^{1}

^{2}

^{3}

^{1}

^{2}

^{3}

^{1}

^{4}

^{3}

^{1}

^{2}

^{5}

^{*}

DWD, JR, DWY, and JMC developed the research plan and all authors discussed the results. DWD formulated the parasite hypothesis and then collected, dissected, and measured syconia and wasps; collated and analyzed the data; drew

The authors have declared that no competing interests exist.

Mutualisms are interspecific interactions in which both players benefit. Explaining their maintenance is problematic, because cheaters should outcompete cooperative conspecifics, leading to mutualism instability. Monoecious figs (

In a biosphere driven by selection at the level of the individual gene [

Fig trees (

Within receptive syconia, the lengths of floral styles are highly variable [

Blue galls contain pollinators, red galls contain parasites, and yellow ovules contain seeds. “Ovule length” is used to estimate the distance of the seed or gall from the syconium wall and is measured as the point the pedicel joins the fig wall to the top of the seed or gall, excluding what remained of the style.

Recently, a fourth hypothesis, based on “optimal foraging” by ovipositing foundresses, has been proposed. Simulations have shown that the fig-pollinator mutualism can be stabilised if ovule profitability is correlated with flower style length, and if some foundresses die before laying all of their eggs [

Although crucial in determining foundress behaviour, the fitness differential for wasp larvae developing in inner versus outer ovules is largely unknown. However, there is evidence that inner ovules develop into larger galls due to increased space near the syconium cavity (the lumen), resulting possibly in larger, more fecund female offspring [

If inner ovules represent enemy-free space for pollinator larvae, we would predict that externally ovipositing parasitic wasps are more likely to kill pollinator larvae in outer ovules. In the syconia of _{1,45} = 4.85, _{5,45} = 17.22, _{1,45} = 6.66,

Controlling for variance in ovule length that can be attributed to the difference between sites (_{5,5856} = 110.96, _{1,5856}= 34.64, _{3,5856} = 622.00,

Exit, exited; Para, parasites.

Small, filled symbols represent the fitted binary logistic regression model, which used parasite or pollinator presence as the dependent variable (large, open symbols).

Over 90% of the nonpollinating wasps we found belonged to two genera,

Mean Frequencies (±s.e.) of the Four Categories of Ovules Present in

Long galls (inner ovules) are on the left, and short galls (outer ovules) are on the right of the

Egg limitation in the pollinator,

The contribution parasitic wasps may make to the overall mechanisms that lead to mutualism stability across

Data from three sites per species have been pooled for ease of comparison.

Exit, exited; Para, parasites; Poll, pollinators.

Finally, we found that pollinator body size did not correlate with ovule length (_{1,55} = 0.44, _{5,55} = 13.19, _{1,55} = 6.75, _{1,55} = 6.69,

Our study is the first to show that pollinating fig wasps may gain a fitness benefit by selecting inner ovules for oviposition, because these ovules have reduced vulnerability to parasitism. The provision of ovules with high variance in profitability to foundresses clearly demonstrates that the larger, more sessile partner in the symbiosis [

The potential role played by parasitic wasps may also help to resolve the evolutionary paradox posed by fig trees having generation times several orders of magnitude longer than those of their pollinators [

Thus, despite the short-term costs posed by parasitic wasps to the mutualists [

We measured both the probability of offspring mortality through parasitism, and the body sizes of female offspring, in relation to ovule position within the syconium. We used a total of 64 syconia from six populations of the Australian fig

In the laboratory, each syconium was sliced into eighths lengthways. Every ovule was then systematically removed from all sections. We measured the total length of every fourth ovule (pedicel + seed or gall, excluding what remained of the style) to the nearest 0.024 mm using an eyepiece graticule attached to a binocular microscope. We did not measure the pedicel length separately for two reasons. (1) Galls or seeds at the extreme outside wall of the syconium do not have pedicels, which would result in a series of zeros in the resulting dataset and subsequent problems with data analysis. (2) In

After dissection, each ovule was assigned to one of four categories: (1) seed—ovules containing seeds; (2) exited—ovules with an exit hole made by a vacated male wasp; (3) parasite—in which the ovule contained a parasitic wasp, and (4) pollinator—ovules containing pollinating wasps (

The larval biology of most species of nonpollinating fig wasps has been divided into three major ecological groupings [

For an estimate of the body size of female

Unless otherwise stated, we transformed all measurements to natural logarithms to normalise the error variances. We compared the mean lengths of each of the four categories of ovules, using a general linear mixed model that included ovule category and syconium volume as predictors. Site was included as a random factor.

To test our hypothesis that parasites can only gain access to pollinators in middle and outer ovule layers, we ran a general linear mixed model that used the mean length of ovules occupied by pollinators to predict the mean length of ovules occupied by parasites. Syconium volume was included as an additional covariate to control for any effects of syconium size on ovule length, and site was again included as a random factor [

We used a binary logistic regression to measure the relationship between ovule length and the likelihood of parasitism. Ovules containing seeds or those that had been vacated by a male wasp were excluded from the analysis. For the dependent variable, we included those ovules known to contain either a parasitic wasp (1) or a pollinator (0). Site and syconium volume were included as additional covariates to ovule length.

We estimated the head area of female pollinating wasps (length × width). To test whether pollinators were distributed nonrandomly within syconia according to their size, we used a general linear model with head area as the dependent variable, site as a random factor, and both pollinator-occupied ovule length and fig volume as covariates.

(28 KB DOC)

We thank Jane deGabriel for organising access to the Hervey's Range site, and Jamie Moore for collecting syconia from Brisbane. Jinyan Yao and Sarah Al-beidh provided help in the field. Stu West, two anonymous referees, and the editor provided constructive comments that enabled us to improve our manuscript.

standard error