The study was carried out at forest sites along the Atlantic Slope in Costa Rica, located within the Central Volcanic Cordillera Conservation Area (CVCCA) in Costa Rica, from January to June 2013. The vegetation in the CVCCA consists predominantly of moist and very moist tropical rainforest, with about 4000 - 6000 mm of annual rainfall [15] .

Psychotria elata (Sw.) Hammel is a distylous shrub with a height of 0.5 to 5 m; glabrous leaves; solitary terminal, inflorescences capitate to semi-hemispheric globose, surrounded by orange red basal bracts. The flowers are surrounded by about 16 mm high glabrous, sessile, white bracteoles. The flowers are visited by butterflies and hummingbirds, have sessile ovary; and blue ovoid fruits that become black at maturity. P. elata is distributed along forest edges and interiors of 30 to 1700 m elevation from Central America to Colombia. It flowers and fruits year-round with flowering peak from January to August [16] .

During January and February 2013, in a visit to the CVCCA in Costa Rica, we selected four natural P. elata populations located at four sites (Tirimbina (TIR) (10˚41'584"N and 84˚12'218"W, 171 m asl), Quebrada Gonzaléz (QUE) (10˚16'192"N and 84˚93'793"W, 538 m asl), Rara Avis (RAR) (10˚28'114"N and 84˚04'28"W, 727 m asl), and Zurqui (ZUR) (10˚04'854"N and 84˚01'456"W, 1570 m asl) (Figure 1). To estimate local population density, in each site we counted all P. elata individuals in a 1600 - 1800 length and 6 m wide plot.

For each flowering plant, we determined the number of flower morphs (thrum or pin) within each sampled plot. The relative difference between the morph frequencies (deviations morphs) was calculated by the formula (Number of short-styled plants―Number of long-styled plants/total flowering plants), as proposed by Brys et al. (2004) [17] . Consequently, the results may vary from 0 (same frequency of both morphs) to 1 (where only one

of the two morphs is present) [17] [18] . Significative differences from an expected 1:1 ratio in each population were tested by a X² test.

Voucher specimens are deposited in the Herbarium of the Universidad de Costa Rica (UCR) and the Herbarium TANG of the University of Mato Grosso, UNEMAT campus Tangará da Serra, Brazil.

In each population, in one-or two-week intervals, from January to June 2013, we recorded the number of inflorescences and of open flowers per inflorescence per plant.

To assess the variation in display of P. elata inflorescences, the capitulum width (including the lateral bracts) was measured. To assess the mean number of flower/inflorescence per morph, we dissected 5 inflorescences per plant of 30 and 22 thrum and pin morph plants, respectively. We avoided sampling of clones, from the same individual, by sampling plants at a distance of at least 3 m from each other [19] .

To estimate pollen viability, 25 thrum and 25 pin buds in pre-anthesis were collected from 10 plants per population and stored in acetic acid. The anthers of each bud were mounted on slides and stained with acetocarmine [20] .

The total number of pollen grains per flower was calculated by multiplying the total pollen counts of one anther by the number of anthers per flower (N = 5).

The herkogamy of each studied population was characterized by a sample of five flowers of up to five different plants per morph (total number of flowers = 175). Of each fully open flower we measured the corolla length (from base to apex of the lobes), stigma height (measured above the inferior ovary to the apex of the bifurcation) and anthers height (above the inferior ovary to the apex of the anther). Measurements were performed using an Olympus microscope SZX12 equipped with an ocular micrometer of 0.01 mm at 90X.

The reciprocity between anther and stigma height in the two morphs was determined by the reciprocity index (RI) sensu Richards and Koptur (1993) [21] .

where A is the mean anther position of the morph and S the mean stigma position of the alternative morph, when reciprocity is perfect RI = 0 [22] . Positive values indicate that the mean anther position is higher than the stigma position, while negative values indicate the opposite [23] .

To determine the incompatibility system and reproductive success per morph in each population, the number of seeds per fruit and number of fruits per inflorescence (40 - 50 days after treatments) was evaluated in the following treatments: i) spontaneous self-pollination (bagged flowers when in bud stage, using one to five plants per morph/population) and ii) open pollination (flowers exposed to natural pollination), applied in all populations. The flowers treated with hand and spontaneous self-pollination of both morphs of P. elata (total N = 15) were collected 24 h after the treatment, fixed in FAA and the pollen tube growth observed by fluorescence microscopy [24] [25] .

The reproductive success was estimated by counting the number of fruits per inflorescence and number of seeds per fruit of the inflorescences (N = 10), originating from natural pollination of up to five plants of each morph per population. Early development and abortion of the two ovules were observed in the fruits with no seeds.

Floral visitors were observed during the day in the months of March to May 2013 in all populations, except ZUR. Despite several attempts, the pollinators of this population could not be filmed, due to rapid and constant climate variations, posing risks to the filming equipment and safety of the researchers. In the other populations, the identification of flower visitors was performed by filming the flowers and inflorescences of both floral morphs with Sony^® digital cameras in about eight hours of footage per day. Aside from filming, additional information was obtained by direct observations.

At every visit, the collected resource and frequency of the visitor were recorded. The visitor frequency per inflorescence/morph was classified as infrequent (1 - 5 visits), frequent (5 - 10 visits), or very frequent (over 10 visits). Visitors that touched the reproductive organs (anthers and stigmas) during the visit were considered pollinators.

Test of differences in the frequency of floral morphs between populations was tested using the chi square test.

One-way analysis of variance (ANOVA) Statistica version 10 [26] was used for the variations in floral traits between populations and the Student t test to check the morphometric variations of flowers and morphs and pollen production and viability between populations. For natural pollination, when significantly different, we performed a post hoc Tukey’s HSD test. To improve the normality and homoscedasticity, data of the number of pollen grains were log-transformed before proceeding with the analysis.

In all populations flowering began in mid-February to early March 2013. Flowering of the floral morphs was synchronous, except in population QUE, where the thrum plants initiated flowering about 30 days after the pin plants. Interestingly, the thrum morph flowers in the RAR population lasted two days, while the pin flowers in the same and other populations lasted one day.

Both floral morphs of P. elata were present in the populations studied. Although no monomorphism was detected in any population, the deviations from the 1:1 ratio in the proportions of pin and thrum morphs were variable (Table 1). The highest differences between floral morph ratios were observed in the populations QUE (1Thrum:4Pin, χ² = 6.38, p = 0.01) and RAR (6.3 Thrum:1Pin; χ² = 9.40, p = 0.002). The other populations did not differ significantly from the 1:1 floral morph ratio in population TIR (χ² = 2.81, p = 0.09) and ZUR (χ² = 2.92, p = 0.09).

The inflorescence width ranged from 3.5 to 10 cm and was similar between morphs in all populations studied. The average number of flowers per inflorescence ranged from 10 to 57 and was significantly higher in thrum morphs of the populations TIR, ZUR and RAR and pin morphs of QUE (Table 1). The number of inflorescences per plant ranged from 1 to 53 and was significantly higher in thrum morphs of RAR, although the number of open flowers per inflorescence/day did not differ statistically between morphs in all populations (Table 1).

The corolla tube length was similar between morphs in the populations TIR (t = 0.186; p > 0.05); RAR (t = 0.150; p > 0.05) and ZUR (t = 0.71; p > 0.05), whereas the corolla tube length of the pin morphs in QUE was statistically different (t = 2.8; p < 0.05).

The height of stigmas and anther within the thrum and pin morphs was similar in the populations studied, except in ZUR, where the variation in the height of the sexual organ (anther and stigma) in both morphs was greater than in the other populations (Figure 2(a) and Figure 2(b)).

The reciprocity index for both morphs of P. elata ranged from 0.00 to 0.08.

The difference between the stamen (thrum) versus style heigth (pin) and stamens (pin) versus style heigth (thrum) was significantly different in TIR (F = 19.41; p = 0.006 and F = 19.41; p = 0.0009, respectively, Figure 3(a) and Figure 3(b)) and ZUR (F = 9.38; p = 0.004 and 29.17; p = 0.0001, respectively, Figure 3(a) and Figure 3(b)). There were no significant differences between the stamen and stigma height of alternative morphs in population RAR (F = 1.74; p = 0.19, and F = 3.33; p = 0.07, Figure 3(a) and Figure 3(b)). Interestingly, in population QUE the stamen height of the pin morph was perfectly equal to the stigma height of the thrum morph, i.e., herkogamy is reciprocal (F = 0.051; P = 0.82, Figure 3(b)), but the level of reciprocity was not the same between the stamen height of thrum morph and stigma height of pin morph (F = 50.28; p = 0.0001, Figure 3(a)).

In a comparison of the populations, pollen production was significantly lower in the morphs of population

+ = predominance of long-styled plants; − = predominance of short-styled plants; ^*Morphs differ significantly (p < 0.005) by the Student t test.

TIR (F = 27.70; p = 0.002) Table 2. The average pollen production per morph and population was highly variable, and pollen viability was high (>90%) in all populations and similar between morphs (Table 2).

A breakdown of incompatibility was observed in Psychotria elata in TIR thrum morphs and RAR pin morphs, confirmed by the treatment of spontaneous hand self-pollination (Table 3) and growth of the pollen tube to the stylus base in both treatments. The fruiting percentages resulting from natural pollination varied from 7.77% to 27.54% for pin morphs and from 7.82% to 33.49% for thrum morphs. The percentage of fruits produced by flower from natural pollination in population TIR was significantly greater from QUE and RAR (Table 3) and may be overestimated, since many of these seeds may have resulted from spontaneous self-pollination.

The number of fruits and seeds of each floral morph in the studied populations are listed in Table 4. The mean fruit productions per individual and per inflorescence were higher in thrum than in pin morphs in the populations TIR (mean thrum = 47.33 ± 29.73 vs. pin 30.66 ± 35.94; and thrum 7.1 ± 4.05 vs pin. 4.60 ± 5.6, respectively); ZUR (mean thrum 43.33 ± 47.72 vs. pin 29.66 ± 28.14; and thrum 9.76 ± 10.4 vs pin 5.31 ± 7.77, respectively); QUE (mean thrum 22.5 ± 7.77 vs. pin 9.5 ± 9.19; and thrum 2.23 ± 2.14 vs. pin 1.9 ± 1.44, respectively). The fruit production per individual from the thrum morphs of population RAR was five times higher than from the pin morph (mean thrum 45.00 ± 32.22 vs. pin 9.5 ± 5.5) whereas the number of fruits per inflorescence was similar between the morphs (thrum 3.48 ± 4.58 and pin 3.8 ± 1.30). One-seeded fruits had normal development, however with the presence of a single embryo. In the populations TIR and ZUR, total seed production was not higher in the morph with greater frequency (thrum and pin, respectively, Table 1). These populations, where the morph ratio was ≤0.45 (Table 1) were those that produced seedless fruits. The populations QUE and RAR, with a morph ratio of ≥0.5 (Table 1), produced the highest percentage of fruits with two seeds (Table 4).

^*Morphs differ significantly (p < 0.005) by the Student t test.

Number of flowers of spontaneous self pollination and the control were obtained from the average number of flowers per inflorescence per morph in the population, multiplied by the number of treated inflorescences/morph per population. Significant differences (p < 0.05) were indicated by letters. Same letters indicates no significant difference between populations by the Tukey test.

In the populations studied, the different guilds of visitors observed in 106 hours of footage were represented by hummingbirds the families Trochilidae, butterflies of the families Nymphalidae, Ridiononidae and bees (family Apidae) (Table 5). Chalybura urochrysia was the only hummingbird visitor in all populations.

The visiting behavior of Phaethornis striigularis differed between the populations TIR and QUE. In the first, this hummingbird served as pollinator, but in the second as nectar thief, because it stole the nectar without inserting the beak into the corolla tube. Although in different frequencies Hyposcada virginiana evanides (Nymphalidae) and Eurybia lycisca (Riodinidae) were observed in all populations, whereas Euglosa sp. (Apidae) were observed onlyin TIR and RAR (Table 5). The thrum morph was preferred by the visitors in TIR (78%) and RAR (92%). In the population QUE most visitations (76%) were recorded to the pin morph.

+++ = very frequent; ++ = frequent; + = infrequent.