Study site and species. Research was conducted in Pha Taem National Park, Ubon Rachathani Province, eastern Thailand (15°23'56" N, 105°30'27" E, 220 m a.s.l.), where four Ceropegia species, C. acicularis, C. boonjarasii, C. citrina and C. tenuicaulis (Fig. 1A-D), all endemic to the park, occur in sympatry in sandy soil in open areas of dry deciduous dipterocarp forest (see Fig. S1 for more details on study site and species)48–50. The permissions to collect flowers of Ceropegia species studied were obtained from Department of National Park Wildlife and Plant Conservation. (DNP), Thailand. All the plant experiments were in compliance with relevant institutional, national, and international guidelines and legislation. Voucher specimens of the four Ceropegia species were deposited in BCU (Thailand): C. acicularis Kidyoo: THAILAND. Ubon Ratchathani: Pha Morn, Pha Taem National Park, 240 m, 13 July 2016, M. Kidyoo 1645 (BCU); C. boonjarasii Kidyoo: THAILAND. Ubon Ratchathani: Pha Morn, Pha Team National Park, 240 m, 5 July 2017, M. Kidyoo 1648 (BCU); C. citrina Kidyoo & A. Kidyoo: THAILAND. Ubon Ratchathani: Pha Morn, Pha Taem National Park, 240 m, 2 August 2017, M. Kidyoo 1651 (BCU); C. tenuicaulis Kidyoo: THAILAND. Ubon Ratchathani: Pha Morn, Pha Taem National Park, 240 m, 5 July 2017, M. Kidyoo 1647 (BCU).
Census of floral visitors. A total of 118 samples of Ceropegia flowers were collected in 70% ethanol in the daytime during 2016‒2019. The insects found inside (which were exclusively Diptera) were identified to family or genus based on morphology using the identification keys in Marshall (2012)51 and the website http://milichiidae.info curated by I. Brake, the world specialist of the Milichiidae.
Specimens were classified into morpho-molecular taxonomic units most likely to correspond to species. Sequences of the cytochrome c oxidase subunit I (COI, 658 bp) were obtained for 41 individual flies (out of 168 collected in total). Specimens were selected to include several individuals of each morphospecies whenever possible. However, amplification failure resulted in four morphospecies not being sequenced (see Fig. S2 for details). DNA was amplified using the primer pairs LCO-1490: 5’-ggtcaacaaatcataaagatattgg-3’ and HCO-2198: 5’-taaacttcagggtgaccaaaaaatca-3’ 52. Maximum likelihood phylogeny constructed with PhyML 3.0 online (http://www.atgc-montpellier.fr/phyml/)53 was used as a guide to refine morphospecies delineation. Differential morphological diagnosis was defined for each final morpho-molecular taxonomic unit and used to classify the specimens that were not sequenced.
Each fly was checked for attached pollinaria. Body length and width of the major pollinators were measured. Specimens are kept in the collections of A. Kidyoo at Chulalongkorn University, Bangkok, Thailand, and Rumsaïs Blatrix at CEFE, Montpellier, France. Non-Metric Multidimensional Scaling (NMDS) based on presence-absence of fly species was applied on flower samples, using the Sørensen distance. Differences among plant species were tested using permutational multivariate analysis of variance (PERMANOVA). Pairwise PERMANOVAs were run and p-values were adjusted for multiple comparisons using the FDR method54. Indicator Species Analysis55 was applied to the fly species (10,000 permutations) in order to identify those specific to each Ceropegia species.
Behavior of floral visitors. Temperature, relative humidity and light intensity were recorded hourly at the study site during 29–31 July 2017 and 14–19 July 2019, using data loggers (iButton DS1923, Maxim Integrated, San Jose, California, USA; HOBO Pendant Temp/Relative Light Two Channel and HOBO U23-001A Temperature/Relative Humidity Data Logger, Onset Computer Corporation, Bourne, Massachusetts, USA). Temporal patterns of flower visitation and behavior of pollinators were recorded with NV-GS75 digital cameras (Panasonic, Matsushita Electric Industrial Co., Ltd., Japan) on the flowers on the first day of anthesis of C. acicularis (n = 10), C. boonjarasii (n = 9) C. citrina (n = 3) and C. tenuicaulis (n = 10) from 06:00 to 18:00 continuously. As visitation of C. boonjarasii flowers did not cease after sunset, additional recording from 18:00 to 06:00 was also made on them. Visitation rate was computed from videos made during the same time period as the recordings of atmospheric data. Only flies that entered the flowers were counted as visits. As morphospecies assignment was not possible from video recordings, fly identity was noted at the family level or occasionally at the genus level when possible.
In addition, behavior of the insects confined inside the basal inflated portion of the flowers of C. acicularis (n = 3, with 3 to 5 hours of observation per flower) and C. tenuicaulis (n = 6, with 1 to 2 hours of observation per flower) was recorded by inserting the tip of a commercial mini camera into the corolla tubes cut transversely just above the basal inflation.
Crossed pollinators experiment. To test whether morphological fit between flowers and flies (mechanical isolation) is involved in pollinator specificity, flies visiting one Ceropegia species were transferred to another Ceropegia species with which they are not naturally associated. Flies were collected from C. boonjarasii and C. tenuicaulis flowers. Those with pollinaria attached to the mouthparts were screened out. In addition to natural pollinators, we used Drosophila melanogaster laboratory strains, which are similar in size to natural pollinators.
Pre-anthesis flowers (recipient flowers) of C. acicularis (n = 7), C. citrina (n = 2) and C. tenuicaulis (n = 3) were bagged (mesh size 0.5 x 0.65 mm) to preclude uncontrolled visitation. At anthesis, the target flies were gently dropped into each flower using a moistened paintbrush, and the neck of the corolla tube was plugged with cotton to prevent insects from escaping and additional pollinators from entering. After the manipulated flowers wilted, they were collected in 70% ethanol. The number of pollinaria removed from the gynostegium and the number of insects carrying pollinaria were counted in the laboratory.
Morphometric analysis of interspecific variation in floral morphology. To evaluate the differences between plant species in quantitative floral morphological traits that may function to restrict access by insects, we measured the following characters on 10 flowers from different plant individuals per species (nine for C. citrina): total flower length, corolla tube length, smallest diameter of corolla tube, basal inflation length and widest diameter of basal inflation. Interspecific variation in floral morphology was tested with multivariate analysis of variance (MANOVA), prior to which the data were standardized, with the five floral characters as dependent variables and plant species as the main factor. The degree of morphological overlap between the plant species was assessed by a principal components analysis (PCA). Moreover, the relationship between the body length and width of insects and the size of different floral characters measured was investigated by a regression analysis with Holm adjusted p-values.
Floral and bug scents. The volatile organic compounds (VOCs) were collected from the flowers on the first day of anthesis (n = 4 to 7 individuals per species) during the time when insect visitation was peak for each Ceropegia species by the dynamic headspace technique30,56,57. Details about the method of collection and analyses of plant VOCs are presented in Kidyoo et al. (2021)30. The cycle of 30 min.-accumulation plus 5 min.-air pulling was repeated two times (thus three cycles in total) for C. boonjarasii and C. citrina, the floral odours of which were hardly detectable by the human nose. In parallel, control extractions for ambient contaminant compounds were carried out using empty bags. The volatile extracts were subsequently analyzed by GC–MS at the ‘Platform for Chemical Analyses in Ecology’ (PACE, Montpellier) technical facilities of the LabEx CeMEB (‘Centre Méditerranéen pour l’Environnement et la Biodiversité,’ Montpellier, France) following the method of Kidyoo et al. (2021)30 and Souto-Vilarós et al. (2018)57.
During field study, we regularly observed Cletus trigonus58 bugs (Coreidae) living in the same habitat as Ceropegia plants. This bug species has a very large distribution. When crushed, this bug smells like the flower of C. acicularis to the human nose. As bugs have been reported as possible models of mimetic odors used by flowers to attract kleptoparasitic fly pollinators, individuals of this species were collected for VOCs analysis and comparison with flowers. Each bug individual (n = 5) was enclosed in a PET (polyethylene terephthalate) bag and crushed from the outside to simulate attack by a predator. The odor was left to accumulate for 15 min. Otherwise, samples were processed as for flowers.
Data were processed using MZmine™ version 2.1859 adapted to GC data processing (customized software available on demand), using the same automated protocol ensuring the consistency of peak integration. By comparing samples to the controls collected on the same day, potential contaminant compounds were subtracted from the samples. We discarded all the VOCs that had a relative amount < 0.1%. The retention times of a series of n-alkanes (standard solution of alkanes, 04070, Sigma Aldrich®, Munich, Germany) were used for converting retention times to Kovats retention indices. Identity of VOCs was determined by comparing the mass spectra with those in the NIST database (NIST 2007 library; Wiley, 9th edition), and by comparing the calculated retention indices with those reported in the literature60. Moreover, when possible, the identity of VOCs was confirmed by comparison with mass spectra and retention times of authentic standards, using the same equipment and method. Non-Metric Multidimensional Scaling (NMDS) based on standardized relative proportions of VOCs was applied on flower samples, using the Bray-Curtis distance. Differences among plant species on their volatile profiles were tested using PERMANOVA (10000 permutations). Pairwise PERMANOVAs were run and p-values were adjusted for multiple comparisons using the FDR method54. Indicator Species Analysis55 was applied to the VOCs (10,000 permutations) in order to identify those specific to each species.
Role of vibratile trichomes of C. boonjarasii flowers in pollinator attraction. We tested whether the vibratile trichomes that remarkably ornament the distal end of the flowers of C. boonjarasii (Fig. 1B) are involved in pollinator attraction by comparing visitation rates between shaved and intact flowers. From 18 pre-anthesis flowers, 11 had the vibratile trichomes completely pulled out by hand and seven were kept intact for use as controls. Flowers were covered with nylon bags (mesh size 0.5 x 0.65 mm) to prevent uncontrolled visitation. Bags were removed during the next day to allow normal visitation of the flowers. At the end of the day, before the flowers bend down and release the insects trapped inside17,29 flowers were collected in 70% ethanol for census of flies inside. The effect of experimental shaving was tested using a Generalized Linear Model (GLM) with number of flies trapped in the flowers as the dependent variable and presence of vibratile trichomes and the individual plants as fixed factors.
Statistical analyses were performed with R 4.0.461.