From March 2014 to May 2017, wild populations of E. pubescens, E. wushanense, and putative hybrid were studied in the Jinchengshan National Forest Park, Nanchong of Sichuan Province, China (1128 m a.s.l., 30°45′N, 106°28′E), an area considered the forefront of the ecological barrier on the upper reaches of the Yangtze River. The region experiences a mean annual temperature of 18.6°C, with average temperatures ranging from 5.4°C in January to 21.7°C in July. Annual precipitation and evaporation average 1063 mm and 705.8 mm, respectively. The soil in this area is red loam soil at a depth of 0–30 cm. The tree and shrub layers of the plots were similar, dominated by secondary conifer and broad-leaf mixed forests, which succeeded the evergreen broad-leaf forest [31]. The habitats of E. pubescens, E. wushanense, and the putative hybrid populations exhibited slight differences (Table 1).

Epimedium species typically have a flowering period from March to May, during which they produce terminal inflorescences. Each panicle generally contains 10–50 flowers. The flowers consist of two layers, with four sepals in each layer, as well as four exterior conical spurs and four stamens surrounding the persistent pistil inside. The outer sepals may exhibit pale green, purple, or yellow colors and shed as the buds open, while the inner sepals are white. The four petals are pale yellow and have a spur-like shape. Nectar is produced by nectaries located at the tip of the spur and stored within the spurs. Each anther contains two chambers, and green or yellow pollen is present upon dehiscence. Detailed floral parameters of the Epimedium species studied at these sites can be found in Table 2.

During the months of March to May in both 2014 and 2016, the flowering phenology of each Epimedium species was meticulously observed and recorded at three subplots, both at the population and individual levels. The parameters observed in this study included the first flowering date, peak flowering date, last flowering date, the flowering amplitude curves, and the flowering intensity. The determination of the first flowering date, peak flowering date, and last flowering date followed the statistical methods described by Dafni [32]. Before the anthesis stage of the Epimedium plants, 30 buds were randomly selected from three populations within each Epimedium subplot. These buds were carefully labeled with notice plates to record the total number of buds per inflorescence. The number of open flowers was subsequently recorded every two days until the end of the flowering period. The first flowering date was defined as the point when approximately 25% of the plants had started to flower. The peak flowering date was reached when over 50% of the plants were in bloom. The last flowering date was considered the point at which over 75% of the plants had faded, marking the end of the flowering period. The flowering amplitude curve and flowering intensity were calculated using the statistical methods developed by Herrera [33]. The flowering amplitude curve represents the number of opening flowers per day and per unit time (from 4:30–18:30 h) at the individual plant level. Flowering intensity, on the other hand, is the ratio of the maximum number of opening flowers produced by a plant on the peak flowering date.

To assess plant height, leaf morphology, and floral morphology, over 30 flowering plants were randomly chosen from three populations within each Epimedium subplot. Various measurements were taken during the peak blooming period of each species (n = 30), including floral diameter, spur length and height, entrance height and width of the spur, stamen length, pistil length, style length, as well as anther length and width. Vernier calipers from Guanglu, Guilin, China were used for these measurements. To determine the volume of nectar produced, a minimum of 40 flower buds from 10 plants in each population were randomly selected during the peak blooming period, labeled, and bagged. Ten flowers were carefully removed from different plants at 9:00 AM on the first, second, third, and fourth days of flowering. The nectar volume of each flower was immediately measured using calibrated 5 μL capillary tubes from Hirschmann Laborgeräte, Germany.

During the peak blooming period, pollination observations were conducted for a duration of 5 days in all three populations of each Epimedium species. The observations involved documenting the type of pollinator, their pollination behavior, visit frequency, and pollination time. Furthermore, images of the pollinators were captured using a Nikon DSLR D7000 camera with 16.2 megapixels. All flower visitors were captured and later identified and measured in the laboratory. To compare the effectiveness of pollinators in each species, the removal and deposition of pollen by a single efficient floral visitor per plant were examined. Sixty flower buds from 10 plants (6 flowers per plant) were randomly chosen at the peak blooming period in each population, labeled, and placed in bags. These bagged flowers were then divided into two groups randomly, with 30 flowers in each group. In one group, the flowers were gently opened and emasculated using tweezers before the anthers disappeared, after which the bags were carefully covered again. The other group did not undergo any treatment. On the first day of flowering, when floral visitors were observed, the bags were opened. Whenever a flower was visited by a potential insect pollinator, it was promptly removed and preserved in a Formalin Aceto-Alcohol (FAA) solution. To determine the mean number of pollen grains per unvisited flower per Epimedium species, we collected 20 unvisited flowers per species, just before anthesis. These were also fixed using FAA and the pollen grains in each were counted in the laboratory using a Leica DM500 binocular microscope (Heerbrugg, Switzerland). By counting the pollen removal from captured insects, they were placed in a high-speed centrifuge to separate the pollen grains from other tissues through centrifugal force. Subsequently, the separated pollen grains were counted. Using these data, a pollen-removal ratio per insect visitor was calculated. Pollen deposition on stigmas by pollinators of the Epimedium species was assessed per insect visitor and reported as the number of individual pollen grains deposited [34].

During the peak blooming period of the three Epimedium species, we conducted five pollination treatments from 2016 to 2017. These treatments included: (1) Natural control: 30 buds from different panicles were labeled and left open for pollination without any treatment until the fruits ripened. (2) Automatic self-pollination: Thirty buds from different panicles were labeled and covered with fine-mesh nylon bags to prevent insect visitors. No other manipulations were performed. (3) Artificial self-pollination: Thirty buds from different panicles were labeled and covered with fine-mesh nylon bags. The stigmas were manually pollinated with pollen from the same flower after the anthers had decreased. The bags were then used to cover the panicles. (4) Conspecific cross-pollination: Thirty buds from different panicles were labeled and covered with fine-mesh nylon bags. Before the flowers opened, the anthers were manually removed using tweezers. Then, the stigmas were pollinated with pollen from other individuals of the same species that were located at least 200 meters away. These panicles were bagged after pollination. (5) Interspecific artificial pollination: Sixty buds from each Epimedium species were labeled and covered with fine-mesh nylon bags. Before the anthers dehisced, the flowers were opened and emasculated using tweezers. Then, the stigmas were pollinated as follows: (a) E. pubescens stigmas were pollinated with pollen from E. wushanense and putative hybrid, (b) E. wushanense stigmas were pollinated with pollen from E. pubescens and putative hybrid, and (c) putative hybrid stigmas were pollinated with pollen from E. pubescens and E. wushanense. These panicles were bagged after pollination. The harvested Epimedium seeds from that year were sown directly in the seedbeds of the study sites, which had humus-rich soil. The seeds were covered with approximately 1 cm of soil and kept moist on the surface. By the beginning of the second spring, the germination rate was counted.

All the collected data, which included the flowering phenology, floral parameters, volume of secreted nectar, fruit set, germination set, and the efficiency of effective pollinators, were subjected to statistical analysis using SPSS v22.0 (SPSS Inc., Chicago, Illinois, USA). The normality and homoscedasticity of variables were evaluated using the Shapiro-Wilk and Levene tests, respectively. Normal and normalizable data were compared using a one-way analysis of variance (ANOVA) with Duncan’s multiple-range test. Furthermore, regression analysis (LSD) was employed to compare the visit frequencies among different Epimedium species. To minimize experimental errors and enhance result reliability, it is important to incorporate randomization and repetition in sampling and experimental design.

Phenological studies of Epimedium sites indicate that the flowering season lasts for 20–30 days, from early March to mid-May annually. However, the first, peak and last flowering dates of each species vary, depending on the plant’s habitat and climate (Fig. 1). The flowering amplitude curves of E. pubescens and E. wushanense exhibit bell curves that peak on the 15th and 26th days of flowering, respectively (Fig. 1a). At the population level, the flowering periods of E. pubescens and E. wushanense overlap for approximately 15 days period. E. pubescens and E. wushanense are largely characterized by a single peak in their flowering amplitude curves, while the putative hybrid tends to bloom gradually in batches. The relative flowering intensities are consistently over 50%, suggesting that the flowering period of these Epimedium species is short and centralized (Fig. 1b).

The longevity of a single flower of each species is 5–7 days, bisexual, zygomorphic, inodorous, and grows invertedly. Intuitively, the floral features of the putative hybrid exhibit intermediate forms between E. pubescens and E. wushanense (Fig. 2). The leaves of E. pubescens are oval-acuminate, with finely spined toothed edges and brown patches, thinly leathery; while E. wushanense leaves are acutely pointed, with toothed edges, fine spines, and a dark green, glossy appearance. The putative hybrid produces acutely pointed, light green leaves with brown patches and the leaf margins have toothed edges with fine spines. The stems may be hairy or glabrous. The flowers of all species display inner and outer sepals, with each layer containing 4 sepals that grow in a symmetrical manner. The outer sepals exhibit a pale purple or green color and gradually shed during the bud stage. Following the large-bud stage, the inner white sepals envelop the petals, as well as the male and female organs, along with the outer sepals, and progressively expand throughout the blooming process (Figs. 2a–2i). Four pale yellow-spurred petals grow symmetrically. The petals of E. pubescens are small and vertical (Fig. 2c), whereas those of E. wushanense (Fig. 2f) and the putative hybrid (Fig. 2i) are either straight or curved, and the nectaries are located at the tip of the spur. From the large-bud stage until the first day of flowering, the pistil is slightly longer than the stamen (Figs. 2b, 2f, 2g). The anthers decrease in length and enclose the stigma on the initial day of flowering. After successful pollination, the milky white stigma transitions to a pale yellow hue and gradually elongates. E. pubescens possesses 4–5 ovules in one ovary, whereas E. wushanense and the putative hybrid have 10–12 ovules.

Differences in morphological parameters among the Epimedium species include plant size, leaf shapes (leaf length and width), and floral features, which are summarized in Table 1. Plant height was significantly different between E. wushanense (47.358 ± 1.121 cm), the putative hybrid (31.698 ± 1.815 cm), and E. pubescens (23.422 ± 1.129 cm) (Table 2, p < 0.05), while leaf length and width were also significantly different between the two species and putative hybrid (Table 2, p < 0.05). The diameter of E. wushanense flowers (37.215 ± 0.418 mm) was significantly greater than that of E. pubescens (19.252 ± 1.101 mm) and the putative hybrid (18.124 ± 0.323 mm) (Table 2, p < 0.05), although the difference was not significant between E. pubescens and the putative hybrid (Table 2, p > 0.05). The length and width of the spur were significantly different between E. wushanense, the putative hybrid, and E. pubescens, respectively (Table 2, p > 0.05). The length and width of the anthers in E. wushanense and the putative hybrid are significantly greater than in E. pubescens (Table 2, p > 0.05), with no significant differences found between E. wushanense and the putative hybrid. Evaluations of plant height, leaf length, leaf width, length of outer sepal, length of filament, length of style, and other floral characters demonstrated that E. wushanense was significantly larger than E. pubescens and the putative hybrid. The morphology of the putative hybrid showed an apparent transitional state between E. pubescens and E. wushanense. The secretion of nectar in the three species took place daily from the large-bud stage until the withering of the flowers, and the volume of nectar in the bagged flowers gradually increased. However, on the second day, the rate of increase slowed Fig. 3. The mean nectar volume per flower between the three study subjects was significantly different, with E. wushanense having ± 38.6 times more nectar than E. pubescens and ± 5.9 times more nectar than the putative hybrid (Fig. 3, p < 0.05).

Insect visitation to the three Epimedium species was more frequent within 2–3 days of a flower opening. Few insects visited plants during rainy weather. At least sixteen different species of insects, mainly Hymenoptera and Diptera, were recorded during the field observation period for the three Epimedium species. By counting the number of pollen grains on each captured insect specimen’s body using a binocular microscope. It was found that various bee species carried more pollen and were therefore arguably effective pollinators (Table 3). Lasioglossum sp. (Hymenoptera: Halictidae) were the most efficient pollinators of E. pubescens in terms of the pollen grain count they carried, visiting flowers for both pollen and nectar (Figs. 4a, 4b). The pollen grains on the stamens of E. pubescens were collected and transported by the legs of bees, which packed them into their pollen baskets. During flower visitation, these bees also deposited pollen grains from their baskets onto the stigma. Apis sp. (Hymenoptera: Apidae) were opportunistic visitors with a low frequency of visits and were occasionally seen on E. pubescens (Fig. 4c). The nectar-feeding Bombus trifasciatus (Hymenoptera: Apidae) and Bombus grahami (Hymenoptera: Apidae) frequently visited the flowers of E. wushanense, and distinct visiting behaviors were observed between the two species of bumblebees. B. trifasciatus showed legitimate pollination behavior when visiting E. wushanense by tightly grasping a spur with their front legs, inserting their proboscis into the spur to extract nectar, pollen grains adhering to neck hairs during visitation (Fig. 4d). When B. trifasciatus visited flowers on separate plants of E. wushanense, neck hair pollen was deposited on stigmas during flower visitation. B. grahami did not display legitimate floral visitation to E. wushanense by piercing holes in the middle of spurs to consume nectar, bypassing the anthers (Fig. 4e). Microscopic examination revealed no pollen on the bodies of these bees. Lasioglossum sp. species showed legitimate flower visitation to E. wushanense, pollen being deposited onto their pollen basket (Fig. 4f). By pollen grain counts, the most efficient visitors to the putative Empodium hybrid were Lasioglossum sp. and Anthophora sp. (Hymentoptera: Anthophoridae. Figs. 4g–4h). Anthophora sp. were only observed visiting the hybrid Epimedium, collecting nectar. B. grahami were opportunistic visitors to the hybrid to collect nectar (Fig. 4i), having a low visitation frequency and with no pollen found on their bodies during microscopic examination. Our results show that in terms of pollen grains removed from anthers and pollen grains deposited onto stigmas, Lasioglossum sp. moved significantly more pollen grains than B. trifasciatus and Anthophora sp. (Table 3, p < 0.05).

The visiting frequency of effective pollinators of the three Epimedium species (with mean visiting frequencies of E. wushanense at 3.08%, E. pubescens at 4.63%, and the hybrid at 1.49%, respectively) was relatively low (Fig. 5). There were significant variations in the visitation rate of effective pollinators throughout the day. Lasioglossum sp. had a higher visit frequency than B. trifasciatus and Anthophora sp., with the peak period for effective pollinator visits occurring from 12:00 to 14:00 (Fig. 5).

The results of the artificial pollination combinations and the corresponding fruit-setting rates and germination-setting rates are presented in Table 4. All three Epimedium species in this study were found to have an outcrossing breeding system. Cross-pollination and open pollination treatments led to the production of fruits, with fruit set rates surpassing 50% and germination setting rates exceeding 70%. Conversely, treatments involving autonomous and artificial self-pollination did not result in any fruit production. When interspecies artificial pollination was conducted among these three Epimedium species, fruit production was observed. When E. pubescens was used as the female parent (♀) and E. wushanense and putative hybrid were used as the male parent (♂), the fruit set obtained was 18.92% and 24.11%, respectively. The seeds obtained from the hybrid also exhibited a certain germination setting rate of 30.04% and 25.85%, respectively. When E. wushanense was treated as the female parent (♀) and E. pubescens and putative hybrid were used as the male parent (♂), the fruit set reached 75.22% and 72.97%, while the germination set reached 80.72% and 88.04%, respectively. When the putative hybrid was used as the female parent (♀) and E. pubescens and E. wushanense were used as the male parent (♂), the fruit set reached 83.33% and 90.63%, with germination set rates as high as 87.73% and 85.31%, respectively.

Floral morphology and pollination biology are essential considerations in the study of floral evolution and plant reproduction. The pollination system serves as a model for understanding the interaction between natural selection and evolution [35,36]. Having a comprehensive understanding of how pollination influences the reproductive success of hybrids is essential for their conservation and recovery. This is particularly important because certain hybrids may give rise to unique genotypes through intraspecific hybridization, potentially enhancing their ability to adapt to the environment [37–39].

The occurrence of hybridization among species is primarily influenced by their geographical distributions [40]. The occurrence of spontaneous hybridization between closely related species in sympatric populations frequently leads to gene introgression, which significantly impacts the genetic structure of natural populations [41]. Epimedium species, categorized as belonging to the Old World temperate distribution type, have a widespread and sporadic distribution within a narrow strip spanning from Japan in Asia to Algeria in North Africa (25°~48°N, 5°~143°E) with each species usually having a narrow distribution [42,43]. Both E. pubescens and E. wushanense have small population sizes and are fragmented in their distribution within the Jincheng Mountain National Forest Park in Sichuan. The population of E. wushanense is significantly smaller than that of E. pubescens [23,24]. The coexistence of Epimedium species in the Jinchengshan National Forest Park, along with the relatively narrow ecological niche of E. wushanense, provides conditions for the possibility of hybridization. Pollen movement also contributes to the potential hybridization of different species [44]. The synchronization of flowering affects the flow of genes within species and the interactions between plants and other trophic levels, such as pollinators [45]. The relative flowering intensity indicates the spatial distribution of plant flower resources, which can influence the movement patterns of pollen gene flow in plants [46]. Although the first flowering dates, peak flowering dates, and last flowering dates of E. pubescens and E. wushanense differ, they both exhibit a short and centralized flowering period that overlaps for approximately 15 days at the population level (Fig. 1), indicating the potential for hybridization.

Hybridization facilitates ecological divergence as hybrids display a mixture and transition of parental traits, and the new gene combinations generated are speculated to contribute to ecological divergence [47]. Suzuki found that once hybrids are formed, they maintain distinctive traits between their genetic parents and inherit the dominant genes [48]. Breeding system methods can easily determine the parent of the hybrid in the sexual reproduction of Epimedium plants, as clonal plants do not form hybrids [49]. The obvious differences between parents are revealed through the morphology of the hybrid, such as inflorescence height, inflorescence number, ovule number, and flower characteristics [28,29]. The morphology of the Epimedium flower has been widely studied in previous research. The evolutionary trend of Epimedium corolla morphology was originally a reduction in the number of petals from five to four and three, and a decrease in the size of the innermost petal to form spurs [25,50]. This trend is thought to be related to the long-tongued pollination syndrome that evolved in insects, allowing more efficient transfer of pollen to the stigma. According to previous research, the floral aromas in the Epimedium genus are usually small and generally do not have a noticeable fragrance. However, some varieties of flowers may have a subtle and mild aroma. Overall, the rewards that attract insects to Epimedium are primarily pollen or nectar [51]. The nectar reward and morphology of the spur play important roles in the attraction of pollinators. The putative hybrids have intermediate traits between their parents in terms of the corolla shape, spur length, and stamen length, which are associated with the reproductive fitness of plants. The putative hybrid had longer corolla tubes and spurs than E. pubescens but shorter than E. wushanense, potentially attracting different pollinators and increasing the efficiency of pollinator-mediated gene flow. The putative hybrid also had an intermediate number of stamens, which may affect its pollen production and transfer. These traits contribute to the adaptability of the putative hybrid and provide a potential mechanism for its successful establishment in nature.

It has been reported that the majority of Epimedium species exhibit self-incompatibility as a mechanism to prevent self-fertilization. This is achieved through dichogamy, herkogamy, and self-incompatibility mechanisms [24]. Pollination experiments with three Epimedium species (Table 4) showed that they belong to an outcrossing breeding system, relying on insect pollination and excluding self-pollination. The continuity of populations is reliant on pollination services, as approximately two-thirds of crops worldwide depend on insect pollination to some degree, with wild pollinators typically contributing to crop yields [52]. Plants attract insects for pollination through color, size, pollen, or nectar reward, influencing pollinator activity, efficiency, and reproductive efficiency of flowers [53,54]. Through the observation of pollinators, it was found that E. pubescens and E. wushanense share the pollinator Lasioglossum sp., indicating the possibility of natural hybridization between them. Furthermore, Anthophora spp. was observed only on the putative hybrid and not on its parents, creating isolation conditions between the hybrid and its parents. Effective pollinator species and their visiting behaviors are crucial for the successful reproduction of plants [55]. The quantity of conspecific pollen grains present on a stigma is frequently indicative of seed production and is utilized to measure the effectiveness of pollinators in plants [56]. In this study, the pollen removal and deposition by effective pollinators varied among Lasioglossum sp. was the most effective pollinator, contributing to the reproductive fitness of the putative hybrid in fragmented habitats. The flower-visiting behaviors of Lasioglossum sp. primarily involved pollen collection, while B. trifasciatus and Anthophora sp. were rewarded with nectar due to the feedback mechanism of the sender on the change of flower characteristics. The diversity of Epimedium species is significantly impacted by pollinator-mediated gene flow.

The experimental results from artificial pollination combinations indicate that different species of Epimedium can pollinate each other successfully and obtain fertile seeds. However, the fruit setting rates and seed germination rates of different combinations vary (Table 3). Hybrid incompatibility and hybrid sterility can prevent hybridization due to the various isolation mechanisms between species [12]. Previous studies have shown that E. pubescens and E. wushanense have a relatively close genetic relationship based on ISSR and RAPD markers and hybridization is not restricted [28]. Suzuki and Du et al. have produced hybrids between large-flower taxa and small-flower taxa Epimedium species under positive and reverse hybridization artificial pollination, showing that large-flower taxa Epimedium species are widely affinitive pollen receptors and are suitable as female parents for interspecific hybridization, while small-flower type Epimedium species have good pollen affinity and are suitable as male parents in crossbreeding [25,57]. The seed germination rate of E. pubescens as the female parent (♀) and E. wushanense and putative hybrid as the male parent (♂) were only 30.04% and 25.85%, respectively, while the other hybrid combinations were above 70.00%, with the highest rate of 88.04%. This indicates that there is a possibility of natural hybridization between them to form species. It further proves that E. wushanense grouped into large-flower taxa is suitable for the female parent in interspecific hybridization, and E. pubescens grouped into small-flower taxa is suitable for the male parent. Moreover, it is possible that putative hybrids can backcross with their parents, providing continuity in the formation of putative hybrid species [16,58].