Current divergent selection may promote floral trait differentiation among conspecific populations in flowering plants. However, whether this applies to complex traits such as colour or scents has been little studied, even though these traits often vary within species. In this study, we compared floral colour and odour as well as selective pressures imposed upon these traits among seven populations belonging to three subspecies of the widespread, generalist orchid Anacamptis coriophora. Colour was characterised using calibrated photographs and scents were sampled using dynamic headspace extraction and analysed using gas chromatography-mass spectrometry. We then quantified phenotypic selection exerted on these traits by regressing fruit set values on floral trait values. We showed that the three studied subspecies were characterised by different floral colour and odour, with one of the two predominant floral volatiles emitted by each subspecies being taxon-specific. Plant size was positively correlated with fruit set in most populations, while we found no apparent link between floral colour and female reproductive success. We detected positive selection on several taxon-specific compounds in A. coriophora subsp. fragrans, whereas no selection was found on floral volatiles of A. coriophora subsp. coriophora and A. coriophora subsp. martrinii. This study is one of the first to document variation in phenotypic selection exerted on floral scents among conspecific populations. Our results suggest that selection could contribute to ongoing chemical divergence among A. coriophora subspecies.

Three populations of A. c. coriophora, three populations of A. c. fragrans and one population of A. c. martrinii – each comprising between 50 and 100 individuals – were sampled in southern France during spring/summer 2016.

In each of the seven populations, 50 to 65 individuals were randomly selected and marked. For each individual, plant height and inflorescence length (distance between uppermost and lowermost flowers) were measured to the nearest 0.5 mm with a tape measure and flowers were counted.

In six populations out of seven (three of A. c. coriophora, two of A. c. fragrans and one of A. c. martrinii), floral colour was characterised using calibrated photographs. For each individual, one of the lowermost flowers was photographed using a standardised procedure (front view and planar position, indirect natural sunlight) with a Nikon D70 digital camera equipped with a Novoflex 35 mm lens. A ruler and two reference greys (23 % and 35 % of average reflectance) were included in each photograph for further scaling and colour calibration, respectively. Each photograph was calibrated using the Image Calibration and Analysis Toolbox plugin (Troscianko and Stevens, 2015) in the software ImageJ (Schneider et al., 2012). After this calibration step, on each photograph, the labellum was manually selected and its colour characterised using (i) RGB values, to provide a measure of floral colour as we humans perceive it, and (ii) the excitation values of the SW (short wavelength) and MW (medium wavelength) photoreceptors of the honeybee (Apis mellifera). For each labellum, we measured one mean and one standard deviation values for each of the R, G, B, SW and MW channels.

In each of the seven populations, floral odour was sampled in situ using headspace dynamic extraction. For each individual, the inflorescence was enclosed in a ~10*5 cm Nalophan® bag for 30 min. ChromatoProbe® quartz microvials filled with 3 mg of a 1:1 mix of Tenax-TA and Carbotrap® were used as adsorbent traps, and 1 µl of internal standard (n-tetradecane, 100 ng µl^-1) was injected in these traps. Two traps – one for mass spectrometry analyses, used for compound identification, and one for flame ionisation detection analyses, used for compound quantification – connected to a pump by silicon tubes were inserted into the bag after 20 min. The air inside the bag was then pumped through the two traps for 10 min at a rate of 40 ml min^-1. After floral odour extraction, each trap was put into a vial and stored at -20°C until further analyses. Floral odour was analysed by gas chromatography (GC)-mass spectrometry (MS) and flame ionization detection (FID) using a Trace 1310 gas chromatograph coupled with a flame ionization detector and an ISQ mass spectrometer (Thermo-Electron, Milan, Italy) and equipped with an OPTIMA® 5-MS-HT capillary column (30 m × 0.25 mm × 0.25 µm, Macherey-Nagel, Düren, Germany). Traps were handled with a Multi-Purpose Sampler (Gerstel, Mülheim an der Ruhr, Germany) and desorbed with a double stage desorption system composed of a Thermal Desorption Unity (TDU) and a Cold Injection System (CIS; Gerstel, Mülheim an der Ruhr, Germany). First, traps were desorbed splitless with a temperature of 250°C on the CIS trap cooled at -80°C using liquid nitrogen. Then, the CIS trap was heated to 250°C with a 1:4 split ratio to inject the compounds into the column. Oven temperature was held at 40°C for 3 min, increased from 40°C to 210°C at a rate of 5°C/min and from 220 to 250°C at a rate of 10°C/min, and held at 250°C for 2 min. The temperatures of transfer line and ion source of the mass spectrometer were 250°C and 200°C, respectively, and acquisition was set from 38m/z to 350m/z, at a 70eV ionisation energy. The FID was heated to 250°C. Retention times of a series of n-alkanes (Alkanes standard solution, 04070, Sigma Aldrich®) were used to convert retention times into retention indices. Compounds were identified on the GC-MS samples based on their retention indices and mass spectra, which were compared to those recorded in databases (Adams, 2007). Peak areas were measured on the GC-FID samples using the software Xcalibur (Thermo-Electron, Milan, Italy) and were converted to absolute (i.e. in ng) and relative (i.e. in %) amounts using the peak area of the internal standard (n-tetradecane).

One month after the measurement of floral traits, the fruit set of each individual was measured and used as a proxy of female reproductive success.

In the floral colour dataset, two measurements were made for three individuals; in this case, the average value was used in subsequent analyses.