Temporal changes in the most effective pollinator of a bromeliad pollinated by bees and hummingbirds

Study site and species

This study was conducted in an area covered by montane Atlantic Forest, located in the Serra dos Órgãos National Park (PARNASO), Rio de Janeiro state, Brazil (22° 52′–22° 54′ S and 42° 09′–45° 06′ W, ca. 960 m a.s.l.) among four reproductive events (from 2014 to 2018). The average annual rainfall at the study site is 2,436 mm, with the rainiest period between December and March and colder and drier months from June to August. The mean annual temperature is 18.6 °C, with minimum and maximum monthly temperatures of 13.7 °C and 22.9 °C (climate data for 2015 to 2018 from the meteorological station located inside the PARNASO). The field research reported here was performed using the required permit (SISBIO Nos. 34882 and 432793).

Edmundoa lindenii (Regel) Leme (Bromeliaceae-Bromelioideae) is a terrestrial, saxicolous, or epiphyte herb, endemic to the Atlantic Forest in south and southeastern Brazil (Martinelli et al., 2008; Zappi et al., 2015). In the study area, this species blooms between December and February and produces fruits between March and April; its flowers are visited by bees and hummingbirds (R.L.B. Leal et al., 2015, personal observations).

Floral biology

We measured inflorescences of E. lindenii (n = 16 individuals) directly in the field with a measuring tap, considering the following traits: scape length, inflorescence diameter and bract length. Flowers (n = 73) from 28 individuals were collected in the field, stored in 70% alcohol and measured in the laboratory with a digital caliper considering the following structures: corolla tube length (i.e., from septal nectary to the opening of the corolla) and the width of the corolla tube opening. We counted the number of ovules in 25 flowers (n = 15 individuals).

To analyze the color quantitatively, we measured the spectral reflectance of petals, sepals and bracts. For this, 12 flowers (n = 6 individuals) were collected in the field, stored in thermal bags containing moist paper and brought to the laboratory, where they were immediately measured (Lunau et al., 2011). We measured the reflectance using an USB2000 spectrophotometer (OceanOptics, Inc., Dunedin, FL, USA) coupled with a deuterium–halogen light source (DH-2000; OceanOptics, Inc., Ostfildern, Germany), with a light emission range between 215 and 1,700 nm. We took all reflectance measurements at a 45° angle in relation to the plant structure and we used barium sulfate as the white standard and black paper as the black standard (Chittka & Kevan, 2005).

We used the logarithm version of the receptor noise-limited model (Vorobyev et al., 2001) to compare the colors of the petals, sepals and bracts. Chromatic distances were calculated according to the trichromatic formulation for bees and the tetrachromatic formulation for hummingbirds (Vorobyev & Osorio, 1998). We modeled spectral sensitivity curves using data from Sephanoides sephaniodes (Herrera et al., 2008) to estimate hummingbird color distances and from Bombus terrestris for bees (Telles & Rodríguez-Gironés, 2015). In all cases, we used standard daylight illumination (D65—Wyszecki & Stiles, 1982). Using these models, we determined the spectral location of each structure in a color space for each pollinator.

The distance between two points in a color space provides an approximation of the perceived color difference (Endler & Mielke, 2005). We evaluated color distances between sepals, petals and bracts. Using the receptor noise-limited model, we estimated that two colors were discriminable if their distance was greater than 0.27 units for bees (Telles & Rodríguez-Gironés, 2015) and 1.0 for hummingbirds (Vorobyev et al., 1998). For representation, we also calculated the color loci of the flower colors in the respective color space models: the color hexagon for bees (Chittka, 1992) and the color tetrahedron for hummingbirds (Vorobyev et al., 1998).

Nectar

We measured the nectar volume in flowers previously bagged in bud stage, with a graduated microliter syringe (Hamilton, NV, USA) and the concentration with hand-held refractometer (Bellingham + Stanley Eclipse, UK). To evaluate nectar production during anthesis, without the effect of nectar removal, 36 flowers (n = 10 individuals) previously bagged at the bud stage were measured once after anthesis onset. In total, we performed measurements at four different times of the day: 7:00 AM (n = 10 flowers); 8:30 AM (n = 10); 10:00 AM (n = 10); and 11:30 AM (n = 6). To evaluate if nectar removal stimulates its secretion, 24 flowers were submitted to four treatments (n = 6 individuals, four flowers per individual—one flower per treatment per individual): R = nectar was measured once at 11:00 AM; R1 = nectar was measured twice (at 10:00 AM and 11:30 AM); R2 = nectar was measured three times (at 8:30 AM, 10:00 AM and 11:30 AM); and R3 nectar was measured four times (at 7:00 AM, 8:30 AM, 10:00 AM and 11:30 AM). Thus, for the treatments R1, R2 and R3, nectar was remeasured on the same flowers. We calculated the total amount of sugar (mg) per flower by multiplying nectar volume (μL) by its corrected concentration (mg/μL) according to Dafni, Kevan & Husband (2005).

Breeding system and pollen limitation

We evaluated the breeding system and pollen limitation (PL) through manual pollination treatments. Floral buds of different individuals were previously bagged with “voile” bags and the flowers submitted to the following treatments: (1) spontaneous self-pollination–20 flowers from 8 individuals, were bagged and not manipulated in 2016; (2) hand self-pollination—49 flowers from 20 individuals, were supplemented manually with pollen from the same flower and bagged in 2016; (3) hand cross-pollination—130 flowers from 47 individuals, were supplemented with pollen from other individual, located at least 10 m away and then bagged. We conducted the cross-pollination treatment in the years 2016 (26 flowers), 2017 (48 flowers) and 2018 (56 flowers); (4) pollination under natural conditions—131 flowers from 36 individuals were marked and kept unbagged. We evaluated the flowers from natural pollination in 2016 (49 flowers), 2017 (20 flowers) and 2018 (62 flowers). At the end of the experiments we collected the fruits and counted the number of seeds per fruit in the laboratory. We calculated the index of pollen limitation as IPL = 1−P_n/P_c where P_n is the proportion of seed-bearing fruits multiplied by the mean number of seeds per fruit of flowers exposed to natural pollination and P_c is the proportion of seed-bearing fruits multiplied by the mean number of seeds per fruit of flowers after hand cross-pollination (adapted from Lloyd & Schoen (1992); Larson & Barrett (2000)). We used as response variable of reproductive success in all analyses such an estimate combining fruits bearing seeds and the number of seeds, which is appropriate because E. lindenii is parthenocarpic (i.e., flowers develop into fruits independent of pollination). Values of IPL ≤0.2 indicate absence of PL, whereas IPL > 0.8 indicates strong PL (Freitas, Wolowski & Sigiliano, 2010). We assessed the self-incompatibility by the index of incompatibility (ISI), a relative measure of the seeds produced after self- and cross-pollination (Zapata & Arroyo, 1978). Species with ISI < 0.30 may be classified as self-incompatible (Ramirez & Brito, 1990).

Frequency and efficacy of floral visitors

We performed focal observations (sensu Dafni, Kevan & Husband, 2005) to evaluate the identity of floral visitors and their frequency of visits, by censuses of 30 min per individual (n = 190 individuals) between 6:00 AM and 12:00 PM, totalizing 184 h of observation. Observations were done in four reproductive events, in the years 2015 (43.5 h, n = 50 individuals), 2016 (39.0 h, n = 42), 2017 (51.0 h, n = 48), and 2018 (50.5 h, n = 50). Images and videos were captured during the visits to evaluate the foraging behavior and the floral resources obtained. The visits were identified as legitimate or illegitimate by the expected mode of pollination, considering the shape and arrangement of the flower parts (sensu Irwin et al., 2010; Freitas, 2018). Specimens of insects were collected for posterior identification. We grouped the floral visitors into three functional groups based on identity, body size and foraging behavior, as following: hummingbirds, large bees (length ≥ 10 mm) and small bees (<10 mm).

We evaluated the efficacy of the three functional groups of floral visitors through experiments of selective exposition. In this experiment, flowers previously bagged at the bud stage were exposed to a single visit by any visitor of the three functional groups, as follows: small bees (48 flowers in 2016), large bees (20 flowers in 2017 and 60 in 2018) and hummingbirds (20 flowers in 2017 and 65 in 2018). Here we also considered the proportion of seed-bearing fruits multiplied by the mean number of seeds per fruit as the response variable of the pollination efficacy by each functional group.

Data analyses

We performed all the analyses in R version 3.4.4 (R Core Team, 2018). We evaluated the production of nectar during anthesis and the effect of nectar removal in nectar secretion by analyses of variance (one-way ANOVA), using the function aov. We assessed the differences between treatments (time of anthesis and number of removals) by Tukey HSD post-hoc test, using the function TukeyHSD.

We conducted a linear model to evaluate if the reproductive success after natural pollination varied between three reproductive events (2016, 2017 and 2018). Prior to analyses, we cubic-root transformed values of the response variable to meet the normality assumptions. We used the reproductive events (three levels: 2016, 2017 and 2018) as fixed effect. We established the model using the function lm and we tested the model assumptions by visual inspection of the residuals using the qqnorm function. We calculated the significance of each term in the model using the function Anova (Type II) from the car package (Kuznetsova, Brockhoff & Christensen, 2017) and the differences between levels of categorical factors using the function lsmeans from lsmeans package (Lenth, 2016).

To evaluate whether hummingbirds and large bees differ in their efficacy, we conducted a linear model. Prior to analyses, we cubic-root transformed values of the response variable to meet the normality assumptions. We used the functional group of pollinators (two levels: hummingbirds and large bees) and the year when the treatments were conducted (two levels: 2017 and 2018) as fixed effects. We established the model using the function lm and we tested the model assumptions by visual inspection of the residuals using the qqnorm function. We calculated the significance of each term in the model using the function Anova (Type II) from the car package (Kuznetsova, Brockhoff & Christensen, 2017) and the differences between levels of categorical factors using the lsmeans package (Lenth, 2016). We did not compare the efficacy of small bees as no seeds were produced after their visits.

To evaluate if the frequency of visits varies between the four reproductive events and the group of floral visitors, we conducted a log-linear model for categorical data (Wood, 2017). We used the relative frequency of visits as a response variable and the reproductive events (four levels: 2015, 2016, 2017 and 2018) and the functional groups (three levels: hummingbirds, large bees and small bees) as fixed effects. We established the model using the glm function with the family set to poisson (Lindsey, 1997). We tested the model assumptions by visual inspection of the residuals. We calculated the significance of each term in the model using the anova function (test = Chisq) and the differences between levels of categorical factors using the emmeans function available in the emmeans package.

Floral biology

The flowers of E. lindenii are grouped in a compound corymboid inflorescence with ca. 100–150 flowers, inserted in the leaf rosette (Fig. 1). Inflorescence diameter reached 121.32 ± 17.01 mm and scape length 296.87 ± 23.86 mm (mean ± SD throughout the text). The flowers are hermaphrodite, with the androecium presenting six stamens included in the corolla and anthers with longitudinal dehiscence (Fig. 1). The gynecium is also included in the corolla and the style ends in a three-lobed stigma (Fig. 1). The inferior and trilocular ovary contained 197.9 ± 54.12 ovules. The length of bracts and sepals was 55.15 ± 6.99 mm and 26.0 ± 4.0 mm, respectively. The corolla is tubular (length: 17.95 ± 2.92 mm) with a narrow opening (3.11 ± 1.17 mm). The flowers have diurnal anthesis, characterized by the presence of exposed pollen grains and receptive stigma. The anthesis began at around 06:00 AM and lasted for about 6 hours when corolla closed. The secretion of nectar started in the beginning of anthesis and it did not increase over time (F = 0.44; df = 3; p = 0.726; Fig. 2A). However, the removal of nectar stimulated new secretion (F = 6.632, df = 3, p = 0.00273, Fig. 2B).

The bracts reflect red wavelengths, whereas the corolla is UV-reflecting white and the sepals are UV-absorbing white (Fig. 3). The color of petals, sepals and bracts, as well as open or closed flowers, is distinguishable by bees and hummingbirds. Flower color was 1–7 times above the discrimination criteria (0.27) for bee vision (petals-sepals 4 ± 1, bracts-sepals 4 ± 2) and 5–15 times above the discrimination criteria (1.0) for hummingbirds (petals-sepals 8 ± 2, bracts-sepals 12 ± 4, bracts-petals 15 ± 7).

Breeding system and pollen limitation

Edmundoa lindenii is self-incompatible (ISI = 0.08; Table 1) and parthenocarpic (Table 1), requiring the action of pollinators for sexual reproduction. We observed that the reproductive success after natural pollination varied between reproductive events (ANOVA: F = 12.9, df = 2; p < 0.001). The lowest reproductive success occurred in the reproductive event of 2017 (contrasts: 2016–2017: t = 5.050; df = 103; p < 0.001; 2016–2018: t = 2.100; df = 103; p = 0.095; 2017–2018: t = −3.840; df = 103; p < 0.001; Fig. 4). Pollen limitation was expressive only in 2017 (PL index: 2016 = −0.21, 2017 = 0.70, 2018 = −0.003).

Floral visitors and temporal variation

The flowers of E. lindenii were visited by 11 species of animals belonging to three functional groups (hummingbirds, large bees and small bees; Table 2). Hummingbirds were the group with the highest species richness with seven species (Table 2; Fig. 1). The small bee T. spinipes was the only visitor that conducted illegitimate visits, resulting in damage of corolla and/or anthers by chewing. Hummingbirds and large bees foraged for nectar acting as legitimate visitors and small bees collected pollen.

The pollinator functional group affected the reproductive success of E. lindenii (Table 3) and hummingbirds had higher efficacy than large bees (contrast: t = 3.015 df = 148, p = 0.003; Table 4; Fig. 5). Small bees did not act as pollinators, as none of the flowers they visited produced seeds. The frequency of visits varied between functional groups of floral visitors with significant interaction between functional group and reproductive event (Table 4; Fig. 6).