Local adaptation to different pollinators is considered one of the possible initial stages of ecological speciation as reproductive isolation is a by-product of the divergence in pollination systems. However, pollinator-mediated divergent selection will not necessarily result in complete reproductive isolation, because incipient speciation is often overcome by gene flow. We investigated the potential of pollinator shift in the sexually deceptive orchids Ophrys sphegodes and Ophrys exaltata and compared the levels of floral isolation vs. genetic distance among populations with contrasting predominant pollinators. We analysed floral hydrocarbons as a proxy for floral divergence between populations. Floral adoption of pollinators and their fidelity was tested using pollinator choice experiments. Interpopulation gene flow and population differentiation levels were estimated using AFLP markers. The Tyrrhenian O. sphegodes population preferentially attracted the pollinator bee Andrena bimaculata, whereas the Adriatic O. sphegodes population exclusively attracted A. nigroaenea. Significant differences in scent component proportions were identified in O. sphegodes populations that attracted different preferred pollinators. High interpopulation gene flow was detected, but populations were genetically structured at species level. The high interpopulation gene flow levels independent of preferred pollinators suggest that local adaptation to different pollinators has not (yet) generated detectable genome-wide separation. Alternatively, despite extensive gene flow, few genes underlying floral isolation remain differentiated as a consequence of divergent selection. Different pollination ecotypes in O. sphegodes might represent a local selective response imposed by temporal variation in a geographical mosaic of pollinators as a consequence of the frequent disturbance regimes typical of Ophrys habitats.
One important mechanism driving angiosperm diversification is pollinator-mediated selection due to pollen limitation and floral isolation (Lowry et al., 2008; Xu et al., 2011). Although the evolutionary role of pollinator-mediated selection remains debated under complex ecological conditions (Kay & Sargent, 2009), pollinator specialization has traditionally been considered one of the prime mechanisms for ecological speciation due to the direct relationship between pollinator specialization and floral isolation (Grant, 1971; Johnson, 2006; Schiestl, 2012).
Pollinator shift frequency in specialized plant lineages depends on the influence of local selective forces (i.e. pollen limitation and pollen transfer efficiency) and the degree of floral isolation and its genetic basis (Schiestl & Schlüter, 2009; Xu et al., 2012a; and references therein). In fact, if the transition to a different pollinator requires changes in several noncorrelated floral traits and attraction signals, then strong and prolonged directional selection is required to promote those changes (Nosil et al., 2009). Such evolutionary transition, involving several traits, will probably proceed gradually. Alternatively, some recently diverged lineages, where pollinator shifts facilitated rapid radiations, are often characterized by related species that attract different specific pollinators because of subtle changes in floral advertisement or shape, rather than large floral structural rearrangements and large changes in signals (Fulton & Hodges, 1999; Schemske & Bradshaw, 1999; Stuurman et al., 2004). These observations suggest that in such specialized plant groups, only one or a few floral traits play a prominent role in preventing pollinator sharing and that even small alterations in key traits may have substantial effects in attracting distinct pollinators or pollinator groups.
Sexually deceptive orchids are typified by a combination of traits mediating specific pollinator attraction, and sexual mimicry of pollinator females achieves pollination specificity (Kullenberg, 1961; Paulus & Gack, 1990a; Bower, 1996). Several recent studies provide strong experimental evidence, demonstrating that floral scent holds the key to specific pollinator attraction, whereas floral display serves a secondary role by increasing floral detection against vegetation background (Vereecken & Schiestl, 2009; Peakall et al., 2010; Xu et al., 2012a). Additional empirical evidence supports these results by showing that artificial manipulation of floral display does not significantly alter pollinator attraction (but see Spaethe et al., 2007). In contrast, floral scent modifications increase or decrease pollinator visitation rates (Xu et al., 2012b). Therefore, a subtle change in floral scent, even in the absence of marked changes in floral display, might potentially attract different orchid pollinators (Xu et al., 2012a).
Recently, the genetic and chemical basis of pollinator specificity in sexually deceptive orchids has been partially elucidated. In Ophrys sphegodes and related species, for instance, the specificity of pollinator attraction is due to quantitative variation in alkenes, which differ in double-bond position and carbon chain length (Mant et al., 2005a; Schlüter et al., 2011a). Therefore, even minor variation in gene expression patterns underlying alkene biosynthesis might be sufficient to produce a different odour bouquet, as in O. sphegodes vs. Ophrys exaltata (Schlüter et al., 2011a; Xu et al., 2012b). Closely related species of sexually deceptive orchids in the same phylogenetic lineage attract different pollinators as a result of qualitative or only quantitative changes in proportions of active odour bouquet compounds. Alterations in compound proportions provide variability for selection to act upon and thus offer the potential for ecological speciation (Schiestl & Ayasse, 2002; Xu et al., 2012a).
Pollinator communities and species population sizes are likely to fluctuate through time, particularly in disturbed and transient habitats as those typically occupied by Ophrys species (Gardiner & Vaughan, 2009; Hutchings, 2010). Indeed, floral specialization has been considered a risky strategy due to strict relationships between the survival/extinction of pollinator species and the plant species dependent on its exclusive pollination service (Johnson & Steiner, 2000). Thus, potential interaction plasticity with the specialized pollinator might allow the plant to attain reproductive success, even under limited presence of the main pollinator, through temporary exploration of locally more abundant secondary pollinators (Waser & Ollerton, 2006). Under these conditions, local selection on plants imposed by a variable geographical and temporal mosaic of potential pollinators could lead to multiple pollination ecotypes (Harder & Johnson, 2009). This adaptive strategy might facilitate pollinator-specialized species survival when short-term pollinator community fluctuations occur as a consequence of habitat alterations, such as those caused by anthropogenic disturbance (Petanidou et al., 2008; Potts et al., 2010).
Whether Ophrys pollination ecotypes are populations, or incipient or actual ecological species (Van Valen, 1976) has been hotly debated among evolutionary biologists (e.g. Bateman et al., 2011; Vereecken et al., 2011). The crucial difference between adaptation to local pollinators (pollination ecotypes) and progenitor-derivative speciation due to a pollinator shift can be identified in the evolution of reproductive isolation associated with the transition to a different pollinator – a transition that is ultimately dependent on the strength of divergent selection and the amount of interpopulation gene flow (Schlüter et al., 2011b).
Ultimately, biological species are defined by reproductive isolation, which is typically weak or absent among geographical races (Slatkin, 1987; Coyne & Orr, 1998). Therefore, reduction in or loss of gene flow (as consequence of the insurgence of some forms of reproductive isolation) between ancestral populations leads to incipient ecological species from local ecotypes, allowing their gene pools to develop independently (Macnair & Gardner, 1998). Nevertheless, because local adaptation often represents the initial stages of ecological speciation, a marked distinction between these sequential stages, based on the level of gene flow between different ecotypes, is often difficult to define (Nosil, 2012). Lexer & Widmer (2008) emphasized the seriousness of this challenge in several ecological species pairs that displayed a considerable amount of interspecific gene flow, and where genomic divergence was only detected at a few loci despite using several genome-wide markers.
To date, few studies have examined the nature and level of reproductive isolation between different ecotypes of a given species (Scopece et al., 2010; and references therein), and geographical isolation is considered the primary isolating factor among races or local ecotypes (Lowry et al., 2008). However, it does not represent an obvious barrier that prevents gene flow between geographically close or adjacent ecotypes/populations (Anderson et al., 2010), a situation that can in principle allow constant gene flow between populations (Räsänen & Hendry, 2008). In such cases, the amount of floral isolation is comparable with the levels of interpopulation gene flow as an indirect estimate of whether different populations represent locally adapted pollination ecotypes that established a form of incipient reproductive isolation (Nosil et al., 2009; Thibert-Plante & Hendry, 2010). In fact, although divergent selection can favour reproductive isolation at a local scale, incipient divergence can be prevented by the homogenizing effect of gene flow between adjacent populations, maintaining species cohesion (Morjan & Rieseberg, 2004). Investigations into transient stages in the continuum between local adaptation and incipient speciation are integral to speciation research. Such studies may serve to identify the selective forces acting on floral traits and establish the evolutionary outcomes of pollinator shift and subsequent floral isolation (Fenster et al., 2004).
In the present study, with the aim of disclosing the effect of changes in preferred pollinators on the establishment of reproductive isolation, we investigated the potential for pollinator shifts in the Mediterranean sexually deceptive orchid Ophrys sphegodes Mill., and compared the levels of floral isolation vs. genetic distance between populations with different dominant pollinators. We estimated intra- and interspecific reproductive isolation and gene flow in O. sphegodes s.s. and the closely related species O. exaltata Ten. The two taxa belong to the same phylogenetic lineage, and Xu et al. (2011) only recently inferred that the species are effectively reproductively isolated.
Materials and methods
The sexually deceptive orchid species Ophrys sphegodes and O. exaltata were chosen for this study. Ophrys sphegodes is a widespread species that inhabits sunny meadows or calcareous grasslands from the British Isles to the southern Mediterranean countries (Delforge, 2006) whereas O. exaltata is more restricted to the Mediterranean region, inhabiting sandy soils in southern France and the central part of the Mediterranean region. In a broad survey of the entire genus, Devey et al. (2008) showed that the two species are phylogenetically closely related. In peninsular Italy, O. sphegodes and O. exaltata frequently occupy similar ecological niches – primarily coastal sandy areas (Delforge, 2006) – and the taxa often occur in ‘mosaic sympatry’ (sensu Mallet et al., 2009). Due to severe habitat fragmentation, particularly where the Apennine Mountain chain separates the Tyrrhenian and Adriatic coasts, both species accommodate a large number of local varieties, either interpreted as local ecotypes or endemic species (cf. Delforge, 2006; Pedersen & Faurholdt, 2007). Overall, the two species exhibit well-established reproductive isolation via different primary pollinators (i.e. Andrena nigroaenea for O. sphegodes and Colletes cunicularis for O. exaltata). However, the species are interfertile and can produce hybrids in the wild (Xu et al., 2011).
We investigated four O. sphegodes populations and three O. exaltata populations along the Italian peninsula (Table 1). For each plant sampled, a piece of leaf tissue was field collected and placed in a plastic bag filled with silica gel. Genomic DNA was extracted using GenElute Plant Genomic DNA Miniprep Kit (Sigma–Aldrich, Milan, Italy). The amplified fragment length polymorphism (AFLP) procedure followed Vos et al. (1995), albeit with modifications (Moccia et al., 2007) and using fluorescent dye-labelled primers. Six primer combinations were chosen from Xu et al. (2011): FAM-EcoRI-AGC/MseI-ACAC, NED-EcoRI-ACC-/MseI-ACTG, HEX-EcoRI-AGC/MseI-ATCG, FAM-EcoRI-ATG/MseI-CGG, NED-EcoRI-AAC/MseI-CGC and HEX-EcoRI-AGC/MseI-CCAA. Fragment separation and detection were conducted on a 3130 Genetic Analyzer (Applied Biosystems, Foster City, CA, USA). GeneScan-500 LIZ (Applied Biosystems) was used as the internal standard. Raw data alignment and fragment-size detection were performed using GeneMapper 3.7 software (Applied Biosystems). Presence or absence of AFLP bands was scored visually. Artefacts and missing data were avoided by only including informative AFLP markers in the binary matrix that could be unambiguously scored for all samples.
Table 1. Examined populations of Ophrys exaltata and Ophrys sphegodes
O. exaltata (O.exaGAR)
Capoiale, Gargano, Puglia, Italy
O. exaltata (O.exaTUS)
Marina di Castagneto, Tuscany, Italy
O. exaltata (O.exaCUMA)
Cuma, Campania, Italy
O. sphegodes ssp. classica (O.sphCLA)
Porto San Stefano, Tuscany, Italy
O. sphegodes (O.sphCUMA)
Cuma, Campania, Italy
O. sphegodes (O.sphGAR)
Capoiale, Gargano, Puglia, Italy
O. sphegodes ssp. argentaria (O.sphARG)
Caldine-Fiesole, Tuscany, Italy
Genalex (Peakall & Smouse, 2006) was run as the macro in Microsoft Excel to calculate genetic distances as the basis for generating a principle coordinate analysis (PCoA) scatter plot. An analysis of molecular variance (amova) was conducted using a matrix of genetic distances between all haplotype pairs. Genetic differentiation was estimated using Φ-statistics, an F-statistics analogue for binary data. Significance of Φ-statistics and variance components was assessed with 999 permutations (Peakall & Smouse, 2006). Pairwise ΦST values based on Jaccard's similarity were calculated from FAMD (Schlüter & Harris, 2006).
structure 2.2 (Pritchard et al., 2000) was applied to investigate population structure. The model implemented in structure set the posterior probability (q) to describe the proportion of an individual genotype originating from each of K categories. Following the method described in the study by Evanno et al. (2005), we tested K from 1 to 7 (i.e. the number of populations sampled) with a burn-in of 50 000 steps, followed by 300 000 MCMC iterations and 10 replicates to confirm stabilization of summary statistics. Estimates were carried out under the admixture model, allowing for correlated allele frequencies on all sampled individuals, while ignoring sampling localities. The output obtained from structure was graphically displayed with distruct (Rosenberg, 2004). The genomic basis of differentiation between O. sphegodes populations from Cuma and Gargano was analysed using a genome scan. Accordingly, to identify putative genomic targets of divergent selection, an FST outlier scan was conducted using the Dfdist package (Beaumont & Nichols, 1996), which applies the Bayesian allele frequency estimate method developed by Zhivotovsky (1999). The analysis was conducted as in other studies (e.g. Minder & Widmer, 2008; Pérez-Figueroa et al., 2010), excluding loci with an allele frequency over 0.99 and using mean FST trimmed at the 30% level. This trimmed mean FST was chosen as the target average for Dfdist to simulate FST null distribution values (50 000 realizations), assuming a θ parameter of 0.05; variation in this θ parameter was reported to have very little impact on results (e.g. Minder & Widmer, 2008, and references therein). FST outliers were identified as data points outside the 95% confidence interval (outside the 0.025 and 0.975 quartiles, and at P <0.025).
The potential effects of geography and pollinator differentiation on genetic structure within O. sphegodes were separated using a generalized linear model (GLM). Pairwise ΦST values were modelled with the explanatory variable geographical distance and shared pollinator, as well as an interaction term between the two variables. A geographical distance matrix was computed with Arcgis 9.2. The analysis was performed in R 2.15.2 (R Development Core Team, 2012), as described in the study by Schlüter et al. (2011b).
Pollinator choice experiments
Pollinator fidelity was tested through pollinator choice experiments between O. sphegodes and O. exaltata individuals. Plant inflorescences were collected from natural populations along the Tyrrhenian (Cuma, Campania) and Adriatic (Gargano, Puglia) coasts where O. sphegodes and O. exaltata are sympatric. The study populations co-flower at the end of March/beginning of April. Pollinator choice plots with individuals of both species from the two coasts were established at Cuma and Gargano. We established 12 choice plots from 17 March to 25 March 2011 available during a two-hour period from 09.00 to 11.00 h, five choice plots were established in Cuma and seven plots in Gargano, intermixed with natural populations. Each plot consisted of four inflorescences, with one individual at a time of each species from both coasts. The inflorescences were randomly placed, 30 cm apart, in flowering bushes along sandy paths. Inflorescences in the choice plots were replaced with new ones after every pollination event or, in the absence of any visit, after 30 min.
Male bees patrolled along sandy footpaths, where numerous nesting places of female solitary bees were observed; the male bees also checked for females in flowering shrubs (e.g. Rosmarinus officinalis, Spartium junceum, Emerus major) where the female bees foraged for nectar. Pollination events were recorded only when the bee was successfully caught after an observed pseudo-copulation leading to pollinarium removal. The bees were later identified by comparison with an Ophrys pollinator reference collection at the University of Zürich, Switzerland.
An additional pollinator-baiting experiment was performed from 10.00 to 13.00 h on the 21 March 2012 on the Tyrrhenian Coast by exposing freshly cut inflorescences of O.sphCUMA during local pollinator hours. Visiting bees were caught only after observation of pseudo-copulation with pollinia removal, and inflorescences were replaced following each pollination event.
Floral hydrocarbons produced by the two species were analysed for plants from all populations for which pollinator choice experiments were performed, that is, O. sphegodes from Cuma, Tyrrhenian Coast (O.sphCUMA) and Gargano, Adriatic Coast (O.sphGAR), and O. exaltata from Cuma, Tyrrhenian Coast (O.exaCUMA) and Gargano, Adriatic Coast (O.exaGAR). For each sampled individual plant (the same plant used for genetic analyses, plus additional individuals bearing up to 20 flowers), one labellum of an unpollinated flower was placed in a 2-mL glass vial (Supelco, Bellefonte, PA, USA) and rinsed in 500 μL hexane (HPLC grade, Fluka, Buchs, St. Gallen, Switzerland) for 1 min and gently shaken. The labellum was subsequently removed from the vial; all scent samples were stored at −20 °C until analysis. Gas chromatography (GC) was performed following Mant et al. (2005b) with minor modifications as detailed in Xu et al. (2011). Several samples were re-analysed for compound identification with a mass selective detector (GC/MSD; Agilent 5975) using the same oven and column parameters. It is noted that (Z)-11 and (Z)-12 alkenes cannot be discriminated with the parameters used. Compound spectrum and retention time were compared with those of a synthetic standard, as reported by Xu et al. (2011). The relative amount of each odour compound was calculated as the proportion of total alkene and alkane amounts with a chain length between 18 and 30 carbons. Principal component analysis (PCA) was used for the analysis of interspecies floral scent variation based on scaled relative amount of hydrocarbons. The within-species differences in floral scent bouquet between populations were analysed using distance-based tests for homogeneity of multivariate dispersions (Anderson, 2006), with 999 permutations. The differences for each individual compound and total alkenes between different populations within species were analysed using Student's t-test or Mann–Whitney U-test, depending upon the results of testing for normality and heteroscedasticity of variance using Leven's and Shapiro–Wilk tests. All statistical analyses for floral scent were performed in R 2.15.2 (R Development Core Team, 2012).
AFLP analysis yielded 322 variable markers. Genetic divergence between population pairs was assessed via ΦST values, yielding the following results: the lowest ΦST values were detected between O.sphARG and O.sphCLA (0.07) and the highest between O.sphARG and O.exaTUS (0.27); O.sphCUMA and O.sphGAR populations showed a ΦST of 0.10. Average ΦST was 0.14 between O. sphegodes populations, and 0.18 between O. exaltata populations. Interspecific ΦST between O. sphegodes and O. exaltata populations averaged 0.21 (Table 2, Fig. 1). amova showed that 87% of genetic variance was partitioned among individuals within a population, whereas the remaining 13% was explained by variance among populations (Table 3).
Table 2. Pairwise ΦST values (coefficient: standard Jaccard. distance transformation: d =1-s) for the examined populations
Table 3. Analysis of molecular variance (amova) for AFLP markers
Source of variation
d.f., degree of freedom; SSD, sum of squared deviations; %, proportion of variance components, SE ≤ 2.93%; Φ, genotypic variation.
PCoA analysis grouped allopatric and sympatric populations based on taxonomic classification (the first two axes explained 25.9% and 19.4% of the variation, respectively; Fig. 2). Only a few individuals from sympatric Cuma and Gargano populations were intermixed between the two groups. PCoA showed that all O. sphegodes populations from the Tyrrhenian Coast (O.sphCLA, O.sphARG and O.sphCUMA) formed a cohesive group, but the species populations also largely overlapped with the O.sphGAR Adriatic population.
The most probable number of genetic clusters (K) present in the data (determined following Evanno et al., 2005) was K =2, corresponding to the assumption that only two species contributed to the sample gene pool. Individuals of the two species exhibited strong assignment to their respective cluster, with the exclusion of a few putative hybrid individuals (Fig. 3).
A genome scan for FST outliers was performed for O. sphegodes populations from Cuma and Gargano using a trimmed mean FST of 0.027 (Fig. 4). Overall, 269 loci were included in the genome scan, which identified four outliers (P <0.025) below or above the respective 0.025 or 0.975 quartiles of the expected FST distribution. All four outlier loci were more strongly differentiated than expected. However, among the four loci, only two were unique to the comparison between O. sphegodes populations from Cuma and Gargano, whereas the other two loci were found as outliers also in other pairwise population comparisons.
Our GLM analysis modelled genetic pairwise population differentiation (ΦST) as a function of geography and shared pollinators. Geography and pollinators were both significant (P <0.05) factors in explaining population differentiation (Table 4).
Table 4. Generalized linear model of pairwise ΦST values among Ophrys sphegodes populations as explained by the factors indicated. To facilitate the analysis, the Cuma population was assumed to have a different pollinator than the remaining O. sphegodes populations
Pollinator activity was negligible in the afternoon, and overall activity in Gargano was notably higher than that in Cuma. The captured pollinator summary is given in Table 5. O.exaGAR attracted 10 C. cunicularius individuals and O.sphGAR attracted four A. nigroaenea individuals; both taxa from Gargano attracted their legitimate pollinators independent of where the plots were located (Gargano and Cuma). Results differed for the two Cuma species. O.exaCUMA attracted four C. cunicularius individuals and two individuals of a yet unidentified bee species in the genus Eucera. The Eucera bees were only captured in the O. exaltata Cuma population, and it is the first time a Eucera bee has been reported to pollinate individuals from the O. sphegodes/exaltata lineage. O.sphCUMA attracted five A. nigroaenea individuals and 46 A. bimaculata individuals. The pollinator-baiting experiment performed in 2012 confirmed that O.sphCUMA attracts A. bimaculata and A. nigroaenea. These species were caught six times and twice, respectively, on freshly cut inflorescences.
The PCA plot of O. exaltata and O. sphegodes from the Tyrrhenian and Adriatic coasts showed a clear separation between the two species (Fig. 5), consistent with previous studies (Xu et al., 2011). The distance-based tests for homogeneity of multivariate dispersions showed that O. sphegodes populations from the Tyrrhenian and Adriatic coasts were significantly different in their floral scent bouquet (P =0.023; 999 permutations). In contrast, the floral scent bouquet of O. exaltata was not significantly different between two populations (P =0.209; 999 permutations). These floral scent differences were mainly due to a higher proportion of total alkenes in Tyrrhenian populations (Fig. 6). For O. sphegodes, the O.sphCUMA population produced 62.8 ± 5.2% total alkenes, whereas the O.sphGAR population only produced 38.7 ± 15.0% total alkenes (P <2.2∙10−16, Mann–Whitney U-test). For O. exaltata, the O.exaCUMA and O.exaGAR populations produced 69.3 ± 8.0% and 59.3 ± 16.2% alkenes, respectively (P =0.019, Mann–Whitney U-test) (Fig. 6). It is noticeable that the O.sphCUMA population not only produced a higher proportion overall of alkenes that are active compounds to A. nigroaenea (the pollinator of O. sphegodes) but also produced three compounds, (Z)-7-C21, (Z)-7-C23 and (Z)-7-C25, that are active compounds to the pollinator of O. exaltata, C. cunicularius (Mant et al., 2005a), but were absent from the O.sphGAR population.
Overall, our results indicated a pollinator shift and showed significant differences in floral scent between populations of O. sphegodes. However, significant intraspecific genetic structuring was not observed between populations. These results suggest the following: (a) the Cuma and Gargano O. sphegodes populations have reached an early stage of divergence and are adapting to different pollinator species, or (b) plant–pollinator relationships in some sexually deceptive orchid species are more geographically variable that has traditionally been proposed (i.e. Kullenberg, 1961).
Our previous pollinator choice experiments with O. sphegodes and the sympatric O. exaltata in semi-natural/disturbed habitats along the Tyrrhenian and Adriatic Italian coasts indicated the absence of pollinator sharing between these two closely related species (Xu et al., 2011). However, our genetic analyses, even if with a small data set, supported previous findings, showing that hybridization can sporadically occur (Xu et al., 2011). Interestingly, our experiments also revealed that the Tyrrhenian Coast O. sphegodes (O.sphCUMA) was pollinated primarily by the bee species A. bimaculata. Andrena nigroaenea is the typical O. sphegodes pollinator throughout most of the species range (Mant et al., 2005b), but this species was only responsible for approximately 10% of pollinia removal at Cuma (Table 5). More importantly, O. sphegodes from the Tyrrhenian Coast (O.sphCUMA) maintained its attractiveness to A. bimaculata when transferred to Gargano (Adriatic Coast), indicating that A. bimaculata attraction is based on specific floral traits – probably floral scent – and not merely on the (potentially) more frequent occurrence of this bee species at Cuma. To date, A. bimaculata has only been reported as a pollinator of the typically abdomen-pollinated O. sicula (Gaskett, 2010) and as a possible pollinator of O. creticola on Crete (Paulus & Gack, 1990b; Paulus & Schlüter, 2007).
Floral scent analysis results indicated stronger differentiation between the O. sphegodes populations than between the O. exaltata populations from the two coastal regions (Figs 5 and 6) and such differences were mainly due to a higher proportion of alkenes produced by the O.sphCUMA population. Three compounds [i.e. (Z)-7-C21, (Z)-7-C23 and (Z)-7-C25] that are active to C. cunicularius (preferred pollinator of O. exaltata) were exclusive to the O.sphCUMA population (Fig. 6). These three (Z)-7 alkenes have not previously been recorded as an EAD-active compound for A. nigroaenea (Stökl et al., 2005), and the addition of a (Z)-7 alkene mix [including (Z)-7-C23, but not (Z)-7-C25 and (Z)-7-C27] was rendering O. sphegodes flowers less attractive to A. nigroaenea. A possible role played by these three compounds in A. bimaculata attraction has yet to be tested. Interestingly, the Tyrrhenian O. sphegodes (O.sphCUMA) scent continued to attract A. nigroaenea, even though its scent bouquet was different both qualitatively and quantitatively from that of the Adriatic O. sphegodes (O.sphGAR). Therefore, the two O. sphegodes populations share A. nigroaenea as pollinator, and no substantial floral isolation should be expected between them.
In spite of the small genetic differences, our data showed that between the seven study populations, levels of intraspecific gene flow were higher than those of interspecific levels. Overall, lower genetic divergence was detected between allopatric O. sphegodes populations than between sympatric O. sphegodes and O. exaltata populations. Despite the different pollinators attracted by each species, the Adriatic O. sphegodes (O.sphGAR) population exhibited a lower genetic distance from the geographically closer O.sphCUMA population than from the more distant O.sphCLA and O.sphARG Tyrrhenian populations. O.sphCLA and O.sphARG probably attract the same pollinator as the Adriatic O. sphegodes, that is, A. nigroaenea. Nevertheless, GLM results suggested that geography and shared pollinators were significant factors in differentiation between O. sphegodes populations. Despite the difficulties in resolving ancestral polymorphisms due to hybridization in very recently diverged populations/species, the low genetic differentiation estimates (i.e. low ΦST values; Table 2) observed between different O. sphegodes populations were comparable with, or even lower than, the values previously reported between other populations of Ophrys species using, as in our study, AFLP markers (Devey et al., 2009; Schlüter et al., 2011b). Therefore, data suggest that O. sphegodes populations continue exchanging alleles despite an apparent change in preferred pollinator species. Nevertheless, the genome scan for FST outliers identified four loci that were more strongly differentiated than expected by chance. Among these four loci, two were consistent with a scenario of divergent selection between O. sphegodes populations from Cuma and Gargano, even if the small sample set and the absence of multiple pairwise comparisons between populations with the same pollinator combination strongly limit the power of this analysis.
Based on high interpopulation gene flow in O. sphegodes, independent of pollinators, we can infer that local adaptation by pollinator shift in the Tyrrhenian O. sphegodes populations has not (yet) generated detectable genome-wide separation from the Adriatic populations. In an alternative (but not mutually exclusive) scenario, the Tyrrhenian and Adriatic populations represent a step towards incipient ecological speciation driven by genic divergence (i.e. loci linked to FST outliers). In this case, even if the entire genome is experiencing gene flow, as long as the one or few genes underlying the floral isolation trait remain differentiated as a consequence of divergent selection, these populations have the potential to maintain reproductive isolation, regardless of gene flow between them (i.e. genic ecological speciation model and porous genomes; Wu, 2001; Lexer & Widmer, 2008).
Our results match studies by Paulus & Gack (1990b), Lorella et al. (2002) and Claessens & Kleynen (2011) that demonstrate a (narrow) range of related Ophrys species (particularly those with a wide distribution) pollinated by different bees and suggest that Ophrys–pollinator interactions are flexible rather than a static one-to-one relationship. Vereecken et al. (2011) observed this flexibility in Ophrys– pollinator relationships, and Bower (1996) reported that so-called minor responders often accompany the main pollinator(s) in the sexually deceptive Australian orchid genus Chiloglottis. Our finding that a Eucera bee species serves as a minor O. exaltata pollinator in the Cuma population suggests that even bees of other genera can act as secondary pollinators.
The different pollination ecotype resolved in the Tyrrhenian O. sphegodes population may represent a response to local selection imposed by variability in the geographical mosaic of pollinators. Genetic drift is an alternative explanation for the observed pattern in floral odour bouquet evolution and its attractiveness to a novel pollinator in the Tyrrhenian populations. However, the low between-population genetic differentiation (Table 3) is indicative of extensive gene flow. In addition, the large effective population size typically found in these terrestrial deceptive orchids (Cozzolino & Widmer, 2005 and reference therein), including our study system (Mant et al., 2005b), indicates that a drift scenario is less likely compared with an adaptive process of pollinator shift.
A classical question in the evolution of specialized pollination mechanisms is how much stringency in the relationship between a plant species and its pollinators can limit the ability of a plant species to adapt to and survive local and temporal changes in the pollinator community. Interestingly, a few studies on the spatial and temporal breadth of pollinator of Ophrys taxa (including this study) have revealed other minor pollinators. An imperfect match in odour between the plant mimic and its model (the female insect) may represent a form of pre-adaptation for a local or temporary shift in specialized pollinators (Bower, 1996; Peakall et al., 2010). Indeed, bioassays have shown that sexual deception is a type of imperfect mimicry, and the pollinator bees actively preferring ‘novel’ signals over the more commonly encountered cues as adaptive responses to promote outbreeding (Vereecken & Schiestl, 2008). Therefore, compound variability involving compounds inactive to some pollinators but active to others, and negative frequency-dependent selection, might provide an advantage to the uncommon odour phenotypes by increasing pollination success. Such selection for imperfect mimicry may facilitate more relaxed signal refinement in mimics to optimally match the signals released by the mimic's specific insect models. It would thereby maintain a source of natural interindividual variation in floral scent, which could represent the standing genetic variation necessary for rapid local adaptive responses to a fluctuating pollinator community in disturbed habitats.
Although compelling evidence supports floral adaptation as the basis for ecological speciation (Kay & Sargent, 2009), a local shift to a different pollinator does not necessarily lead to speciation, particularly if the selective pressure is transient or fluctuating. Ophrys populations adapted to different pollinators may well represent a very interesting case study of incipient speciation by local adaptation. How easily this transient change in preferred pollinators between locally adapted populations can turn into a ‘permanent’ form of reproductive isolation and speciation probably depends on the tempo and mode of divergent selective forces working on local populations and on the degree and duration of the disturbances generating the local ecological differences.
This work was supported by the PhD program of the University of Naples, Salerno Camera Commercio grant to SC, ETH Zurich (TH0206-2) to FPS and Swiss National Science Foundation (SNF 31003A_130796) to PMS.
We thank Donata Cafasso and Rosita Rinaldi for help with molecular analyses. Giuseppe Pavarese gave support with the pollinator experiments in Cuma and Nicolas Vereecken helped with the identification of pollinating bee species. The authors also thank Michael Hutchings for his constructive comments on the manuscript and the Write Science Right company for language editing. The authors are very grateful to Richard Bateman and another anonymous referee for their extremely constructive comments on an earlier version of the MS.