Of 19 European orchid species representing 9 genera (Table 1), n-alkanes (saturated hydrocarbons) of chain lengths 21 – 29 were present in labella of all species, but in varying relative and absolute abundance (Table 2). Odd chain-lengths alkanes with 23 carbon atoms or more were especially abundant (Table 2). Alkenes (unsaturated hydrocarbons), although generally less common, were also widespread among the investigated species (Fig. 1). In total, 17 species produced at least trace amounts of alkenes in their cuticular waxes (Table 2). In only two species (Orchis provincialis, Neotinea lactea), no alkenes could be detected. In 11 species, alkenes constituted more than 10% of all analysed compounds (Table 2). Some species (Serapias lingua, S. cordigera, Gymnadenia conopsea, Neotinea ustulata) produced 50% or a higher proportion of alkenes in their wax layer. The highest relative abundance of a single alkene was found in N. ustulata, where 11-tricosene (C23en) represented 42% of all the compounds found in this species. The South African Disa bivalvata, used here as outgroup, produced 81.9 μg (80.6%) alkanes and 19.7 μg (19.4%) alkenes in a single labellum.

The strict consensus tree among the six resulting cladograms (length: 401 steps), with a topology broadly congruent to those reported in previous phylogenetic analyses with broader datasets 404142, is shown in Fig. 1. As expected, in spite of the fact that the majority of the terminal clades have bootstrap values above 50% (bootstrap values > 50% are shown in Fig. 1), the relationships among the major clades into which the ingroup is split are weakly supported, as observed in previous published analyses 404142.

The MacClade reconstruction of alkenes production unequivocally indicated alkenes presence as the primitive character state in the group and with only two independent losses in single species (O. provincialis, N. lactea). Concordantly, Bayesian phylogenetic analysis (data not shown), constrained with the model of nucleotide evolution that fits the data best, (TrN, -lnL = 3378.8591 43; gamma distribution shape parameter α = 0.808; base frequencies: freqA = 0.2639, freqC = 0.1830, freqG = 0.2151, freqT = 0.3381; substitution rate: [A-C] = 1.000, [A-G] = 2.425, [A-T] = 1.000, [C-G] = 1.000, [C-T] = 3.919, [G-T] = 1.000), was topologically similar to the tree topology of parsimony analysis and strongly supported the presence of alkenes as the ancestral state for the character.

In our analyses, pollinator category had a significant effect on the absolute amounts of alkenes (Fig. 2; ANOVA, F_2,14 = 12.05, P = 0.001) and numbers of alkenes (Kruskal-Wallis, Chi² = 9.0, d.f. = 2, P = 0.01) produced by the flowers. Species pollinated mostly or solely by male bees (Ophrys sphegodes, Serapias lingua, S. cordigera, Anacamptis papilionacea; group 2) produced significantly higher absolute amounts of alkenes (LSD Post Hoc, group 1–2: P < 0.001; 2–3 P = 0.002; Fig. 2) and greater number of alkenes with different double bond position (Mann-Whitney U-test: group 1–2: P = 0.01; 2–3: P = 0.006) than the other groups. The "female bees" group was not significantly different from the "moth, beetles, flies" group for amount of alkenes (LSD Post Hoc, P = 0.174; Fig. 2) and number of alkenes (Mann-Whitney U-test, P = 0.84). The amount of alkanes did not differ significantly among the groups (ANOVA, F_2,14 = 3.69, P = 0.051); however, number of alkanes was different (Kruskal Wallis Chi² = 6.8 d.f. = 2 P = 0.03), but post-hoc tests did not show any significant differences among the groups.

The relative amounts of alkanes/alkenes were also significantly different between the pollinator categories (ANOVA, F_2,14 = 21.35, P < 0.001). Comparison among groups revealed all groups to be different from each other (LSD Post Hoc, P < 0.001) except "male bees" and "moth, beetles, flies" (P = 0.17). Species with relatively low absolute amounts but unusually high proportions of alkenes were Neotinea ustulata, Dactylorhiza maculata, Gymnadenia conopsea, G. odoratissima, and G. rhellicani (Table 2). Neotinea ustulata produced unusually high proportions of 11-tricosene and 11-pentacosene. Dactylorhiza maculata produced mostly 5-pentacosene and 5-heptacosene. Gymnadenia conopsea, G. odoratissima, and G. rhellicani produced mostly 7- and 5-heptacosene (Table 2).

The concept of preadaptation in floral traits presumes that existing traits have acquired a new function, which can be in the form of a "transfer exaptation" (new function replaced the old) or "addition exaptation" (new function added to the old) 36. The identification of floral exaptations is dependent on a combination of phylogenetic analyses and ecological studies investigating functions of floral traits 44. A well known example is the tropical vine genus Dalechampia, where resins are produced primarily for herbivore deterrence 35; this trait was a preadaptation for the evolution of pollination systems based on resin reward 3436. Preadaptations of pollinators have also been assumed to influence the evolution of specialized pollination systems, such as that observed in the yucca – yucca moth mutualism 45. However, some traits of pollinators in this system are, on the contrary, synapomorphies inferred to be key innovations 46. For the evolution of sexual deception within Orchidinae, we suggest an evolutionary scenario similar to Dalechampia. The ability to produce alkenes, (i.e. the genes encoding the necessary enzymes) is likely an ancestral trait – a preadaptation with vegetative function in the cuticular wax layer. The transition into male-pollinator attraction was achieved by producing increased amounts and different double-bond varieties of alkenes, since female bees produce large amounts of alkenes in specific patterns of double-bond positions 31. Additionally, morphological changes in the labellum evolved to achieve mimicry of shape, colour and pilosity of female bees 10.

Plants of the genus Ophrys generally produce large amounts of different alkenes. Erdmann 47 found alkenes in all 18 analysed species, including O. speculum, that uses oxo- and hydroxy acids for pollinator attraction 48. In this species, even buds and pollinia produce alkenes. We found that large amounts and diversity of alkenes with different double bond positions in Ophrys was also shared among other species of Orchidinae, namely the two outcrossing representatives of Serapias, and Anacamptis papilionacea. Although several evidences, including chromosome number, suggest that Ophrys, Serapias, and Anacamptis are closely related, the exact relationship cannot be inferred by the existing phylogenies that have low bootstrap support for the basal relationship among genera 4042. It is thus not clear whether the production of large amounts of alkenes in Ophrys is an evolutionary novelty in the genus or inherited from a common ancestor with Serapias. Considering the latter possibility, the elaborate morphology of Ophrys labella, with often dark color and UV-reflecting "mirror-patterns", that is unique to the genus, should be considered as the key innovation for sexual deception. However, earlier studies suggest that Serapias is pollinated by sexual deception as well 49. Thus, a better understanding of the pollination system in Serapias, as well as phylogenies with better resolution of relationships among the major groups are necessary to resolve this issue.

The pattern of strong alkene production in the non-rewarding Ophrys, Serapias, and Anacamptis papilionacea is paralleled by a common use of primarily male bees as pollinators in these orchids. A strong alkene production, as shown by the significantly higher absolute amounts in these species, is the likely prerequisite for these low volatile compounds to act as semiochemicals. Since male bees use alkenes for chemical communication 31, these orchids likely use alkenes to manipulate the behaviour of their pollinators.

In Serapias, pollination by male bees sleeping in the flowers has been proposed 50, but earlier studies suggested that males of the bee, Ceratina cucurbita, are sexually attracted to the flowers and search for females within them 49. These flowers may thus mimic either a nest entrance or virgin females sitting in the flowers, to entice males to enter the flowers. The high abundance and diversity of alkenes suggest that alkenes mediate such signalling. Anacamptis papilionacea is pollinated by patrolling male bees of the genus Eucera 51, and the alkenes produced by the flowers may also render the flowers more attractive for the males. In both species, further investigations are necessary to clarify details of the pollination system and the inferred role of floral odor.

Besides their role as pheromones in bees, alkenes are also important semiochemicals in flies. This may explain the unusually strong production of 11-tricosen and 11-pentacosene in Neotinea ustulata, that is pollinated specifically by the tachinid fly, Tachina (Echinomyia) mangnicornis 52. Many fly species use alkenes as mating signals; one example is Drosophila virilis, with 11-pentacosene as its major sex pheromone component 53. Female flies also use host-pheromones for host location, but little is yet known about the chemical identity of these signals 54. It is possible that the flower-produced alkenes are also involved in the attraction of the pollinator via false host- or mating signals, and future investigations on the pollination system in this species may also prove interesting. The moderate to high production of alkenes in the rewarding orchid clade (including Gymnadenia conopsea, G. odoratissima, G. rhellicani, Platanthera bifolia, together with the nectarless Dactylorhiza maculata), may be explained as retention of the primitive state of alkene production.

Male insects invest more time and energy in mating behaviour than females, which spend more time feeding and collecting food for brood care 55. Therefore, females are generally considered as more efficient pollinators 56. Whole guilds of plants, however, are specialized for pollination by male bees, e.g. Euglossine pollinated plants that produce fragrance as a reward for the pollinators 5758, Oncocyclus irises being pollinated primarily by night sheltering Eucera males 59, and orchids that mimic specific model plants, such as Cephalanthera rubra and Diuris maculata 6061. Since the behaviour of male and female bees differ, the resulting pollination patterns can be assumed to be different, too. Whereas food-seeking female bees are more efficient pollinators, males conduct longer-distance visits and thus contribute relatively more to outcrossing 56. Indeed, it has been predicted that fragrance-collecting male euglossine bees, as well as males pollinating sexually deceptive orchids, mediate long distance pollen flow 62636465. As self-compatibility and inbreeding depression are widespread in orchids 66, distance of pollen flow may impact strongly on the quality of seeds produced 67. Additionally, pollination by exploitation of reproductive behaviour of males is often more specific, particularly in sexually deceptive orchids and orchids pollinated by fragrance-collecting euglossine bees 6869. Specificity may decrease pollen loss, a factor that is likely important in orchids, given their highly efficient pollination via pollinia. Collectively, pollination by male bees may be advantageous by mediating specific and long-distance pollen flow, and selection may thus favour floral signals that attract primarily males.

We suggest that the evolution of male bees as pollinators in the deceptive pollination systems developed by Ophrys, Anacamptis papilionacea and possibly also Serapias was achieved by "sensory exploitation". This concept has been developed in the framework of sexual selection and states that sensory preference in females drives the evolution of male signals 70. A similar scenario can be proposed for the evolution of pheromone-imitations in response to male pollinators. To attract male pollinators by pseudo-pheromones, the orchids hitchhike on pre-existing, intraspecific communication channels. The existing reception system and behavioural preference of males for alkenes may have enabled the orchids to exploit this pollinator resource relatively easily, by increased production of alkenes, with pre-existing enzyme systems.

For all scent (hexane extracts) collection, fresh, unpollinated flowers were used. Seven species (plants originating from Cilento and Vallo di Diano National Parks, Southern Italy) were kept in the Naples Botanical Gardens in an insect-tight cage; flowers of the other 13 species were collected from wild populations. One labellum of five individuals (if possible) or five labella of one individual (Serapias cordigera, Himantoglossum hircinum, Platanthera bifolia, Neotinea ustulata, Neotinea lactea) were extracted in 200 μl of hexane (Merck Uvasol) by shaking in a 2 ml vial for 1 min. Afterwards, the labella were removed and the sample stored at -20°C until analysis. Species, sampling locations and their pollination mode are indicated in Table 1 (for the nomenclatural authority of all listed orchid species see 42.

Phylogenetic analysis of nuclear ribosomal ITS1 and 2 sequences of the selected taxa was carried out as described in 41 by using Disa bivalvata (Diseae, Disinae, Orchidaceae) as outgroup. Ingroup and outgroup species choice was limited to those taxa for which scent data have been collected. Briefly, Genbank available ribosomal sequences were reduced to only ITS1 and ITS2 and alignment was accomplished by using Clustal W with GAPOPEN and GAPEXT parameters set to a value of 4 (for details see 4041. The resulting matrix was then subjected to a parsimony analysis by using the software package MEGA 3.2, and bootstrap percentages 71 were calculated with 500 replicates. As a framework for character reconstruction, we used the strict consensus tree obtained in this analysis, and MacClade 4.0 72 was used to examine alkene production evolution within the clades. MacClade determines character reconstruction by parsimony, or if more than one state can be assigned to branches, MacClade displays an equivocal pattern on those branches 72. Alkene production was designated as unordered, with no assumptions of transformations between states 72. To estimate posterior probabilities for individual nodes, a Bayesian phylogeny was inferred using a variant of the Markov chain Monte Carlo (MCMC) algorithm as implemented in the MrBayes software v3.1.2 73. Five Markov chains (four heated, one cold) were run for 150,000 generations using random starting trees and fixed model employed in branch length calculations. Trees were sampled every 100 generations. To set prior probabilities for the analysis (stationary frequencies of character states, character substitution rate matrix, proportion of invariable sites and shape parameter of the gamma distribution of the variation), hierarchical likelihood ratio tests (hLRTs) were performed in MODELTEST v3.7 74 and the best-fitting model given the dataset was chosen.

Before gas chromatographic analysis, 100 ng of octadecane (purity 99.8%, Fluka, Buchs, Switzerland) were added to all samples as an internal standard. One μl of the sample was injected splitless into an Agilent 6890 N gas chromatograph (GC; Agilent Technologies, Palo Alto, USA) equipped with a HP5 column (5% Phenyl-methylpolysiloxane, 30 m × 0.32 mm ∅ × 0.25 μm film thickness, Agilent Technologies). The column was equipped with a 5 m × 0.53 mm diameter deactivated retention gap. The GC was equipped with a flame ionization detector (FID). Hydrogen served as carrier gas (2 ml/min, constant flow mode) and nitrogen was used as make-up gas. The injector temperature was kept at 300°C. The oven was kept at 50°C for one minute and then heated to 300°C at a rate of 10°C per minute and kept at 300°C for 20 minutes. Chromatogram outputs were recorded by the Chemstation program (Agilent Technologies, Palo Alto, USA) for qualitative and quantitative analysis. The internal standard method was applied to calculate absolute amounts of scent compounds. To calculate relative amounts, the absolute amounts of each compound was divided by the sum of all compounds and multiplied by 100.

For identification of compounds, one μl of each sample was injected on column into a Trace GC Ultra with DSQ II mass spectrometer (MS; Thermo Electron Corp., Milan, Italy), equipped with the same column used for quantitative analysis. Helium served as the carrier gas (2 ml/min, constant flow mode). The oven was kept at 45°C (1 min) and then heated to 280°C at a rate of 10°C/min. The transfer line to the MS was heated to 220°C. The ion-source of the MS was heated to 250°C, the MS was run in full scan mode, starting after 5 min; 1.31 scans/s were done at a scan rate of 500.3 amu/s, and the mass range was 50 – 420. Compounds were identified by comparison of mass spectra and retention times with those of synthetic reference compounds. On a HP-5 column, alkenes elute before the corresponding alkanes. Alkenes with the same chain length but different position of the double bond have reproducibly different retention times, so that (Z)-11 elutes first, and (Z)-5 last. Rather than identifying all compounds present in the extracts, we focussed on straight chain saturated hydrocarbons (n-alkanes) and unsaturated (n-alkenes) hydrocarbons with double bond positions 5, 7, 9, or 11 and of chain lengths 21 – 29. These compounds have been shown previously to be responsible for pollinator attraction in many Ophrys species (summarized by 4). The isomeric configuration of the alkenes was not determined in this study; however, until now, Z-alkenes were primarily found in plant cuticles 21.

For the analysis of the relation between pollination syndrome and hydrocarbon production, the species were classified into three main pollinator categories (Table 1) according to the available information 49757677. The Bee-group was split into female and male, since male bees are important and exclusive pollinators of some orchids. Group (1) "female bees", comprised only deceptive species, with generalized food deception being predominant. Group (2) "male bees", comprised sexual deception (Ophrys), sleeping hole (the outcrossing Serapias spp.), and pollination by patrolling Eucera males (A. papilionacea). Group (3) "moths, beetles, flies", with both rewarding and deceptive species. For most species, these categories should be seen as quantitative estimations, since few Mediterranean orchids are highly specialized 77. Two species were excluded from the analysis: Serapias parviflora, which is an autogamous species 78, and Neotinea lactea, where the pollinators remain unknown. For the comparison of absolute amounts of alkane/alkene, all amounts were divided by the number of labella sampled, to normalize the amounts to a single labellum. The size of the labellum was estimated, by using published data on average length and width of labella 79, calculating the product and dividing it by two (Table 1). The absolute amounts of alkane and alkene were normalized to the size of the flowers by dividing it by this value for labellum size. The resulting values were ng/mm². An ANOVA with LSD post-hoc tests was used to compare absolute amounts of alkanes/alkenes among pollination systems. A Kruskal-Wallis test with Mann-Whitney U-test for post-hoc comparisons with the level of significance set to 1% (Bonferroni correction) was used to compare the number of alkanes/alkenes produced.