pollinators: the importance of mouthpart dimensions in effective pollen removal and deposition

Pollination studies often use visitation frequency of potential pollinators as an indicator of their importance, but this is only one component and may not reflect actual pollen transfer rates. In this study, we determine the most effective pollinator group of Angadenia berteroi, a tropical perennial subshrub with large yellow flowers that set few fruits. We determined visitation frequency and pollen transfer effectiveness of the four most common groups of visitors (longand short-tongued bees, and skipper and non-skipper butterflies). Using potted plants, we exposed flowers to single visits from different types of pollinators to measure fruit set. We demonstrate that A. berteroi is most effectively pollinated by long-tongued bees, though many other species visit the flowers; the most frequent visitor group is not the most important pollinator, because they neither carry nor deposit much pollen, as the width of their proboscis is small compared with long-tongued bees. In this system, the width of the proboscis of the pollinators correlates with pollen transfer efficiency. Our results demonstrate the importance of pollen removal, pollen deposition, and fruit set, in determining the most effective pollinators, rather than visitor frequency. The distinctive morphology of these flowers, with a large bell and a narrow, short tube, suggests that other flowers of this shape may similarly benefit more from visitors with mouthparts shorter than previously considered optimal.


Introduction
The evolution and diversification of the perianth has been associated with pollinator attraction (Proctor et al. 1996;Inouye and Kearns 1997;Richards 1997) and mechanical fit (Fenster et al. 2004). Pollination syndromes are groups of floral traits present in distantly related plants that share similar pollinators: floral morphology, phylogenetic position and floral reward characteristics are all important in predicting what pollinates a given plant species (Johnson and Steiner 2000;Etcheverry and Alemá n 2005;Ollerton et al. 2007). The pollination syndrome concept reflects selection response to a functional group of pollinators that can exert similar selective pressures collectively on floral design (Fenster et al. 2004). Within communities, the majority of plant species are visited by a variety of pollinator groups, but visitation does not necessarily imply pollination; not all flower visitors are important and effective pollinators (Stebbins 1970;Waser et al. 1996;Fenster et al. 2004;Ne'eman et al. 2010). A pollinator's overall importance to a plant's reproduction involves its efficiency (successful dispersal of pollen grains deposited on conspecific stigmas) and visitation frequency (Armbruster 1985;Waser et al. 1996;Armbruster et al. 2000;Reynolds and Fenster 2008;Reynolds et al. 2009).
Several studies have reported that the body structure of floral visitors, especially the feeding apparatus associated with the dimensions and the morphology of the flowers, is one of the factors determining which visitors can effectively function as pollinators (Inouye 1980;Waser et al. 1996;Anderson et al. 2002;Castellanos et al. 2004;Ibanez 2012;Moré et al. 2012;Miller et al. 2014). Proboscis length is an important determinant of pollination efficiency during foraging for bumblebees (Inouye 1980;Dohzono et al. 2004;Arbulo et al. 2011). In hawkmoth-pollinated plants, floral tube length determines which species may transfer pollen; hawkmoths with tongues that are too short or too long will not pick up pollen effectively Anderson et al. 2002;Moré et al. 2012). Flower width has also been associated with pollen transfer and pollination efficiency (Galen 1989;Campbell et al. 1996).
Angadenia berteroi (Apocynaceae, Apocynoideae) is a perennial subshrub listed as threatened by Florida's Department of Consumer Services, Division of Plant Industry (Gann et al. 2002). Its tubular flowers have a complex flower structure (Fig. 1); unlike asclepioid Apocynaceae that disperse large numbers of pollen grains together in pollinia, these flowers have a mechanism to glue and disperse many pollen grains onto the mouthparts of floral visitors (Fig. 2). The anthers form a conical structure surrounding the style head, depositing their pollen on the non-receptive distal portion of the style head; the receptive stigma is at the base of the style head, and the glue-bearing area is in the middle of the style head (Galetto 1997;Lipow and Wyatt 1999;Fig. 1). This type of secondary pollen presentation suggests a somewhat specialized pollination system (Yeo 1993;Torres and Galetto 1999). Flowers in the Apocynaceae have apparently evolved to attract insects with mouthparts (or other body parts) long enough to reach the nectar (Endress 1994;Proctor et al. 1996;Darrault and Schlindwein 2005;Pinto et al. 2008). Lengths of the proboscides of the pollinators are apparently related to the lengths of the floral tubes (Proctor et al. 1996), affecting the mechanical fit between floral tube and pollinators as well as pollination efficiency (Nilsson 1988;Boberg et al. 2014).
Having observed a variety of visitors to A. berteroi, we undertook this study to determine which floral visitors are the most effective pollinators.
We address the following questions: (i) What is the diversity and abundance of animals visiting the flowers? (ii) Are all visitors equally effective at transferring pollen to conspecifics? And if not, why? We used a variety of methods to determine the effectiveness of each type of floral visitor in terms of pollen removal, simulated pollen deposition and fruit set.

Study species
Angadenia berteroi bears large, showy, yellow, tubular flowers. They have the typical complex floral arrangement of the non-asclepiad Apocynaceae . The flowers have no notable fragrance, and offer viscous nectar as a pollinator reward, with the sugar concentration of the nectar ranging from 30 to 67 % . Flowering begins in early April in South Florida and continues until late June; the flowers open early in the morning (prior to sunrise) and last ,24 h ( ). Our field observations revealed that the natural level of fruit set in A. berteroi is low: average fruit set at six sites ranged from 3.3 to 26.4 % (16.6 % on average). The species is self-incompatible relying on pollinators to set fruit . Though visitors were observed infrequently, Lepidoptera were by far the most common.

Study sites
We conducted our field studies of A. berteroi in four pine rockland fragments and one fire management unit in Everglades National Park. We chose the sites based on the presence of many individuals of the study species ).

Flower visitors
We conducted pollinator watches weekly, for 3 h per week per site (12 intervals of 15 min per day), from 9:00 am to 12:00 pm (the hours with the highest visitation rates, B. B. Roque, personal observations) during the flowering period (April -June). We haphazardly selected plants with open flowers for each of the intervals. Flower visitors were counted and some representatives of each group were captured at all sites using aerial nets. Floral visitors were divided into four groups, namely longtongued bees (Apidae and Megachilidae), short-tongued bees (Halictidae), skipper (Hesperiidae) and non-skipper butterflies (Pieridae and Nymphalidae). We identified each visitor to functional pollinator group and, where  (Table 1). Voucher specimens will be archived at the Florida State Collection of Arthropods (Gainesville, FL).
Pollen grains were collected from the insect bodies to see whether visitors carried A. berteroi and/or other pollen. The percentage of individuals in each visitor group with pollen on the proboscis was estimated. To determine the identity of pollen grains, we used our reference collection of pollen from pine rockland plants.
To quantify the efficiency of each visitor group, we estimated pollen on the visitor mouthparts as the average number of pollen grains per individual visitor of each group. Pollination watches were performed to monitor flower visitors on plants in five different sites where A. berteroi is present, to provide a representation of visitors over the natural distribution of this species. Foraging behaviour was categorized by following visitor movements after they visited A. berteroi flowers. Visitation frequency of the floral visitors was estimated by counting the number of visits of each of the visitor groups to A. berteroi flowers during the observation periods where at least one visitor was seen, and calculating the corresponding percentage of the total visits observed in those periods. (Periods where no visitors were observed to flowers were not recorded, so standard frequency was not calculated.) Time spent on a flower was clocked for each insect visitor observed during our watches.
Visitation frequency and pollen on the visitor mouthparts allowed us to rank the importance of each visitor species to the reproduction of A. berteroi (visitation frequency × pollen on the visitor mouthparts, as modified from Reynolds and Fenster 2008). We were unable to accurately calculate pollinator importance confidence intervals as we only estimated the relative frequency data (see above).
We used the Kruskal -Wallis test to detect differences among visitor groups for the average length of the visit, as the data were not normally distributed, and then used the Mann-Whitney test ( post hoc) to determine differences among the groups. We used the sequential Bonferroni correction to control type I error for all pairwise comparisons. Statistical analyses were performed using SPSS 21 (SPSS 2014).

Pollinator effectiveness
We wished to determine how the thickness of the proboscides of each of the visitor groups affects pollen transfer efficiency. We first measured the length and width of the proboscis of each captured flower visitor using a dissecting microscope (Leica MZ12 5). We then used monofilament nylon fishing line of four different diameters (4 lb, 0.20 mm diameter; 6 lb, 0.23 mm diameter; 8 lb, 0.28 mm diameter and 25 lb, 0.53 mm diameter) inserted into single flowers to simulate flower visits. The different diameters were chosen to approximate the average size of the mouthparts of the four different pollinator groups (from narrowest to widest: skippers, non-skipper butterflies, short-tongued bees and long-tongued bees). Each simulated visit consisted of inserting a single 4-cm-long piece of fishing line into the corolla tube of a fresh flower until it reached the bottom of the tube (potentially contacting the middle portion of the style head where it might contact the style's sticky secretions), followed by its careful withdrawal to prevent dislodgment of any adhering pollen grains; the number of pollen grains adhering to the line was then counted. Fifty simulated visits were performed for each line diameter. This was to test for a possible relationship between the numbers of pollen grains removed and the thickness of the fishing line.
To examine a possible relationship between line thickness and pollen deposition, we hand pollinated fresh flowers using fishing line of four different diameters. Each thickness was inserted into a fresh flower to the bottom of the corolla tube to collect pollen (as above); we then stained entire length of the fishing line with methylene blue to stain the adhering pollen grains and introduced the stained portion into another new fresh flower. Flowers thus hand-pollinated were then carefully dissected, and the length of the stigmatic surface that was stained on the second (recipient) flower was recorded. This measurement (length of the stigmatic surface stained blue) was indicative of the proportion of the Based on our observations, both long-tongued and short-tongued bees were the only pollinator groups that consistently entered the bell of the flower to insert their proboscides into the pollen chamber. To estimate how far into the corolla tube these two types of pollinators could reach, we measured the distance from the apical part of the pollen chamber to the corolla walls using a Bausch & Lomb measuring magnifier ( Fig. 1) for flowers from 30 individual plants (63 flowers total). We also measured the width of the head of the two bee groups (n ¼ 4 for short-tongued bees and n ¼ 17 for long-tongued bees) using a dissecting microscope.
To determine the most effective pollinator, we placed 15 greenhouse-grown potted plants in the field to quantify pollination success at Site 3, the site with the highest visitation frequency (B. B. Roque, personal observations). On 20 different days during the flowering period, we compared the qualitative effectiveness of the different pollinator groups by allowing a single visit to individual flowers on the potted plants. Flowers that were ready to open prior to observation periods were bagged, while in bud, to exclude visitors. At the time of observation (from 9:00 am to 12:00 pm), bags were removed and flowers exposed to foraging insects. For a specific flower, pollinator visits were restricted to a single visit by one individual from one of the four groups of pollinators. After visitation, a flower was labelled (by pollinator group) and bagged to exclude subsequent visitors. Pollination success was quantified for these same flowers by recording the fruit production of visited flowers on the potted plants, maintained in the greenhouse. We placed at least 15 plants per day, each with one to three open flowers, over 20 days of observations, using a new set of plants each day. We recorded a total of 69 visits to over 400 flowers observed in this potted plant placement experiment. Pollination efficiency of each visitor group was assessed by comparing the percentage of visited flowers that produced fruit after a single visit.
Analysis of variance was used to test for differences among visitor groups for proboscis length and width, and length of the stigmatic surface stained with methylene blue; post hoc tests were conducted using Tukey's honestly significant difference to test for differences between pairs of visitor groups. Pollen loads on the fishing line were compared using the Kruskal -Wallis test, with the Mann -Whitney test ( post hoc) to determine differences in pollen loads among fishing line sizes, as the data were not normally distributed. We also evaluated the differences between the two groups' head widths using Student's t-test. We used the sequential Bonferroni correction to control type I error for all pairwise comparisons. Statistical analyses were performed using SPSS 21 (SPSS 2014).
Pollination efficiency and visitation frequency allowed us to rank the significance of each visitor group to the reproduction of A. berteroi. Along with fruit set from flowers visited, we determine the most effective pollinator of A. berteroi.

Flower visitors
We observed a total of 153 insect visits to A. berteroi flowers. We captured 56 insect visitors at the 4 sites, belonging to 12 species in the orders Lepidoptera and Hymenoptera (Table 1). Within each of these groups, there were two subgroups: Lepidoptera were skippers (Hesperiidae) and non-skipper butterflies (Pieridae and Nymphalidae) and Hymenoptera were long-tongued (Megachilidae and Apidae) and short-tongued (Halictidae) bees. Skippers were the most frequent visitors, followed by long-tongued bees (Table 2). Lepidoptera were often observed to visit more than one flower on a plant, whereas long-tongued bees were much less likely than non-skipper butterflies and skippers to either visit consecutive A. berteroi flowers on the same plant or to return to a previously visited flower. The duration of visits differed significantly among pollinator groups (Kruskal -Wallis test, x 2 3 = 12.3, P ¼ 0.006, n ¼ 153, Table 2). Long-tongued bees were the only visitors that carried large quantities of pollen on their proboscides; 76.5 % of all captured long-tongued bees carried many pollen grains on their mouthparts (Table 3). In all cases, the number of pollen grains carried by long-tongued bees (82.9 + 69.8) was in excess of the number of ovules (61.3 + 15 ovules per flower; B. B. Roque, personal observation). Additionally, the pollen adhered to the proboscides of the visitors in aggregated clumps that were difficult to dislodge from the insect's proboscis during manipulation. The few non-skipper butterflies we found carrying pollen had proboscides with only few pollen grains (Table 3). Long-tongued bees were the most important pollinators of A. berteroi, with higher estimates of pollinator importance than non-skipper butterflies.

Pollinator effectiveness
Proboscis width and length differed significantly among visitors groups (F 3, 51 ¼ 41.11, P , 0.0001; F 3, 44 ¼ 85.3, P ¼ 0.002, respectively, Table 4). Non-skipper butterflies and skippers had the longest proboscides (Table 4). Longtongued bees were the visitor group with the widest proboscides (0.70 mm, species of long-tongued bees (Megachile georgica and Melissodes communis communis) did not differ significantly in proboscis width or length (t ¼ 1.12, df ¼ 15, P ¼ 0.26; t ¼ 21.24, df ¼ 11, P ¼ 0.24, respectively). In the visitation simulations using nylon fishing line of different diameters, pollen quantity differed significantly among the widths (Kruskal -Wallis test, P , 0.001, n ¼ 172, Fig. 3), with the widest fishing line having the greatest pollen load. There was no significant difference in the number of pollen grains extracted in simulated visits of the three other diameters, using fishing line sizes 0.20, 0.23 and 0.28 mm, which represented the proboscides of the skippers, butterflies and short-tongued bees, respectively.
In the pollen deposition simulation experiment, the length of the stigmatic surface stained with methylene blue was also influenced by the diameter of the fishing line (F 3, 51 ¼ 14.19, P , 0.0001). Size 0.53 fishing line contacted a significantly larger proportion of the stigmatic surface than the smaller diameter lines (Fig. 4).
The mean (+SD) distance from the apical portion of the pollen chamber and the corolla walls was 2.6 (+0.4) mm. The mean width of the long-tongued bees'   The diameter of 0.20 mm approximates the diameter of the proboscis of skippers, 0.23 mm represents non-skipper butterflies, 0.28 mm represents short-tongued bees and 0.53 mm represents longtongued bees. Sample size ¼ 50 in each group. To aid interpretation, the number of ovules ranges from 46 to 76 ovules per flower. Diameters with the same letters are not significantly different with Kruskal-Wallis analysis.
heads was 3.9 (+0.22) mm (Fig. 1), while that of shorttongued bees' heads (Halictidae) was 1.8 (+0.07) mm. The final test of pollinator effectiveness, whether flowers visited by the different pollinators set fruit, provided clear results. Placing potted plants in the field, observing visits, bagging, tagging and following subsequent fruit set showed that only flowers visited by long-tongued bees set fruit following a single visit. Of the 44 flowers visited by long-tongued bees, 16 (36.4 %) set fruit. None of the flowers visited by the other pollinator groups (4 by nonskippers butterflies, 19 by skippers and 2 by short-tongued bees) produced any fruit. Fruit set differed significantly among pollinator groups (Kruskal-Wallis test, x 2 3 = 111.7, P ¼ 0.009, n ¼ 69). Though the sample sizes were small for non-skipper butterflies and short-tongued bees due to their lower rate of visitation (Table 2), we also found that these groups of visitors do not carry significant amounts of pollen on their proboscides; consequently, we conclude that these two groups of visitors are much less effective pollinators of A. berteroi than are long-tongued bees.

Discussion
Pollination efficiency is a function of multiple interacting characters and behaviours, including flower shape and size as well as animal behaviour and morphology (Ollerton et al. 2007). In the pine rocklands of South Florida, A. berteroi is effectively pollinated by two native solitary bees (Megachile georgica and Melissodes communis). Though many other species visit the flowers, the ones that are the most frequent (i.e. skippers like Polites baracoa) neither carry nor deposit pollen.
Many studies of other plant -pollinator systems provide evidence that the morphological match between the pollination apparatus and the length of the proboscis is associated with pollination effectiveness (Inouye 1980;Waser et al. 1996;Castellanos et al. 2004;Moré et al. 2012;Miller et al. 2014). Our experiment with fishing lines of various thicknesses, however, suggests that the width of the proboscis of the pollinators of A. berteroi is more important than the length in determining pollen transfer efficiency.
Flowers in the Apocynaceae tend to be pollinated by long-tongued pollinators (Endress 1994), with many reports of butterflies and hawkmoths pollinating species of this family (Haber 1984;Darrault and Schlindwein 2005;Sugiura and Yamazaki 2005;Moré et al. 2007). We found, surprisingly, that skippers and non-skipper butterflies did not carry much pollen on their proboscides, nor did they deposit pollen on stigmas of A. berteroi: it appears that they were acting as nectar thieves. Both skippers and non-skipper butterflies have been described as nectar thieves in other systems (Adrienne et al. 1985). Hoc and Garcia (1999) found Lepidoptera to be nectar robbers for Phaseolus vulgaris. Castro et al. (2013) reported that for flowers of Polygala vayredae, several species of Lepidoptera behave as nectar thieves.
The complex flower morphology of A. berteroi is similar to the morphology described for other Apocynaceae . The flowers restrict access to only those visitors with mouthparts long enough to reach the nectar at the base of the floral tube. Furthermore, the sugar concentration of the nectar (30-67 %) is within the range of values reported for flowers pollinated by long-tongued bees ( 40 %; Proctor et al. 1996). Although Pascarella et al. (2001) stated that A. berteroi is visited exclusively by Lepidoptera, our field observations showed that long-tongued bees were common floral visitors, as well as skippers. Skippers were the most frequent and constant visitors, often visiting numerous flowers of the same species in a row.
We have observed that skippers and non-skipper butterflies often revisit the same flowers, while longtongued bees rarely return to a previously visited flower. Insects revisiting the same flowers could have negative consequences, as A. berteroi has a late-acting selfincompatibility mechanism . In many self-incompatible Apocynaceae, flower revisitation increases the probability that self-pollen is deposited onto the stigma, leading to ovule and fruit abortion Wyatt 1999, 2000;Wyatt et al. 2000;Wyatt and Lipow 2007). Abortion interferes with ovules in those fruits developing from cross-pollination and wastes those potential progeny (Lipow and Wyatt 1999;Lopes and Machado 1999). The principal pollinators of A. berteroi in the pine rocklands of south Florida are two native bees; bee pollination (mostly by Euglossine bees) has been previously reported for this family (Lopes and Machado 1999;de Moura et al. 2011). Even though their proboscides are slightly shorter than the length of the narrow portion of the floral tube (i.e. 6 mm, , long-tongued bees carry large quantities of pollen on their proboscides. Evidently, long-tongued bees push their mouthparts firmly against the anthers in their effort to reach the nectar at the bottom of the floral tube and thereby pick up much more pollen on the wide proboscis base than the narrow, longer mouthparts of Lepidoptera that apparently miss the reproductive parts of the flowers entirely. Many studies have highlighted a close match between the length of the flowers and the length of pollinator mouthparts of members of the family Apocynaceae (Endress 1994;Proctor et al. 1996;Lopes and Machado 1999;Darrault and Schlindwein 2005;Moré et al. 2007). In contrast, we observed no association between the lengths of the proboscis of the pollinators with pollen removal, but we did find a correspondence between the width of pollinator mouthparts and pollination efficiency.
Our results contrast with the findings of Moré et al. (2007) and de Araú jo et al. (2014), who reported that flowers of Mandevilla spp. were pollinated exclusively by pollinators with long, thin proboscides. Additionally, de Araú jo et al. (2014) reported that Agraulis vanillae and Ascia monuste are the effective pollinators of Mandevilla tenuifolia; coincidently, these two taxa were the same non-skipper butterfly visitors of A. berteroi in southern Florida. The floral morphology of A. berteroi, however, with a relatively wide and large throat (bell) and a short narrow basal tube, allows visitors with mouthparts shorter than those of butterflies to enter the flowers, reach the nectar and contact the pollen-bearing, glue-applying and receptive portions of the stigma in the process. In our study, the long-tongued bees' heads were wider than the apical portion of the pollen chamber, causing the bees to touch the reproductive parts of the flowers; short-tongued bees, with their thinner heads, entered more deeply into the corolla, but did not contact the gynostegium. The thicker fishing line used to simulate the size of the mouthparts of the long-tongued bees removed twice as much pollen as the thinner lines, and we observed a similar pattern with our pollen deposition experiments, in which the thicker fishing line touched the stigmatic surface much more extensively than thinner ones. Using a similar approach, Darrault and Schlindwein (2005) observed that proboscis width played an important role in pollen transfer efficiency in Hancornia speciosa (Apocynaceae).

Conclusions
By looking more closely into the mechanics of pollen removal and deposition, and by allowing single visits by the four groups of visitors, our results demonstrate that pollen removal ability, and single-visit fruit set data, can be more important for determining effective pollinators than simple visitation frequency. The relatively shorter mouthparts of longtongued bees appear to be more than adequate to effectively garner nectar and effectively transport pollen, despite previous assumptions to the contrary.