Differentiation during fig ontogeny suggests opposing selection by mutualists

Abstract Dioecy allows separation of female and male functions and therefore facilitates separate co‐evolutionary pathways with pollinators and seed dispersers. In monoecious figs, pollinators' offspring develop inside the syconium by consuming some of the seeds. Flower‐stage syconia must attract pollinators, then ripen and attract seed dispersers. In dioecious figs, male (“gall”) figs produce pollen but not viable seeds, as the pollinators' larvae eat all seeds, while female (“seed”) figs produce mostly viable seeds, as pollinators cannot oviposit in the ovules. Hence, gall and seed figs are under selection to attract pollinators, but only seed figs must attract seed dispersers. We test the hypothesis that seed and gall syconia at the flower stage will be similar, while at the fruiting stage they will differ. Likewise, monoecious syconia will be more similar to seed than gall figs because they must attract both pollinators and seed dispersers. We quantified syconium characteristics for 24 dioecious and 11 monoecious fig species and recorded frugivore visits. We show that seed and gall syconia are similar at the flower stage but differ at the fruit stage; monoecious syconia are more similar to seed syconia than they are to gall syconia; seed and gall syconia differentiate through their ontogeny from flower to fruit stages; and frugivores visit more monoecious and seed syconia than gall syconia. We suggest that similarity at the flower stage likely enhances pollination in both seed and gall figs and that differentiation after pollination likely enhances attractiveness to seed dispersers of syconia containing viable seeds. These ontogenetic differences between monoecious and dioecious species provide evidence of divergent responses to selection by pollinators and seed dispersers.

. A major advantage of dioecy is that it allows separation of female and male functions, providing opportunities for more independent responses to selection by mutualists (Gleiser et al., 2019;Patel & McKey, 1998). This benefit of dioecy is usually framed with respect to pollination (i.e., male and female floral structures), but in a very unique system, Ficus (Moraceae), it also applies to the next step of angiosperm reproduction, seed dispersal (Dumont & O'Neal, 2004;Lambert, 1992;Weiblen, Lomascolo, Oono, & Dumont, 2010).
In Ficus, adaptations associated with dioecy are not straightforward because traits that are advantageous for pollination may be disadvantageous for seed dispersal, and vice versa. For example, flower-bearing syconia of dioecious Ficus need to be similar between sexes, because both sexes need to attract the same pollinators, but dissimilar at the "fruiting" or seed dispersal stage, as only female syconia must attract seed dispersers (see below). Here we provide evidence that Ficus species have responded to opposing selection pressures between sexes and between stages of reproduction.
Before presenting five predictions, we describe the relevant life history of monoecious and dioecious figs.
In the genus Ficus, monoecy is ancestral (Rønsted, Weiblen, Clement, Zerega, & Savolainen, 2008;Weiblen, 2000). Dioecy occurs in a little over half of all Ficus species and is restricted to the paleotropics (Berg, 1989). All figs have an extremely tight interaction with pollinators as, in general, each species of fig is predominantly pollinated by one species of wasp in the family Agaonidae (for exceptions, see Cook & Rasplus, 2003;Weiblen, 2002). In monoecious fig species, pollination is effected by pollen-loaded, fertilized female wasps, which enter the closed syconium through a hole called the ostiole, tightly closed by overlapping involucral bracts (Datwyler & Weiblen, 2004), pollinate female flowers inside and oviposit in the ovules of some of the flowers. Wasp larvae feed on developing seeds.
After completing development, wasps emerge through the seed coat and copulate within the syconium. Fertilized females collect pollen from male flowers and leave the syconium through a hole drilled by male wasps to find another receptive fig, where their life cycle starts again. Flowers pollinated by the incoming wasps produce viable seeds that are animal-dispersed when the syconium develops into a "fruit" (Figures 1 and 2). Male wasps are wingless and hence never disperse from the syconium in which they emerged (Janzen, 1979;Weiblen, 2000). In an evolutionary sense, monoecious figs "pay" for viable seeds with seeds consumed by their pollinators.
In dioecious species, male and female functions are separated in different trees. In some trees, pollen-loaded fertilized female wasps enter syconia and oviposit in every flower they pollinate; seeds do not develop as wasp larvae develop in their place. Consequently, syconia at the mature fruit stage contain practically no viable seeds.
Those trees are technically monoecious, but only the male function is successful. Hence, they are commonly called male figs, or "gall" figs, because they pass their genes to the next generation only via male gametes, dispersed in the pollen collected and distributed by the female wasps (Weiblen, 2002) (Figures 1 and 2). Given that they essentially never produce viable seeds, gall figs are not under selection to attract seed dispersers. They "pay" for pollen dispersal with tissue that would have developed into mature seeds. In "female" trees, female wasps enter syconia and pollinate the flowers. However, they are unable to oviposit because the flowers have styles longer than the wasps' ovipositor; hence, female wasps die within syconia without leaving any descendants. Syconia on female trees are a deadly trap for the pollinating wasps, yet they produce viable seeds that benefit from seed dispersal (Figures 1 and 2). Female figs are called "seed" figs. In an evolutionary sense, seed figs pay nothing for pollination-they "cheat." They are, however, under selection to attract female wasps at the flowering stage and later to provide a nutritious reward to seed dispersers.
Monoecious and seed figs are dispersed by many types of frugivores (Shanahan, So, Gompton, & Gorlett, 2001), and syconia have evolved to attract seed dispersers in a variety of ways (Lomascolo, Levey, Kimball, Bolker, & Alborn, 2010). As described above, gall figs are presumably the exception because they contain no viable seeds.
Perhaps more important, consumption of gall syconia would jeopardize the pollination mutualism, as all pollinators of dioecious figs require gall syconia for reproduction.
These complexities of reproductive natural history led Lambert (1992) and Dumont, Weiblen, and Winkelmann (2004)  to the contrary, they should be structured to avoid consumption at all stages. Extending this rationale a step further, Lambert (1992) and  hypothesized that female plants should very closely mimic male plants at the flower stage to counteract strong selective pressure among female wasps to differentiate between "safe" gall syconia and "death trap" seed syconia. Yet, Grafen and Godfray (1991) propose that benefits of mimicry at the flower stage are reciprocal for seed and gall figs. It may be argued that gall figs also benefit from similarity because gall figs depend on emerging female wasps successfully finding and pollinating seed figs, as it is the only way in which the genes of gall figs will be represented in the next generation. Pollen-loaded female wasps that enter another gall fig will pollinate the gall flowers, but no seeds will develop and, hence, genes from the pollen donor will not pass to the next generation.
Indeed, it has been shown that intersexual mimicry in volatile organic compounds (VOC), responsible for flower and fruit odor, does occur in at least some fig species whose gall and seed figs flower synchronously (Hossaert-McKey et al., 2016). The role of selective pressure by pollinating wasps is further supported by the same authors' observation that mimicry does not occur in species whose gall and seed figs do not flower synchronously. Moreover, the VOC profile of fruit-stage seed syconia is clearly different from that of gall syconia in a species of figs dispersed by mammals, which have keen olfaction (Borges, Bessière, & Hossaert-McKey, 2008). In the same study, species dispersed by birds (which have generally poor olfaction) did not differ in VOCs between gall and seed figs, suggesting another stage-the fruit stage-at which figs can evolutionarily respond to differences in mutualists. Volatiles, however, may not tell the complete story. Although frugivores seem to prefer seed over gall figs (Chen et al., 2017;Lambert, 1992), the evidence is mostly anecdotal and at least some frugivorous bats do find and consume figs that lack the traits typical of mammal-dispersed species, including fruit odor .
We suggest a thorough test of Lambert's (1992) and Dumont et al.'s (2004) conceptual framework is lacking. Such a test would examine multiple monoecious and dioecious species at the flower and Pollinated flowers produce viable seeds that are animal-dispersed when the syconium ripens. Male wasps die without ever leaving the syconium. In male, or "gall" syconia of dioecious figs, pollen-loaded fertilized female wasps enter the syconium and oviposit in every flower they pollinate; essentially, all developing seeds are consumed by wasp larvae, so the gall syconia generally contain no viable seeds. Therefore, they should not be under selection to attract seed dispersers. In female, or "seed" syconia, pollen-loaded, fertilized female wasps enter syconia and pollinate the flowers, but cannot oviposit in the long-styled flowers. Female wasps die within syconia without reproducing. Seed figs produce viable seeds that benefit from seed dispersal and are therefore under selection to attract seed dispersers. Also illustrated are the five predictions tested in this study. The life stage and sex to which each prediction applies is marked with a blue arrow and its corresponding number fruit stages and would integrate several traits known to influence fig choice by pollinators and frugivores with quantitative measures of frugivore consumption among syconia types.
Previous tests of the framework's predictions have yielded mixed support (Chen et al., 2017;Lambert, 1992;Weiblen et al., 2010). We extend those studies by testing a series of predictions with data on traits known to be used by pollinators and frugivores to find resources and by standardized observations of feeding activity at fruiting figs. At the flower stage, we analyze syconia color, odor, and size because these traits have been shown to affect pollinator attraction Patel, Anstett, Hossaert-McKey, & Kjellberg, 1995). At the fruit stage, we add contrast against the background, pulp softness, and length of the peduncle (a structure that exposes the syconium away from the branch and trunk, thereby increasing accessibility to frugivores), which are also reported to influence frugivore attraction Lambert, 1992;Lomascolo et al., 2010;Schmidt, Schaefer, & Winkler, 2004). We propose five predictions. Because we could not collect data on all stages of maturity for all species included in this study, we test these predictions with the most appropriate set of species in each case. Our predictions, illustrated in Figure

| Color
We quantified syconium color using a USB2000 portable spectrometer (Ocean Optics, Inc) and a PX-2 Pulsed Xenon light source, which took reliable readings between 300 and 740 nm. This range includes wavelengths visible by humans and ultraviolet wavelengths. We scanned syconia using a sensor that had five optical fibers illuminating the target surface and a sixth fiber that returned the reflected light to the spectrometer. The scanning angle was fixed at 45° by using a black metal stand with a hole at that angle. The metal stand also blocked external light. We fastened a non-UV-filtering microscope slide to the opening of the hole where the sensor was introduced to keep constant the distance between the end of the sensor and the syconium. To obtain an average reflectance spectrum for a given syconium, we scanned it three times in three different spots on the syconium. We calculated reflectance as the proportion of a certified white Spectralon Diffuse Reflectance Standard (Labsphere). All spectra for each species were averaged every 5 nm and further smoothed with a smoothing function in pavo, an R package for quantifying color (Maia, Eliason, Bitton, Doucet, & Shawkey, 2013;R Core Team, 2018). We obtained a quantitative measure of syconium color for each species with this same R package (of all the options included in pavo, we chose those that we specify in parenthesis), including hue, which is the wavelength of maximum reflectance (H1), mean relative brightness across the entire spectrum (B2), and chroma, which represents the saturation of the spectrum (S8) and is calculated as the wavelength of maximum reflectance minus the wavelength of minimum reflectance, divided by brightness (B2) (Maia et al., 2013;Ordano et al., 2017).
We calculated color contrasts as the Euclidean distance between the color of a syconium and the color of the structure against which it is seen by a mutualist, which consisted of leaves, bark or nearby unripe fruits when syconia were at the fruit stage, depending on the species.
We normalized spectra to the same brightness by dividing the reflectance at each wavelength by the total reflectance for each species.
This represents contrast due to the color of the syconium and not to brightness, and is called chromatic contrast .
Qf is the color spectrum of the syconium and Qb is the color spectrum of the background structure; λ is the wavelength in nm and the sum corresponds to the complete spectrum, 300-740 nm.

| Volatile compounds
To quantify odor, we collected syconia in the field and extracted the volatile compounds in the laboratory. In situ extraction of volatiles was not possible because plants were found in some very remote sites and some species were 30 m-tall trees, which had to be carefully climbed to collect syconia. To ensure equal treatment of all species and both sexes, syconia were brought back to the laboratory, where volatile extraction started no more than 3 hr after collection. Syconia were placed inside 1-gallon Reynolds oven baking bags that had a carbon filter at one end to clean air entering the bag, and a filter con- Only the total amount of volatile compounds was used as a measure of how odorous the syconia were as, due to budget constraints, we were not able to run all samples through the mass spectrometer to identify the volatiles of all the species, sexes, and stages of maturity.
We standardized the total amount of volatiles contained in a given sample by dividing it by the total surface area of the syconium used to collect that sample. Total surface area was calculated using the mathematical formula for the surface area of a sphere, using the average diameter of the syconia of each species, and multiplying this number by the number of syconia included in the sample. We acknowledge that some syconia were not perfectly spherical, but shape variation had little impact on our standardized estimates of total volatiles.

| Syconium size, softness, and peduncle length
For syconium size, we measured diameter at the widest point with a caliper to the nearest 0.5 mm and used it as an estimate of overall size. For softness, an arbitrary scale between 1 (hardest, syconium surface did not indent when pressed between thumb and index finger) to 4 (softer, syconium surface easily indented when pressed identically) was developed by the authors and measured by the same person throughout the study . Peduncle length (±0.5 mm) was measured between the base of the syconium and the point where the peduncle attached to the branch or trunk.

| Visitation to fig trees
We quantified visits to monoecious, seed, and gall figs using two video cameras (Sony DCR-HC40) positioned 3-5 m from a fruiting tree and recording simultaneously. Videotaping started at 6 a.m. and ended at 10:30 a.m., and then again from 6:30 p.m. until 11:00 p.m., which allowed us to detect both diurnal and nocturnal frugivores.
Nocturnal recording was done using an infrared light (Sony HVL-IRH2). We aimed one camera at the ground to record terrestrial frugivores and one at either a branch or the trunk, depending on where most ripe figs occurred on the tree. We recorded 68 individual seed figs (1,077.3 hr) of 28 species (Table 1) and 20 gall figs (339.4 hr) of seven species (Table 1). Monoecious species were less common in our study site, so we were able to record four monoecious trees (68.8 hr) of four species (Table 1).

| Difference in traits of syconia at different developmental stages
To test prediction 1, whether gall and seed figs are indistinguishable at the flower stage, we applied a generalized linear model (GLM)specifically, a logit model with a binomial family error (Gelman & Hill, 2007;Piquer-Rodríguez et al., 2018;Rojas, Vergara-tabares, Valdez, Ponzio, & Peluc, 2019), with sex as the dependent variable and color (hue, brightness, and chroma, Endler, 1990), total amount of volatile compounds, and diameter as the predictor variables. For this test, we had complete data on six species of gall figs and seven seed figs (Table 2). To test prediction 2-that syconia from seed and gall figs will be distinguishable at the ripe fruit stage-we performed the same GLM between gall and seed figs (n = 24 species of each; Table 3), adding chromatic contrast against the background, pulp softness, and peduncle length as predictor variables because they are known to influence frugivore preference  and likely have little influence on wasp behavior. To test prediction 3-that seed and monoecious figs at the ripe fruit stage are more similar to each other than they are to gall figs-we used the same predictor variables as for prediction 2 and performed two tests, hereafter 3A and 3B. Test 3A: At the ripe fruit stage, we performed a binomial logistic regression between seed (n seed = 24 species) and monoecious figs (n mono = 11 species) (Table 3).
Test 3B: We performed a GML of the multinomial logit type, comparing seed (n seed = 24 species), gall (n gall = 24 species) and monoecious (n mono = 11 species) figs with the same variables and sets of species as the previous analyses (Table 3). Because we had only three data points for each maturity stage for each sex (one set of three points for each of three species), statistical power was very low. We therefore compared the Euclidean distances qualitatively, intending the result to inform future studies.

| Differential visitation to seed and gall figs by frugivores
To test prediction 5--that frugivores visit monoecious and seed figs more often than gall figs--we used a chi-squared test to compare the number of visited and not-visited monoecious, seed, and gall trees. We applied William's correction due to the small sample sizes (McDonald, 2014). We also performed a separate analysis just for female and male trees, as it seemed a cleaner test given that they were the same species of both sexes. For this analysis, we applied a Yate's continuity correction (McDonald, 2014). Any frugivore that entered the tree was assumed to have been attracted to the tree and its presence was recorded as a visit, unless it was clearly observed for the entire visit and did not consume a fruit during that time. A tree was recorded as visited if at least one frugivore visited it.

| Differentiation of gall and seed figs at the flower stage (prediction 1)
Gall and seed figs at the flower stage were largely indistinguishable, regardless of which variables were included in the logistic models (Table 4). The model that included only color variables had the lowest AIC value, which was more than two units lower than that of the full model, but less than two units different than the model with both color and odor variables. We calculated the mean confusion matrix for the model with lowest AIC, averaged over 100 runs (Table 5), and it showed that seed syconia were classified almost randomly as either seed or gall, while gall syconia were a bit more accurately classified as gall. Overall, however, the classification error, calculated as 1-(# of correct predictions/total observations), was quite high: 40% (averaged over 100 runs, Table 4). This result means that each time a new syconium of unknown sex at the flower stage is classified based on the variables included in the best model, it will be incorrectly classified 40% of the time. If gall and seed syconia were identical, the classification error would be 50% (i.e., random).

| Gall versus seed syconia (prediction 2)
In all three models comparing gall and seed syconia at the mature fruit stage, at least two variables significantly or marginally

14.85
Note: The variables included in the analysis are listed in columns 3-8 and are syconium brightness (amount of light reflected by the fruit, B2), hue (wavelength of peak reflectance, generally interpreted as the color of the syconium, H1), saturation (how "pure" a color looks, represented by the slope of the curve approaching peak reflectance, S8), syconium softness (a measure of the range from dry, fibrous syconia, to soft, moist and fleshy), diameter at the widest point, and total amount of volatiles quantified by gas chromatography. The last six columns represent the number of individuals of each species included in the analyses of morphology, color, and extraction of volatile compounds, as well as the mean and standard deviation of number of syconia per individual. significantly differentiated between gall and seed syconia. The model with the lowest AIC included all the variables. The mean confusion matrix averaged over 100 runs (Table 5) showed that seed and gall syconia at the ripe fruit stage tended to be classified much more accurately into their correct category than seed and gall syconia at the flower stage (mean classification errors 17% and 40%, respectively).

| Seed versus monoecious versus gall syconia (prediction 3)
Test 3A Seed and monoecious syconia at the ripe fruit stage did not differ significantly; no variable was a significant predictor in any of the three models. The model with the lowest AIC included color and odor variables, but no model had an AIC that was less than two units below that of the model with only color variables. We calculated the confusion matrix and misclassification error for the simplest model, which included only color variables. Misclassification error, 27%, was higher than that for gall and seed syconia (17%). The average confusion matrix shows that, although seed syconia are quite well classified, classification of monoecious syconia had a high error rate.
Overall, these results indicate that monoecious and seed syconia at the ripe fruit stage are similar in appearance.

Test 3B
When comparing gall versus seed versus monoecious syconia with a multinomial regression, the categories were significantly differentiated by at least one of the variables, and up to three variables in some of the models (Table 4). Again, as predicted and in line with results from analyses reported above, the largest differences were between gall and the other two types of syconia, as the calculated probability of being misclassified was much lower for gall syconia than for ripe seed or monoecious syconia (77%, 51%, and 1%, respectively). The highest probability of misclassification was between ripe syconia of seed and monoecious figs, supporting the idea that they tend to look more similar to each other than to gall syconia (Table 5).

| Differentiation of gall and seed syconia with ontogeny (prediction 4)
The first two principal components (PCs) explained 60% of variance in syconia and generally reflected the differences one would expect in the maturation of any fruit (e.g., increases in size and volatile compounds, softening of the pulp, change of pigmentation; Table 6). Confirming the results of our previous analyses between gall and seed syconia at the flower and ripe fruit stages, differentiation between gall and seed syconia appears to occur during the fruits' ontogeny after pollination, especially at ripening; the mean Euclidean distance in a bivariate space defined by the first two Note: We tested the predictions that monoecious and seed figs would be similar to each other, as they are both under selection to attract frugivorous seed dispersers. These two fig types would be different from gall figs, which do not produce viable seeds and, thus, are not under selection to attract seed dispersers. The variables included in the analysis are listed in columns 3-10 and are syconium brightness (amount of light reflected by the fruit, B2), hue (wavelength of peak reflectance, generally interpreted as the color of the syconium, H1), saturation (how "pure" a color looks, represented by the slope of the curve approaching peak reflectance, S8), syconium softness (a measure of the range because dry, fibrous syconia, to soft, moist and fleshy), diameter at the widest point, total amount of volatiles quantified by gas chromatography, length of the peduncle that attaches the fruit to a branch of trunk, and chromatic contrast against the structure against which the syconium was exposed (foliage or trunk). The last six columns represent the number of individuals of each species included in the analyses of morphology, color, and extraction of volatile compounds, as well as the mean and standard deviation of number of syconia per individual.
PCs between gall and seed syconia of the same species showed an increasing trend from flower, to unripe, to ripe syconia (Table 7A), with most of the differentiation occurring at maturation. The variables that define the third PC do not obviously match traits one would expect to change with fruit maturation. However, because the third PC explains a fairly substantial amount of variance (24%), we included it in a similar analysis based on Euclidean distance in trivariate space. The overall trend is similar to that with the bivariate Euclidean distance, except that one of the species, F. congesta, does not follow the expected pattern (Table 7B). Because the fourth PC explained only an additional 10% of the variance, we did not include it in distance calculations.

| Frugivore visits to figs (prediction 5)
Three out of four monoecious trees (75%) were visited by frugivores during video recording, 22 out of 68 seed trees (32%) were visited, and only one out of 20 gall trees (5%) was visited. The one gall tree that was visited had only a single visit by a single disperser, a bandicoot (unidentified to species, family Preamelidae). The bandicoot was not observed consuming a fruit but, following protocol, we recorded its presence as a visit because we could not clearly observe it the entire time in the tree. In support of prediction 5, monoecious and seed trees were visited significantly more often than gall trees (χ 2 = 9.76, df = 1, p = .008). A comparison between

Model description (dependent and independent variables) AIC
McFadden's Pseudo R 2

Mean misclassification error
Variables included in the model (Significant variables in bold; **p < .01; *p < .05; ǂ p < 0.1) Note: The best models, based on AIC and simplicity (in cases where the difference in AIC was less than two), are marked with an asterisk. For those models, we calculated a mean misclassification error by averaging 100 runs of the model done with random subsets of 40% of the data (testing set).
The model was built with the other 60% of the data (training set). We also report McFadden's pseudo R 2 as a different estimate of model fit. The last column provides more detail of the variables included in each model and indicates the level of significance of those with the most influence. Models are numbered according to the predictions that they test, as described in the main text and Figure 2. seed and gall figs indicated that the former received more visits than the latter (χ 2 = 4.30, df = 1, p = .038). Visitors consisted of a variety of birds and bats (mostly unidentifiable to species) and very few bandicoots.
Visitation rates to seed figs were 0.021 visits/hr; to gall figs, it was 0.003 visits/hr; and to monoecious figs, it was 0.058 visits/hr.
We consider these rates to be rough estimates and conservative because, as mentioned previously, when several frugivores visited the tree in a short period of time, we counted them as one visit even though they could have been by different individuals. Also, it is almost certain that some visits were undetected by our cameras.

| D ISCUSS I ON
Our results support most predictions derived from Lambert's (1992) and Dumont et al.'s (2004)  gametes to produce viable seeds (Grafen & Godfray, 1991). In any case, the mimicry between seed and gall figs is apparently effective; pollinators visit gall and seed syconia at the flower stage with equal frequency Patel et al., 1995;Weiblen, Yu, & West, 2001).
Syconia at the ripe fruit stage differed significantly between seed and gall figs in traits known to affect frugivore consumption of figs . Seed syconia tended to show darker and more saturated colors (related to greater amount of pigments), contrasted more against the background, were softer, and emitted more VOCs (related to stronger odor). These results align with previous studies that described differences between seed and gall figs in other traits that often attract seed dispersers, such as syconia nutritional content Weiblen et al., 2010) and diameter Lambert, 1992). Differences in color and odor were also assessed by , although qualitatively, and likewise found to differ between seed and gall syconia. As predicted, seed syconia generally exhibited traits that should make them more attractive to seed dispersers. Summarizing results from all studies, including ours, seed syconia at the fruit stage are more nutritious, have more pigments, are more odorous, and are softer than gall syconia. The only exception we could find is reported by , who showed that hardness of syconia did not differ between seed and gall figs. This difference between our results and those of Dumont et al. may be real (i.e., the different species of figs TA B L E 5 Mean confusion matrices resulting from cross-validation tests of each of the models described in Because the traits that distinguish seed syconia are those known to attract seed dispersers of Ficus, because those seed dispersers consume seed syconia far more often than gall syconia (this study, Chen et al., 2017, but see Phua & Corlett, 1990), and because gut passage of fig seeds is known to enhance their germination (Chen et al., 2017), the differences between seed and gall syconia are consistent with selection by frugivorous seed dispersers. This idea is further supported by our finding that ripe seed syconia tend to have traits that are more similar to those of ripe monoecious fig species, which should also be under selective pressure to attract seed dispersers, than to their conspecific gall figs. We also show that frugivores are much more often attracted to monoecious and seed figs than they are to gall figs, which suggests that differentiation in syconia traits at the fruit stage leads to enhancement of the female function in seed and monoecious figs. It remains to be tested whether consumption by frugivores leads to differences in fitness.
Including odor as a variable that may have evolved differently in gall versus seed figs is important because volatiles are known to influence the behavior of both pollinators and seed dispersers. However, a limitation of our study is that we used a relatively crude measure of volatiles, total amount, not the amounts and identities of individual volatile compounds, many of which are known to be important to mutualists. (Borges et al., 2008;Hossaert-McKey, Soler, Schatz, & Proffit, 2010;Raguso, 2008). We acknowledge that a thorough test of differentiation between gall and seed syconia at the flower stage would include the Our results also demonstrate, as predicted, that seed dispersers are attracted far more frequently to monoecious and seed figs than to gall figs. Although previous studies suggest a preference of frugivores for seed figs over gall figs (Chen et al., 2017;Lambert, 1992 (Lambert, 1992).
A limitation of our study is that predictions for differences between seed and gall syconia were based on potential selective pressure of seed dispersers in general even though, for example, birds and mammals may exert very different, potentially conflicting selective pressures on fruit traits. We did not differentiate between bird and mammal-dispersed species because we did not have enough data for each disperser type for all stages of maturity. This renders our analysis conservative and even so, we found that fruit-stage seed syconia are more similar to each other than to fruit-stage gall syconia. Moreover, although bats often find and consume figs without the typical VOC profile or the overall characteristics associated with mammal dispersal , we found that they did not visit gall figs. Thus, it seems reasonable for purposes of this study to have combined all seed figs into a single group attractive to frugivores.
TA B L E 7 Euclidean distance between gall and seed figs (Ficus sp.) at the different stages of syconium ontogeny in a (A) bivariate space defined by the first two principal components, and (B) trivariate space defined the first three principal components Note: Only the three species for which we had data for all three stages of ontogeny were included. Variables included in the analysis are pulp softness, diameter, total volatile compounds, hue, chroma, and brightness. In two out of three species included gall and seed syconia differentiate well at the mature fruit stage, while at the flower and immature fruit stages they do not differentiate well. Results from F. congesta do not show this trend as clearly throughout its development.
In summary, our results and those from previous studies suggest that figs' response to opposing selection by pollinator and seed disperser mutualists leads to overall similarity of syconia at the flower stage and differentiation at the ripe fruit stage, ensuring pollination of all fig types Lambert, 1992;Lomascolo et al., 2010) and to differential attractiveness to seed dispersers between gall and seed syconia.
In a more general context, our results extend prior work on the evolutionary advantages of dioecy. Dioecy is thought to be facilitated by pollinator-mediated selection for floral dimorphism and driven by avoidance of selfing and optimization of resource allocation to male and female structures (Ashman, 2000;Charlesworth, 1999;Freeman, Doust, El-Keblawy, Miglia, & McArthur, 1997). Worldwide, dioecy is associated with climbing growth form, abiotic pollination, and animal-mediated seed dispersal (Renner & Ricklefs, 1995). These associations do not generally occur in our sample of Ficus species in Papua New Guinea, a region of especially high Ficus diversity.
Although some of our species start out as climbers, they afterward become free-standing trees (most species strangle their host), and all of them are insect-pollinated. Consistent with worldwide patterns, however, all are animal-dispersed. In Ficus, the exceptions to traits generally associated with dioecy are almost certainly related to its unusual life history. In particular, effective breeding populations are extraordinarily large (i.e., gene flow is high and selfing presumably low) and males produce fruit-like structures, which uniquely require allocation of parental resources (Herre, Jander, & Machado, 2008).
Mutualisms may be defined as antagonisms in equilibrium, as both partners exploit the other to maximize their own benefits; tight interactions evolve only if benefits exceed costs for both partners (Bronstein, 2015;Morris, Vazquez, & Chacoff, 2010). and more generalized ones (figs and seed dispersers).

CO N FLI C T O F I NTE R E S T
None declared.

AUTH O R CO NTR I B UTI O N S
SBL conceived the idea, designed, and performed fieldwork and statistical analyses. DJL helped to shape the idea, design the methods and fieldwork, and frame the study in theory. SBL and DJL wrote the manuscript.

DATA AVA I L A B I L I T Y S TAT E M E N T
All data needed to repeat the analyses in this paper are presented in Tables 1-3. Additional data will be freely available upon request.