The extent of isolation among closely related sympatric plant species engaged in obligate pollination mutualisms depends on the fitness consequences of interspecies floral visitation. In figs (Ficus), interspecific gene flow may occur when pollinating wasps (Agaonidae) visit species other than their natal fig species. We studied reproductive isolation in a clade of six sympatric dioecious fig species in New Guinea. Microsatellite genotyping and Bayesian clustering analysis of the fig community indicated strong reproductive barriers among sympatric species. A total of 1–2% of fig populations consisted of hybrid individuals. A new experimental method of manipulating fig wasps investigated the reproductive consequences of conspecific and heterospecific pollinator visitation for both mutualists. Fig wasps introduced to Ficus hispidioides pollinated and oviposited in receptive figs. Seed development and seedling growth were largely comparable between conspecific and heterospecific crosses. Heterospecific pollinator fitness, however, was significantly less than that of conspecific pollinators. Heterospecific pollinators induced gall formation but offspring did not develop to maturity in the new host. Selection on pollinators maintaining host specificity appears to be an important mechanism of contemporary reproductive isolation among these taxa that could potentially influence their diversification.

The observation that some closely related plants co-occur as distinct species, even in the presence of hybridization, has long motivated the study of reproductive isolating mechanisms (Stebbins 1950). Reproductive isolation in plants can involve prepollination mechanisms, such as phenological, mechanical, and ethological barriers, or postpollination mechanisms, such as pollen incompatibility, genetic incompatibility, or hybrid inferiority (Grant 1971). In animal-pollinated taxa, pollinators may act as agents of reproductive isolation by means of species-specific floral constancy in foraging behavior (Oyama et al. 2010). Additionally, trait mismatching in heterospecific visitation can prevent heterospecific pollen deposition or the collection of pollinator rewards (Kephart and Theiss 2004; Kay 2006).

When reproduction of pollinators and host is interdependent, selection for specialization may be intense, especially where similar hosts occur in sympatry. Indeed, the evolution of extreme pollinator specialization is observed in cases where plants provide “nursery” rewards for pollinators (Ollerton 2006). A recent natural experiment on the obligate pollination mutualism between Joshua trees and yucca moths (Smith et al. 2009) revealed that phenotype matching plays a role in reproductive isolation of parapatric host tree varieties. In spite of hybridization between varieties in a contact zone, fitness differences between pollinators ovipositing in the host variety from their native range compared to the host variety from outside their range selects for specificity and serves as a potential mechanism of reproductive isolation between Joshua tree varieties.

The obligate pollination mutualism between figs (Ficus, Moraceae) and fig wasps (Agaonidae, Hymenoptera) is similarly specialized (Janzen 1979; Weiblen 2002; Cook and Rasplus 2003). Life cycles of pollinating fig wasps and their Ficus host plants are completely interdependent. Volatile signals attract female fig wasps to a receptive fig of a host (Hossaert-McKey et al. 1994; Chen and Song 2008) where they gain access to the unisexual flowers of the enclosed inflorescence through a narrow, bract-covered opening, called the ostiole. Females oviposit in pistillate flowers and deposit pollen in the process. In most cases, pollinators do not have a second chance at reproduction as they cannot leave the fig upon entering (Moore et al. 2003). In monoecious figs, flowers either set seed, form galls that nourish fig wasp offspring, or produce pollen that eclosing wasps later transport to other receptive figs. In dioecious species, reproductive functions are separated between plants bearing either seed-producing inflorescences (female figs) or inflorescences producing wasps and pollen (male figs).

Ficus is known not only for its obligate pollination mutualism but also for extreme species richness (>750 species), and diversity in sympatry. In centers of endemism, such as New Guinea, recent radiations of closely related species coexist in sympatry (Berg and Corner 2005). Although the highly specific host associations of wasps have been implicated in fig speciation, observations of fig pollinators visiting multiple host species in sympatry (Molbo et al. 2003; Machado et al. 2005; Marussich and Machado 2007; Su et al. 2008) and molecular evidence of fig hybridization (Parrish et al. 2003; Renoult et al. 2009) have called into question the role of pollinators in historical and contemporary fig reproductive isolation. Recent studies have inferred ancient host switching or host conservatism from phylogenetic patterns (Lopez-Vaamonde et al. 2002; Weiblen and Bush 2002a; Jousselin et al. 2003; Machado et al. 2005; Marussich and Machado 2007; Jackson et al. 2008; Ronsted et al. 2008; Azuma et al. 2010; Moe and Weiblen 2010). Ecological mechanisms affecting gene flow and reproductive isolation in this system have not been investigated directly. Given that the nature and extent of isolating mechanisms among sympatric fig species depends on the reproductive consequences of pollinator host choice, experiments are needed to examine the fitness of both mutualistic partners. Such experiments can provide direct evidence on the nature of contemporary reproductive isolation and potential mechanisms for diversification in the past.

This study investigated the extent of hybridization and mechanisms of reproductive isolation among sympatric fig species in New Guinea. We applied Bayesian clustering methods based on microsatellite data to identify putative natural hybrids and estimate rates of heterospecific gene flow. We employed a new method for introducing pollinators to novel hosts that allowed us to compare fig and pollinator reproduction following conspecific visitation versus heterospecific visitation and separate host choice behavior from subsequent pollination and oviposition behaviors.



Ficus bernaysii King, Ficus congesta Roxb., Ficus hahliana Diels, Ficus hispidioides S. Moore, Ficus morobensis C.C. Berg, and Ficus pachyrrhachis K. Schum. and Lauterb. are members of dioecious Ficus subgenus Sycomorus section Sycocarpus. They comprise the majority of a clade estimated to have originated in New Guinea at least 15 million years ago and within which precise phylogenetic relationships are unresolved (Silvieus et al. 2008). These New Guinea endemic taxa often occur at high density in patches of secondary regrowth. They have abundant cauliflorous figs growing along the length of the trunk and are pollinated by Ceratosolen species (Wiebes 1963). Most fig species in this study system have uniquely associated pollinator species. For example, Ceratosolen notus pollinates F. congesta, whereas a distinct but unnamed Ceratosolen pollinates F. pachyrrhachis (Silvieus et al. 2008).

Landowners at the study site recognized two entities distinguished by the length and density of epidermal hairs on young shoots and the persistence of stipules that were referred to as F. bernaysii morphotypes A and B in a previous publication on pollinator sharing (Moe et al. 2011). Our examination of type specimens determined that these entities correspond to F. hahliana Diels and F. bernaysii. The taxon referred to as F. bernaysii in previous work at the study site (Weiblen 2000, 2001, 2004; Novotny et al. 2002; Weiblen and Bush 2002b; Weiblen et al. 2006, Weiblen et al. 2010) and subsequently as F. bernaysii“A” (Moe et al. 2011; Moe and Weiblen 2011) rather corresponds to F. hahliana. Confusion about the identity of F. hahliana dates from at least Wiebes (1963), where type material for Ceratosolen hooglandii, the pollinator of F. bernaysii, included specimens reared from both F. bernaysii (Hoogland 4890) and misidentified F. hahliana (NGF 12471). Moe et al. (2011) identified two cryptic clades of C. hooglandii associated with F. hahliana and F. bernaysii, respectively. No such confusion concerns the other species included in our study.


Experimental technique was developed over eight months and two field seasons in 2007–2008. Experiments were performed May–August of 2009 at Ohu village in the Madang district of Madang Province, Papua New Guinea (latitude 5°13′38″ S, longitude. 145°40′44″ E). Experimental trees were located in a 400 ha patchwork of secondary regrowth and undisturbed forest. Fifty individuals of each focal species were sampled for microsatellite analyses. Young leaf tissue was collected from each individual and dried over silica gel, and later stored at −80°C prior to DNA extraction. A voucher specimen was also collected from each individual, alcohol preserved, and later dried for long-term storage.


DNA was extracted from each individual using a Qiagen DNeasy Plant Tissue extraction kit. We amplified and genotyped individuals at 14 loci (Table 1). Among the microsatellite loci used, four primer pairs had been developed for Ficus montana (FM4–15 and FM3–64) and Ficus septica (FS4–11 and FS3–31) by Zavodna et al. (2005), four for Ficus racemosa (Frac86) and Ficus rubiginosa (Frub29, Frub38, and Frub436) by Crozier et al. (2007) and six for F. hahliana (B30, B47, B83) and F. pachyrrhachis (P164, P211, P215) by Moe and Weiblen (2011). Amplification of microsatellite loci was performed in an Eppendorf Mastercycler in a total volume of 10 μl using 0.2 mM fluorescent end-labeled forward primer and unlabeled reverse primer, 0.2 mM buffer solution, 0.2 mM of each dNTP, 0.8 mM BSA, 0.3 units of TaKaRa Ex Taq polymerase (TAKARA BIO Inc., Shiga, Japan), and 20–50 ng template DNA. PCR conditions are indicated in Table 1. Microsatellite alleles were visualized using an ABI 377 Sequencer along with a ROX 500 (Applied Biosystems) size standard and scored by hand using Genotyper 2.5 software (Applied Biosystems, Foster City, CA).

Table 1.  Microsatellite loci analyzed. The total number and length of alleles observed across all species, PCR conditions: annealing temperature (Ta) and number of cycles, and the original primer note references are shown.
PrimerNo. of allelesLength (bp) T a No. of cyclesReference
  1. 1Indicates number of touchdown cycles starting 10°C above the annealing temperature.

FM3–64 9267–29154101+20 Zavodna et al. 2005
FM4–15 18 232–298 53 30 Zavodna et al. 2005
FS3–31 8219–24354101+20 Zavodna et al. 2005
FS4–11 11 279–357 54 101+20 Zavodna et al. 2005
Frac8610141–18350151+20 Crozier et al. 2007
Frub29  6 179–199 54 101+20 Crozier et al. 2007
Frub3825172–13250151+20 Crozier et al. 2007
Frub436 15  97–131 53 30 Crozier et al. 2007
B3045215–34760101+20 Moe and Weiblen 2010
B47 16 171–219 53 30 Moe and Weiblen 2010
B8316165–1955330 Moe and Weiblen 2010
P164 18 227–288 60 101+20 Moe and Weiblen 2010
P21115 99–1275330 Moe and Weiblen 2010
P215 18 212–244 53 30 Moe and Weiblen 2010


We performed kinship analysis using Kinalyzer (Berger-Wolf et al. 2007; Ashley et al. 2009) to identify siblings among our samples. One individual from each sibling group was randomly chosen and included in tests for linkage disequilibrium and deviations from Hardy–Weinberg equilibrium using GENEPOP (Raymond and Rousset 1995; Rousset 2008). We applied Sequential Bonferroni corrections for multiple tests (Holm 1979). Data were analyzed in Microchecker (Van Oosterhout et al. 2004) to test for the presence of null alleles.

In describing genetic differentiation among focal species, we used all 14 loci to calculate Fst, a measure based on allele identity (Weir and Cockerham 1984), and Rho (Valdes et al. 1993), a measure based on allele size and an estimate of Rst (Slatkin 1995).

Bayesian clustering analyses were implemented in STRUCTURE (Pritchard et al. 2000). This method assigns individuals to one or more ancestral populations based on their allelic genotypes. Hybrids can be identified by their partial assignment to more than one ancestral population. To test whether the six named plant species correspond to genetically distinct clusters, we ran five independent iterations in STRUCTURE with ancestral population number (K) set for K= 4, K= 5, K= 6, and K= 7 and without using a priori species identifications. Each Markov chain included a 100,000-generation burn-in and ran for 106 additional generations. We used an admixture model and allowed for correlated allele frequencies among clusters. Using the method described in Evanno et al. (2005), we identified the most appropriate K value as six, which resulted in consistent cluster assignments over five iterations. These results along with the observation of six morphological species suggest that the mostly likely number of ancestral populations is six (Pritchard et al. 2000). Assuming K= 6, we used prior information on species assignment, based on morphological identification, and ran the analysis again at three values of interspecies migration rate, v= 0.01, 0.05, and 0.10. STRUCTURE estimated the posterior probabilities of each individual being (1) a nonhybrid, but with an incorrect a priori species assignment, (2) an F1 hybrid, or (3) an F2 hybrid. Eight individuals with a higher probability of being either misidentified or having hybrid ancestry than having nonhybrid ancestry were singled out for reexamination. Leaf vouchers and DNA reextraction and genotyping confirmed identity for these individuals. Analyses were run again at K= 6 with corrected genotypes and one misidentified individual reassigned to the correct species. The STRUCTURE results reported are from this second round of analyses.

An additional analysis identifying hybrid individuals was implemented in BayesAss (Wilson and Rannala 2003). BayesAss uses Bayesian and Monte Carlo–Markov chain methods to estimate recent migration rates among populations and estimate each individual's ancestry. Individuals are classified as an immigrant from a specific population, a nonimmigrant, or the offspring of an immigrant and a nonimmigrant (hybrid). The program assumes unlinked loci and a relatively low rate of migration among populations (less than 1/3), but allows for deviations from H-W equilibrium. A 10,000-iteration burn-in, followed by 3,000,000 iterations and default delta values were used. Individuals assigned as hybrids were noted. The mean and 95% confidence interval (95% CI) for estimated pairwise migration rates between focal species were recorded.



Ficus hispidioides served as the pollen recipient for all experiments. A nonreciprocal experimental design, permitting only fitness comparisons associated with conspecific and heterospecific pollination in one species, was necessitated by field logistics and natural history. The relatively accessible, abundant, and large figs (3–4 cm in diameter at receptivity) of F. hispidioides supported manipulations that were not feasible for other species. Five of six study species were included as pollen donors due to the limited availability of figs from F. bernaysii during the study period. We collected phenological and life-history data on F. hispidioides to appropriately design pollination experiments. We estimated numbers of foundresses typically encountered in receptive figs of F. hispidioides, so a comparable number of wasps could be introduced to experimental figs. A total of 250 receptive figs yielded an average of five foundresses per fig (SD = 4.06). We chose to introduce six pollinators per experimental fig to account for wasp mortality associated with the method of introduction. We also determined the minimum diameter at which figs are accessible to pollinators to design an effective pollinator exclusion treatment. We collected 120 figs, measured their diameter, and classified them as prereceptive (no pollinators inside), receptive (live pollinators inside), or postreceptive (dead pollinators and/or developing seed or galls inside). The average diameter of receptive figs was 33. 8 mm (range 25–39.5). Only figs ≤24 mm in diameter were treated.

Prereceptivity treatment

Seven functionally female and seven functionally male F. hispidioides trees were chosen as experimental trees on the basis of having large clusters of accessible and unreceptive figs. Figs with diameters ≤24 mm were tagged loosely around the peduncle. A ring of Tanglefoot pest barrier was applied around the ostiole of each fig and organza fabric was adhered to the pest barrier (Fig. 1A, B). This treatment allowed fluid to escape from the fig interior and for expansion of the fig during development while excluding pollinators from entering the ostiole (Fig. 1C). Fig diameters were measured and recorded every two days and the seal around the ostiole checked for integrity. If a seal was discovered broken and an ostiole exposed for any period of time, the fig was considered potentially contaminated and removed from the experiment. Figs were considered receptive upon reaching a diameter ≥30 mm and when numerous fig wasps had adhered to the Tanglefoot.

Figure 1.

Experimental method. (A–D) Prereceptivity treatment excluded pollinators and provided for controlled introduction. (E–G) Receptivity treatment. Pollinating wasps collected from receptive figs were introduced to experimental figs through Pasteur pipets.

Receptivity treatment

When figs reached receptivity, a cylindrical section of the fig wall running perpendicular to the shoot apex was cored using a 3 mm diameter stainless steel borer. A 3 cm section of glass Pasteur pipet plugged with cotton fiber was inserted into the hole (Fig. 1D). After two days, pollinators were manually introduced to treated figs that had not aborted. Figs from all five pollen donor species were collected and brought to experimental F. hispidioides trees (Fig. 1E). We first attempted to introduce winged wasps collected from ripe, functionally male figs of the five focal species that invariably failed to enter experimental figs when deposited in unplugged Pasteur pipets using a fine-tipped watercolor paintbrush. In 894 trials, no wasp demonstrated taxis along the length of the pipet toward the fig interior. However, when we removed foundresses engaged in active pollination from untreated receptive figs, and introduced them to experimental figs through Pasteur pipets, these wingless wasps, regardless of species, readily entered experimental figs (Fig. 1F, G). In the conspecific pollination treatment, we introduced six Ceratosolen dentifer wasps collected from F. hispidioides figs. In the four heterospecific treatments, we introduced six pollinators from one of four fig species to figs of F. hispidioides. After wasps had actively entered a fig, the pipet was replugged with cotton. Local availability of receptive, untreated figs determined the number of replicates performed. Control figs received no pollinator introduction. Figs were checked every two days until all figs aborted or reached maturity (approximately six weeks per tree).

Seed and gall formation

Figs were collected, split open, and checked for seed or gall development either upon abortion or maturity. A 3 mm wide longitudinal section of each fig was cut and pistillate flowers were counted under a dissection microscope and categorized as either undeveloped or setting seed in the case of female trees or forming galls in the case of male trees.

Seed viability

To determine viability and survivorship of interspecific crosses in comparison with nonhybrid F. hispidioides, we measured germination, growth, and survival rates for each pollen donor treatment. As the focal species is a pioneer of primary succession in forest gaps, growth rate could be an important aspect of fitness. Seed from experimental figs was germinated in a light chamber with 40–100 seeds from each fig in separate petri dishes. The proportion of seeds germinating per fig was recorded. Germinated seeds were randomly assigned to planters in 12 × 6 arrays in a chamber at 26.6°C, grown with 12 h of light per day and fertilized biweekly. Growth was monitored by length of the longest leaf, measured once a week, and seedling height, measured once a month for 139 days.

Seedling genotyping

We extracted DNA from one seedling grown from each experimental fig that produced viable seed. These DNAs were screened for microsatellites and subjected to Bayesian clustering analysis to confirm that seedlings were in fact hybrids resulting from the experimental treatments.


Data from functionally male and female trees were analyzed separately because gall development in male figs contributes to wasp fitness whereas seed development in female figs is directly associated with plant fitness. Control treatments assessed the effectiveness of pollinator exclusion. The development of a few control figs indicated rare events where naturally occurring pollinators circumvented our exclusion treatment. Although genotyping of seedlings distinguished experimental introductions from contamination in the case of seed figs, it was not possible to genotype wasps in galls. Therefore, to test whether gall formation in treated figs was significantly more frequent than in control gall figs, we performed pairwise Fisher's exact tests on counts of developed and undeveloped figs. Pairwise Fisher's exact tests also compared the proportion of figs that initiated gall and seed development in each heterospecific treatment against development in the conspecific treatment where F. hispidioides served as the pollen donor.

We performed ANOVA to determine the effects of pollen donor species and maternal tree on seed germination rates and survivorship after 139 days with the fig as the unit of replication. We then performed 2 × 2 contingency (Fisher's exact) tests to compare each heterospecific treatment against the conspecific treatment. We also performed ANOVA to determine the effect of pollen donor and maternal tree on plant height and maximum leaf length as measures of growth with the seedling as the unit of replication.



Kinship analysis revealed a large number of siblings in our samples that reduced the effective sample size of each species from 50 to 17–23 individuals (Table S1) for the calculations of genetic differentiation, tests of Hardy–Weinberg equilibrium, and linkage disequilibrium. After sequential Bonferroni multiple test corrections, linkage disequilibrium was not detected among the loci we sampled overall. However, in four of six species (F. hahliana, F. congesta, F. hispidioides, F. pachyrrhachis), a different locus was found to be significantly heterozygote deficient in each species after multiple test corrections (Tables S1 and S2), suggesting the presence of null alleles (Table S1). Mean Fst values for each species ranged 0.1875–0.2346 and mean Rho values for each species ranged 0.2349–0.4227 (Table S1). Genotypes are archived in the Dryad Digital Repository (doi: 10.5061/dryad.c3h3v).


Seven out of 300 individuals were identified with high posterior probability of hybrid ancestry using the highest migration rate prior (m= 0.10; Table 2, Fig. 2). Two of these also exhibited hybrid ancestry at the lowest migration rate (m= 0.01). These same individuals and a third were identified as putative hybrids through BayesAss analysis (Table 2). Pairwise migration rates among species estimated in BayesAss were very low (0.11–0.64%). The highest estimated migration rates were from F. pachyrrhachis to F. morobensis at 0.41% (95% CI ≤ 0. 01–2.26%), and vice versa at 0.64% (95% CI ≤ 0. 01–2.50%). All other pairwise estimates were < 0.3% with upper bounds of 95% CIs < 1.6%.

Table 2.  Posterior probabilities of assignment of seven individuals identified as putative hybrids from Bayesian clustering analysis (K= 6) using a priori species identity information with migration rate priors of 0.01, 0.05, and 0.10 are shown. Bold type values indicate the ancestry assignment with the highest posterior probability. Bold type individuals are most likely of hybrid ancestry at all tested migration rates.
Hybrid species assignmentMigration rateNon hybridF1 hybridF2 hybrid
  1. 1Hybrids identified by BayesAss.

F. hahliana×bernaysii F20.01 0.931 0.0060.063
  0.05 0.668 0.027 0.305
 0.100.4100.051 0.536
F. hahliana×pachyrrhachis F2 0.01 0.507 0.004 0.489
 0.050.1450.007 0.848
  0.10 0.066 0.007 0.927
F. morobensis×pachyrrhachis F110.010.027 0.899 0.074
  0.05 0.003 0.926 0.071
 0.100.001 0.930 0.069
F. morobensis×pachyrrhachis F21 0.01 0.920 0.008 0.071
 0.05 0.658 0.0340.307
  0.10 0.470 0.051 0.521
F. morobensis×hahliana F20.01 0.933 0.0110.054
  0.05 0.676 0.057 0.267
 0.100.4330.105 0.462
F. morobensis×pachyrrhachis F2 0.01 0.784 0.000 0.216
 0.050.3860.001 0.612
  0.10 0.222 0.002 0.776
F. morobensis×pachyrrhachis F210.010.4520.003 0.545
  0.05 0.128 0.005 0.867
 0.100.0620.005 0.933
Figure 2.

Barplots of ancestral population assignments for 300 individuals from Bayesian clustering analysis assuming six, color-coded ancestral populations (K= 6). Priors were used on population information based on morphological identification and an assumed migration rate of (A) 0.10, (B) 0.05, and (C) 0.01. Asterisks indicate individuals identified with the highest probability as either first or second generation hybrids.


Data from five male and six female trees were collected whereas data from three additional experimental trees were lost to foraging fruit bats. Among 563 gall figs and 345 seed figs of F. hispidioides receiving the pre-receptivity treatment, we observed abortion rates of 17.9% and 31.3%, respectively. A total of 410 gall figs and 216 seed figs that survived the coring treatments either received experimental pollinator introductions or served as controls. Heterospecific pollinators of functionally male figs induced gall formation in a significantly lower proportion of figs than conspecific pollinators (Fisher's exact test P < 0.01), but only figs that received the conspecific treatment produced mature galls and adult wasp offspring (Table 3). One of 62 control figs developed galls, indicating effective but not absolute exclusion of pollinators. Only figs treated with pollinators from F. hispidioides, F. congesta, and F. pachyrrhachis initiated gall formation in a significantly greater number of figs than controls (Fisher's exact P < 0.01).

Table 3.  Experimental gall fig treatments. Pollinator species, pollen donor species, and numbers of Ficus hispidioides figs treated, figs that initiated gall development, and mean (± SE)% flowers galled per fig.
Introduced pollinatorPollen donor n treated n initiated n matured
  1. 1Denote a significant difference from control figs (pairwise contingency test, Fisher's exact P < 0.01).

  2. 2Denotes a significant difference from F. hispidioides (pairwise contingency test, Fisher's exact P < 0.01).

C. dentifer F. hispidioides  90511251
C. notus F. congesta  81 121,2  0
C. hooglandii F. hahliana  36 22 0
C. sp. ex Ficus morobensis F. morobensis  17  22  0
C. sp. ex Ficus pachyrrhachis F. pachyrrhachis 125411,2 0
Control None  62  1  0

In functionally female figs, seed set was observed in all treatments except F. hahliana (Table 4), including one of 33 control figs (Table 4). Microsatellite genotyping of seedlings allowed the identification and exclusion of nonexperimental pollinations from further analyses (see “Seedling genotyping” in Methods).

Table 4.  Experimental seed fig treatments. Pollinator species, pollen donor species, and numbers of Ficus hispidioides figs treated, figs that initiated seed development, figs that matured seed, and mean (± SE) seed set.
Introduced pollinatorPollen donor n treated n initiated n matured
  1. 1Significant difference from F. hispidioides (pairwise contingency test, Fisher's exact P < 0.05).

C. dentifer F. hispidioides 541110
C. notus F. congesta 45  4  4
C. hooglandii F. hahliana 23 0 1 01
C. sp. ex Ficus morobensis F. morobensis 16  1  1
C. sp. ex Ficus pachyrrhachis F. pachyrrhachis 52 7 6
Control None 33  1  1


More than 50% of seed resulting from all successful crosses germinated (Table 5) and ANOVA showed no significant effect of pollen donor or maternal tree (P= 0.138 and P= 0.189, respectively). However, pairwise 2 × 2 contingency tests showed that F. congesta×hispidioides seed germinated at a significantly lower rate than F. hispidioides (Fisher's exact P < 0.01). Nonhybrid F. hispidioides had the highest survivorship among germinated seeds (Table 5) and ANOVA did not detect a significant effect of pollen donor treatment (P= 0.284) or maternal tree (P= 0.307). However, 2 × 2 contingency tests showed that F. morobensis×hispidioides and F. pachyrrhachis×hispidioides seed had significantly lower survivorship than nonhybrid F. hispidioides (Fisher's exact P < 0.01). Survivorship of seedlings was generally low across all seedling types due to fungal infections, followed by overly dry conditions in the growth chamber.

Table 5.  Germination and survival of hybrid and nonhybrid seeds. Totals are pooled over all experimental figs/replicates that developed seed. Ficus hispidioides was the maternal parent for all crosses. As such, all rows correspond to hybrid seed except that of F. hispidioides.
Pollen donorTotal n seeds collectedTotal n seeds germinatedTotal n germinated seeds plantedTotal n seedlings survived
F. congesta 177115 8232
F. hahliana 0 0  0  0
F. hispidioides 371327 8036
F. morobensis  61  52  45  2
F. pachyrrhachis 27824513631

Seedlings of all types grew at comparable rates (Fig. 3, growth data archived in the Dryad Digital Depository doi: 10.5061/dryad.c3h3v). Pollen donor species had no significant effect on the height (P= 0.958) or length of the longest leaf (P= 0.569) after 139 days. The maternal tree also had no effect on the height (P= 0.975) or length of the longest leaf (P= 0.918) after 139 days.

Figure 3.

Seedling growth. (A) Average seedling height after 139 days. (B) Average length of the longest leaf after 139 days. Error bars are 1 SD.


As in other studies on hybridization in dioecious figs (Ramirez 1994; Parrish et al. 2003), we found little evidence of gene flow among sympatric fig species. Bayesian clustering analysis of microsatellite genotypes demonstrated little admixture among sympatric species and that naturally occurring hybrids are rare in the study area. Approximately 1–2% of the sampled genotypes could be regarded as putative hybrids assuming a liberal migration rate of 10%. With a more conservative migration rate of 1%, which aligns more closely with Bayesian estimates, less than 1% of individuals are considered hybrids. Given that the majority of individuals were assigned to only one ancestral population with high probability and estimated migration rates were low, the pattern suggests limited backcrossing and introgression that might be explained by the results of our pollination experiment.

Our experimental method of bypassing host recognition and passage through the ostiole resulted in hybridization in three of four heterospecific crosses. Comparable seed development in F. hispidioides from conspecific and heterospecific crosses involving several close relatives argues against postpollination barriers to heterospecific fertilization or embryogenesis. Growth rates of hybrid and nonhybrid seed appeared to be comparable although we admit that statistical power to separate the effects of individual pollen donors and maternal trees was weak. Germination of F. congesta×hispidioides and survivorship of F. morobensis×hispidioides and F. pachyrrhachis×F. hispidioides was in fact lower than nonhybrids in the greenhouse. Selection against hybrids might indeed act as a mechanism of reproductive isolation in nature but our experiments on functionally male figs suggest a more immediate reproductive isolating mechanism.

Cross-pollinating wasps failed to achieve fitness in a nonnatal host species. Wasps induced gall development in a novel host but their offspring did not reach maturity. That gall formation was initiated in some proportion of all heterospecific treatments suggests that cross-pollinating wasps were at least capable of oviposition in the novel host. Reduced rates of gall formation in F. hispidioides by species other than C. dentifer could be attributed to morphological mismatches between ovipositors and style lengths (Weiblen 2004), whereas physiological mechanisms might explain why galls initiated in heterospecific treatments failed to reach maturity. Such mechanisms could include the failure of larvae in a novel host to induce the proliferation of the nucellus or failure to feed upon it, as required by the diet. Host sanctions have also been detected in other fig lineages (Jander and Herre 2010) and there is recent speculation that figs may selectively abort figs that have been self-pollinated (Gates and Nason 2012).

In any event, the failure of pollinators to colonize a novel host suggests that species-specific recognition of suitable hosts may be strongly selected. Even newly emerged female resident pollinators of F. hispidioides refused to enter experimental figs through a Pasteur pipet, whereas females removed from receptive figs while in the act of pollination readily entered experimental figs. This suggests two aspects of wasp behavior. First, attraction to flowers and subsequent oviposition and pollination behaviors are conditioned upon having identified a suitable host and passed through an ostiole. Second, after a wasp has passed through, it does not discriminate among host species. The chemosensory attraction and discriminatory behavior of wasps, therefore, only occurs prior to oviposition and indeed prior to entering a fig. As antennal segments bearing sensillae are torn from wasps’ heads in the process of entering the fig, we might not expect to find discriminatory behavior beyond this life-history stage. This idea is also supported by the low number of heterospecific wasps found in naturally pollinated syconia of these sympatric dioecious figs (Weiblen et al. 2001; Moe et al. 2011).

Volatile chemical cues are known to play a role in pollinator host choice. Studies have shown that fig volatile cues at receptivity are species-specific (Ware et al. 1993; Grison-Pige et al. 2002b; Proffit et al. 2009) and wasps are attracted to the volatile bouquet of particular hosts (Bronstein 1987; Ware and Compton 1994; Grison-Pige et al. 2002a; Chen et al. 2009). Recently, Lu et al. (2009) found molecular evidence of selection on a gene influencing olfactory reception in Ceratosolen solmsi, the pollinator of dioecious, Ficus hispida. Species-specific chemical signals might serve to reinforce reproductive isolation among sympatric species whose hybrids are less fit. Alternatively, pollinator behavior in response to variation in fig chemistry may be one of few isolating mechanisms among genetically compatible sympatric species, which implies that pollinator specificity could potentially play a role in diversification, as modeled by Kiester et al. (1984). If the reproductive consequence of selecting an unsuitable host is as dire as it appears for Ceratosolen in F. hispidioides, selection on wasp behavior alone may be sufficient to reproductively isolate host figs. This may be counted among few specific examples of the often-speculated potential for pollinator adaptation to affect plant diversification in general (Grant 1971; Moe et al., in press).

A previous study on pollinator sharing rates (Moe et al. 2011) did not detect pollinator sharing between F. morobensis and F. pachyrrhachis, but the only two hybrids identified in the field shared these parental species. Artificial hybrids of F. hispidioides and F. congesta grew and survived at comparable rates to nonhybrids and yet pollinator sharing between these parental species is rare (Moe et al. 2011). This suggests that behavioral barriers could be stronger than postreproductive barriers. However, lifetime fitness estimates of hybrids are needed to evaluate the potential for reinforcement to shape pollinator behavior.

Critical insights on host specificity in the fig pollination mutualism (Machado et al. 2005) has motivated the reinterpretation of examination of phylogenetic and cophylogenetic patterns (Haine et al. 2006; Marussich and Machado 2007; Jackson et al. 2008; Jousselin et al. 2008; Su et al. 2008; Renoult et al. 2009; Azuma et al. 2010; Moe et al. 2011). These patterns invite simple explanation, but it is possible that multiple processes yield similar patterns and a particular process may result in diverse patterns (Irwin 2002; Revell et al. 2008; Cavender-Bares et al. 2009; Crisp and Cook 2009). Working hypotheses on host specificity and hybridization developed from phylogeny require independent testing through other lines of inquiry, experimentation, and analysis. For example, evidence of gene flow among Ficus species has been interpreted as evidence of host switching and low pollinator specificity (Renoult et al. 2009). Our cross-pollination experiment, although unidirectional by necessity, offers an alternative explanation. We found that gene flow could occur among sympatric species in at least one direction without requiring the successful colonization of a new host. This finding highlights the importance of examining ecological context and the reproductive consequences for mutualistic partners in concert with patterns of cophylogeny and contemporary host-pollinator associations.

Our results invite the further speculation that patterns of host specificity and codiversification in a system often touted as a textbook example of coevolution need not be maintained by reciprocal selection. Extreme species specificity (Weiblen et al. 2001; Moe et al. 2011) and patterns of codivergence in New Guinea Sycomorus figs (Weiblen and Bush 2002b; Silvieus et al. 2008) can be explained by selection on pollinating wasps imposed by host figs, without comparable selection on host figs imposed by pollinating wasps, similar to adaptive deme formation (Mopper 1996). Highly specific wasp behavior could be an effective reproductive isolating mechanism without significantly reduced fig hybrid fitness. Additional fitness differences may exist which this study was not able to detect, such as infertility of hybrids, or reduced attractiveness of hybrids to pollinators. However, in theory, these differences need not exist to explain the high specificity of pollinators, the rarity of hybrids, or congruent phylogenetic patterns. The development of evolutionary models of unidirectional selection imposed by one organism on another, where lifecycles of the interacting organisms are interdependent, could evaluate whether reciprocal selection is actually necessary to explain patterns long regarded as products of coevolution.

Associate Editor: C. A. Buerkle


This work was supported by The American Society of Plant Taxonomists, The American Philosophical Society's Lewis and Clark Exploration Fund, The Garden Club of America's Award in Tropical Biology, a Carolyn Crosby Fellowship, a Bernard and Jean Phinney Graduate Fellowship, a Myrna G. Smith International Fellowship, and The Bell Museum of Natural History's Dayton Research Fellowship. Thanks to the New Guinea Binatang Research Center and the Ohu Bush Laboratory for logistical support, to B. Isua, M. Brus, D. Sau, F. Pius, and R. Fafen for assistance in field experimentation, to E. Treiber for molecular work, to C. Berg for consultation on Ficus identification, to R. Shaw, G. Heimpel, P. Tiffin, D. Althoff, K. Segraves, J. Nason, and two anonymous reviewers for comments and suggestions.