Nectary tracks as pollinator manipulators: The pollination ecology of Swertia bimaculata (Gentianaceae)

Abstract Floral nectaries are closely associated with biotic pollination, and the nectar produced by corolla nectaries is generally enclosed in floral structures. Although some Swertia spp. (Gentianaceae), including S. bimaculata, evolved a peculiar form of corolla nectaries (known as “gland patches”) arranged in a conspicuous ring on the rotate corolla and that completely expose their nectar, little is known about the pollination of these plants. Two hypotheses were made concerning the possible effects of gland patches: visual attraction and visitor manipulation. The floral traits, mating system, and insect pollination of S. bimaculata were examined, and the pollination effects of gland patches were evaluated. A comparative study was made using Swertia kouitchensis, a species with fimbriate nectaries. Swertia bimaculata flowers were protandrous, with obvious stamen movement leading to herkogamy in the female phase and to a significant reduction in nectary–anther distance. The species is strongly entomophilous and facultatively xenogamous. The daily reward provided per flower decreased significantly after the male phase. The most effective pollinators were large dipterans, and the visiting proportion of Diptera was significantly higher in S. bimaculata than in S. kouitchensis. Most visitors performed “circling behavior” in S. bimaculata flowers. Removing or blocking the nectaries caused no reduction in visiting frequency but a significant reduction in visit duration, interrupting the circling behavior. The circling behavior was encouraged by nectar abundance and promoted pollen dispersal. Visitor species with small body size had little chance to contact the anthers or stigma, revealing a filtration effect exerted by the floral design. These results rejected the “visual attraction” hypothesis and supported the “visitor manipulation” hypothesis. The nectary whorl within a flower acted like a ring‐shaped track that urged nectar foragers to circle on the corolla, making pollination in S. bimaculata flowers more orderly and selective than that in classically generalist flowers.


| INTRODUCTION
Floral nectaries are specialized structures in angiosperms that are closely associated with biotic pollination (Nicolson, Nepi, & Pacini, 2007). Nectar is easy for plants to produce and easy for animals to intake and digest, and it is often the primary offering of a flower, enhancing the reproductive success of plants (Willmer, 2011). Plants invest substantial amounts of sugar and water to supply nectar as a readily available energy source for animals (De la Barrera & Nobel, 2004), which makes nectar secretion a major determinant of the interaction between flowers and visitors (Willmer, 2011).
Nectaries are usually located more or less basally within the flower, with the nectar stored in spurs, concealed in corolla tubes, or enfolded (at least partially) by some other floral structures to protect against evaporation, nectar theft, fungal spores, bacteria, and/or accidental removal (due to wind, rain, or gravity); thus, the chance of anther contact would increase, owing to the inserting action the visitor must perform to reach the nectar (Bernardello, 2007;Corbet, Willmer, Beament, Unwin, & Prys-Jones, 1979;Pacini, Nepi, & Vesprini, 2003;Willmer, 1983). However, fully exposed nectar can also be observed in some species with abundant pollen (e.g., certain members of Apiaceae and Asteraceae), in which case the visitors scrabble over the surface of the flower or inflorescence, spreading pollen in the process, thus leading to "mess pollination" (Corbet, 2006;Willmer, 2011). Some Gentianaceae species (i.e., some Swertia spp.) evolved a rather peculiar form of corolla nectaries with a flat spot-like shape, which are located in the middle of the corolla lobe instead of at its base. These nectaries, which usually differ in color from the rest of the corolla, are arranged in a conspicuous ring on the rotate corolla, and the nectar secreted by these "gland patches" is entirely exposed ( Figure 1). There is no record of the floral visitors or pollinators of these species in the literature, and hence, knowledge of their pollination ecology is limited. Gland patches are derived from primitive forms of nectaries that occur at the base of the corolla lobe and are found in most species of Swertia (Chassot, Nemomissa, Yuan, & Küpfer, 2001;He, Xue, & Wang, 1994). These nectaries are generally concave and have fimbriate margins (Xue, He, & Li, 2002). Little has been published on the pollination of Swertia. According to the literature, the most frequent visitors and primary pollinators of Swertia spp. with fimbriate nectaries are probably bees Khoshoo & Tandon, 1963).
In the traditional sense, the rotate corolla of Swertia spp. displays a generalist pollination syndrome (Figure 1b,c), that is, plants with this corolla shape are usually pollinated by "mess pollination" by a range of fairly generalist visitors (Willmer, 2011). It is possible, though, that the characteristic gland patches play an important role in the flower-visitor interaction, influencing the preference, behavior, and/or pollination effectiveness of visitors, resulting in a relatively selective and ordered pollination pattern different from typical generalist pollination. We hypothesized that gland patches may have two possible effects on pollination: (1) visual attraction, as the yellow-green nectaries are arranged in a conspicuous ring-like pattern on the corolla (Figure 1b-d), which might be identified by floral visitors within a proper distance range and consequently attract more potential visitors to the flower; and (2) localization and guidance, that is, manipulation of the visitors, where the nectary whorl acts like a ring-shaped track, urging the feeding insects to keep moving on the corolla and using nectar as a food reward.
In this study, we tested these hypotheses by investigating the pollination ecology of Swertia bimaculata Hook. f. & Thomson ex C.
B. Clarke, a species with gland patches (Figure 1a), to understand the role that gland patches played in pollination. A comparative study was made between S. bimaculata and a congener, Swertia kouitchensis Franch., which has fimbriate nectaries, to examine morphological and functional differences between the two types of nectaries. Specifically, this study addressed the following questions: (1) How does pollination differ between species with these two types of nectaries? (2) How do gland patches affect visitor preference, behavior, and pollination effectiveness? and (3) Have other reproductive strategies associated with gland patches been developed by S. bimaculata?

| Study site
Swertia bimaculata is an annual plant species that is widely distributed in southern China, and its habitat elevation ranges from 250 to 3,000 m above sea level (a.s.l.). Several natural populations inhabiting Mount Guanmen, located on the eastern Ta

| Sexual system
The sexual system investigation was conducted in 22 successive days  . (e) A flower that was bagged for one day, showing nectar drops. (f, g) Two common Diptera visitor species with typical behavior, that is, crawling along the nectaries and circling on the corolla so that the insect body is constantly outside the androecium circle. (h, i) Two common Hymenoptera visitor species with the typical behavior. (j, k) A Diptera visitor with atypical behavior, which leads part of the insect's body to get into the androecium circle (j, top view; k, side view). a, anther; bf, bisexual/monoclinous flower; ff, female flower; n, nectary; nd, nectar drop; p, pollen; s, stigma. Bars = 50 mm (a) and 10 mm (b-k)

| Floral lifespan and morphology
The lifespan of 30 monoclinous flowers randomly selected from the population was observed. The timing of each floral event was recorded, including the opening/closure of the corolla, the beginning/ ending of pollen dispersal, and the exposure/deactivation of the stigma, which was defined by the physical appearance of the stigma (e.g., coloration and texture). The duration of male, interval, and female phases were then calculated using these data.
A set of 68 flowers was randomly selected from the population to collect morphological data for floral organs, including corolla diameter, nectary whorl diameter, stamen length, androecium diameter, and pistil length. Measurements between different types of flowers (e.g., between male-and female-phase monoclinous flowers) were compared. A similar investigation was conducted in S. kouitchensis, during which 30 randomly selected S. kouitchensis flowers were used for morphological measurements. Another 50 randomly selected monoclinous S. bimaculata flowers (including both male-and female-phase monoclinous flowers) were measured to quantify positional relationships among the nectary whorl, androecium, and stigma (i.e., distances between each pair); the measurements were grouped based on floral age and sexual phase, and mean values between different groups were compared. All floral measurements were taken using a Vernier caliper and then converted to geometrical standards using the appropriate algorithm (e.g., the distance between two nonadjacent anthers in a flower was measured using the caliper and then converted to the androecium diameter using the law of sines).

| Mating system
In total, 18 monoclinous buds were randomly collected from the S. bimaculata population in 2012 and 2013. The five anthers from each bud were ground in a centrifuge tube as 1 ml of water was gradually added, until the pollen grains were fully released. The tubes were well shaken, and 12-25 samples (1.5 μl each) of the pollen suspension were transferred onto a microscope slide using a pipette. Pollen grains in each sample were observed and photographed under a microscope (Olympus BX43, Tokyo, Japan). The number of pollen grains in each sample was counted from one field of view, and the number of pollen grains per flower (P) was calculated. The ovary of each bud was dissected under a stereomicroscope (Olympus SZX2, Tokyo, Japan), all ovules were photographed, and the number of ovules (O) was counted from the photograph. The pollen/ovule coefficient (P/O) for each bud was then calculated as an indication of the mating system (Cruden, 2000).
To better characterize the mating pattern of S. bimaculata, six treatments were established for six groups of monoclinous flowers:   sugar content (S, mg) was then calculated as V × C, as the density of low-concentration sugar solutions can be roughly regarded as 1 mg/ μl. If nectar was obviously secreted (at least 1 μl), the bag was replaced over the flower; otherwise, another first-day flower was randomly selected and bagged to keep a constant total number (20) of bagged flowers. This procedure was repeated daily for 10 days, but after the fifth day, no new flowers were added to the sampling set. In total, 33 flowers were recorded during the 10 days. Daily nectar volume, sugar concentration, and sugar content were compared between floral dates and between male-and female-phase monoclinous flowers. Over the two periods of data collection, the insect species visiting S. bimaculata flowers and their behavior were recorded through photographs and field notes. Based on observations, behaviors were categorized into several modes according to the feeding course in the flower (e.g., the landing place and the feeding order among the petals) and the way of movement between petals (e.g., flying or crawling).

| Insect pollination and effects of nectaries
Visitor species were divided into several groups based on taxonomy and body size.  daily nectar volume; sugar concentration; and sugar content between each pair of floral dates. Mann-Whitney: morphology measurements between male-and female-phase monoclinous flowers; the numbers of matured and total ovules and the seed-set coefficients between each pair of groups in the mating pattern experiment; daily nectar volume, sugar concentration, and sugar content between male-and female-phase monoclinous flowers; and pollination effectiveness between visitor groups. Kruskal-Wallis: the number of matured and total ovules; and the seed-set coefficient among groups C, D, E, and F in the mating pattern experiment). Otherwise, parametrical tests were used to compare mean values between groups: independent sample t tests for two groups and one-way ANOVA for multiple groups (two-sample If the data did not conform to a normal distribution, Spearman's ρ was used as the correlation coefficient; otherwise, Pearson correlation coefficients (r) were used. Values are presented as means ± standard deviation (SD) in the results, unless stated otherwise.

| Sexual system
Both monoclinous and female (i.e., male sterile) flowers were observed in Swertia bimaculata populations. Stamens of the latter flower type were usually absent, although a minority had rudimentary and infertile stamens ( Figure 1d). The proportion of female flowers in the experimental population averaged 9.7% ± 3.3%.

| Floral lifespan and morphology
Once the corolla opened, anthers of the monoclinous flowers dehisced in 0.9 ± 0.8 hr (N = 19). When anthers dehisced and pollen began to disperse, the flower was considered to be in the male phase, which lasted for 0.5 ± 0.5 days (N = 14) under natural conditions, until all pollen inside the anthers was dispersed. The male phase was followed by the interval phase, when the stigma was not yet exposed, and this phase lasted for 1.0 ± 0.4 days (N = 15). The flower entered the female phase as soon as the stigma was exposed (two stigma lobes opened, exposing the receptive surface) and could receive pollen, which lasted for 2.1 ± 0.7 days (N = 9). The corolla then gradually closed over the next 2.0 ± 0.8 days (N = 10). As the rate of pollen dispersal varied with insect visiting frequency, the duration of the male phase was strongly influenced by external factors (e.g., weather); this was noted in field observations. Therefore, the phase between anther dehiscence and stigma exposure was considered a generalized "male phase" in this study. The duration of this generalized male phase was stable, as it was not substantially influenced by external factors. The division of sexual phases does not apply to female flowers because they have no fertile male organs. Hereafter, "male-phase" or "femalephase" flowers refer to monoclinous flowers in their generalized male or female phase, unless stated otherwise.
In S. bimaculata, female flowers had significantly smaller diameters of corolla, nectary whorl, and androecium than monoclinous flowers, and their pistil lengths were shorter (Table 1; Figure 1d). Androecium diameter was significantly larger in female-than in male-phase flowers (p < .001), indicating an obvious stamen movement away from the stigma after anther dehiscence in monoclinous flowers (Table 1; Figure 1b,c). In addition, during anthesis, S. bimaculata nectaries were always more distant from the flower axis than the androecium was (Table 1; Figure 1b,c), contrary to what was found in S. kouitchensis (Table 1). These results indicated that the floral morphology in S. bimaculata and S. kouitchensis was quite different, especially as reflected in the positional relationship between nectaries and other floral structures.
As the flower aged, anthers (a) clearly moved away from the stigma (s), that is, the distance from a to s significantly increased (male-to interval phase: t = 3.49, df = 30, p = .002; interval-to female phase: t = 3.34, df = 35, p = .002), which would lead to distinct herkogamy at the beginning of the female phase ( Figure 2; Table 2). Additionally,
In the mating pattern experiment, fruit-set coefficients, seed numbers per capsule, and seed-set coefficients of the six groups of monoclinous flowers were obtained (Table 3). None of the flowers in group A (bagged and emasculated) set fruits, suggesting that apomixis did not occur. A minority within group B (bagged flowers without emasculation) set fruits with insufficient seed-set rates.

| Nectary morphology and nectar secretion
Gland patches were observed as flat structures that slightly bulged out from the corolla. Adaxial epidermal cells of the corolla had similar F I G U R E 2 Positional relationships between nectaries, anthers, and stigmas in male-phase, interval, and female-phase Swertia bimaculata flowers. Fifty monoclinous flowers were measured. For each flower, an anatomically longitudinal section that passed through the flower axis and a nectary was chosen and fixed at coordinates (0, 0). The anatomically horizontal/vertical direction was then defined as the X/Y-axis, and the coordinates of the stigma were recorded. Because stamens alternated with corolla lobes, the section plane can only pass through an anther on the opposite side of the floral axis.
The symmetry point of this anther with respect to the floral axis was selected as the coordinate of the "anther". Thus, the nectary-anther distance in this figure is exactly equal to the shortest distance from the nectary whorl to the androecial circle (see the diagrammatic sketch in the figure, which shows the longitudinal section of a flower). a, anther; f-p, female phase; int., interval; m-p, male phase; n, nectary; s, stigma Nectar secretion of a monoclinous flower lasted for 3-8 days (4.9 ± 1.6 days, N = 23), during which a total of 14.4-124.7 μl (57.2 ± 29.1 μl, N = 19) of nectar was secreted ( Figure 5). The average daily nectar volume before stigma exposure, that is, in the first 2 days of anthesis, was significantly higher than that after stigma exposure (before: 13.9 ± 8.0 μl, N = 52; after: 8.5 ± 6.2 μl, N = 89; Z = −3.93, p < .001), indicating a highly significant difference in the rate (volume per unit time) of nectar production between male and female phases ( Figure 6).
Sugar concentration in S. bimaculata nectar ranged from 1.6% to 22.4% (7.6 ± 4.1%, N = 122). It averaged 10.8 ± 4.2% (N = 37) in the first 2 days of anthesis and 6.2 ± 3.1% (N = 85) in the following days, indicating a highly significant difference in sugar concentration between the male-and female-phase nectar (Z = −6.06, p < .001) ( Figure 5). In addition, the first and only significant decline of daily sugar concentration occurred at the beginning of the female phase:  Figures 5 and 6). Based on these results, the daily abundance of food reward for pollinators s-a n-a n-s T A B L E 2 Relative distances between nectary (n), anther (a), and stigma (s) in Swertia bimaculata  and there was a highly significant difference between these groups (t = 3.80, df = 94, p < .001; Figure 7b).

| Insect pollination and effects of nectaries
In nectary effect experiment 2, each flower in groups A (control), B (spotted areas covered), and C (gland patches covered) was visited by 5-13 (9.5 ± 3.4, N = 4), 6-10 (7.5 ± 1.7, N = 4), and 5-13 (7.6 ± 3.4, N = 5) insects in 15 min, respectively. There was no significant difference between these three groups (F 2,10 = 0.58, p = .58; B, and C, respectively, and a highly significant difference among these three groups was found (F 2,103 = 11.724, p < .001; Figure 7d). Tukey's HSD indicated that the difference was not significant between groups A and B (p = 1.000), but it was significant between groups A and C (p < .001) and between groups B and C (p < .001). Dunnett's t test indicated no significant difference between the control group (A) and group B (p = .999) and a significant difference between the control group and group C (p < .001).
Gland patches and spotted areas were the two main floral factors Nearly all visitors performed a "circling" behavior when feeding on S. bimaculata flowers, for example, searching along the nectaries successively to collect nectar, moving clockwise or anticlockwise on the corolla. When doing so, most visitor species crawled between petals, whereas others (mainly some Syrphidae spp.) flew from one petal to another. The former behavior was defined as the "typical" behavior for S. bimaculata pollinators because it was far more common and more likely to generate contact between the insect's body and the anthers (Figure 1f-i). Furthermore, a visitor was designated "large" if the species had this typical behavior and the individual's body length was greater than 10 mm.
The circling behavior of insects was also observed in S. kouitchensis, which has fimbriate nectaries. The main difference of insect behavior observed between S. bimaculata and S. kouitchensis is that in S. kouitchensis, the radial distance from nectary to pistil is short, allowing a pollinator to contact the stigma as well as the anthers when circling along the nectaries, whereas the long nectary-pistil distance in S. bimaculata flowers makes it almost impossible for a typically behaved pollinator to come in contact with the stigma (Figure 1f (Table 4).
Furthermore, in each type of flower, visiting frequency was not substantially different between bees and flies (Table 4). In all three types of flowers, the colonial stability (indicated by the number of effective visits per flower per hour) of flies was higher than that of bees, as was the colonial effectiveness (indicated by the number of times the anthers or stigma was touched per flower per hour). It is noteworthy that large flies contributed to more than 40% of the anther contacts and more than 70% of the stigma contacts in monoclinous flowers, although they only accounted for approximately 12% of all insect visitors (Table 4). Nevertheless, no large flies were recorded visiting female flowers and only a few were registered in field observations. Incidentally, pollen-eating behavior only happened four times among the 1,168 visits, always by bees.
In S. kouitchensis, visiting frequency was 2.9 times flower −1 hr −1 , which was less than that for S. bimaculata (17.1 times flower −1 hr −1 ), and the visiting frequency of bees (2.2 times flower −1 hr −1 ) was higher than that of flies (0.7 times flower −1 hr −1 ). The proportion of fly visitors to S. bimaculata was significantly higher than that to S. kouitchensis (p < .001).
A significantly higher individual pollination stability (indicated by the probability of an effective visit) in flies than in bees was revealed in both male-phase flowers ( flies: 2/11, p = .227), which was possibly due to the small sampling size (Table 5) Table 5).
The correlation between pollinator species' body size and the spe-   the mean value of P remained above three when Int was less than 5 s, and even in post-female-phase flowers (the corolla had not closed but nectar secretion had completely stopped, with absolutely no available nectar), the average P was still above two ( Figure 10).
This showed that there was also a long-term "taming" effect contributing to the maintenance of circling behavior, especially under a high visiting frequency.
According to the above, nectary tracks play a key role in manipulating the pollinators' behavior, encouraging them to pass through most of the petals in a flower (instead of a minority of them), reflected in the typical circling behavior. However, it might be more interesting to verify whether promoting circling behavior could benefit the pollination of S. bimaculata (i.e., will average pollination effectiveness increase if the pollinator probes more petals in a flower?). Video and F I G U R E 1 0 Correlation between P (number of petals the visitor probed per visit) and Int (visiting interval) when insects visited Swertia bimaculata monoclinous flowers. Mean values and standard errors (SE) of P are displayed using columns and bars, respectively. Visiting interval is defined as the time between the departure of the previous visitor and the arrival of the subsequent visitor. If the previous visitor was still in the flower when the subsequent visitor arrived, Int was regarded as zero. Data from 1,083 visits to monoclinous flowers were collected and used in the first four sets; these data were divided by the three Int quartiles of 24 s, 74 s, and 182 s, so that the four data sets had similar sampling sizes (268, 272, 272, and 271). The fifth data set was drawn from the first set by adjusting the upper limit of the visiting interval to 5 s. The last set of data was collected from 26 visits to post-female-phase flowers, whose nectar secretion had completely stopped, disregarding Int. Different annotation letters indicate a statistically significant difference at α = 0.05 (Mann-Whitney test) F I G U R E 1 1 Correlation between A or S (number of times that the anthers/stigma were/was touched per visit) and Int (visiting interval) when large flies visited Swertia bimaculata monoclinous flowers. Mean values and standard errors (SE) of A or S are displayed using columns and bars, respectively. Visiting interval is defined as the time between the departure of the previous visitor and the arrival of the subsequent visitor. If the previous visitor was still in the flower when the subsequent visitor arrived, Int was regarded as zero. Data from 71 visits to male-phase flowers and 62 visits to female-phase flowers were collected and used in the two panels, respectively. Data were divided into four sets by the three Int quartiles (male-phase flowers: 17 s, 33 s, and 64 s; female-phase flowers: 19.5 s, 58.5 s, and 178.25 s), so that the four data sets had similar sampling sizes (male-phase flowers: 17, 18, 17, and 19; female-phase flowers: 16, 15, 15, and 16). In each panel, the fifth data set was drawn from the first set by adjusting the upper limit of the visiting interval to 5 s. Different annotation letters indicate a statistically significant difference at α = 0.05 (two-sample t test for A and Mann-Whitney test for S) field observation showed that when an effective pollinator (e.g., a large fly) fed on nectar on successive petals, one side of the insect's body contacted the anthers successively; thus, numerous pollen grains adhered to the hairy parts of the insect (Figure 1f,g). The significant positive correlation between P and A or S (the number of times that the anthers/stigma were/was touched in a visit) in large flies revealed that promoting P would benefit both pollen dispersal and pollen receipt in S. bimaculata (Figure 9), although pollen receipt may be indirectly improved by the circling behavior (i.e., the typical circling behavior itself may not lead to stigma contact, but with more circling movement during a visit, the insect would have a greater chance to exhibit an atypical behavior that would result in stigma contact).
The correlation analysis between Int and A or S was a more direct examination of available nectar volume's effect on pollination effectiveness. The results showed a significant correlation between Int and A in male-phase flowers, for all visitor species and for large flies ( Figure 11), but in female-phase flowers, no significant correlation between Int and S was found. This suggested that the amount of available nectar mainly facilitated pollen dispersal instead of pollen receipt.
Although the nectaries in S. bimaculata monoclinous flowers are closer to the androecium than to the stigma, the nectary-anther distance is not short enough to allow all nectar-feeding insects to touch the anthers. The results showed that more than 63% of insect visits to male-phase flowers were ineffective (i.e., no anthers were touched), and there was a positive correlation between pollinator species' body size and the species' average pollination effectiveness in male-phase flowers ( Figure 8). Therefore, the combination of nectary track and floral design (e.g., anther location) in male-phase flowers can be regarded as a spatial filtration mechanism that gives larger visitors a greater chance to contact anthers, reducing pollen waste on smaller visitors who have less chance to contact stigmas in female-phase flowers

| Gland patch comparison with other types of corolla nectaries
Floral nectaries can occur in virtually all parts of a flower, and they can be divided into receptacular, hypanthial, perigonal, calyx, corolla, androecial, or gynoecial nectaries based on their location (Schmid, 1988).
Therefore, nectar of these species is usually stored inside spurs, hidden in corolla tubes, or enfolded by other floral structures such as inner petals. In this context, the gland patch is a peculiar form of corolla nectary because it fully exposes nectar on the visually attractive surface.
The microscopically rough surface sculpture of the adaxial surface of the S. bimaculata corolla revealed in the present study (Figure 4b,c) would generate a "lotus effect", described as ultrahydrophobicity that makes water form spherical droplets (Barthlott & Neinhuis, 1997). In S. bimaculata, this effect might help to ensure that the nectar droplet is restricted to the gland patch region (Figure 1e) instead of spreading over the corolla, which might cause nectar loss and/or confusion in visitor manipulation. According to the literature, plant species with corolla nectaries would attract several groups of insect pollinators, including bees (Huang, Guo, Pan, & Chen, 1999;Suzuki, 1984), beetles (Thien, 1974), and flies (Schneider & Jeter, 1982;Silberbauer-Gottsberger, Gottsberger, & Webber, 2003); however, none of these reports addresses pollination in species whose nectar is fully exposed on the petal surface, so the present study is important for a comprehensive understanding of the pollination effect of floral nectaries.

| Nectar presentation and pollinator preference
In S. bimaculata, the fully exposed nectar presentation suggests that short-tongued pollinators are preferred visitors (Corbet, 2006;Nicolson, 2007), and nectar on the flat surface of the gland patch may be easy to collect by sponging mouthparts that uptake nectar by capillary adhesion rather than suction (Kingsolver & Daniel, 1995;Krenn, Plant, & Szucsich, 2005). These inferences were supported by our records regarding insect behavior and visiting frequency. Field observation showed that, most of the time, only a small volume of nectar was present in a flower because of the high visiting frequency, making bees less efficient at gathering the nectar through their chewing or siphoning mouthparts (Figure 1h). The feeding efficiency of flies, on the contrary, was not affected by the small amount of nectar (Figure 1f,g,j,k;Cruden, Hermann, & Peterson, 1983;Faegri & van der Pijl, 1979).

| The economics of nectar feeding and nectar production in S. bimaculata
Although pollination is usually regarded as a mutualism, there is always a conflict of interest between the plant and the pollinator. Both participants should be trying to balance their costs against the rewards and hence assessing the net benefits gained (Willmer, 2011).
Nectar is not only an essential reward to the pollinator but also a substantial cost to the plant (De la Barrera & Nobel, 2004), which makes it a key factor to understand plant-pollinator interactions. From the economic point of view, S. bimaculata is "tricky" for presenting a fully dispersed pattern of nectar within a single flower, which makes the visitor tend to move intraflorally instead of interflorally. For a visitor, flying from a petal in one flower to a petal in another flower will cost far more energy than walking to the adjacent petal in the same flower (cf. Voigt & Winter, 1999), whereas the expected reward gained by the two choices is equal. Therefore, it can be hypothesized that a nectar forager can achieve the best energy budget if it feeds on all five petals within a flower before moving to the next flower. Video analysis showed that an average insect (or a large fly) visitor would walk across 4.1 (or 4.2) petals in a monoclinous flower, which is consistent with the previous hypothesis.
The present study showed that a male-or female-phase flower of S. bimaculata produced 13.9 or 8. These results implied a potential "market mechanism" under which the visiting insects could expect equal rewards of nectar from two different sexual phases of flowers by adjusting their visiting probability (although the actual rewards might differ owing to differences in sugar concentration). According to Faegri and van der Pijl (1979) and Cruden, Hermann, & Peterson (1983), such a level of mean reward is more similar to that of a typical fly-pollinated generalist flower (less than 0.05 μl) than a bee-pollinated flower (0.10-10.00 μl). Notably, mean nectar reward volume under natural circumstances may differ from our estimates, which were based on daily measurements from bagged flowers.
Producing nectar with such a low C may be a strategy to prevent the nectar from becoming too concentrated and viscous for the pollinators to ingest (cf. Willmer, 1983). It is also possible that the C was diluted to some extent by moist air or dew in the mornings during the investigation.
According to De la Barrera and Nobel (2004), the production of nectar often peaks with maximum pollen availability, but sometimes it peaks with the maximum stigma receptivity; male fitness is often more strongly associated with nectar production than is female fitness (Aizen & Basilio, 1998;Mitchell, 1993;Pleasants & Chaplin, 1983). Carlson (2007) found that the nectar accumulation rate (indicated by sugar mass) of Chrysothemis friedrichsthaliana (Hanst.) H. E. Moore (Gesneriaceae) was greater during the male phase than the female phase, with a difference of 83%. The present study showed a similar result in that the volume, concentration, and sugar content of nectar produced daily by a S. bimaculata flower were all higher in the male phase than in the female phase, with a difference of 64%, 74%, and 139%, respectively. Moreover, both daily nectar concentration and daily nectar sugar content declined significantly at the beginning of the female phase, indicating a male-biased nectar production schedule.
The present study revealed that, in S. bimaculata, nectar rewards were usually very abundant in only a few flowers (see the outliers in Figure 5). This might represent an evolutionarily stable strategy to produce both nectarful and nectarless flowers within a plant species; wherein, the cost to the plant can be minimized through automimic flowers that produce no nectar (Bell, 1986;Brink, 1982;Thakar, Kunte, Chauhan, Watve, & Watve, 2003;and references in Gilbert, Haines, & Dickson, 1991).

| Nectar production and floral design tend to enhance male function in S. bimaculata
A widely tested theory in pollination ecology is that in plants with monoclinous flowers, the evolution of floral attractive traits may be driven primarily by selection on male function because increased pollinator visits may be more beneficial to male function than female function (Aizen & Basilio, 1998;Bell, 1985;Burd & Callahan, 2000;Campbell, 1989;Carlson, 2007;Devlin & Stephenson, 1985;Galen & Stanton, 1989;Lloyd & Yates, 1982;Melendez-Ackerman & Campbell, 1998;Stanton, Snow, & Handel, 1986;Willson, 1994), based on the assumption that male fitness is most strongly limited by access to mates, whereas the strongest limiting factor on female fitness is usually resources (Bateman, 1948;Charnov, 1979;Darwin, 1871). The present study indicated that S. bimaculata supports this theory: The manipulatory effect of the nectary track nearly exclusively promoted the chance of anther contact, not stigma contact, showing a male-biased floral design. The nectar production schedule is also male-biased, implying a male-biased resource allocation regarding floral rewards. Furthermore, the male-phase flowers had a higher frequency of insect visits and effective visits and a greater probability of effective visits than female-phase flowers, which are correlated with the male-biased floral design and nectar production schedule.
In S. bimaculata, the bias toward male function may have made the gynomonoecious sexual system an evolutionarily stable strategy (cf. Charnov, Smith, & Bull, 1976). In female flowers, the abortion of stamens could eliminate interference from the androecium, hence facilitating the female flowers achieving greater female fitness than monoclinous flowers with an equal investment of ovule resources (i.e., higher seed-set rates). The results of this study provide evidence for this inference, as the natural seed-set rate in female flowers was significantly higher than that in monoclinous flowers, whereas the female flowers had significantly lower visiting frequencies.

| Comparison of pollination in S. bimaculata and plants with generalist/specialist pollination
Both the open design of the corolla and the fully exposed nectar presentation in S. bimaculata suggested a generalist pollination syndrome (Corbet, 2006;Faegri & van der Pijl, 1979). Pollination in generalist flowers has been described as "catering for the mass market" by Proctor, Yeo, and Lack (1996) or "mess pollination" by Willmer (2011).
This concept was supported by the numerous species of floral visitors observed in the present study. Nevertheless, almost all visiting insects exclusively exhibited circling behavior on S. bimaculata flowers, reflecting an essential difference from mess pollination, in which case the visitors scrabble on the flower or inflorescence in a disorderly manner. In classically generalist flowers, pollen is easily accessible for most floral visitors, whereas, in S. bimaculata, most insect visitors were too small to contact the anthers or stigma while feeding on nectar.
Based on the comparison made above, it could be derived that S. bimaculata flowers were pollinated in a more ordered and selective way than typical generalist flowers. Furthermore, nectary tracks, which were not observed in generalist flowers, were essential in manipulating and filtering the visitors. Thus, pollination in S. bimaculata does not fit a generalist syndrome.
Notably, the selective strategy used by S. bimaculata in pollinator filtration was quite different from that used by specialist flowers.
Specialist flowers usually present a specific syndrome, ensuring that only the specific pollinators can be attracted to the flowers and access pollen and/or nectar (Baker & Hurd, 1968;Faegri & van der Pijl, 1979;van der Pijl, 1961), whereas, in S. bimaculata, various insect species are attracted to the flowers owing to the easily accessible food reward, although the majority of them have little chance to contact the anthers. In other words, the filtration process for effective pollinators is "prearrival" in specialist species and "postarrival" in S. bimaculata.

| The mating strategy of S. bimaculata, in comparison with other reported Gentianaceae spp.
It is widely known that organisms with sexual reproduction benefit from outcrossing: Outcrossed progeny is expected to be more heterozygous, hence individually more adaptive and with more fitness than selfed progeny (Agrawal, 2006;Bell, 1982;Maynard Smith, 1978). However, in flowering plants, pure outcrossing (e.g., selfincompatible) may bring potential risks to the number of offspring because cross-fertilization relies on an external agent for pollen transfer, for example, animals and wind, which are often unpredictable and/or Dafni, Hou, Duan, He, & Liu, 2005;Lennartsson, Oostermeijer, van Dijk, & den Nijs, 2000;Webb & Pearson, 1993) are frequently reported mechanisms for avoiding autogamy and sexual interference. Nevertheless, highly selfed mating system has evolved along with the loss of protandry and/or herkogamy in some Gentianaceae spp. (Fischer & Matthies, 1997;Machado, Sazima, & Sazima, 1998;Webb & Pearson, 1993), probably in response to the pollen limitation caused by unstable pollinator abundance and/ or constancy (Duan, Zhang, & Liu, 2007;Dudash, 1993;Petanidou, den Njjs, & Oostermeijer, 1995;Petanidou, Ellis-Adam, den Njjs, & Oostermeijer, 1998). Some Gentianaceae spp. grown in harsh habitats have developed mechanisms of spontaneous self-pollination as a delayed mechanism for reproductive assurance  or even as a main contributor to natural seed set (Petanidou, Ellis-Adam, den Njjs, & Oostermeijer, 1998). In the present study, floral lifespan and morphology investigations showed that S. bimaculata was both protandrous and herkogamous (herkogamy was especially distinct in the female phase), and the pollen dispersal experiment confirmed the strictness of dichogamy, as the free-pollinated anthers were normally emptied by pollinators in less than one day, long before the exposure of stigma. According to these results, it is most likely that autogamy (selfing within a flower) does not occur in the S. bimaculata population under natural circumstances. Thus, this species can only be pollinated by insect pollinators with geitonogamous or xenogamous pollen. The occasional fruit set of the bagged flowers in the mating system experiment was probably caused by accidental self-pollination, where the bag was blown by the wind, causing successive contact with anthers and stigma. During the field experiment, in the female phase of these bagged flowers, most pollen was still adhered to the anthers or the inner surface of the bag because insect visitors were excluded. The P/O ratio coefficient also suggested a highly outcrossed mating system, probably with a relatively high outcrossing rate among Gentianaceae species; according to , the P/O ratio of S. przewalskii is approximately 250-300, which is much lower than that of S. bimaculata (approximately 900-1000). Despite the strict prevention of self-pollination, there were considerable fruit-(over 90%) and seed-set rates (over 80%) in the free-pollinated S. bimaculata flowers. Furthermore, no significant pollen limitation was revealed, suggesting that being both protandrous and herkogamous was a successful mating strategy for S. bimaculata because it helped the species to produce offspring with both quality (highly outcrossed) and quantity (high seed-set rate). The possibility of being highly outcrossed without losing seed set is based on the premise of stable, effective, and ordered pollination, owing to high pollinator abundance in the habitat and the extraordinary design of nectary tracks.