Mimicry implies that an organism gains fitness by resembling a model species, and one example is rewardless plants that attract pollinators by resembling co-flowering species that provide rewards. While trait matching between mimic and model has been characterised in many cases of putative floral mimicry, few have demonstrated that resemblance is adaptive and dependent on model presence. 

Sun orchids (Thelymitra) are believed to mimic flowers of buzz-pollinated rewarding plants by displaying false anthers. To test the adaptive value of the false anthers we examined whether fruit production of T. crinita and T. macrophylla was reduced when anthers were experimentally removed or obscured, and whether the reduction was stronger when putative model plants were abundant. We also assessed visual flower similarity of both orchids and their putative model plants according to bee colour perception and identified shared pollinators and whether their behaviour on T. crinita was similar to that on buzz-pollinated model plants. 

Fruit production of both sun orchids was strongly reduced (60-71%) by removal or painting of false anthers but was not affected by the abundance of model plants. Sun orchid flowers closely matched flower colour of co-flowering pollen-rewarding species, and T. crinita shared pollinators with the rewarding species. Visiting bees attempted to buzz and manipulate the false anther, with a behaviour similar to that observed on model plants. 

The experimental results demonstrate that the false anther is an important adaptation to pollination in sun orchids. Striking visual flower similarity and shared pollinators between orchids and models suggest that sun orchids are pollinated by bees that mistake orchids for buzz-pollinated rewarding plants. The adaptive value of the false anther did not depend on model plant abundance in the local population, indicating that the relevant spatial scale is larger, or that the effects of the model species are weak in comparison to effects of other rewarding species, i.e., that magnet effects of nectar-rewarding species are dominating. 

False anthers are widespread in the genus Thelymitra, and this “mimicry trait” seems to represent an evolutionary novelty that offers unique opportunities to explore adaptations to pollination in deceptive plants.

2.1 Study system

Thelymitra is a genus of over 100 nectarless orchid species primarily found in Australia. These terrestrial, perennial herbs have inconspicuous roots, oval-shaped tubers, and a single leaf near the base of the plant (Western Australian Herbarium, 1998). Most species within the genus are believed to be bee-pollinated via food deception, although some may be autogamous (Beardsell & Bernhardt, 1983; Edens-Meier et al., 2014). Unlike typical orchids, Thelymitra's labellum resembles the other petals, and flowers have an almost radially symmetric perianth. The flower's sexual parts are fused to a short column ending with a hood-like structure (mitra) or pseudanthery (Pridgeon et al., 2001; Fig. 1), hereafter ‘false anther’. Sun orchids, as their common name suggests, typically open their flowers in response to daylight, warm temperatures, and humidity. Here, we focus on two common species, T. macrophylla and T. crinita. The two species have similar flower morphology (Fig. 1), but T. macrophylla flowers earlier than T. crinita. In a previous study, bees of the genera Leioproctus and Lasioglossum were identified as pollinators of T. macrophylla, while only Leioproctus bees pollinated T. crinita (Edens-Meier et al., 2013). It is not known whether bees attempt to buzz flowers, and whether pollinators vary among populations.

2.2 Study sites

We collected data in autumn 2023 and 2024 in Southwestern Australia. The distribution of T. macrophylla and T. crinita ranges from Perth to Albany, and from Gingin to Esperance, respectively. Study populations of T. macrophylla were in Banksia woodland habitats within Warwick and Shanton Park bushland reserves, and study populations of T. crinita were found in forests dominated by Eucalyptus marginata in the Darling Range on the Perth hills (Fig. S1). Due to an unusual heatwave that caused flowering of T. crinita in the Perth region to end early, we also included a population in the Margaret River region, at Flat Rock bushland, Augusta (Fig. S1). Both Thelymitra species are abundant in the study regions, with typical population sizes ranging from 20 to 100 plants. The flowering time ranges from September to October for T. marcrophylla and from October to November for T. crinita (Western Australian Herbarium, 1998). Both orchids co-occur with a range of rewarding species with similar flower colour, including both pollen-rewarding buzz-pollinated species and nectar-rewarding species that are not buzz-pollinated (Edens-Meier et al., 2014; Brundrett et al, 2024). In the T. macrophylla study populations, main co-flowering species were Orthrosanthus laxus (Iridaceae) and Sowerbaea laxiflora (Asparagaceae) (Fig. 1 C,D that both are pollen-rewarding and buzz-pollinated. In the T. crinita populations, the main co-flowering species were pollen-rewarding Thysanotus manglesianus and Agrostocrinum hirsutum (Asphodelaceae) (Fig. 1 G,H), and nectar-rewarding Lechenaultia biloba, Dampiera linearis, and Scaveola calliptera (Goodeniaceae).

2.3 Field observations and experiments

To determine flower colour similarity, we measured flower colour in one population of each orchid species, including all main co-flowering rewarding species (Table 1). To identify orchid pollinators and their behaviour on orchids and model plants, we conducted pollinator observations in three populations of T. crinita (Table 1). We initially included both orchid species in our observations, but due to very low insect activity in T. macrophylla populations, we focused on T. crinita. To test the adaptive significance of false anthers and whether it depends on rewarding context, we used two experimental approaches, anther removal and anther painting. With anther removal, physical damage may influence bee behaviour on flowers. With anther obscuration by paint, false anthers remained intact, and only the visual signal was removed. First, we examined the effect of removing the false anther in two populations of T. macrophylla, one with putative model plants (O. laxus, S. laxiflora) and one without (Table 1). Second, we examined the effect of obscuring the false anther using paint in two populations of each of the two species (Table 1). In the anther-painting experiment on T. macrophylla, we could not identify any workable sites with variation in model abundance, and the experiment was conducted in two population that both had abundant model plants. In T. crinita, we conducted the anther-painting experiment in two populations that differed quantitatively in local abundance of putative model plants. In both populations, we quantified abundance of model and magnet plants by estimating the number of flowers per species in five randomly distributed 5 × 5 m plots per population. We counted all flowering stems and all flowers on ten stems per plant, and estimated total number of flowers by multiplying number of stems by mean number of flowers per stem. We included all rewarding species with blue-violet flower colour as potential model (A. hirsutum, T. manglesianus which are pollen-rewarding) and magnet plants (D. linearis, L. biloba, S. calliptera which are both nectar and pollen-rewarding) regardless of the reward type, because buzz pollinators forage on both nectar- and pollen-rewarding flowers. The mean number of rewarding flowers per plot was 3.9 times higher in the population with high model abundance (mean ± SD = 87.6 ± 15.0) compared to the population with low model abundance (mean ± SD = 22.6 ± 12.4), and the ratio of mimics to models were approximately half in the former (0.45 and 0.89, respectively).

2.4 Flower colour

To characterise flower colour in the two deceptive orchids and the co-flowering pollen- and nectar-rewarding species, we collected inflorescences in the field and brought them to the laboratory, where we measured six flowers per species using a field spectrometer (JAZ, Ocean Insight, Orlando, FL). On each flower, we took one measurement on the surface of the tepals in the orchids and of the petals in the co-flowering plants. To quantify colour similarity according to bee visual perception, we used the conventional bee vision model (colour hexagon space) based on photoreceptor spectral sensitivity of honeybees (Chittka, 1992). We quantified colour similarity by calculating the mean Euclidian distance between hexagon colour loci of Thelymitra and putative models and magnets. In addition, we used false colour photography as per Lunau et al. (2021), to identify colour patterns and details in the central area of the orchid flowers. This technique combines colour and UV photographs and is a robust method for obtaining detailed information on flower colour patterns under natural light conditions in the field (Verhoeven et al., 2018; Lunau et al., 2020). The technique splits colour photos into the three colour channels blue, green, and red, and by combining the green- and blue-channel photos with the UV photos, the resulting assemblage illustrates how bees perceive flower colour, i.e., including UV and excluding red.

2.5 Pollinator observations

To identify pollinators of T. crinita, we assessed the approach of all visitors to the flowers (attraction to the flower without necessarily landing), their behaviour and potential landing, and the capability to remove and deposit orchid pollinia, combining direct pollinator observations with camera recordings. Because natural visitation rate was very low, we created experimental arrays of T. crinita inflorescences close to rewarding species, to increase the likelihood of observing visitors (Scaccabarozzi et al., 2020). Arrays were formed by three cut orchid inflorescences placed in glass vials (one inflorescences per vial, each with 4–6 flowers) spaced 10 cm apart, and located ca 1 m from flowering individuals of the main rewarding species, T. manglesianus and A. hirsutum. In Perth hills, we conducted 80 and 20 ten-minute trials in the first and second population, respectively, and in the Augusta population, we conducted 200 ten-minute trials. This resulted in 50 hours pollinator observations in total. In each period, we also recorded insect behaviour using an EOS M video camera (Canon, Tokyo, Japan) for subsequent viewing in slow motion. Observations were done between 9 am and 3 pm (typical flower anthesis) during sunny or partly sunny days, when temperatures were above 22°C (measured 20 cm above the ground with a Smartsensor AR827, Sinosource Scientific Company, China).

To determine if T. crinita shares pollinators with putative model species, we also conducted pollinator observations and camera recordings of T. manglesianus and A. hirsutum, during some of the orchid arrays. For both species, we performed 21 ten-minute observation periods, yielding a total of 3 hours and 30 min observations per species. When possible, pollinators of both T. crinita and the putative model plants were identified in the field. In other cases, specimens were netted and later identified at the Museum of Western Australia.

2.6 False anther manipulations

a)     Anther removal experiment

On 24 September 2023, we randomly tagged 50 plants in each of two T. macrophylla populations separated by approximately 1 km within the Warwik Bushland Reserve, Perth. Putative model plants were present in one of the populations. In each population, we randomly assigned half of the plants to an anther removal treatment, while the other half were used as controls (n=25 per treatment). In the anther removal treatment, we carefully excised the false anthers (i.e., the upper part of the column, Fig. S2) from all open flowers using scissors. Most flowers were already open at the time of the experiment, and any remaining buds were removed. Prior to conducting manipulations, we recorded number of flowers per individual and examined if flowers had been visited. All flowers with pollinia removal or deposition were marked and excluded from the treatments (no difference between treatments; P > 0.05). In both populations, we recorded fruit production of all plants two weeks after removing false anthers.

To test if anther removal affected the ability to produce fruits, we supplementally hand-pollinated all flowers on 16 additional plants in another population in the Warwick bushland reserve, where eight had their false anther removed as described above, and eight were unmanipulated controls. We recorded fruit production two weeks later. Anther removal in T. macrophylla did not affect the ability to produce fruits (one-way ANOVA; F_1,12 = 0.35, P = 0.56).

b)     Paint experiment

On 4 October 2023, we randomly tagged 40 plants at flowering peak in each of two T. macrophylla populations (both with abundant model plants) in Shenton Bushland Reserve, Perth. In each population, we randomly assigned half of the plants to an anther paint treatment, while the other half were used as controls (n=20 per treatment). In the paint treatment, we obscured false anthers in all open flowers by applying a colour that matched the colour of the column (Eraldo Di Paolo, Acrylic Paint, Warm Blue; colour match examined by spectrometer measurements and false colour photography; Fig. 2A). To control for potential side-effects of painting, we also applied paint to the backside of the corolla of all flowers in the control treatment. On 17 October 2023, we randomly tagged 50 plants at flowering peak in each of two T. crinita populations (high vs. low abundance of model plants) on the Darling Scarp, Perth Hills. In each population, we randomly assigned half of the plants to an anther paint treatment, while the other half were used as controls (n=25 per treatment). Treatments were conducted as described for T. macrophylla. Prior to manipulations, we recorded number of flowers per individual and marked and excluded all flowers with pollinia removal or deposition for both orchids (no difference between treatments; both P > 0.05). In both experiments, we recorded fruit production of all plants two weeks after the anther manipulations.

To test the ability for autogamous pollination in both orchids, we bagged the inflorescences on ten individuals of each species at the bud stage, and recorded fruit production four weeks later. None of the bagged inflorescences in any of the two species developed fruits.

2.7 Statistical analysis

We initially checked if number of flowers per plant differed between treatment groups (anther removal or paint vs. control) or populations at the onset of the experiments, by using two-way ANOVAs including the interaction between population and treatment. Number of flowers did not differ between treatments or populations in any of the experiments (all P > 0.31).

To examine the effects of treatment (anther removal or paint vs. control), population, and their interaction on number of fruits produced in each of the three experiments, we used Generalised Linear Models with Poisson distribution and log link. Data were in some cases zero-inflated, and we used a negative binomial distribution when we detected overdispersion of the residuals. We used the package glmm TMB in R Studio (version 4.2.3).