Do floral and ecogeographic isolation allow the co‐occurrence of two ecotypes of Anacamptis papilionacea (Orchidaceae)?

Abstract Ecotypes are relatively frequent in flowering plants and considered central in ecological speciation as local adaptation can promote the insurgence of reproductive isolation. Without geographic isolation, gene flow usually homogenizes the allopatrically generated phenotypic and ecological divergences, unless other forms of reproductive isolation keep them separated. Here, we investigated two orchid ecotypes with marked phenotypic floral divergence that coexist in contact zones. We found that the two ecotypes show different ecological habitat preferences with one being more climatically restricted than the other. The ecotypes remain clearly morphologically differentiated both in allopatry and in sympatry and differed in diverse floral traits. Despite only slightly different flowering times, the two ecotypes achieved floral isolation thanks to different pollination strategies. We found that both ecotypes attract a wide range of insects, but the ratio of male/female attracted by the two ecotypes was significantly different, with one ecotype mainly attracts male pollinators, while the other mainly attracts female pollinators. As a potential consequence, the two ecotypes show different pollen transfer efficiency. Experimental plots with pollen staining showed a higher proportion of intra‐ than interecotype movements confirming floral isolation between ecotypes in sympatry while crossing experiments excluded evident postmating barriers. Even if not completely halting the interecotypes pollen flow in sympatry, such incipient switch in pollination strategy between ecotypes may represent a first step on the path toward evolution of sexual mimicry in Orchidinae.


| INTRODUC TI ON
Adaptation to different environmental conditions or habitats may promote the evolution of genetically different forms of a species, that is, ecotypes (Turesson, 1922). The process is relatively frequent in flowering plants with wide distribution and its role in plant speciation has been widely recognized, as ecotypes may represent a first step in the accumulation of reproductive isolation along the so-called speciation continuum (Lowry & Gould, 2016;Nosil, 2012). Typically, different ecotypes are adapted to different ecological conditions hence geographic isolation is the main barrier for preventing their meeting and, eventually, intermixing (Baack et al., 2015). However, when different ecotypes come into secondary contact and/or occur in proximate/geographically close habitat that may allow a large amount of gene flow among ecotypes, introgression and admixture, rather than reinforcement of phenotypic divergence, are the most likely outcomes (Sancho et al., 2018;Zitari et al., 2012). This process is considered a reversal along the speciation continuum and is among the main causes of loss of allopatrically acquired biodiversity (Seehausen et al., 2008).
Adaptation to different habitats may also drive the evolution of partially reproductively isolated ecotypes that can persist in sympatry or parapatry despite some gene flow. This is often due to ecologically based prezygotic mechanisms that can involve, for instance, flowering time or pollination strategy (Briscoe Runquist et al., 2014;Lowry et al., 2008). Plant species with wide geographic ranges may experience different pollinator set (Johnson, 2006(Johnson, , 2010Van der Niet et al., 2014), varying in absolute or relative pollinator composition. This may lead to a different strength and direction of the pollinator-mediated selection and, consequently, to a geographic variation in floral traits subject of selective pressure (Anderson et al., 2010;Newman et al., 2014). Indeed, local adaptation of ecotypes to different pollinator species can initiate speciation (Van der Niet & Johnson, 2009;Sobel & Streisfeld, 2015). In this circumstance, recently diverged lineages can have accumulated local adaptation to different pollinators that impedes or slow down their intermixing, hence, an intraspecific polymorphism can be established with the coexistence of two or more morphs phenotypically differentiated (Leimar, 2005). Partial reproductive isolation between ecotypes can help the maintenance of genetic differences in sympatry as long as there is strong pollinator-mediated divergent selection that overwhelm the presence of some ongoing gene flow (Rymer et al., 2010). Accordingly, most of the studies that have examined transition between different ecotypes of the same plant species highlighted the importance of pollinator-assortative mating (Anderson et al., 2010;Newman et al., 2014). At the same time and as expected within species, the presence of postpollination/postzygotic barriers has only been found between ecotypes that were also differing in ploidy level (i.e., cytotypes; Pegoraro et al., 2016;Husband & Sabara, 2004; but see Richards & Ortiz-Barrientos, 2016 for some notable exceptions).
Here, we investigated the factors that allow the maintenance of phenotypic divergence in the sympatric orchid ecotypes A. papilionacea subsp. papilionacea and A. papilionacea subsp. grandiflora (hereafter referred to as A. p. papilionacea and A. p. grandiflora, respectively).
The pollination strategy of A. p. papilionacea has been investigated in previous studies that found an important contribution of male hymenopterans of different species (Scopece et al., 2009;Vogel, 1972;Vöth, 1989). Dressler (1981) suggested that this strategy of male insect attraction could represent the first step in the evolution of sexual mimicry. The pollination of A. p. grandiflora remains fairly little known but the two ecotypes show evident phenotypic differences that suggest different pollination strategies. In particular, several floral clues and the lack of nectar (true for both ecotypes) indicate a generalized food-deceptive strategy for A. p. grandiflora, commonly found among other members of the genus Anacamptis ( Van der Cingel, 1995). Accordingly, and differently from A. p. papilionacea, A. p. grandiflora shows a large flattened labellum and prominent nectar guides that are typical flower traits involved in generalized food deception (Johnson & Schiestl, 2016). While pollination by sexual mimicry relies solely on male pollinators, fooddeceptive pollination mostly relies on female pollinators, foraging for nectar or pollen reward. Female pollinators also show a very different behavior than males when landing on the flowers (Ne'eman et al., 2006). Thus, the adoption of female versus. male pollinators may enhance assortative mating and produce strong disruptive selection between morphs, for example, on flower morphology and scent. Concordantly, Scopece et al., (2015) found differences in pollen transfer efficiency when examining allopatric populations of A.
p. papilionacea and A. p. grandiflora suggestive of different pollinator behavior.
In this study, we aim to understand how the two A. papilionacea ecotypes remain distinct even when coexisting. To fulfill this aim, we addressed the following specific questions: 1. Are there differences in pollination strategy between the two ecotypes? 2. What is the extent of phenotypic differentiation between A. p. papilionacea and A. p. grandiflora and is it maintained in sympatric populations?
3. Which factors contribute to reproductive isolation? 4. Is there any difference in ploidy level between the two ecotypes? 5. Are there any intrinsic pre-and postzygotic barriers that contribute to the maintenance of the two ecotypes?

| Study system and study areas
Anacamptis papilionacea is a Mediterranean orchid within a clade of food-deceptive species (Aceto et al., 1999). It is self-compatible but needs insects to transfer the pollen (Scopece et al., ,2007(Scopece et al., , , 2009). This species contains two most common ecotypes, A. papilionacea subsp.
papilionacea and A. papilionacea subsp. grandiflora (Figure 1), that are mainly allopatric but coexist in some Mediterranean regions. In some of these contact zones (e.g., in Southern Italy; Scopece et al., 2009), A. papilionacea populations show a clear prevalence of one ecotype while in others, as on Sardinia island, both ecotypes co-occur without any detectable genetic differences (Arduino et al., 1995).
A. p. papilionacea inflorescences narrow toward the top and contain 4-15 flowers with reddish-violet labellum in the outside part and that tend to disappear in the lucid center. Overall, A. p. papilionacea displays some typical floral traits for generalized food deception (as a long spur and a large colored labellum). However, differently from other food-deceptive species, A. p. papilionacea has been found to be primarily pollinated by male hymenopterans, for example, Eucera tuberculata males in a population on Elba Island (Vogel, 1972), Eucera bidentata males in some Greek populations (Vöth, 1989), Eucera nigrescens males (Cozzolino et al., 2005) or Anthophora crinipes males (Scopece et al., 2009) in southern Italy. This unusual attractiveness for male hymenopterans suggests pollinator attraction based on some sexual signals and Faegri and Van der Pijl (1979) coined the term of "rendezvous attraction" for this peculiar pollination mechanism. Because volatile signals are key stimuli in the sexual behavior of bees (Kullenberg & Bergström, 1976), the preferential attraction of males by A. p. papilionacea suggests that some olfactory signals are likely involved. Schiestl and Cozzolino (2008), by analyzing A. papilionacea floral scent, found a prominent production of chemical compounds (alkanes and alkenes) similar to those produced by flowers of sexually deceptive species of the related genus Ophrys (Ayasse et al., 2003), supporting the assumption that chemical signals related to mating behavior may be involved in the attraction of pollinators by A. p. papilionacea. Nonetheless, evidence summarized in Van der Cingel (1995) shows that A. papilionacea sensu latu (i.e., A. p. papilionacea and A. p. grandiflora) is also pollinated by other insect classes, even butterflies (Vogel, 1972).
The A. p. grandiflora ecotype mainly differs from A. p. papilionacea by its larger flower size and a very clear pattern of dark reddishpurple venation (nectar guides) on the whitish/pink labellum. Also, the A. p. grandiflora labellum is often larger and flatter than that of A. p. papilionacea. In contrast to A. p. papilionacea, the pollination system of A. p. grandiflora has never been investigated thoroughly, but the presence of marked nectar guides and a large labellum suggests a generalized food deception for this ecotype. A. p. grandiflora has habitat requirements like A. p. papilionacea, yet with a higher altitudinal range, growing up to 2000m altitude. Its distributional range is western Mediterranean and includes Southern France, Spain, Portugal, Morocco, Algeria, Tunisia, Sicily, and Sardinia, but is less commonly found in southern Italy (Baumann, 2006;Ketzschmar, 2007). The present study was conducted in different allopatric (where a single ecotype was present) and sympatric (where both ecotypes coexist) populations on Sardinia island where both ecotypes are common.

| Morphological differentiation
Morphometric analyses were conducted in two sympatric (Campuomu and Magomadas) and two allopatric populations (Porto Alabe for A. p. papilionacea and San Michele for A. p. grandiflora). For each individual included in morphometric analysis, inflorescence height was measured, the number of flowers counted, and two flowers were picked and stored in EtOH 70%. To obtain floral trait measurements, sampled flowers were dissected, and floral parts were placed between two transparent plastic film sheets. These sheets were subsequently scanned to obtain digital images in a 300 dpi TIFF format with a coordinate millimeter paper on the back for reference; measures of floral traits were later obtained using ImageJ 1.33 software (Rasband, National Institutes of Health, USA). We measured F I G U R E  For sample preparation and analysis, we followed a two-step protocol (Dolezel et al., 2007) as described in Xu et al., (2011). The two pollinia were chopped and mashed together with approximately 25 mm 2 leaf material of Phaseolus coccineus (2n, 1C = 1.01 ± 0.4 pg;

| Ploidy level
Bennett & Leitch, 2005) which served as internal standard (IS). The data were processed by using the ratio of integrated peaks of the study organisms and P. coccineus.   of Magomadas. Scent extraction was conducted by picking the labellum from the flower and dipping it for 30 s in 0.2 ml of hexane and removing it thereafter (Schiestl & Cozzolino, 2008). The analysis of the scent and identification of compounds was done using a gas chromatograph (GC) with FID detector and DB-5 column, following previous studies (Schiestl et al., 2000;Schiestl & Cozzolino, 2008;Schiestl & Marion-Poll, 2002). Compounds were identified based on few samples being analyzed with GC with mass selective detection and comparison of mass spectra and retention times of synthetic standards and compounds in natural samples. Absolute amounts of compounds in the samples were calculated by using an internal standard that was added before the analysis.

| Pollinators and floral traits
Double bond position in alkenes was determined using synthetic alkene standards. Alkenes with different double bond positions have different retention times on the DB-5 column used in this study, thus double bond position can be inferred by comparing retention times of natural compounds and synthetic standards (Mant et al., 2005). The cis/trans configuration was not analyzed, but the cis-configuration was inferred because of its more common occurrence in unsaturated fatty acids, the precursors in the biosynthesis of alkenes (Schlüter et al., 2011).
Data were analyzed statistically through a hierarchical clustering on principal component (HCPC) analysis using the R package

| Flowering time estimation and calculation of phenological isolation index (RI phenology )
Phenological data were recorded in 2012 from the middle of March until the end of the flowering time in the sympatric population of Magomadas. This population was visited periodically and the number of flowering individuals for each ecotype was recorded. Phenological isolation index was calculated as described in Lowry et al., (2008).

| Pollen staining experiment and calculation of floral isolation index
In 2011 (Peakall, 1989;Xu et al., 2011). Inflorescences were controlled daily with a 10x magnifying glass to detect pollen removal and deposition and were replaced with new inflorescences every five days.

| Hand-pollination experiment and calculation of postmating isolation indices
Hand-pollination experiments were performed in the Botanic Garden of Cagliari (Sardinia), during the spring of 2017, to estimate postmating isolation indices.
Plants were collected from natural allopatric populations of A. p.
papilionacea and A. p. grandiflora and, to prevent uncontrolled pollinations, were placed in cages covered with a thin nylon net prior to flowering. Ripe fruits, when produced, were collected and stored in silica gel at 4°C. Seeds were subsequently observed under an optical microscope as described in Scopece et al., (2007).

| Morphological differentiation
The three resulting PCAs showed a similar overlap both in sympatric and allopatric populations ( Figure 2).

| Ploidy level
We observed no significant differences in the ratio of internal

| Characterization of pollination strategy
We  In bold significant differences.  (Z)-11-C31). In terms of relative amounts, significant differences were found for many compounds present in both ecotypes (Table 4).
In particular, large differences in relative amount between ecotypes were detected for C27, C28, and (Z)-7-C25. The HCPC analysis con-  p. papilionacea being included in that of A. p. grandiflora ( Figure 5).

TA B L E 3 (Continued)
RI ecogeography , an index which ranges between 0 (no isolation) and 1 (complete isolation), was 0.81 in A. p. grandiflora and 0.25 in A. p.

| Flowering time estimation and calculation of phenological isolation index (RI phenology )
A. p. papilionacea and A. p. grandiflora show slightly different flowering times with A. p. grandiflora flowering earlier than A. p. papilionacea ( Figure 6). RI phenology was 0.07 in A. p. papilionacea and 0.08 in A.

| Floral isolation
Experimental plots with stained pollen staining built in the two years showed similar results with a higher proportion of intra-than interecotype movements (80.2% in 2011 and 75.6% in 2017, respectively; see Table 6). Therefore, we found nonrandom mating and some degree of floral isolation between ecotypes (0.62 in A. p. papilionacea and 0.71 in A. p. grandiflora, respectively; Table 5).

| Postmating isolation
Percentage of fruit production was equal in intra-and interecotype

| D ISCUSS I ON
In this study, we investigated the mechanisms that allow coexistence of two ecotypes (A. p. papilionacea and A. p. grandiflora) of the circum-Mediterranean orchid species Anacamptis papilionacea.
Overall, as expected for ecotypes of one species, postmating barriers were very weak or absent between ecotypes. However, A. p.
papilionacea and A. p. grandiflora showed significantly different geographic distribution and pollination mode, with a distinct prevalence of male pollinators found in A. p. papilionacea. These differences are thus considered key in maintaining ecogeographic and floral isolation between the ecotypes and may represent the initial step toward evolution of sexual mimicry within an orchid lineage with food deception as plesiomorphic pollination system. Still, such difference in geographic distribution and pollination mode is only an initial stage F I G U R E 3 Pollinator sharing between the two Anacamptis papilionacea ecotypes. Pollinator network of A. p. papilionacea (PAP) and A. p. grandiflora (GRA) visualized using the "networklevel" function in the bipartite package68 in R (http://www.R-proje ct.org) of ecological divergence between the two ecotypes, nor we cannot predict if it can move forward.
Ecogeographic isolation is often the earliest reproductive barrier arising among plant ecotypes and incipient species (Sobel, 2014).
The geographic distribution of the two ecotypes in Sardinia reflects their different ecological habitat preferences and build up a strong but highly asymmetric ecogeographic barrier (Table 5). A. p. papilionacea appears to be more strictly linked to a Mediterranean climate ( Figure 5), with a distribution influenced by temperature and precipitation seasonality, while A. p. grandiflora has a wider altitudinal range, occurring up to 2000m altitude, and a broader niche on Sardinia.
These differences in ecogeographic preferences can slow down intermixing of the two ecotypes because geographic partition can decrease the chance of random mating. However, this mechanism appears to contribute more in A. p. grandiflora that has a more exclusive distribution, rather than in A. p. papilionacea whose distribution is almost included within that of A. p. grandiflora. Nevertheless, microhabitat preference of ecotypes (not estimated here) can still contribute to strength the ecogeographical isolation at local scale.
In the Mediterranean basin, the frequent changes of land connection and insularity due to geological events and the habitat fragmentations due to intense anthropogenic pressures promote the frequent occurrence of secondary contact of allopatrically diverged lineages (Feliner, 2014;Pavarese et al., 2011;Zitari et al., 2011). In bold significant differences after one-way ANOVA.

TA B L E 4 (Continued)
pollinator availability) cannot be excluded in allopatric populations in spite of their geographic proximity and overlapping phenology.
Such difference in pollinator attraction (and morphology), however, also occurs in the sympatric populations where the two ecotypes are certainly exposed to the same pollinator regime. Usually, phenological or floral isolation is weak in generalized food-deceptive orchids species pairs because they attract a wide range of different pollinators and thus have a high chance of pollinator overlap and low  (Cozzolino et al., 2005). In that specific case, we argue that the two ecotypes achieve floral isolation by adopting two partly different pollination strategies. This difference is mirrored by the difference found in pollen transfer efficiency in the sympatric population of San Priamo under a common pollinator regime. Our result concords with previous records of pollen transfer efficiency for allopatric A. p. papilionacea and A. p. grandiflora (Scopece et al., 2015), being higher in the former than in the latter ecotype. Difference in pollen transfer efficiency was found to be strictly linked to the exploitation of different pollination strategies, with sexually deceptive species experiencing higher values than food-deceptive species (Scopece et al., 2010).
The two ecotypes also differ in floral hydrocarbon bouquets ( Figure 4) both in terms of presence/absence of specific compounds and in terms of relative amounts of other compounds (Table 4). These hydrocarbons (Mant et al., 2005) and other types of compounds (as polar compounds, Cuervo et al., 2017) are often used by the closely related sexually deceptive Ophrys species for attract their specific male pollinators (Ayasse et al., 2011;Schiestl et al., 2000).
Attraction of male pollinators may be initially facilitated by olfactory signals and only secondarily by morphological adaptation of the flower. There are several cases of orchids, as for instance Disa atricapilla and Disa bivalvata, that look like food-deceptive orchids, yet sexually attract male pollinators by emissions of specific scent bouquets (Steiner et al., 1994).
In other examples, such as in Orchis galilaea, which shares flower shape and color with its allied food-deceptive sister species, only males Halictus bees were found as pollinators (Bino et al., 1982). As hypothesized by Schiestl and Cozzolino (2008), the widespread occurrence of unsaturated hydrocarbons (alkenes) which have a fundamental role in sexual mimicry in the genus Ophrys, may be an exaptation for the evolution of "incipient" sexual deception (sensu Johnson & Schiestl, 2016). Electroantennographic detection (GC-EAD) studies and behavioral assays are needed (Schiestl & Marion-Pol, 2002) for demonstrating whether some olfactory cues play a similarly important role in A. p. papilionacea as found in sexually true deceptive orchids such as the genus Ophrys (Ayasse et al., 2011;Schiestl et al., 2000). As an alternative hypothesis, scent differences between A. p. papilionacea and A. p. grandiflora may be nonadaptive and just the consequence of different patterns of genetic drift or trait correlations (Juillet et al., 2011).
Our pollen staining experiment showed that attraction of different pollinators could lead to floral isolation between the two ecotypes. Indeed, we found that most of pollen movements (around 80%) were intraecotype. In our study system, thus, floral isolation is likely to be the main isolating mechanism between ecotypes in sympatric populations. Nevertheless, even if ecogeographic isolation (mainly for A. p. grandiflora) and floral isolation (for both ecotypes) certainly contributed to the isolation of the two ecotypes, the amount of interecotype pollen movements (about 20%) in sympatry is still not negligible and likely sufficient to lead to a quick genetic homogenization of the allopatrically gained phenotypic divergence.  systems (Hermann et al., 2013;Zu et al., 2016), and/or the presence of a master gene/supergene (i.e., as those controlling local mimicry polymorphism in butterflies, Le Poul et al., 2014)

ACK N OWLED G M ENTS
The authors thank Dr. Pietro Niolu for help in insect classification, Brian Osterwalder for help in data collection, Dr. Philipp Schlueter for help with flow cytometry analysis, and Lucrezia Laccetti for help with data analysis.

CO N FLI C T O F I NTE R E S T
The authors of this manuscript declare no conflict of interest, financial, or otherwise.