Oecologia (2004) 138: 405413 DOI 10.

1007/s00442-003-1443-9

PL ANT ANI MA L IN TERACTION S

Susan Aragn . James D. Ackerman

Does flower color variation matter in deception pollinated Psychilis monensis (Orchidaceae)?

Received: 17 September 2003 / Accepted: 27 October 2003 / Published online: 10 December 2003 # Springer-Verlag 2003

Abstract Negative frequency dependent selection (FDS) had been proposed as a mechanism for the maintenance of the high levels of variability in floral traits of rewardless plants. Thus far the evidence has been equivocal for discontinuous traits. We experimentally tested the FDS hypothesis for continuous variation in flower color of Psychilis monensis, a rewardless, epiphytic orchid of Mona Island, Puerto Rico. P. monensis flowers all year long with a peak in June and July. Plants are selfincompatible. Individual flowers last 915 days if unpollinated. Over a 9-month observation of an unmanipulated population, a high percentage of plants showed pollinarium removals (79%) and fruit initiation (63%), but the actual percentage of flowers visited was very low (12%) and final fruit set was as low as the lowest recorded for orchids (2.4%). In a FDS experiment using a Latin Square design, we manipulated flower color in three populations. Over 50% of the variation in either male or female reproductive success was explained by time and site with no significant effect of treatment except as part of a three-way interaction of time site treatment. Paired comparisons with the controls gave equivocal results for both male and female measures of reproductive success. Major community changes had occurred during the FDS experiment with flower activity falling dramatically and by the third run of the experiment, only P. monensis was in flower. Coincidentally, the numbers of effective visits increased with time, presumably as pollinators became less discriminating in search of new food resources. Thus, negative frequency dependent selection is either sporadic or non-existent. Reproductive success was statistically
S. Aragn (*) . J. D. Ackerman Department of Biology, University of Puerto Rico, P.O. Box 23360 San Juan, PR 009313360, USA e-mail: saragongeo@yahoo.com Tel.: +1-508-7937525 Fax: +1-508-7938881 Present address: S. Aragn Graduate School of Geography, Clark University, Worcester, MA 01610, USA

related to locality and date, which was reflected in the flowering phenology of the local communities. High natural levels of color variation may be more influenced by drift than selection. Keywords Puerto Rico . Reproductive success . Visitation rates . Natural selection . Flowering phenology

Introduction
For plants that rely on biotic pollination, the ability to attract animals to their flowers can be an important component of fitness (Waser 1983; Stanton et al. 1986; Galen 1989). Flowers signal pollinators by different means: color, shape, size, and fragrance. Pollinators take these signals as indicators of quality or quantity of reward (Collias and Collias 1968; Waser 1983; Menzel 1985; Melndez-Ackerman et al. 1997); thus, variation on these traits may have consequences on the visitation rates and in turn, on plant reproductive success. Those floral characters that are more attractive may be strongly influenced by selection. Many studies have examined the effects of floral color on pollinator attraction (Waser and Price 1985; MelndezAckerman et al. 1997; Melndez-Ackerman and Campbell 1998; reviews in Waser 1983 and Chittka et al. 2001). Flower color has been seen as an important component of pollination syndromes (a set of characteristics that are associated with a pollinator type), which assume that certain color preferences are specific or innate to a pollinator or a guild of pollinators (Faegri and van der Pijl 1979). Current studies have shown that flower visitors have a wide range of visual capabilities and that insects in particular are not restrained by their vision to locate floral resources (Waser 1983; Chittka and Menzel 1992; Waser et al. 1996). The current view for function of color is that it serves mostly as an advertisement that pollinators learn to associate with the quality of floral reward (Chittka and Menzel 1992; Melndez-Ackerman et al. 1997; Waser and Chittka 1998) and that color preference will depend on the

406

spatio-temporal context of the species (Endler 1990; Waser et al. 1996). Deceit pollination systems provide a unique opportunity to study the role of color as an advertisement uncoupled from reward and to determine whether or not the mechanisms of selection on color in such systems differ from those where reward is offered. Variability in floral characteristics is expected to be high in deceptive plants. There is evidence of high variation in color (Nilsson 1980; Cropper and Calder 1990; Aragn and Melndez-Ackerman, unpublished data), shape and size (Ackerman and Galarza-Prez 1991), and fragrance (Moya and Ackerman 1993; A.A. Cuevas, unpublished data) in deceptive orchids, but few studies have explored the consequences of variability of floral traits on visitation rates and consequently, on reproductive success, through a spatio-temporal frame (Sabat and Ackerman 1996; Ackerman et al. 1997). High variation in floral traits in rewardless orchids has been hypothesized to be maintained by pollinator behavior. After visiting flowers with no energetic return, floral visitors will tend to fly greater distances (Dafni 1987; Dukas and Real 1993), change direction (Heinrich 1979), or visit a flower of different color (Plowright and Laverty 1984; Smithson and Macnair 1997a). This behavior would generate a negative frequency dependent selection, in which rare morphs would be favored since they will be overvisited in comparison to common colors, increasing the chance to be pollinated and thereby increasing their reproductive success (Smithson and Macnair 1997a; Gigord et al. 2001). A corollary of this hypothesis would be that a rewardless plant surrounded by conspecific plants with flowers possessing different floral traits would have a higher reproductive success than one surrounded by flowers with the same traits. We examined the ecological and potential evolutionary effects of intrapopulation variation in floral color that may influence male and female reproductive success in Psychilis monensis Sauleda, an endemic orchid of Mona Island, Puerto Rico. We concentrated on the study of floral color since this trait is probably the most reliable long distance signal that diurnal insects use to locate floral resources (Kevan 1978; Chittka and Menzel 1992; Gegear and Laverty 2001). To this end, we first examine the breeding system, phenology, floral biology and natural pollination in an unmanipulated population. Then we experimentally test the hypothesis that pollinators exert frequency dependent selection on floral color. We manipulate flower color variation by painting flowers in natural populations. We expect that populations with uniform flower color would attract fewer pollinators and have lower reproductive success than controls.

Materials and methods

Study site Mona Island is located about halfway between Puerto Rico and the Dominican Republic, at 1805N and 6753W. It has an area of about 55 km2, most of which is a plateau, approximately 40 m above sea level. There is a small coastal plain, to the south, but nearvertical cliffs bound the rest of the island. The vegetation is classified as Subtropical Dry Forest (Ewel and Whitmore 1973). The flora and fauna show affinities with dry areas on adjacent islands (Wiewandt 1979; Trejo-Torres and Ackerman 2002) and include several rare, threatened and endemic species. The mean annual temperature is 25C, and the annual rainfall is about 800 mm. Most rainfall occurs in two periods, one from May to June and the other from August to November. Winds are predominantly from the east-northeast. Periodically, the island is subjected to hurricanes; the most recent was Hurricane Georges in September 1998. The study was carried out in four sites scattered in the western side of the island. Antena, Monitoreo and Caobos populations are located on the plateau and are dominated by Bursera simaruba (L.) Sarg., Euphorbia petiolaris Sims, Reynosia uncinata Urban, Antirhea acutata (DC.) Urban, Corchorus hirsutus L., Plumeria obtusa L., and Tabebuia heterophylla (DC.) Britton. Carabineros is located on the coastal plain at the border of a mahogany plantation, and is dominated by Phyllanthus epiphyllanthus L., Coccoloba microstachya Jacq., Bourreria succulenta Jacq., Erithalis fruticosa L., and Metopium toxiferum (L.) Krug & Urban ex Urban.

Study species Psychilis monensis is epiphytic or lithophytic and produces a cylindrical pseudobulb each year, which bears 13 leaves. Each pseudobulb produces a long peduncle with a racemose inflorescence, which may produce flowers intermittently for several years by forming new racemes from the upper nodes of the scape. One to ten flowers are displayed simultaneously on the inflorescence. There is high variation in flower color, size and fragrance besides other characteristics. Like other members of the genus Psychilis, the flowers are pollinator-dependent yet offer no pollinator reward (Ackerman 1989). Pollinators are unknown but are assumed to be bees. The anterior petal or labellum of P. monensis accounts for most of the color variation in the perianth and is the most conspicuous part of the flower. Color of the labellum was recorded with a S2000 spectroradiometer (OceanOptic, Dunedin, Fla.) with illumination from a Fiber Optic illuminator (Fiber Lite Model 190, Dolan-Jenner Industries, Woburn, Mass.) In general P. monensis labella spectra reflects in the UV region (between 350400 nm), and reflected strongly at short wavelengths (reflectance peak around 460 nm, human blue) and long wavelengths (steep increase in reflectance close to 600 nm that plateaus around 600650 nm, human red) but had little reflectance in comparison at intermediate wavelengths (a sharp depression around 550600 nm, human green and yellow). Flower color variation is continuous rather than separated in distinct color morphs, thus flower color varies from pale pink to deep purple (Fig. 1). A detailed study of the labellum color variation will be published elsewhere (Aragn and Melndez-Ackerman, unpublished data), but details are available in Aragn (2002).

Breeding system To determine the breeding system and the importance of insects for pollen transfer, we randomly selected 21 plants at the Antena site and bagged them before they began to flower. The bags were constructed with fine-mesh veil around a cylindrical wire frame and tied to a supporting stick. To confirm whether or not P. monensis is

407 Carabinero were chosen because they had similar density and visitation regimes (based on a pilot study). At each site, two adjacent areas (replicates) of approximately 50 m long and from 7 to 10 m wide were sampled and all plants with active inflorescences were tagged. Inflorescences that flowered later were added to replace those that ceased to flower and to keep the population treatments uniform at each site. The three most recently opened flowers from all tagged plants were assigned to one of the following treatments: Treatment 1 (PAINTED) consisted of a monochromatic population where all the variability in color was eliminated by painting the mid lobe of the labellum of all flowers with water based lilac acrylic craft paint (a combination of Lilac 3215, White 3218, and Transparent medium, in the following proportions 4:3:2, Apple Barrel Colors, Plaid Enterprises, Ga.) This paint was chosen because it causes no withering in the flowers and color choice was based on the previous observations of high pollination rate of flowers of lilac color, in this way we assured that the difference that we see were due to the frequency differences and not for the color choice. Treatment 2 (TRANSPARENT) consisted of painting the mid lobe of the labellum with a transparent water based acrylic medium (Apple Barrel Colors, Plaid Enterprises, Ga.), to detect any influence of the paint on insect visits. Treatment 3 (CONTROL) the population was left unmanipulated, therefore maintaining a polychromatic population. All plants were allowed to receive natural pollinator visits for 46 days, and then checked for pollinarium removals and depositions. This was repeated three times. The starting dates for the experimental runs were on 11 June, 11 July, and 4 August 1999 and at each run the color treatments were rotated among sites such that by the end of the experiment each site had experienced all treatments. To standardize floral displays among treatments, older flowers and the two forthcoming buds were removed. The experimental design was a replicated 33 Latin Square Design with sites as columns, times as rows with two replicates at each site. We tested the hypothesis that pollinators exert negative frequency dependent selection by over-visiting the rare phenotypes and avoiding the common ones; therefore, populations with high variation in labellum color were expected to receive more visits than populations that are uniformly colored. Flowers in the monochromatic population (lilac PAINTED, treatment 1) were expected to experience less pollinarium removal (male reproductive success) and fewer pollinia depositions (female reproductive success) than flowers in the polychromatic population (CONTROL, treatment 3). The polychromatic (CONTROL) population (treatment 3) and the paint-control (TRANSPARENT painted) population (treatment 2) were expected to show similar values of reproductive success. Although fitness differences are an individual phenomenon, we chose to test the hypothesis of FDS (or the likelihood of FDS) indirectly by examining the cumulative effects expressed at the population level for several reasons. First, we knew from studies of related species (Ackerman 1989) and our preliminary observations that pollinator visits were expected to be infrequent requiring very large sample sizes. Secondly, we were interested in the mechanisms of maintenance of high levels of color variation, a population phenomenon, and the test requires experimental removal of such variation. Lastly, color variation is continuous and the determination of the frequency for each phenotype would require spectroradiometric assessment of such a large number of flowers it would have been impractical to do. Because we chose to assess the effects of different color variation regimes at the population level, we likely lost power to detect FDS. Our results may thus be viewed as conservative. We analyzed the ratio of pollinarium removals over the number of flowers per plant as a measure of male reproductive success, and pollinarium depositions over the number of flowers per plant for female reproductive success. These variables were arcsine transformed to normalize the data and analyzed with a Latin square ANOVA using the SAS computer program (SAS 1990). We followed the procedure outlined by Steel and Torrie (1988, p. 219) using replicates to increase the degrees of freedom of the error in a 33 Latin square design. The Latin square design is an elegant and powerful technique to test effects with the minimum number of replicates; however, it requires the assumption of an absence of

Fig. 1 The reflectance spectra of Psychilis monensis flowers (130 unmanipulated, 5 painted with lilac acrylic and 5 painted with the transparent medium) are given as a function of the wavelength. Relative reflectance range (minimum and maximum relative reflectance values) from flowers in different locations is given to show the high variability encountered in this orchid (thick lines). Relative reflectance spectra were recorded using a S2000 spectroradiometer, (OceanOptic, Dunedin, Fla.). Reflectance measurements were taken from a 2 mm diameter area at the center of the labellum midlobe, within a 400 750 nm spectral range, and expressed as relative values to a white standard (Labsphere, North Sutton, N.H.). Continuous lines represent the spectra from lilac painted flowers. Dashed lines represent spectra from flowers painted with transparent medium autogamous, six plants were bagged without any additional manipulation. To assess whether or not the species is selfcompatible, we self-pollinated 25 flowers on seven plants. We also cross-pollinated 23 flowers on the remaining eight plants with pollinaria gathered from donor plants at least 5 m apart. Each pollination involved all four pollinia of the pollinarium which is what normally occurs naturally. After 2 months fruits were censused.

Floral biology and natural visitation. Two hundred sixty plants were tagged in early May 1999 and then another 73 plants were added in June. Monthly censuses of flower and fruit production were taken from June 1999 to February 2000, except for October and January. In each census the number of pollinaria removed and deposited was assessed to estimate natural levels of reproductive success and the natural history of flowering of this species. Total flower production was estimated by multiplying the number of flowers counted in each census by 2.5 or the number of 12-day-periods between censuses (this is the average flower lifespan) and then summing up the census subtotals. Fruit set percentage (a measure of female reproductive success) was calculated as: (total fruit production / total estimated flowers per plant) 100. The percentage of flowers with their pollinarium removed (a measure of male reproductive success) was calculated as (total number of pollinarium removed 2.5 / total estimated flowers per plant) 100. Multiplying by 2.5 extrapolates the census data to a monthly estimate.

Frequency dependent selection experiment To determine the effect of flower color on pollinator visitation and plant reproductive success, we created populations with different color variability regimes. The sites of Antena, Caobos and

408 interactions. We did get an interaction so our results should be taken with caution. We further reanalyzed the data by partitioning it in two subsets (painted vs control and control vs transparent paint) for each experimental run. We tested the hypotheses that (a) a plant in the lilac painted population should have less reproductive success than one in the control population, and (b) the reproductive success of plants in the control and transparent painted populations should be similar. We used a 2-sample Wilcoxon nonparametric test, with normal approximation performed in JMP v. 4 (SAS 2000). To assess the diversity and availability of other floral resources around the experimental sites, ten transects of 101 m were established randomly within each site at the beginning of each experimental run. We recorded the number of species in flower at each time. The Shannon-Wiener diversity index (Brower et al. 1997) was calculated based on the total number of species found flowering in each site. Plant voucher specimens are deposited in the herbarium of the University of Puerto Rico, Rio Piedras Campus (UPRRP). Table 1 Morphological, reproductive and ecological characteristics for Psychilis monensis inflorescences in 1999 (n =333). Estimated total no. of flowers (ETF) was calculated for each plant = [number of flowers censused / (days in the intercensus period / mean flower lifespan)], No. of visits was calculated using Removals + Fruits, % of fruit set (fruits/ETF/plant)100, % removal per flower (removals/ ETF/plant)100 Character No. of flowers censused Estimated total no. of flowers (ETF) No. of visits No. of pollinarium removals No. of pollinarium depositions % of fruit set % removal per flower No. of fruits Floral display Mean 11.3 25.98 2.7 1.2 0.2 6.7% 9.6% 1.58 2.6 SD 7.1 16.54 2.6 1.4 0.5 8.7% 11.0% 1.91 1.4 Range 042 094 012 07 03 050% 067% 014 110

Results
Breeding system Psychilis monensis is self-incompatible and completely dependent on pollinators for pollination. None of the six bagged and otherwise unmanipulated plants set fruit. Selfpollinated flowers initiated fruits (7 plants, 25 flowers) but either aborted in the first month or developed empty capsules. The cross-pollinated plants set fruits (8 plants, 23 flowers, 21 fruits, fruit set / plant =88.3%, SD=21.3%). Floral biology and natural visitation Psychilis monensis bloomed throughout the observation period but peaked in June and July (Table 2). The lowest flowering activity occurred from December to February. During this time, the lower activity was in December when 41 of the 333 marked plants had active inflorescences and these produced 112 flowers (2.7 flowers / plant). The greater levels of flowering occurred in July when 270 plants presented active inflorescences producing a total of 1,112 flowers (4.1 flowers / plant) (Table 2). P. monensis plants produced an average of 2617 flowers (range =1 94) in 9 months with an average display of 2.6 flowers (Table 1). Flowers lasted from 9 to 15 days if they were
Table 2 Visitation and fruiting failure in Psychilis monensis. Unmanipulated population (n =333 plants). Poll. pollinia, N Number of plants with active inflorescences,% failure percentage of plants Month censused June July August September November December February
a

not pollinated. Once pollinated, the perianth withers in 2 3 days, and an incipient swelling of the ovary is already noted by the third day. Fruits develop in 2131 days until full maturity and are retained up to 5 months on the plant, releasing minute seeds through longitudinal slits in the capsule. Visitation frequencies Approximately 2 pollinarium-removal events occurred per pollination. This result is expected if each pollinator effectively visited 2 flowers in a given foraging bout with the first visit a removal event and the second a removal and pollination event. However, census data revealed that 2.6 pollinations occurred for each fruit matured (fruit initiated / fruits dehisced). Thus, for every fruit matured, approximately 5.2 pollinarium removals occurred (2 removals 2.6 pollinations). In an intercensus period (3058 days), an inflorescence had a 50% chance of being visited; this probability increased to 79% after 245 days (duration of the monitoring), with only 71 of 333 plants failing to receive any visit over this period of time. The number of plants
that fail to have any visits (nor pollinarium removal neither pollinia deposition), Pollination efficiency pollinia depositions /pollinaria removals

N 259 270 233 195 138 41 74

Flowers Pollinia % plants with Pollinia % plants with Fruits % plants % Failure Pollination removals poll. removalsa depositions poll. depositionsa with fruits efficiency 942 1,112 538 466 429 112 179 11 189 104 55 23 10 47 3.4 43.0 33.5 24.1 10.9 9.8 40.5 1 26 25 8 6 6 20 0.4 7.4 8.2 2.1 0.7 7.3 16.0 56 35 211 121 45 19 37 13.1 10.7 54.5 36.4 23.2 31.7 33.3 83.8 51.5 29.6 45.6 68.8 58.5 41.9 0.09 0.14 0.24 0.14 0.26 0.60 0.43

Some plants have more than one pollinarium removal, pollinia deposition or fruit

409

with no visits at all varied greatly among months. The highest failures were in June and November. The month in which the most plants had pollinaria removed was July (43%) and that corresponded with the highest proportion of plants with new fruits 1 month later (54.5%, Table 2). The highest percentage of plants with pollinarium depositions occurred in February, which was accompanied by a high removal percentage, even though the absolute numbers were very low (Table 2). Approximately 20% of the plants in an inter-census period had more than one visit, and this increased to 52% over the 9 months of monitoring.

Fruit set Fruits were initiated in 209 plants (63%) but only 95 matured fruits (28.5%). Approximately 6% of the estimated number of flowers initiated fruits, and final fruit set was only 2.4%. Plants successfully matured 39% of the initiated fruits. The cause of fruit loss usually could not be discerned since most fruits disappeared between censuses. However, for some there was evidence of fruit predation (3.6%) or fruit abortion (15.5%). Frequency dependent selection experiment Both transparent- and lilac-painted flowers fell within the range of natural color variation in P. monensis. However, both also exhibited less variation than unpainted flowers
Table 3 Species diversity and number of flowering species at the experimental sites. Shannon-Wiener diversity index (H) was calculated based in ten transects of 101 m on each site Species diversity (H) Number of flowering species Time 1 Antena a Caobos b Carabinero
a

Time 2 2 3 0

Time 3 0 0 0

2.34 2.88 1.66

15 16 11

Fig. 2 Percentage pollinarium removals per flower per plant. For each experimental run (Time 1, 2, 3) and treatment (Paint all flowers painted with a water-based lilac paint; Transparent = all flowers painted with the transparent paint medium; Control flowers left unmanipulated) (Boxes represent average and lines range)

Species flowering at Antena site: Time 1: (1) Croton lucidus L., (2) Croton betulinus Vahl, (3) Croton discolor Willd., (4)Melochia pyramidata L., (5) Melochia tomentosa L., (6) Exostema caribaeum (Jacq.) Schult., (7) Randia aculeata L., (8) Antirhea acutata (DC.) Urban, (9) Jacquemontia pentanthos (Jacq.) G. Don. (10) Tabebuia heterophylla (DC.) Britton, (11) Comocladia dodonea (L.) Urban, (12)Plumeria obtusa L., (13)Corchorus hirsutus L., (14) Lantana involucrata L., (15) Centrosema virginianum (L.) Benth. Time 2: (2), (4) b Species flowering at Caobos site: Time 1: (1), (2), (3), (4), (6), (7), (10), (11), (12), (13), (14), (15), (16)Stigmaphyllon emarginatum (Cav.) A. Juss., (17) Pentalinon luteum (L.) Hansen Wunderlin, (18) Erithalis fruticosa L., (19) Galactia dubia DC. Time 2: (2), (4), (13) c Species flowering at Carabinero site: Time 1: (1), (2), (6), (7), (13), (14), (20) Phyllanthus epiphyllantus L., (21) Coccoloba diversifolia Jacq., (22) Swietenia macrophylla King, (23) Thrinax morrisii H. Wendl., (24) Urena lobata L.

(Fig. 1). Transparent-painted flowers in general had higher reflectance than lilac-painted flowers, particularly at wavelengths above 500 nm. Thus, the color in the lilacpainted flowers was more saturated as was expected, but the transparent paint treatment did not entirely mimic the reflectance spectra of unmanipulated flowers as we had hoped. The site with the highest diversity of species in flower was Caobos with 16 species (Table 3). Over the course of the study the number of flowering species in the community decreased rapidly. For the second experimental run less than one-fifth of the species originally flowering remained in flower, and for the last run no species but P. monensis had flowers. The number of plants with active inflorescences used in each experimental run varied from 289 to 389, with total

410

Fig. 3 Percentage pollinarium deposition per flower per plant (from one to four pollinia deposited). Abbreviations as in Fig. 2

flower production of approximately 2,000 flowers for each population. We used an average of 34649 plants in Antena, 31033 plants, in Caobos, and 32635 plants in Carabinero per experimental run. The number of pollinarium removals were higher at time 2 and 3 (July and August) at all sites (Fig. 2) and the
Table 4 Latin Square ANOVA for male reproductive success. The analysis is based on two replicates of 3 times 3 treatments 3 sites design (Steel and Torrie1988). Data were previously arcsine (X +1) transformed to attain normality

Antena site showed the highest number of removals. The Latin square ANOVA model on male RS (pollinarium removals / flower / plant) accounted for 93% of the variation observed. Significant variation was accounted for by time (52%) and site (25%), but not by treatment (Table 4). The analysis showed a significant treatment time site interaction. Lilac-painted populations of Antena and Caobos received fewer removals in the first and second run relative to the control but the difference was significant only at time 1 (Wilcoxon test, time 1: z =0.26, P =0.79; time 2: z =6.16, P <0.001). During the third run the painted population received the highest number of visits although the difference was again not significant (Wilcoxon test: z =0.31, P =0.75). The transparent-painted populations at time 1 and 2 had significantly more pollinarium removals than either the controls, but at time 3 it had significantly fewer (Wilcoxon test, time 1: z =2.40, P =0.01; time 2: z =3.02, P =0.003; time 3: z =6.69, P <0.001). More pollinia depositions occurred at time 2 and 3 (July and August) at Antena and Caobos (Fig. 3). The Latin square ANOVA model accounted for 95% of the observed variation for female reproductive success with significant amount of the variance resulted from both time and site (31% each) (Table 5). Color treatment had no significant effect on pollinations (Table 5), but just as in pollinarium removals, there was a significant treatment time site interaction. The lilac-PAINTED populations had a lower average female RS at time 1 and 2 than the CONTROL and TRANSPARENT populations as predicted in a FDS scenario, but the difference was significant only at time 2 (Wilcoxon test, time 1: z =0.79, P =0.42; time 2: z =3.82, P <0.001). At time 3, the lilac-PAINTED population had the highest female RS, but again, the difference was not significant relative to the controls (Wilcoxon test, time 3: z =0.31, P =0.75). The TRANSPARENT-painted populations received relatively more pollinations than the controls at times 1 and 2, but the differences were significant only at time 2 (Wilcoxon test; time 1: z =1.46 P =0.14; time 2: z =5.04 P <0.001). At time 3, the TRANSPARENT-painted population had significantly less female RS than the controls (Wilcoxon test; time 3: z =5.15, P <0.001).

Male RS (pollinarium removals/flowers/plant) Sources Replicate Site Time Treatment Replicatetime Replicate treatment Timesitetreatment Error Total SS 0.0149 0.1153 0.2363 0.0074 0.0054 0.0044 0.0554 0.0148 0.4541 df 1 2 2 2 2 2 2 4 17 MS 0.0149 0.0576 0.1182 0.0037 0.0027 0.0022 0.0277 0.0037 F 4.03 15.58 31.94 1.00 0.74 0.60 7.50 P 0.12 0.01** 0.004** 0.44 0.53 0.59 0.04*

411 Table 5 Latin Square ANOVA for female reproductive success. Analysis is based on two replicates of 3 times 3 treatments 3 sites design (Steel and Torrie1988). Data were previously arcsine (X +1) transformed to attain normality Female RS (pollinarium depositions / flowers / plant) Sources Replicates Site Time Treatment Replicate time Replicate treatment Time site treatment Error Total SS 0.0012 0.0268 0.0270 0.0021 0.0013 0.0003 0.0229 0.0045 0.0860 df 1 2 2 2 2 2 2 4 17 MS 0.0012 0.0134 0.0135 0.0010 0.0006 0.0001 0.0115 0.0011 F 1.07 11.80 11.89 0.92 0.57 0.11 10.11 5.53 P 0.35 0.02* 0.02* 0.46 0.60 0.89 0.02*

Discussion
Breeding system Psychilis monensis is self-incompatible yet self-pollinated flowers did not immediately abort. They initiated fruits and even developed into seedless capsules. This phenomenon was also shown in P. krugii even though there was a demonstrable cost to capsule formation (Ackerman 1989). One might expect strong selection and an evolutionary response against such wasteful investment of resources but only if geitonogamous pollinations are sufficiently common. The frequency of pollinations in P. monensis was low (6.1% of the total flowers) but plants with multiple fruits generally had more abortions. Perhaps prolonged flowering and small display sizes, characteristic of P. monensis, is an evolutionary response to unproductive geitonogamous pollinations. Floral biology and natural visitation The phenology of plants in seasonally dry forests is highly dependent on their capacity to obtain water (i.e., root length; Van Schaik et al. 1993). In general, plants of tropical dry forests tend to flower in months with higher irradiance (sunniest and drier months) unless they are prevented by water stress (Wright 1991). P. monensis does indeed have a flowering peak in the drier months, but also produces flowers throughout the year. As in many plants, the temporal pattern of flowering in P. monensis may be explained as the composite effect of abiotic influences (mainly irradiance and water stress) and biotic interactions (pollination, seed predation and herbivory) (Van Schaik et al. 1993). In addition to its relationship to geitonogamy discussed above, the continuous production of flowers by P. monensis and the long lifespan of the individual flowers extends the period during which visitation may occur, an important strategy if pollinators or resources are unpredictable (Cole and Firmage 1984; Montalvo and Ackerman 1987; Ackerman 1989), both of which exist on Mona Island.

Visitation Orchid flowers usually receive few visits. Approximately 12% of the flowers of P. monensis had pollinarium removals, which resembles the percentage of removals in a closely related species P. krugii (Ackerman 1989). Other tropical orchid species with deception pollination systems have removal rates that range from 3 to 61% (Nilsson et al. 1986; Ackerman 1989; Zimmerman and Aide 1989; Tremblay et al. 1998; Salguero-Faria and Ackerman 1999) making P. monensis one of the species most severely visitation-limited on a per-flower basis. Nonetheless, a given plant has a high probability of receiving at least one visit (79%). This is due in part to the long lifespan of inflorescences, which may produce flowers continuously for 6 months and sporadically for several years. Pollination efficiency (pollinations per pollinarium removed) was highly variable (0.090.60) (Table 2), with an average of 0.27, which is intermediate among tropical orchid species (0.161.00) (Ackerman et al. 1997 and references therein). Low pollination efficiency ratios are thought to drive selection for acquiring mates (Nilsson et al. 1992), but the variable values for P. monensis and other species for which there are estimates for time intervals or different populations indicate that selection is not constant in time and space (i.e., Ackerman 1989; Ackerman et al. 1994, 1997). Fruit set Zimmerman and Aide (1989) noted that one source of variation within and among populations of epiphytic orchids is a random element produced by biparental seed crops, low fruit set, and few reproductively successful individuals. P. monensis seed crops are indeed biparental and fruit set is low. However, over half the plants were effectively visited (removals and/or depositions) over a 9month period, and the number of plants that matured fruit was nearly 30%. Plants of P. monensis are relatively longlived so the probability that a plant would produce a fruit over its lifetime may be high, which would weaken the Zimmerman and Aide argument.

412

Frequency dependent selection experiment If pollinators visit flowers according to their frequency distribution within a population, then they will exert a positive frequency dependent selection in which common phenotypes will be selected, leading to stabilizing selection of floral traits. Laboratory experiments have shown that bumblebees visit the more common phenotypes in a rewarding scenario (Real 1990; Smithson and Macnair 1996, 1997b; reviewed in Smithson 2001). Under field conditions the pattern is less clear; only some of the experiments support a frequency dependent selection behavior by pollinators (Mogford 1978; Epperson and Clegg 1987; reviewed in Smithson 2001). In the case of rewardless flowers the pattern is reversed; bumblebees switch from common morph preference when flowers are rewarding, to a rare morph preference when flowers are deceptive (Smithson and MacNair 1997a). This type of negative frequency dependent selection could cause the maintenance of polymorphism within populations and may maintain the extraordinary floral variability in rewardless orchids (Heinrich 1975; Ackerman 1981, 1986; Ackerman and Galarza-Prez 1991; Smithson and MacNair 1997b; Girgord et al. 2001). Field experiments with the epiphytic orchid Tolumnia variegata failed to detect negative frequency dependent selection on variation in floral fragrance production (Ackerman et al. 1997); however, working with fragrances could pose a problem, since fragrances are a more complex advertisement than color, being more variable in time and space (Kunze and Gumbert 2001). On the other hand, Smithson (2001) reported that for Dactylorhiza sambucina a decrease in reproductive success occurs as the frequency of the rare color morph increases, and Girgord et al. (2001) in field experiments with the same orchid showed that rare-color morphs have reproductive advantage through male and female RS, demonstrating for first time that negative FDS through pollinator preference for rare morphs can cause the maintenance of a flower-color polymorphism. Our system differs from others that have examined FDS in that color variation in P. monensis is continuous forcing us to create patches of flowers with reduced levels of variation and then monitoring the effects at the population level. We had expected that when flowers look more similar to one another, pollinator visitation rates would drop. However, our Latin square design revealed that most of the variation in both male and female reproductive success was attributed to time and site with no significant effect of treatment. Furthermore, there was a significant treatment time site interaction. Over the duration of the experiment, flowering activity of the plant community dropped radically until by the third phase of the experiment when the only species left bearing flowers was P. monensis and flower production of these was about half of the two previous runs. Such phenological dynamics would certainly affect competition for pollination. The highest pollinator activity for P. monensis was actually during that third phase of the experiment, perhaps when pollinators were desperately seeking nectar

sources. Plant community composition at Carabineros was somewhat different from that of the other two sites used in the experiment and may have further exacerbated both time and site effects. The paired comparisons among treatments gave equivocal results for both removals and depositions. One simply cannot make generalizations from our results other than to emphasize that variability in reproductive success among treatments and control is strongly dependent on time and location. Negative frequency dependent selection, if operational at all, may have occurred only during the second run of the census when flowering activity in the plant community was at its peak. Flower color variation in populations of P. monensis was either not influential to pollinators or selection operates spasmodically. Either scenario may lead to higher than normal levels of flower color variation.
Acknowledgements The authors thank the Puerto Rico Department of Natural Resources and Environment for the transportation to and logistical support in Mona Island Reserve; C. Trejo, T. Otero, E. Lasso, A. Manrique and M. Morales for field assistance; and E. Melndez-Ackerman, M. Aide, J. Zimmerman, C. Trejo and two anonymous reviewers for comments on the manuscript.

References
Ackerman JD (1981) Pollination of Calypso bulbosa var. occidentalis (Orchidaceae): a food deception system. Madroo 28:101 110 Ackerman JD (1986) Mechanisms and evolution of food-deceptive pollination systems in orchids. Lindleyana 1:108113 Ackerman JD (1989) Limitations to sexual reproduction and the evolution of deception pollination in Encyclia krugii (Orchidaceae). Syst Bot 14:101109 Ackerman JD, Galarza-Prez M (1991) Patterns and maintenance of extraordinary variation in the Caribbean orchid, Tolumnia (Oncidium) variegata. Syst Bot 16:182194 Ackerman JD, Rodrguez-Robles JA, Melndez EJ (1994) A meager nectar offering by an epiphytic orchid is better than nothing. Biotropica 26:4449 Ackerman JD, Melndez-Ackerman EJ, Salguero-Faria J (1997) Variation in pollinator abundance and selection on fragrance phenotypes in an epiphytic orchid. Am J Bot 84:13831390 Aragn S (2002) Variation and frequency dependent selection on flower color in Psychilis monensis (Orchidaceae). Master s Thesis. University of Puerto Rico, Rio Piedras Campus, Puerto Rico Brower JE, Zar JH, von Ende CN (1997) Field and laboratory methods for general ecology. 4th edn. WCB/McGraw-Hill, Boston, Massachusetts, pp 180182 Chittka L, Menzel R (1992) The evolutionary adaptation of flower colours and the insect pollinators colour vision. J Comp Physiol 171:171181 Chittka L, Spaethe J, Schmidt A, Hickelsberger A (2001) Adaptation, constraint, and chance in the evolution of flower color and pollinator color vision. In: Chittka L Thomson JD (eds) Cognitive ecology of pollination. Cambridge University Press, Cambridge, pp 106126 Cole FR, Firmage DH (1984) The floral ecology of Platanthera blephariglottis. Am J Bot 71:700710 Collias NE, Collias EC (1968) Annas hummingbird trained to select different colors in feeding. Condor 70:273275

413 Cropper SC, Calder DM (1990) The floral biology of Thelymitra epipactoides (Orchidaceae), and the implications of pollination by deceit on the survival of this rare orchid. Plant Syst Evol 170:1127 Dafni A (1987) Pollination in Orchis and related genera: evolution from reward to deception. In: Arditti J (ed) Orchid biology: Rev Perspect 4:79104 Dukas R, Real LA (1993) Learning constrains and floral choice behavior in bumble bees. Anim Behav 19:566574 Endler JA (1990) On the measurement and classification of colour in studies of animal colour patterns. Biol J Linn Soc 41:315352 Epperson BK, Clegg MT (1987) Frequency-dependent variation for outcrossing rate among flower-color morphs of Ipomoea purpurea. Evolution 41:13021311 Ewel JJ, Whitmore JL (1973) The ecological life zones of Puerto Rico and the U.S. Virgin Islands. U.S. Forest Service Research Paper ITF-18. Institute of Tropical Forestry, Puerto Rico Faegri K, van der Pijl L (1979) The principles of pollination ecology, 3rd edn. Pergamon, Oxford, England Galen C (1989) Measuring pollinator-mediated selection on morphometric traits: bumblebees and the alpine sky pilot, Polemonium viscosum. Evolution 43:882890 Gegear RJ, Laverty TM (2001) The effect of variation among floral traits on the flower constancy of pollinators. In: Chittka L, Thomson JD (eds) Cognitive ecology of pollination. Cambridge University Press, Cambridge, pp 83105 Gigord LDB, Macnair M, Smithson A (2001) Negative frequencydependent selection maintains a dramatic flower color polymorphism in the rewardless orchid Dactylorhyza sambucina (L.) So. Proc Natl Acad Sci USA 98:62536255 Heinrich B (1975) Bee flowers: a hypothesis on flower variety and blooming times. Evolution 29:325344 Heinrich B (1979) Resource heterogeneity and patterns of movement in foraging bumblebees. Oecologia 40:235245 Kevan PG (1978) Floral coloration, its colorimetric analysis and significance in anthecology. In: Richards AJ (ed) The pollination of flowers by insects. Academic Press, New York, pp 51 78 Kunze J, Gumbert A (2001) The combined effect of color and odor on flower choice behavior of bumble bees in flower mimicry systems. Behav Ecol 12:447456 Melndez-Ackerman EJ, Campbell DR (1998) Adaptive significance of flower color and inter-trait correlations in an Ipomopsi s hybrid zone. Evolution 52:12931303 Melndez-Ackerman EJ, Campbell DR, Waser NM (1997) Hummingbird behavior and mechanisms of selection on flower color in Ipomopsis. Ecology 78:253241 Menzel R (1985) Learning in honey bees in an ecological and behavioral context. In: Hlldobler B, Lindauer M (eds) Experimental behavioral ecology. Gustav Fisher, Stuttgart, Germany, pp 5574 Mogford DJ (1978) Pollination and flower colour polymorphism, with special reference to Cirsium palustre. In: Richards AJ (ed) The pollination of flowers by insects. Academic Press, London, England, pp 191199 Montalvo AM, Ackerman JD (1987) Limitations to fruit production in Ionopsis utriculariodes (Orchidaceae). Biotropica 19:2431 Moya S, Ackerman JD (1993) Variation in the floral fragrance of Epidendrum ciliare (Orchidaceae). Nord J Bot 13:4147 Nilsson LA (1980) The pollination ecology of Dactylorhiza sambucina (Orchidaceae). Bot Not 133:367385 Nilsson LA, Johnson L, Rason L, Randrianjohany E (1986) The pollination of Cymbidiella flabellata (Orchidaceae) in Madagascar: a system operated by sphecid wasps. Nord J Bot 6:411 422 Nilsson LA, Rabakonandrianina E, Razananaivo R, Randriamanindry JJ (1992) Long pollinia on eyes: hawk-moth pollination of Cynorkis uniflora Lindley (Orchidaceae) in Madagascar. Bot J Linn Soc 109:145160 Plowright RC, Laverty TM (1984) The ecology and sociobiology of bumblebees. Annu Rev Entomol 29:175199 Real LA (1990) Predator switching and the interpretation of animal choice behavior: the case for constrained optimisation. In: Hughes RN (ed) Behavioral mechanisms of food selection. Springer, Berlin Heidelberg New York, pp 121 Sabat A, Ackerman JD (1996) Fruit set in a deceptive orchid: the effect of flowering phenology, display size, and local floral abundance. Am J Bot 83:118186 Salguero-Faria JA, Ackerman JD (1999) A nectar reward: is more better? Biotropica 31:303311 SAS / JMP (2000) User s Guide Version 4. SAS, Cary, N.C. SAS / STAT (1990) User s Guide, Version 6, 4th edn. SAS, Cary, N. C. Smithson A (2001) Pollinator preference, frequency dependence, and floral evolution. In: Chittka L, Thomson JD (eds) Cognitive ecology of pollination. Cambridge University Press, Cambridge, England, pp 237258 Smithson A, Macnair MR (1996) Frequency-dependent selection by pollinators: mechanisms and consequences with regard to behavior of bumblebees Bombus terrestris (L.) (Hymenoptera: Apidae). J Evol Biol 9:715723 Smithson A, Macnair MR (1997a) Negative frequency-dependent selection by pollinators on artificial flowers without rewards. Evolution 51:715723 Smithson A, Macnair MR (1997b) Density-dependent and frequency-dependent selection by bumblebees Bombus terrestris (L.) (Hymenoptera: Apidae). Biol J Linn Soc 60:401417 Stanton ML, Snow AA, Handel SN (1986) Floral evolution : attractiveness to pollinators increases male fitness. Science 232:162527 Steel RGD, Torrie JH (1988) Bioestadstica: principios y procedimientos. McGraw Hill Interamericana, Mexico, pp 219220 Trejo-Torres JC, JD Ackerman (2002) Composition patterns of Caribbean limestone forests: are parsimony, classification and ordination analyses congruent? Biotropica 34:502515 Tremblay RL, JK Zimmerman, Lebrn L, Bayman P, Sastre I, Axelrod F, Alers-Garca J (1998) Host specificity and low reproductive success in the rare endemic Puerto Rican orchid Lepanthes caritensis. Biol Conserv 85:297304 Van Schaik CP, Terborgh JW, Wright SJ (1993) The phenology of tropical forests: adaptive significance and consequences for primary consumers. Annu Rev Ecol Syst 24:353377 Waser NM (1983) The adaptive nature of floral traits: ideas and evidence. In: Real L (ed) Pollination biology. Academic Press, Orlando, pp 241285 Waser NM, Chittka L (1998) Bedazzled by flowers. Nature 394:835836 Waser NM, Price MV (1985) The effect of nectar guides on pollinator preference: experimental studies with montane herb. Oecologia 67:121126 Waser NM, Chittka L, Price MV, Williams NM, Ollerton J (1996) Generalization in pollination systems, and why it matters. Ecology 77:10431060 Wiewandt TA (1979) La gran Iguana de Mona. Nat Hist 88:5765 Wright SJ (1991) Seasonal drought and the phenology of understory shrubs in a tropical moist forest. Ecology 72:164357 Zimmerman JK, Aide TM (1989) Patterns of fruit production in a neotropical orchid: pollinators vs. resource limitation. Am J Bot 76:6773