Pollinator responses to floral colour change, nectar, and scent promote reproductive fitness in Quisqualis indica (Combretaceae)

Floral colour change is visual signals for pollinators to avoid old flowers and increase pollination efficiency. Quisqualis indica flowers change colour from white to pink to red may be associated with a shift from moth to butterfly pollination. To test this hypothesis, we investigated Q. indica populations in Southwest China. Flowers secreted nectar continuously from the evening of anthesis until the following morning, then decreased gradually with floral colour change. The scent compounds in the three floral colour stages were similar; however, the scent composition was different, and the scent emission rate decreased from the white to red stage. Dichogamy in Q. indica prevents self-pollination and interference of male and female functions. Controlled pollinations demonstrated that this species is self-incompatible and needs pollinators for seed production. Different pollinators were attracted in each floral colour stage; mainly moths at night and bees and butterflies during the day. Observations of open-pollinated inflorescences showed that white flowers had a higher fruit set than pink or red flowers, indicating the high contribution of moths to reproductive success. We concluded that the nectar and scent secretion are related to floral colour change in Q. indica, in order to attract different pollinators and promote reproductive fitness.

Scientific RepoRts | 6:24408 | DOI: 10.1038/srep24408 Quisqualis indica is an Asian tropical climber that undergoes a floral colour change from white to pink to red 24 . Although Eisikowitch and Rotem 25 described the role of orientation and colour change in pollinator attraction under field conditions in Israel, no detailed studies have been carried out on the pollination biology of this species. Our preliminary field observations indicated that floral colour change is related to increased nectar secretion and a very strong scent. In many cases, flowers change colour, but only attract one group of pollinators 15,17,18,21 ; however, the pollination syndromes of this species is assumed to attract two kinds of long tongued insects for pollination 26 , moths at night and butterflies during the day. Therefore, floral colour change may be related to a shift from moth to butterfly pollination. To evaluate this assumption, we examined the potential relationship of floral colour change with nectar and scent as well as the resulted changes in pollinator attraction and reproductive success, and aimed to address the following questions: (1) Is the nectar secretion and scent intensity pattern related to the floral colour change? (2) Is the floral colour change related to pollinator shift? and (3) (Table 1). Flowering occurred mainly from early April to May, whereas sporadic inflorescences were produced until the end of June.
Floral colour change. The petal light reflectance in the white, pink, and red stage ( Fig. 1) of Q. indica is shown in Fig S1. At anthesis (19:00-20:00), the flowers were white with a round corolla (13.89 ± 0.23 mm in diameter, N = 88), which was much smaller than that of the pink (P 1 < 0.0001) and red (P 2 < 0.0001) flowers, and had a delicate scent. All flowers were oriented obliquely upward with the petals first deflexed and then expanded. The following day, the flowers turned slightly pink at sunrise, changed to dark pink by midday, whereas after 14:00, they turned red and became pendulous with the corolla reaching the maximum diameter (16.79 ± 0.25 mm, N = 88; P 3 = 0.015). The older red flowers could persist on the inflorescence for 4-6 d.
Pollen viability and stigma receptivity. Pollen viability was high at anthesis and remained high (approximately 80%) throughout the white floral stage. Then, it decreased in the pink floral stage by approximately 60%, whereas about 20% of the pollen was still viable in the red floral stage for 2-3 d (Fig. 2). Stigma receptivity was very low at anthesis, increased quickly, reached a maximum at 11:00 the following day in the pink floral stage, and declined gradually at d 3 in the red floral stage (Fig. 2).
Floral nectar. The flowers produced nectar continuously in the white floral stage from anthesis until 08:00 in the following morning (3.7 ± 1.8 μl, N = 15) (Fig. 3). Later, there was a sharp decline in nectar volume, when floral colour changed from pink to red and secretion stopped at approximately 14:00. Concurrently, the nectar sucrose concentration decreased gradually from 12.6 to 9.0 mol L −1 . Variations in air temperature and relative humidity on flowering days are also shown in Fig. 3.

Flower visitors and pollinators. Flowers were visited by bees (mainly Apis dorsata and a few Apis cerana)
in the first hour after anthesis, and then, only hawkmoths (Macroglossum sp.) made subsequent visits until the following morning, but with a very low frequency. The highest visitation rates for moths in XTBG were 0.06 ± 0.05 (N = 13) visits flower −1 h −1 at 21:00-22:00, whereas for bees 1.02 ± 0.68 (N = 6) visits flower −1 h −1 in the second morning from 07:00-08:00 (Fig. 4). Flowers were visited by two species of butterflies (Papilio sp. and Pieris rapae) at 10:00-14:00 in the pink floral stage, whereas very few butterflies visited the flowers in the red floral stage. All bees, moths, and butterflies were effective pollinators, because they consistently contacted both anthers and stigmas and had pollen grains deposited on their bodies. Moths visited single flowers to rapidly collect nectar rate and then flew away; butterflies visited single flowers to collect nectar and briefly stayed in an inflorescence; whereas bees visited many newly opened flowers to collect pollen grains and stayed longer in the same inflorescence or different inflorescences within the same plant, which could cause geitonogamy and pollen discounting due to the self-incompatibility in this species.
Fruit set in different floral colour stages. In the Tongliang, Chongqing (CQTL) population, the fruit set rate of control inflorescences (21.2 ± 10.5%, N = 13) was the highest, followed by that of inflorescences exposed to natural pollination in the white stage (15.5 ± 7.1%, N = 14), whereas those exposed in the red stage had the lowest fruit set rate (4.2 ± 2.9%, N = 14). However, differences in the fruit set rates within the CQTL population were not significant probably because of the small sample sizes and large variation (Fig. S2). In the XTBG population, the fruit set of each treatment was lower than that in the CQTL population. The highest fruit set rate was observed in the white floral stage (8.9 ± 2.2%, N = 64), followed by the control (6.5 ± 1.4%, N = 30), and both were significantly higher than those in the pink floral stage (1.8 ± 1.2%, N = 60; P 1 = 0.000 and P 2 = 0.045, respectively) and the red floral stage (0.6 ± 0.6%, N = 58; P 3 = 0.011 and P 4 = 0.000, respectively) (Fig. S2).

Discussion
This study revealed that Q. indica has an unusual pollination system. Its floral colour change attracts different sets of pollinators, including moths and bees in the white floral stage, bees and butterflies in the pink floral stage, and butterflies in the red floral stage, with all sets facilitating cross-pollination. The flowers secreted nectar continuously and maintained high nectar volumes and sugar concentrations in the white stage, which then decreased gradually from the pink to the red stage; thus, the secretion pattern was related to changes in floral colour. Pollen is viable before stigma receptivity, and this short temporal dichogamy can prevent self-pollination. In this study, the scent profiles changed, and the scent emission rate decreased with the floral colour change from white to pink to red, suggesting some functional relation between scent emission and colour change in order to achieve the best pollination attraction in different floral colour stages. This species is self-incompatible, and the results also showed that fruit set rates were higher when inflorescences were exposed to natural pollination in the white floral stage compared with the pink and red floral stages, suggesting that moths and bees had higher contribution to reproductive success than other insects. Overall, our results indicated that nectar and scent secretion along with floral colour change in Q. indica affect pollinator behaviour and promote reproductive fitness. Floral colour change has been recognized as a special pollination strategy, whereas the overall floral display is enhanced without increasing pollen wastage. The flowers of Q. indica are typically pollinated by hawkmoths in the white floral stage 25,27,28 , in which the large amount of nectar and relatively low sugar concentration meet the known reward levels of the preferred flowers of hawkmoths 28,29 . The flowers in the pink and red floral stages become larger and long-tubed and are mainly preferred by butterflies 28,30 . In this study, the pollinator composition in the three floral colour stages was different from that described by Eisikowitch and Rotem 25 . According to pollination syndrome concept 26 , Q. indica would not be pollinated by short-tongued bees. However, bees visited both white and pink flowers with a quite high frequency, regardless flower orientation, and also multiple flowers within the same inflorescence, probably causing lead to pollen discounting. Moreover, they often visited white flowers before the moths, negatively affecting reproductive success by collecting pollen before the effective pollinators. The synchronous mass flowering in Q. indica can cause geitonogamy 31 , which leads to low fruit set rates under natural conditions 32 ; however, dichogamy may reduce the risk of geitonogamy and prevent self-pollination 31 .
Floral nectar rewards can affect pollination by manipulating pollinator behaviour 33 , and floral scents are used by flowers to attract pollinators by simultaneously attracting some pollinator taxa while repelling others 34 . In the white stage, the flowers of Q. indica secreted nectar and emitted fragrance continuously to attract moths, whereas from the pink to red stage, nectar volume, sugar amount, and scent emission rate reduced considerably, and pollinators shift to bees and butterflies. This study showed that the flowers 'advertise' their nectar rewards in the pink and early red floral stages, when pollen viability has declined. These findings were in disagreement with previous studies [14][15][16][17][18]21 , which showed that flowers changed colour maintaining a relatively high pollen viability and preserving male and female fitness for longer time. Red flowers persist on the inflorescence for 4-6 d, increasing their floral display size and creating a "cloud" of scent, which probably acts as a long distance signal to attract more pollinators 3,11,[13][14][15][16] . Floral nectar secretion and scent changes during floral colour change in Q. indica may be part of the flower senescence process. In this study, we could not clarify if nectar and scent changes influence pollinator behaviours separately; thus, further study is needed to determine their individual roles in pollinator attraction.
Floral colour and scent are interlinked traits in plants 35 . Flower scents are classified by their biosynthetic origin into terpenes, phenylpropanoids, and fatty acid derivatives 36,37 , and the phenylpropanoid pathway determines floral colour and scent 38 . Some studies demonstrated that these volatiles are derived from two biochemical pathways: one produces acyclic monoterpenes and the oxides, and the other, benzoate and its derivatives 39 . Q. indica has a strong floral scent in the white floral stage, when it is pollinated by moths, and then its scent emission rate declines in the pink and red floral stages, when it is pollinated by butterflies. It is possible that the scent contributes more in moth attraction during the night, whereas distinct colours contribute more in bee and butterfly attraction during the day. This transition is consistent with the phenylpropanoid pathway, but light may also be a triggering factor (Zhang et al., unpublished data) that stimulates anthocyanin pigment metabolites [40][41][42][43] .
In conclusion, Q. indica is a dichogamous and self-incompatible species; therefore, seed production is pollinator dependent. The combination of floral colour change, nectar secretion, and scent releasing pattern in Q. indica probably influence pollinator behaviours and plant reproductive success. Additionally, floral colour change may be a type of adaption to pollinator shift from nocturnal to diurnal. All sets of pollinators play a role in reproductive success, but pollinators in the white floral stage contribute more to fruit production than those  in other stages, suggesting that diurnal pollination is a form of compensation, in cases that nocturnal pollination is not successful. Overall, our results indicated that nectar and scent secretion patterns reflect the floral colour change rhythm in Q. indica, and multiple signals attract and manipulate pollinators in order to promote reproductive fitness. However, further research is needed to better define the interactions between two key floral traits, colour change and scent. We used a haemocytometer to estimate pollen production per flower as described by Dafni 44 . We used pollen and ovule numbers to calculate the mean P/O of the flowers. To investigate floral colour change, we monitored five flowers from different inflorescences every 2 h for 2 d and recorded the time and flower colour, which was assessed visually and then measured with a spectrophotometer (Ocean Optics, USA). In addition, we also measured the petal length of three flowers from each inflorescence of 30 plants in the white, pink, and red floral stage using vernier calipers (0.01 mm; Guanglu, China) to determine any changes in the corolla size. We used 0.1% 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyl-tetrazolium bromide to test the presence of dehydrogenase as an assay for pollen viability 45 . A total of 64 flowers from four study plots were covered using nylon mesh bags prior to anthesis to prevent insect visits, and then four flowers were used to test pollen viability every 3 h from anthesis to flower wilting. The same assay was used to assess stigma receptivity 45 , using the same number of experimental flowers from the same plants. All tests were carried out under favourable weather conditions. Every day, we randomly selected three inflorescences from each of three study plants located approximately 500 m apart and covered them with nylon net bags in the afternoon prior to anthesis. Three flowers in each inflorescence were used to measure nectar secretion every 3 h from 20:00 to 17:00 the following day. The nectar volume was measured with 10 and 20 μl 'micro-cap' calibrated capillary tubes (Sigma-Aldrich, USA), and nectar sucrose concentration with a hand-held, temperature-compensated refractometer (Bellingham + Stanley Ltd., UK). This experiment was repeated for 3 d  Flower visitors and pollinators. We observed flower visitors to Q. indica in 2012 and 2013 in MLD and XTBG for a total of 81 h and 144 h, respectively, under favourable weather conditions. Observations were made continuously from 20:00 (beginning of anthesis) to 8:00 the following morning by video recording during the night and visually during the day. We recorded the total number of visitors per individual flower and the number of flowers visited by each kind of insect. The visiting frequency of each kind of insects was calculated as the number of visits per flower per hour. We assigned insects as visitors or pollinators, based on their behaviour and likelihood of mediating pollination. Visitors were insects observed on inflorescences, whereas pollinators were insects that consistently contacted both anthers and stigmas and had pollen grains deposited on their bodies. We photographed every type of flower visitor, and voucher specimens of insects were preserved in the insect collections of XTBG.

Study species and sites. Q. indica Linn (Combretaceae) is an Asian tropical climber that is mainly distrib
Field manipulative experiment. To investigate the reproductive contribution of different floral colour stages, we conducted a manipulative field experiment in the XTBG and CQTL populations using inflorescences exposed to natural pollination. We set up four treatments with 30 plants per treatment using 1-3 randomly chosen inflorescences per plant. All inflorescences were covered with nylon net bags to prevent insect visits prior to treatment. The treatments included inflorescences exposed to natural pollination in (1)  (1) open-pollination (control), (2) hand selfing, (3) hand outcrossing, (4) hand geitonogamous pollination, and (5) bagging (autonomous self-pollination). For each treatment, except for the control, we covered inflorescences with bags prior to anthesis to prevent pollinator access, and the flowers were emasculated to prevent self-pollination. We performed hand cross-pollinations using pollen from plants up to 500 m away and randomly selected from 10 inflorescences. Not all flowers within an inflorescence were cross-pollinated because of the technical difficulties related to bud emasculation in this species.
Collection and identification of floral scents. Floral scents were collected in situ from flowers in the white, pink, and red stage from four Q. indica plants using the dynamic headspace adsorption method 46 , and also included inflorescences with a combination of white and red flowers or pink and red flowers. For each collection, an inflorescence with 5-25 flowers was covered with mesh bags to prevent pollinator access. Some leaves close to the inflorescence were removed from the branch at least 24 h prior to the scent collection experiment. For each sample, the average number of flowers used in the white, pink, and red stage was 16, 16, and 15.3, respectively. We enclosed a whole inflorescence within an odourless polyethylene terephthalate bag (Kalle Nalo GmbH, Germany) for the volatile collection. Airflow was maintained through the bag by a battery-driven air pump (Qihai Machinery and Electric Co., China). The air was purified by active charcoal, introduced into the bag at a flow rate of 400 ml min −1 , pumped out of the bag through a glass cartridge (7 mm internal diameter), and filled with 300 mg of Super Q adsorbent (80-100 mesh size; ARS Inc., USA) at a flow rate of 300 ml min −1 . The adsorbent filters were eluted three times with 100 μl of hexane. To detect any environmental contamination during the volatile collections, ambient air was collected at the same place using the same dynamic headspace technique. The leaf branch was used as control in some cases that only a few leaves were included in the inflorescent treatment. The collection was carried out for approximately 3 h in each floral colour stage. Any volatile compounds that were common with the control treatment were removed before the analysis. To calculate the absolute amount of floral volatiles, two internal standards, octane and decyl acetate, were added to each sample as described by Chen et al. 46 . The extract samples were then stored at − 20 °C until analysis.
The extracts were analysed using a coupled gas chromatography mass spectrometer (GC-MS) system (Agilent, USA) equipped with an HP-5MS column (30 m × 250 μm × 0.25 μm). Helium was the carrier gas, and ionization occurred by electron impact (70 eV; source temperature 230 °C). For each sample, 1 μl was injected at an injector temperature of 250 °C. The initial column temperature was 40 °C, and then increased by 3 °C min −1 up to 100 °C, by 5 °C min −1 up to 200 °C, and finally by 20 °C min −1 up to 250 °C, which was maintained for 10 min. Compound identification was based on matching mass spectra in the NIST08 Mass Spec library and confirmed by the retention index (RI) in the NIST online library (http://webbook.nist.gov). The absolute values of all compounds were estimated using the average peak area of the two internal standards as a reference scale. The relative proportions of each compound were also calculated.

Data analysis.
To assess the differences of corolla diameter between floral colour stages, we carried out one-way analysis of variance (ANOVA) in conjunction with Fisher's least significant difference test to identify pairwise differences. The mean fruit set was calculated in all treatments before statistical analysis. We used Kruskal-Wallis test to identify differences in fruit set rates among different treatments in XTBG and CQTL in 2013 and independent-sample t test to identify differences in fruit set under different pollination treatments in XTBG and MLD in 2013 and XTBG in 2014. All analyses were performed using SPSS 20.0 (IBM, USA). A matrix of the relative proportions of all the detected compounds in the three floral colour stages was used to conduct multivariate analysis in R 3.0.0 (https://www.r-project.org/) with the 'vegan' package 47 . Scent composition among different floral colour stages was also compared. MRPP was performed based on the matrix of mean dissimilarities with 999 permutations to test the null hypothesis (scent profiles among floral colour stages had no differences). NMDS was used to find the best two-dimensional representation of the distance matrix. To evaluate each configurations that produces the distance matrix, we tested different stress values. The fit of the produced