundress relatedness 52,55 , especially since in some bethylids adults can recognize kin 57 probably using variation in cuticular hydrocarbon profiles 58 . Close relatedness between foundresses may be enhanced by codispersal and foraging by females from a given natal patch 34 and would be expected to select for more greatly female-biased sex ratios 9,37,53,59,60 . Cofoundress relatedness, which will be highest when females derive from broods produced by fewer mothers 61 , should also select, via inclusive fitness gains, for altruistic brood care 7 , and this may explain why female S. harmandi that have not laid eggs do not commit ovicide 16,17 .
In summary, social behaviour in many hymenopterans is highly advanced 3,4,6,7 , but most parasitoid wasps are socially solitary. In parasitic bethylids, both subsociality (brood guarding 11 ) and the much less-common quasisociality (cooperative brood care (Fig. 1)) can be explained by direct fitness gains to foundress females: selfish defence of own offspring 11 and mutually beneficial host suppression, respectively. Female-biased sex ratios in subsocial bethylids can be largely explained by LMC 35,36,46 but the ultrabiased sex ratios of the quasisocial S. harmandi are better explained by LRE via multifoundress mutually beneficial host exploitation, an unusual scenario among parasitoid wasps. These results do not preclude that the social behaviours of S. harmandi may also be mediated by additional effects of kinship between foundresses or that sex allocation may be contingent on some degree of mate competition between males; indeed these influences are expected 9,36,57,58 and would be especially likely if female offspring disperse from depleted natal hosts in cohesive groups 32,60 .

Methods
Parasitoids and hosts. The life histories of parasitoid wasps have been instrumental in stimulating evolutionary and ecological theory, such as sex ratio evolution 9,10,12,33,35,62 , but most are socially solitary 10 , lacking parental care. Sclerodermus species (members of the hymenopteran family Bethylidae) are unusual among parasitoids in that multiple females may attack and oviposit on a single host and then remain together tending the developing brood 13,14,[16][17][18]21,25,[30][31][32] . In common with its congeners, S. harmandi (Buysson), which has been considered synonymous with S. guani 63 , lays clutches of eggs onto the hosts integument, typically after first feeding on the host 26,27 with clutch sizes ranging up to around 100 eggs, dependant on the size, developmental stage and species of the attacked host 26,28,44 (Fig. 2). The lifetime fecundity of individual females ranges from 30 to 200 eggs 64 .
Progeny sex ratios are biased towards females 28,32 (Fig. 5). Males typically mate with maturing broodmate females when these emerge or prior to their emergence by chewing entrances into their cocoons 16,64 . Adult wing dimorphisms (alate and apterous forms) occur in both sexes in the genus 13,45,65 . In S. harmandi apterous males are rare 32 but possession of wings does not necessarily indicate a propensity or an ability to fly 13,18 . Females are larger in body size than males (female: 3.5 mm in body length, apterous and 3.2 mm, alate; male: 2.1 mm, alate) 65 . Males live for around 1 week but females typically live for 2-7 months 32 . Mated females overwinter in groups in host-made tunnels or cavities in trees 21,32 .
S. harmandi, a widely used biological control agent in forest pest-control practices in mainland China 66 , mainly parasitizes multiple wood-boring longhorned beetles (Cerambycidae) in nature 65 , which varies widely in larval body size depending on host species (for example, larval Monochamus alternatus from 200 to 700 mg in body weight were parasitized in the field 28 ). These hosts often have life histories of 1 year or more (uni-voltinism and semi-voltinism), but S. harmandi is multivoltine 64 . Natural rates of host availability and parasitism are poorly known due to difficulties in sampling within woody tissues; available estimates range from 0.015 (ref. 20) to 4% (ref. 32). Following inundative releases of S. harmandi in biological pest-control programmes, parasitism of hosts in the field may reach 50-80% (refs 21,64,66).
For this study, S. harmandi was cultured at the Forestry Institute of Jiangsu Province, PR China, where it is mass produced for the biocontrol of the pine sawyer beetle Monochamus alternatus Hope (Coleoptera: Cerambycidae), a pest of conifers which vectors nematodes causing pine wilt disease 21,26,27,64,66 . M. alternatus hosts were collected from pine forests in Liyang County during the winter of 2011, and maintained in the refrigerator (ca. 10°C).
Experiment. Laboratory experiments were conducted at 25°C and 60-80% r.h. in 2012 at Nanjing Agricultural University. Varying numbers of adult female S. harmandi (foundresses) were presented with a larval M. alternatus host, weighed to an accuracy of 0.0001 g (Mitler, AL204-IC) in a glass vial (1.0 cm diameter, 5.0 cm long): foundress numbers were 1, 2, 4, 6 or 8 (n ¼ 60, 55, 45, 30 and 30, respectively, giving 220 replicates and adequate overall statistical power). Hosts and parasitoids were examined twice daily and the occurrence of oviposition, the time to oviposition, the number of eggs laid, the duration of offspring development and number of male and female offspring produced were recorded. The average weight of females within offspring groups was also recorded to an accuracy of 0.0001 g.
Statistical analysis. Generalized linear modelling (in Genstat, V14.1, VSN International) was used to explore the influences of host size and foundress number on offspring production and sex allocation, using backwards elimination of explanatory variables from initial statistical models and with overdispersion taken into account, via empirical estimation of scaling parameters, where appropriate 67,68 . Subanalyses within host-weight categories used Kruskal-Wallis tests with post hoc multiple comparison testing 69 .