Floral scent and species divergence in a pair of sexually deceptive orchids

Abstract Speciation is typically accompanied by the formation of isolation barriers between lineages. Commonly, reproductive barriers are separated into pre‐ and post‐zygotic mechanisms that can evolve with different speed. In this study, we measured the strength of different reproductive barriers in two closely related, sympatric orchids of the Ophrys insectifera group, namely Ophrys insectifera and Ophrys aymoninii to infer possible mechanisms of speciation. We quantified pre‐ and post‐pollination barriers through observation of pollen flow, by performing artificial inter‐ and intraspecific crosses and analyzing scent bouquets. Additionally, we investigated differences in mycorrhizal fungi as a potential extrinsic factor of post‐zygotic isolation. Our results show that floral isolation mediated by the attraction of different pollinators acts apparently as the sole reproductive barrier between the two orchid species, with later‐acting intrinsic barriers seemingly absent. Also, the two orchids share most of their fungal mycorrhizal partners in sympatry, suggesting little or no importance of mycorrhizal symbiosis in reproductive isolation. Key traits underlying floral isolation were two alkenes and wax ester, present predominantly in the floral scent of O. aymoninii. These compounds, when applied to flowers of O. insectifera, triggered attraction and a copulation attempt of the bee pollinator of O. aymoninii and thus led to the (partial) breakdown of floral isolation. Based on our results, we suggest that adaptation to different pollinators, mediated by floral scent, underlies species isolation in this plant group. Pollinator switches may be promoted by low pollination success of individuals in dense patches of plants, an assumption that we also confirmed in our study.

species usually requires a combination of different types of barriers, and their hierarchical importance is often taxon-specific (Coyne & Orr, 1998;Rieseberg & Willis, 2007).
Such competition can take the form of a negative association between pollination success and density or population size of conspecifics using the same kinds of pollinators. This is expected to be a more common situation in deceptive plants, where pollen limitation is often severe and pollinators learn to avoid plants after unsuccessful visits (Fritz & Nilsson, 1994;Johnson & Schiestl, 2016).
While the emphasis in sexually deceptive orchids has been on pollinator attraction and its role in reproductive isolation, little is known about the effects of mycorrhizal fungi on reproductive isolation and speciation (Roche et al., 2010). For germination, orchids strongly depend on soil fungi. Their small seeds lack starch reserves and only germinate upon colonization by a soil fungus (Dearnaley, Perotto, & Selosse, 2016) that provides them with nutrients supporting germination until they eventually become photosynthetic. Orchids often depend on specific fungi (e.g., Tulasnellaceae or Serendipitaceae), collectively called rhizoctonias (Dearnaley, Martos, & Selosse, 2013).
Recent studies have hypothesized that associations to specific mycorrhizal fungi may act as an extrinsic post-zygotic barrier by preventing the germination of hybrid seeds through the lack of a proper fungal partner (Jacquemyn, Brys, Cammue, Honnay, & Lievens, 2011;Scopece et al., 2008). Changes in mycorrhizal fungi have thus the potential to drive orchid speciation (Bateman et al., 2014;Otero & Flanagan, 2006;Waterman & Bidartondo, 2008), and there is evidence, although limited, that the sharing of similar fungi is prerequisite for successful establishment of hybrids in orchids (Schatz et al., 2010).
In this study, we investigated multiple potential reproductive barriers and inferred mechanisms of diversification in two species of the Ophrys insectifera group. We focused on the following specific questions: (1) Which reproductive barriers maintain species boundaries?

| Study species and site
Within the sexually deceptive orchids, the Ophrys insectifera group offers a unique system for investigating reproductive barriers between groups with different pollinators. The monophyletic O. insectifera group consists of three species, namely O. insectifera (Figure 1a), O. subinsectifera, and O. aymoninii ( Figure 1b) (Breitkopf, Onstein, Cafasso, Schlüter, & Cozzolino, 2015;Devey, Bateman, Fay, & Hawkins, 2008). Ophrys insectifera has a wide distribution and is pollinated by males of two digger wasp species (Argogorytes mystaceus and A. fargeii; Figure 1c) (Delforge, 2005;Kullenberg, 1951). Ophrys aymoninii is a narrow endemic found in the southern Massif Central in France and pollinated by males of the solitary bee Andrena combinata ( Figure 1d). It regularly occurs in sympatry with the geographically widespread O. insectifera. A time-calibrated maximum clade credibility tree (Breitkopf et al., 2015) supports a very recent divergence between Ophrys insectifera and Ophrys aymoninii (i.e., in the last 500,000 years).
Our study was performed in the Parc Naturel Régional des Grands-Causses in Aveyron, France during May/June 2010-2013 where the two species flower simultaneously. In total, seven populations were studied in these 4 years (Table S1). For a better visualization of the sympatric occurrence, we collected GPS points of randomly selected plants of both species in the mixed populations (Fig. S1).

| Pre-pollination pre-zygotic isolation: floral isolation (RI floral )
In our study, we mainly focused on ethological floral isolation.
Morphologic isolation does occur in Ophrys, namely through positioning of pollinia on either the head or abdomen of a pollinator, but it is often not sufficiently precise to prevent cross-pollination Vereecken, Cozzolino, & Schiestl, 2010). In the here investigated Ophrys species, pollinia are deposited on the head of the pollinator (Figure 1). Differences in flower size between the species (Triponez et al., 2013) (Table S1). Every plot contained two Ophrys plants from each species, set up in 15-ml falcon tubes filled with water, and positioned in a square (0.2 m distance to the next plant). The pollinia of each species were stained with a distinct color using histologic stains: 2% (weight/volume) Trypan red (72210-25G, Sigma-Aldrich) for O. insectifera and 1% (w/v) brilliant green (B6756-100G, Sigma-Aldrich) for O. aymoninii, as described in Xu et al. (2011). The distance between each plot in a transect was 20 m. The plants were examined every 5 days for pollinia removal as well as deposition of stained massulae (pollen packages) on stigmata. Floral isolation (RI floral ) was calculated as 1 − (total number of interspecific pollination events/total number of intraspecific pollination events) (Scopece, Musacchio, Widmer, & Cozzolino, 2007). This value can vary between 0 (no floral isolation) and 1 (total floral isolation). In 2010, two experimental transects were performed at the location Avey3 and one in Avey2. In 2011, one experimental transect was performed at the locations Avey2 and Avey3. In 2012, one experimental transect was performed each at the locations Avey2, Avey3, Avey4, and Avey6.

| Post-pollination pre-zygotic isolation: fruiting success (RI fruting)
To measure post-pollination pre-zygotic barriers, manual intra-/interspecific crosses were performed between the two species using 10 plants of O. insectifera and 7 plants of O. aymoninii. Intraspecific and interspecific crosses were performed with each of the two species (no plant was self-pollinated). Post-pollination pre-zygotic isolation was quantified by counting the number of fruits (fresh fruits with seeds) on inter-and intraspecific crosses. RI fruiting was calculated as 1 − (mean number of fruits in interspecific crosses/mean number of fruits in intraspecific crosses). In cases where interspecific crosses performed better than the intraspecific crosses (resulting in a negative value for RI), the reproductive isolation value was set to zero (Scopece et al., 2007). Finally, fruits were collected when they were ripe and dried in silica gel (Fluka).

| Post-zygotic isolation-embryo development (RI embryo )
To measure post-zygotic isolation in the form of embryo development, seeds of the fruits were used for quantification of developed embryos.
A random sample of 300 seeds from each fruit was examined under a light binocular microscope (Olympus SZH-ILLD) at 64× magnification. Seeds with a well-developed embryo and those without or with weakly developed embryos were counted. Well-developed embryos F I G U R E 1 Pictures of the species investigated in this study. Flowers of (a) Ophrys insectifera, and ( Scopece et al. (2007). In cases where interspecific crosses performed better than the intraspecific crosses, the reproductive isolation value was set to zero (Scopece et al., 2007).

| Molecular barcoding of mycorrhizal fungi
We  As above, 1 μl of the natural sample and 1 μl octadecane solution

| Flower odor sampling and chemical analysis
(1 ng/μl) as an internal standard were injected into the GC-MS (gas chromatography-mass spectrometry) system. For tentative identification of natural compounds, their mass spectra were compared with data reported in the NIST library and by Francke et al. (2000). For unequivocal structure assignments, mass spectra and gas chromatographic retention times of natural products were successfully compared with the following standards: tricosane, tetracosane, pentacosane, hexacosane, heptacosane, nonacosane (all purchased from Sigma-Aldrich); (Z)-9pentacosene, (Z)-9-heptacosane, (Z)-9-nonacosene (all from the stock collection of WF); octyl palmitate and nonyl palmitate (synthesized by WF through the reaction of palmitoyl chloride and the two primary alkohols following standard procedures). Additionally, four unknown compounds and docosenamide were included in the quantitative analysis due to their high abundance. Volatiles that were consistently detected in good signal-to-noise levels and all those that elicited EAD responses (in total 16 volatiles) were used for statistical comparison of relative amounts (amounts of individual components in relation to the total amounts of those 16 target compounds). Because heptacosane (C 27 ) and nonyl palmitate were found to co-elute on a DB-5 column, all samples were run on a J&W 123-7032 DB-Wax (30 m × 0.25 μm) column with splitless injection at 50°C (1 min), followed by a programed increase in the oven temperature to 230°C at a rate of 10°C/min; hydrogen was used as carrier gas with a flow rate of 2.0 ml/min. The J&W 123-7032 DB-Wax column was used for heptacosane (C 27 ) and nonyl palmitate to elute at different retention times, which were additionally identified and confirmed by running standards of both compounds. Based on the ratios of peak areas of the two compounds in these samples, the relative amount of each compound in all natural samples was estimated. Switching System, Agilent Technologies, Palo Alto, CA, USA) was used to direct 50% of the eluate, which was admixed to a purified and humidified air stream, over the excised antenna. EAD signals and FID responses were simultaneously recorded using Syntech software.

| Electrophysiological recordings
EAD responses were judged "real" if reproducible in at least four bee individuals. Compounds releasing EAD responses were identified by comparison of retention times of samples with those of synthetic standard compounds.

| Behavioral assays
Behavioral assays were conducted to test whether the production of key volatiles can induce a pollinator switch. For all assays, freshly

| Phylogenetic analysis
Bayesian analysis was conducted with a single runs of a Markov-chain Monte Carlo (MCMC) chain for one million generations with tree sampling every 500 generations (CBSU BioHPC, Cornell University). Runs converged at split frequencies below 0.01 after 600,000 generations.
The combined dataset had a length of 957 base pairs and Bayesian inference produced a single tree.

| Ploidy level
To exclude differences in ploidy as a reproductive barrier, the rela-  (Table 1) between the species, a general linear model was run with each plant parameter as dependent, species as fixed, and population as random factor. To analyze the impacts of all measured parameters on fecundity, a general linear model was calculated with relative fruit set as a dependent variable, species as fixed, population as random factor, and "number of conspecifics," "plant height," and "total number of flowers" as covariates. The interaction between "no.
of conspecifics" and species was also included, to assess whether density-dependent fruit set differs between species. All covariates were z-transformed (mean = 0, SD = 1) on species and population level before analysis. Because neither species nor population had a significant effect of relative fruit set, we also calculated a multiple linear regression with relative fruit set as dependent, and "no.

| Floral scent
Differences in the relative amounts of individual floral scent compounds between the two species were analyzed using Mann-Whitney U-tests. In addition, we transformed our matrix of relative amounts of compounds (originally in % of the total blend) with a Hellinger transformation, which is a relativization by row (sample unit) totals, followed by taking the square root of each element in the matrix, to make the floral scent data that contained many zero values

| Post-pollination pre-zygotic isolation (fruiting success)
From 26 hand crosses, the six intra-and six interspecific crosses with O. aymoninii as pollen receiver led to equal fruit set (Figure 2b), resulting in a RI fruiting value of 0. In O. insectifera, the six interspecific crosses had an even higher fruiting success than the eight intraspecific crosses resulting in a negative RI fruiting value of −0.333, subsequently set to zero. Thus, no reproductive barrier at this stage was found (Figure 2b).

| Post-zygotic isolation (embryo development)
In both species, interspecific crosses showed a tendency to higher yield of seeds with well-developed embryos than the intraspecific crosses, albeit not significant (Figure 2c). For both species, a negative RI embryo T A B L E 1 Mean (±SD) values of traits measured in the two species in six natural populations. None of the traits was consistently different between the species, but several (maked with an asterisk) showed a significant interaction between species and population (GLM, p ≤ .001)
Based on this large overlap, differences of mycorrhizal partners are unlikely to form a reproductive barrier.

| Fruit set and density
None of the measured traits (plant height, fruits, and density) differed consistently between the species, but for several of them, a highly significant interaction between species and population was found (Table 1). In our general linear model, "number of conspecifics" (=density) was the only factor with a significant effect on relative fruit set (Table 2). In a multivariate regression, it was shown that the effect of density on relative fruit set was significantly negative (Table 3).

| Floral scent and GC-EAD
Of all floral volatiles in the samples, the wax esters, octyl palmitate and nonyl palmitate, as well as the alkenes, (Z)-9-pentacosene and (Z)-9-heptacosene, were found to elicit EAD responses in Andrena combinata males (Fig. S3); one additional compound, tricosane, also elicited reproducible EAD responses, but this compound was found in both species in the same amounts (Table 4) and was thus not considered for the bioassays. Overall, of the 16 most abundant floral volatiles, 12 were chemically identified (Table 4). Of these 16 compounds, 13 were found to differ significantly between species  Table 4). The overall bouquet was different, too ( Figure 4). The most striking differences were found within the relative amounts of EADactive esters and alkenes. Octyl palmitate and nonyl palmitate, as well as (Z)-9-pentacosene and (Z)-9-heptacosene, were found to be present in significantly higher amounts in O. aymoninii than O. insectifera (Table 4).

| Phylogenetic analysis
Our phylogenetic analysis using three nuclear markers (BGP, LACS, and LFY) from 18 specimens of the O. insectifera group showed no clear species clustering between the species. This was as expected from the morphology-based taxonomic classification and suggests a close relationship between the three members of the O. insectifera group (Fig.   S4). In contrast, a moderate geographical clustering was evident in the T A B L E 3 Multiple linear regression with relative fruit set as dependent variable, and number of consepcifics, number of flowers, and plant height as explanatory variables. Number of conspecifics, used as a proxy for density, had a significant negative effect on relative fruit set. Statistically significant values are given in bold analyses. Results found here mirror those already found by Breitkopf et al. (2015) in all terminal clades of their Ophrys phylogenetic analysis and point toward an incomplete lineage sorting scenario as consequence of very recent radiation of species groups in this genus.

| DISCUSSION
Experimental investigations on the evolution and nature of reproductive isolation barriers can provide insights into the process of diversification (Coyne & Orr, 1998Moyle et al., 2004;Schemske, 2010;Scopece et al., 2007Scopece et al., , 2008Widmer et al., 2009). In our study, floral isolation mediated by floral scent appears to be the only significant barrier to gene flow between two recently diverged orchid species.
Although the predominant importance of ethological floral isolation in sexual mimics has also been shown in other species (Scopece et al., 2007;Sedeek et al., 2014;Xu et al., 2011), our study adds information on the traits underlying floral isolation and shows a negative association between fruit set and plant density, a situation that may favor a pollinator switch. In addition, it considers mycorrhizal fungi as a factor for species isolation, which has rarely been done in orchids and never before in the genus Ophrys.
Species-specific mycorrhizal fungi may mediate isolation in two ways: first, as a post-zygotic barrier, hybrids may suffer low fungal recruitment success and hence low germination or seedling survival (Jacquemyn et al., 2011); second, non-randomly distributed fungal species may also influence the habitat preference of their host species, leading to ecological segregation. Our investigations detected a broad sharing of mycorrhizal fungi, with a marked preference for one Tulasnellaceae species. Using species delineation based on 3% ITS divergence is unlikely to have masked cryptic Tulasnellaceae species: first, this is a usual threshold and ITS species delineation is validated in Tulasnellaceae by the fact that it is congruent with other genes (Linde, Phillips, Crisp, & Peakall, 2014); second, lowering the threshold to 1.5% did not change OTU delineation in our work. This fungus family is common in several Ophrys species (Jacquemyn, Brys, Waud, Busschaert, & Lievens, 2015;Pecoraro, Girlanda, Liu, Huang, & Perotto, 2015). Only a small difference in mycorrhizal partners was found earlier in closely related species of the genus Orchis in sympatry (Jacquemyn et al., 2011), and sexually deceptive orchids of the genus Chiloglottis were shown to share a narrow taxonomic group of Tulasnella fungi (Roche et al., 2010). Furthermore, a recent study in the sexually deceptive orchid Caladenia was also showing a strong overlap in mycorrhizal partners suggesting little contribution to reproductive isolation (Phillips, Barrett, Dalziell, Dixon, & Swarts, 2016). The consequence of the sharing of mycorrhizal fungi makes specificity of mycorrhizal symbiosis unlikely to contribute to reproductive isolation, or to enhance pre-zygotic barriers.
Alternatively, adaptation to new pollinators may be fueled by competition for pollination in large plant populations, in the form of negative density-dependent fecundity (Waser & Campbell, 2004). In our study, relative fruit set was indeed negatively associated with number of conspecifics growing close by, suggesting plants have better fruit set when growing isolated or being rare, given similar pollinator abundances. Lower fruit set in dense patches can be explained by negative associative learning of pollinators that unsuccessfully attempted to copulate with flowers Peakall, 1990;Wong & Schiestl, 2002). The avoidance of patches of plants may lead to less visits to each individual plant in a dense population compared to sparsely distributed individuals. Such competition for pollination may promote a pollinator switch because individuals attracting a new pollinator are necessarily rare in the beginning of this process, and thus may enjoy increased pollination success.
Our data suggest that the requirement for the attraction of a novel Andrena-pollinator in O. insectifera is a mutation leading to elevated alkene/ester production. As yet we do not know, however, whether elevated alkenes/esters also reduce the attraction of the pollinator of O. insectifera, which is a necessary prerequisite for isolation against backcrossing into wild-type O. insectifera. In previous experiments in sexual mimics of the genus Ophrys and Chiloglottis, however, it has been shown that hetero-specific scent clearly reduces pollinator attraction . Nevertheless, some overlap in pollinators is likely during the switching phase, unless antagonistic pleiotropy between attractive scent compounds would prevent a phenotype emitting a blend of both. This major obstacle to a sympatric speciation scenario, namely recombination through gene flow breaking down associations between co-adapted alleles (Coyne & Orr, 2004), would be prevented by a mono-or oligogenetic basis of the trait mediating floral isolation, in our case the alkene/ester production.

| CONCLUSION
This study shows that floral isolation, that is, specific attraction of pollinators through floral odor, apparently acts as the sole reproductive barrier for maintaining species integrity in the O. insectifera group. Moreover, this study indicates that female reproductive success was negatively associated with density and that few scent compounds can induce-at least occasional-copulation attempts by a new pollinator. However, for a better understanding of speciation scenarios in sexual mimics, a better resolved phylogenetic framework is desirable, to confidently assign recently diverged pairs of sister species within the flock of genetically often very similar species (Breitkopf et al., 2015). Furthermore, in our specific study system, a better understanding of the chemical ecology of the pollinator of O. insectifera is needed to predict its behavioral responses to variation in floral scent. Such data would allow to more confidently predict patterns of introgression during the establishment of distinct scent types. Finally, a better understanding of the molecular background of key traits for floral isolation (Sedeek et al., 2016) will help unravel origin and maintenance of floral scent differences even in the face of occasional gene flow, and thus better understand speciation in this intriguing group of plants.