Abstract
Most species of Dalechampia vines (Euphorbiaceae) attract bee pollinators with terpenoid resins secreted by a gland-like structure in the inflorescence. In some species, pollinating bees appear to preferentially visit inflorescences (blossoms) with large resin-producing glands, whereas in other species bees preferentially visit blossoms with large involucral bracts. In this study, the reliability of bract and gland size as signals of the quantity of resin produced in one species, D. scandens, was assessed. Whether resin secretion has a cost with respect to the number or mass of the seeds produced by a blossom was also examined.
Measurements were made of bract size, gland size and the amount of resin secreted by blossoms of D. scandens maintained in a common environment, and the relationships between these traits were analysed. Resin production was also manipulated, and the effects of the manipulation were tested on seed set and seed mass.
The amount of resin produced was better predicted by the size of the gland than by the size of the bract. Furthermore, when the effect of gland size was accounted for, bract size only weakly predicted the amount of resin produced. Neither an increase in resin secretion (by daily removal of the resin) nor a decrease (by removal of the resin gland) affected seed set or seed mass detectably, but resin production correlated positively with mean seed mass at the individual level once the size of the resin gland was accounted for.
Gland size is a better indicator of the amount of reward than bract size, although the latter remained an honest signal of the quantity of resin produced. Resin secretion has no detectable cost in terms of seed production, but may be condition dependent, as suggested by a positive correlation with seed mass at the individual level.
INTRODUCTION
Understanding the evolution of pollinator rewards in flowering plants requires insight into both the communication processes between plants and pollinators and the balance between costs and benefits associated with the production of the reward. Because flowers often conceal the reward, and the reward could have been removed by recent visitors, pollinators usually have to base their foraging decisions on floral traits that are correlated with the presence of rewards (e.g. Cresswell and Galen, 1991). This relationship creates the opportunity for plants to advertise to pollinators with signals that can range from honest to dishonest (Schaefer et al., 2004). Pollinator rewards presumably have significant costs, and these may affect the degree to which plants will honestly advertise the amount of reward offered. These costs may accrue in terms of energy (i.e. carbon bonds) or elements invested (i.e. nutrients; Bloom et al., 1985). It is expected that investment in a given amount of reward increases fitness (via increasing attraction of pollinators and therefore dispersal and receipt of pollen) at least as much as the cost (in lowered seed and/or pollen production) decreases it. Hence, the evolution of pollinator rewards should follow the cost–benefit investment dynamics seen in other trade-off systems (e.g. Charnov, 1982). Estimates of the costs of reward production have provided mixed results, however. On the one hand, at least four studies have shown that increased nectar production, induced by nectar removal, has a cost in reducing seed number and/or weight (Pyke, 1991; Ashman and Schoen, 1997; Ordano and Ornelas, 2005; Ornelas and Lara, 2009). On the other hand, four studies failed to detect any cost of nectar production in terms of seed production (Zimmerman and Pyke, 1988; Leiss et al., 2004; Ornelas et al., 2007) or growth (Golubov et al., 2004).
All studies investigating costs of pollinator rewards to date have focused on nectar. However, it would be worth knowing more about the costs of other kinds of rewards, because they may represent different ratios of investment in terms of energy (carbon bonds in photosynthates, or ‘carbon’) vs. mineral elements (i.e. ‘nutrients’, e.g. N, P or K). The distinction between carbon and nutrient costs is thought to be critical in plant allocation, because usually one or the other is limiting (Bryant et al., 1983; Bloom et al., 1985). Following this line of reasoning, in low-light environments, the availability of photosynthates (carbon), rather than mineral nutrients, limits growth and reproduction. Conversely, in high-light environments, the availability of mineral nutrients is more likely to limit growth and reproduction. Thus, plant products (organs, secretions, defences, etc.) that contain large amounts of carbon would be expensive in low-light environments but cheap in high-light environments, while plant products that contain N, P or K would be expensive in high-light environments, but relatively cheap in low-light environments (Bryant et al., 1983; Coley et al., 1985).
While most temperate flowers offer nutritive rewards in the form of nectar or pollen, tropical and subtropical flowers sometimes offer oils (free fatty acids), or non-nutritive rewards such as fragrances (collected by male bees to attract females) or resins (collected by female bees for nest construction). Nectar is rich in photosynthates, but usually also contains significant amounts of mineral nutrients, specifically nitrogen in the form of amino acids (Baker and Baker, 1986). No study has assessed costs in the context of the mineral nutrient vs. photosynthate (carbon) content of the nectar, although these vary markedly across pollination ‘syndromes’ (Baker and Baker, 1986). Pollen is rich in starch and/or oils (carbon cost), proteins (nitrogen cost), and nucleic acids (nitrogen and phosphorus costs). Fatty acids, fragrances and resins are all photosynthate-based rewards. Extension of the carbon/nutrient-balance hypothesis for plant defence (Bryant et al., 1983; Coley et al., 1985) to rewards indicates that carbon-rich rewards should be cheap in high-light environments (i.e. no effect on seed production) and relatively expensive in low-light and/or nutrient-rich environments. A first step in assessing this line of reasoning is to determine if reward production is plastic, and if so, whether high production has a detectable cost in terms of reproductive fitness.
In this study, we analysed production of the pollinator reward in Dalechampia scandens (Euphorbiaceae), a neotropical vine that has a purely carbon-based reward system (resin). Although the resin gland in Dalechampia is fully visible to pollinators, previous studies have shown that pollinators often base their foraging decisions on the size of the bracts subtending the blossom rather than the size of the resin gland (Armbruster et al., 2005; R. Pérez-Barrales et al., University of Portsmouth, UK, unpubl. res.; but see Bolstad et al., 2010; Armbruster et al., 2011). We therefore assessed the signal reliability of the size of the bract and resin-producing gland in terms of their ability to predict the amount of reward available to pollinators. We then estimated the costs of resin production by comparing seed set and seed mass between blossoms that were manipulated either to increase or to decrease resin production. Because nectar generally contains significant amounts of amino acids, previous studies have examined rewards reflecting investments in both mineral nutrients and photosynthates, but very few studies, if any, have tested the costs of a purely photosynthates-based reward.
STUDY SYSTEM
Floral resins are purely carbon-based (plus hydrogen and oxygen) pollinator rewards. They are secreted by the flowers (or associated structures) of members of three unrelated groups of plants: Maxillaria (Orchidaceae), Dalechampia (Euphorbiaceae), and Clusia and related genera (Clusiaceae; Armbruster, 1984; Gustafsson and Bittrich, 2002). These resins, comprising mixtures of oxygenated triterpenes (Dalechampia) or polyisoprenylated benzophenones (Clusia, Clusiella, Tovomitra), are collected by various species of Apidae and Megachilidae bees for use in nest construction (Armbruster, 1984). Several advantages are associated with the production of resin as a pollinator reward. First, the sticky nature of the resin prevents pollinators from completely depleting the reward (Armbruster, 1984). This may be important because the resin reward is generally presented openly, in full view of the pollinators, and flowers with recently depleted resin might otherwise fail to get visits (Armbruster et al., 2005; Bolstad et al., 2010). Second, relatively few insect species collect resins, resulting in relatively specialized pollination which may increase pollination efficiency and quality of arriving pollen.
Dalechampia scandens (Euphorbiaceae) is a neotropical vine pollinated by female bees in the genera Eulaema, Eufriesea, Euglossa (Apidea: Euglossini) and Hypanthidium (Megachilidae: Anthidiini), and/or worker Trigona (Apidae: Meliponini), depending on the geographical location of the population. Local populations of D. scandens usually show a degree of adaptation to one type of bee (Armbruster, 1985; Hansen et al., 2000), thus indicating a fairly specialized pollination system. The blossom of D. scandens comprises a pair of male and female subinflorescences subtended by two showy bracts (Fig. 1), which also have a protective role, closing at night to protect the flowers and during the period of fruit maturation. The female subinflorescence comprises three female flowers containing three ovules each. Therefore, each blossom can produce a maximum of nine seeds. The male subinflorescence comprises ten staminate flowers plus a gland composed of modified bractlets producing terpenoid resin (Fig. 1). When the blossom opens for the first time, only female flowers are receptive. After 2–3 d, the first (terminal) male flower opens, followed by the opening of the other male flowers in succession over a period of approx. 1 week. The plant is self-compatible, and the blossom can self-pollinate during the bisexual phase, the distance between the anther and the stigma affecting the frequency of self-pollination (Armbruster, 1988). The floral resin is collected by bees that use resin in nest construction (Armbruster, 1984). As a nesting resource, resin seems to be particularly valuable, perhaps even limiting, in the reproductive success of bees that rely on it (see Howard, 1985).
Blossom of D. scandens. (A) Entire blossom showing involucral bracts (photo: C. Pélabon). (B) Close-up of the male and female flower clusters with the resin-producing gland (photo: P. H. Olsen). Note the presence of transparent resin on the gland surface. (C) Drawing representing the traits measured. We measured the area of the gland (GA) as the gland width (GW) multiplied by the average height of the two half left and right portions (GHl, GHr). The upper bract area (UBA) was measured as the product of the bract length (UBL) × bract width (UBW).
The size of the blossom and the resin gland vary markedly among Dalechampia species. Previous studies suggested that variation in gland size affects the amount of resin offered to the pollinators and consequently the maximum size of the pollinator that can afford to visit the blossoms (Armbruster, 1984; Armbruster and Herzig, 1984), as a result of the energy-cost/resin-benefit balance (Heinrich and Raven, 1972). To better understand interspecific variation in gland size, Armbruster (1990) calculated adaptive surfaces for different blossom traits, including gland size, and showed that despite the benefits associated with large glands in terms of increased pollinator visitation rates, Dalechampia species were scattered along a ridge of increasing gland size as blossom size increased (Armbruster, 1990, figs 3 and 4). He further hypothesized that the absence of species with small blossoms and large glands resulted from selection either for specialization or against ‘excessive’ costs associated with the production of large amounts of resin (Armbruster, 1990, 1991). The hypothesis about costs of resin remains, however, untested.
Blossom size and gland size also vary within species (Armbruster, 1985, 1991, 1996; Hansen et al., 2003a), and phenotypic selection studies have shown that bees prefer to visit blossoms with either large bracts (Armbruster et al., 2005; R. Pérez-Barrales et al., University of Portsmouth, UK, unpubl. res.) or large glands (Bolstad et al., 2010; Armbruster et al., 2011). Although these two traits are positively correlated, both phenotypically and genetically (Hansen et al., 2003b), it remains uncertain how much information each trait conveys regarding resin secretion and availability. Furthermore, the absence of large glands in species with small blossoms could reflect either indirect selection on gland size (via the correlation with bract size) generated by antagonists such as seed predators (R. Pérez-Barrales et al., University of Portsmouth, UK, unpubl. res.), or excessive costs associated with resin secretion, balancing the fitness benefits of larger glands. Such costs have never been estimated, however.
MATERIALS AND METHODS
Study population and experimental treatments
Individuals used in this study were the fifth greenhouse generation from seeds collected originally in the state of Quintana Roo, Mexico (20 °13′N, 87 °26′W). The greenhouse population was started with seeds collected from 75 maternal plants and subsequently maintained by outcrossing for five generations.
In this experiment, we investigated the effect of increased resin production (by daily harvest of the resin) or complete cessation of resin production (by removal of the resin gland) on seed production at the blossom level. For each plant, we randomly chose blossoms when they first opened (day 1) and allocated them to one of the six following treatments.
(1) Removal of the male inflorescence (staminate flowers + resin gland)
In this treatment we aimed to remove all costs associated with resin production (and unavoidably, costs, if any, of maturing staminate flowers). We removed the whole male inflorescence, because it was not easy to remove the resin gland alone. To identify the costs specific to resin production we compared this treatment with treatment 2, below.
(2) Removal of the staminate flowers
In this treatment, we only removed the staminate flowers, but left the resin gland intact. The difference in seed set or seed size between this treatment and the first treatment should approximate the effect of removing only the resin gland.
(3) Removal of the upper bract
In this treatment, we cut the upper bract close to its insertion point on the stem. This treatment may impose a carbon cost in terms of reduced photosynthetic area during fruiting. It also provides a control for the manipulation performed in the first two treatments.
(4) Removal of the resin daily, over 5 d
In this treatment, we removed the resin from the gland every day over 5 d to assess the rate of resin replenishment and the costs of resin production. The visitation rate observed in a natural population of D. scandens is commonly less than one visit per day during the female phase (R. Pérez-Barrales, unpubl. res.). Therefore, this experimental manipulation represents a rate of resin removal slightly higher than that normally experienced by blossoms in natural populations.
(5) Removal of the resin at day 5
In this treatment, we removed the resin only at the end of the 5-d period.
(6) Control
In this treatment, we only measured, pollinated and bagged the blossom.
Measurements and resin collection
On day 1 (first day the blossom is open), we measured the area of the resin gland (GA = gland height × gland width) for each blossom in treatments 4 and 5 and the area of the upper bract (UBA = bract length × bract width) of each blossom in all treatments (see Fig. 1 for trait definitions). The area of the upper bract represents a proxy of the overall blossom size. Three repeated measurements of bract length and bract width were made, and we used the mean of these three measurements in the analyses.
Resin was collected with a scalpel blade and deposited in a small plastic cup of known weight. The cup and resin were weighed to the nearest 0·1 mg, the amount of resin being estimated as the difference between the two measurements. Particular care was taken to remove any residual resin on the scalpel blade between blossoms using a non-polar solvent.
Pollination and seed harvest
After the blossom was measured and exposed to one of the treatments, we hand-pollinated the female flowers with pollen from a blossom of another randomly chosen individual from the same population. We took care to apply an excess of pollen on the stigma of each of the three pistillate flowers to facilitate pollen competition and avoid variation in seed set or seed mass due to differences in pollen load and intensity of pollen competition (Armbruster and Rogers, 2004). In the treatments where the male flowers were not removed, self-pollination remained unlikely because ample pollen from the sire plant was applied to stigmas 2–3 d before the first male flower opened. Seeds take about 1 month to mature and they are dispersed by explosive dehiscence. At the end of each treatment, 7–10 d after pollination, pollinated blossoms were bagged so that seeds could be collected after dehiscence. Seed set was the number of seeds collected. All seeds from one blossom were weighted together on a precision balance (to the nearest 0·1 mg), and average seed mass was calculated as the total mass of the seed set divided by the number of seeds. Sixty-one plants were included in this experiment, and all six treatments were applied to 45 of these.
Statistical analyses
Resin production and replenishment
We used analysis of covariance (ANCOVA) to estimate the effect of gland area on resin-secretion rates and to test the effect of resin removal on resin production. The analysis was conducted on log-transformed data because the different scales on which variables were measured (mm2 and mg) could induce a non-linear relationship. In this analysis the total amount of resin produced was the response variable, the treatment (resin daily harvested or not) was the predictor variable, and resin-gland area was the covariate. To assess changes in daily rates of resin secretion, we fitted a mixed-effect model between daily resin production and day, with individual plant as random factor. Levels of variation in blossom size or seed mass were investigated using variance component analyses to partition the variance within and among individuals. We obtained the 95 % confidence intervals (CI) of the variance components by resampling the posterior distribution of the parameters of the variance-component analyses using Markov Chain Monte Carlo methods.
Honesty of reward signalling
To investigate how well bract and gland area predict resin production, we compared the phenotypic correlation between bract area, gland area and resin production after the effect of the treatment on resin secretion has been accounted for. We also conducted path analyses between the two morphological traits and resin production to determine whether bract area carried information about resin production beyond its correlation with gland area. Path analyses were conducted via multiple regression on standardized (zero mean and unit variance) variables for the two treatments in which resin production was measured.
Treatment effects on seed set and seed mass
To test the effect of the treatment on seed set and seed mass, we first fitted mixed-effect models including either seed set or seed mass as response variable, treatment as predictor variable and individual identity as random factor, because the same individual plants appeared in the different treatments. For seed mass, we also included the upper bract area as covariate. Seed set is a count and was analysed with a Poisson error structure. We selected the best model using the Akaike information criteria (AIC). We further compared specific treatments by specifying the appropriate contrasts in the mixed-effect model on seed mass including treatment as fixed factor and individual as random factor. All the statistical analyses were performed in R, version 2·10·0 (R Development Core Team, 2010).
RESULTS
Resin production and replenishment
When harvested 5 d after the blossom has first opened, resin glands produced on average 4·22 ± 0·25 mg of resin. When the resin was harvested daily, the total production over the 5-d period increased by 39 %, to 5·85 ± 0·28 mg. Resin production was strongly correlated with the area of the resin gland (Fig. 2); the slopes of the relationship between the total amount of resin produced and the gland area were very similar for both treatments (daily or single collection of resin). The daily rate of resin secretion increased with blossom age (Fig. 3). Finally, once the effects of gland area and harvesting were accounted for, we observed individual differences in rates of resin secretion with 26 % (95 % CI: 0·0–26·9 %) of the variance in relative resin secretion occurring at the among-individual level.
Relationship between the log gland area and the log resin production over 5 d with resin harvested either every day (grey circles) or only once after 5 d (black circles). Results of the ANCOVA on log-transformed data revealed no interaction between treatment and gland area (F1,105 = 0·41, P = 0·53), while the effects of the treatment and gland area were both statistically significant (gland area: F1,105 = 7·89, P = 0·006; treatment: F1,105 = 140·87, P < 0·001). The common slope of the relationship between gland area and resin production for the two treatments was 1·28 ± 0·12. The difference between intercepts of the two treatments was 0·185 ± 0·066 log mg. Note that the slopes presented on the graph were fitted for each treatment separately. For the whole model, r2 = 0·58.
Mean (± s.e.) daily resin-secretion rate. The slope of the mixed-effect model with individual as random factor was β = 0·064 ± 0·019 (model with day as predictor variable AIC = 473·3; without day predictor variable AIC = 475·8).
Honesty of reward signalling
Variation in log bract area explained 19 % of the variation in log resin secretion (r = 0·43; 95 % CI = 0·26–0·57), while variation in log gland area explained 40 % (r = 0·63; 95 % CI = 0·51–0·73). However, the path analysis including the effects of both bract area and gland area on the resin production showed that gland area was the main factor influencing resin production and that the correlation between bract area and resin production was the result of an indirect relationship (via gland area; Fig. 4).
Path diagram showing estimated direct effects of gland area and bract area on the amount of resin produced by the blossom in D. scandens. The double-headed arrow indicates the phenotypic correlation between the two morphological traits. *P <0·05, ***P< 0·001. Un., unexplained source of variation.
Treatment effects on seed set and seed mass
Blossoms produced an average of 8·07 ± 0·077 seeds weighing on average 41·48 ± 0·28 mg each. Individuals differed in average seed mass, with 24 % (95 % CI: 0·5–29·8 %) of the total within-population variance in seed mass generated by among-plant variation. We found no effect of any of the treatments on seed set (AIC for the model with the treatment = 1374; AIC for the model without the treatment =1135; Table 1). Similarly, the different treatments had no detectable effect on seed mass (Tables 1–3). Note that we found only a very weak correlation between seed mass and seed set (r = 0·07; 95 % CI: –0·037 to 0·181), and therefore we did not correct the variation in seed mass by the seed set.
Summary statistics for the seed set and seed mass in the different treatments
| Treatment | n | Mean (± s.e.) seed set (no. of seeds) | Mean (±s.e.) seed mass (mg) |
|---|---|---|---|
| 1. Removal of ♂ subinfloresence | 56 | 8·16 ± 0·22 | 41·13 ± 0·76 |
| 2. Removal of staminate flowers | 56 | 8·12 ± 0·16 | 42·15 ± 0·69 |
| 3. Removal of upper bract | 56 | 8·09 ± 0·17 | 41·04 ± 0·59 |
| 4. Resin harvested daily | 51 | 8·00 ± 0·19 | 41·61 ± 0·66 |
| 5. Removal of resin at day 5 | 45 | 8·00 ± 0·24 | 41·36 ± 0·66 |
| 6. No manipulation | 54 | 8·02 ± 0·18 | 41·60 ± 0·72 |
| Treatment | n | Mean (± s.e.) seed set (no. of seeds) | Mean (±s.e.) seed mass (mg) |
|---|---|---|---|
| 1. Removal of ♂ subinfloresence | 56 | 8·16 ± 0·22 | 41·13 ± 0·76 |
| 2. Removal of staminate flowers | 56 | 8·12 ± 0·16 | 42·15 ± 0·69 |
| 3. Removal of upper bract | 56 | 8·09 ± 0·17 | 41·04 ± 0·59 |
| 4. Resin harvested daily | 51 | 8·00 ± 0·19 | 41·61 ± 0·66 |
| 5. Removal of resin at day 5 | 45 | 8·00 ± 0·24 | 41·36 ± 0·66 |
| 6. No manipulation | 54 | 8·02 ± 0·18 | 41·60 ± 0·72 |
Summary statistics for the seed set and seed mass in the different treatments
| Treatment | n | Mean (± s.e.) seed set (no. of seeds) | Mean (±s.e.) seed mass (mg) |
|---|---|---|---|
| 1. Removal of ♂ subinfloresence | 56 | 8·16 ± 0·22 | 41·13 ± 0·76 |
| 2. Removal of staminate flowers | 56 | 8·12 ± 0·16 | 42·15 ± 0·69 |
| 3. Removal of upper bract | 56 | 8·09 ± 0·17 | 41·04 ± 0·59 |
| 4. Resin harvested daily | 51 | 8·00 ± 0·19 | 41·61 ± 0·66 |
| 5. Removal of resin at day 5 | 45 | 8·00 ± 0·24 | 41·36 ± 0·66 |
| 6. No manipulation | 54 | 8·02 ± 0·18 | 41·60 ± 0·72 |
| Treatment | n | Mean (± s.e.) seed set (no. of seeds) | Mean (±s.e.) seed mass (mg) |
|---|---|---|---|
| 1. Removal of ♂ subinfloresence | 56 | 8·16 ± 0·22 | 41·13 ± 0·76 |
| 2. Removal of staminate flowers | 56 | 8·12 ± 0·16 | 42·15 ± 0·69 |
| 3. Removal of upper bract | 56 | 8·09 ± 0·17 | 41·04 ± 0·59 |
| 4. Resin harvested daily | 51 | 8·00 ± 0·19 | 41·61 ± 0·66 |
| 5. Removal of resin at day 5 | 45 | 8·00 ± 0·24 | 41·36 ± 0·66 |
| 6. No manipulation | 54 | 8·02 ± 0·18 | 41·60 ± 0·72 |
Model selection for the effects of the treatments and blossom size (estimated by the upper bract area, mm2) on average seed mass (mg)
| Model | AIC | ΔAIC | k | Weight |
|---|---|---|---|---|
| Seed mass ∼ constant | 1900 | 0 | 1 | 0·54 |
| Seed mass ∼ treatment | 1901 | 1 | 6 | 0·33 |
| Seed mass ∼ upper bract area | 1910 | 10 | 2 | 0·004 |
| Seed mass ∼ treatment + upper bract area | 1912 | 12 | 7 | 0·001 |
| Seed mass ∼ treatment × upper bract area | 1948 | 48 | 12 | 0 |
| Model | AIC | ΔAIC | k | Weight |
|---|---|---|---|---|
| Seed mass ∼ constant | 1900 | 0 | 1 | 0·54 |
| Seed mass ∼ treatment | 1901 | 1 | 6 | 0·33 |
| Seed mass ∼ upper bract area | 1910 | 10 | 2 | 0·004 |
| Seed mass ∼ treatment + upper bract area | 1912 | 12 | 7 | 0·001 |
| Seed mass ∼ treatment × upper bract area | 1948 | 48 | 12 | 0 |
Models are mixed-effects models where treatment was entered as fixed factor, upper bract area as covariate and individual identity as random factor. The best model in bold includes only the constant (see Table 1 for the mean ± s.e. of the different treatments).
Model selection for the effects of the treatments and blossom size (estimated by the upper bract area, mm2) on average seed mass (mg)
| Model | AIC | ΔAIC | k | Weight |
|---|---|---|---|---|
| Seed mass ∼ constant | 1900 | 0 | 1 | 0·54 |
| Seed mass ∼ treatment | 1901 | 1 | 6 | 0·33 |
| Seed mass ∼ upper bract area | 1910 | 10 | 2 | 0·004 |
| Seed mass ∼ treatment + upper bract area | 1912 | 12 | 7 | 0·001 |
| Seed mass ∼ treatment × upper bract area | 1948 | 48 | 12 | 0 |
| Model | AIC | ΔAIC | k | Weight |
|---|---|---|---|---|
| Seed mass ∼ constant | 1900 | 0 | 1 | 0·54 |
| Seed mass ∼ treatment | 1901 | 1 | 6 | 0·33 |
| Seed mass ∼ upper bract area | 1910 | 10 | 2 | 0·004 |
| Seed mass ∼ treatment + upper bract area | 1912 | 12 | 7 | 0·001 |
| Seed mass ∼ treatment × upper bract area | 1948 | 48 | 12 | 0 |
Models are mixed-effects models where treatment was entered as fixed factor, upper bract area as covariate and individual identity as random factor. The best model in bold includes only the constant (see Table 1 for the mean ± s.e. of the different treatments).
Comparisons of the seed mass between treatments
| Comparison | Effect | n | Mean (±s.e.) difference (mg) |
|---|---|---|---|
| 1 vs. 2 | Cost of resin secretion | 51 | –0·66 ± 0·54 |
| 1 vs. 6 | Costs of male flowers + resin gland | 51 | 0·44 ± 0·69 |
| 3 vs. 6 | Cost of manipulation (removal of the bract) | 50 | –0·53 ± 0·54 |
| 5 vs. 6 | Cost of resin production when harvested once | 45 | 0·06 ± 0·72 |
| 4 vs. 6 | Cost of resin production when harvested daily | 47 | 0·17 ± 0·59 |
| 4 vs. 5 | Cost of an increase in resin secretion by daily removal for 5 d | 45 | 0·11 ± 0·56 |
| Comparison | Effect | n | Mean (±s.e.) difference (mg) |
|---|---|---|---|
| 1 vs. 2 | Cost of resin secretion | 51 | –0·66 ± 0·54 |
| 1 vs. 6 | Costs of male flowers + resin gland | 51 | 0·44 ± 0·69 |
| 3 vs. 6 | Cost of manipulation (removal of the bract) | 50 | –0·53 ± 0·54 |
| 5 vs. 6 | Cost of resin production when harvested once | 45 | 0·06 ± 0·72 |
| 4 vs. 6 | Cost of resin production when harvested daily | 47 | 0·17 ± 0·59 |
| 4 vs. 5 | Cost of an increase in resin secretion by daily removal for 5 d | 45 | 0·11 ± 0·56 |
The effects tested are explained in the column ‘Effect’ and the mean ± s.e. differences between treatments are presented in the final column. These means were obtained by specifying the contrasts in the mixed-effects model including treatment as fixed factor and individual as random factor.
Comparisons of the seed mass between treatments
| Comparison | Effect | n | Mean (±s.e.) difference (mg) |
|---|---|---|---|
| 1 vs. 2 | Cost of resin secretion | 51 | –0·66 ± 0·54 |
| 1 vs. 6 | Costs of male flowers + resin gland | 51 | 0·44 ± 0·69 |
| 3 vs. 6 | Cost of manipulation (removal of the bract) | 50 | –0·53 ± 0·54 |
| 5 vs. 6 | Cost of resin production when harvested once | 45 | 0·06 ± 0·72 |
| 4 vs. 6 | Cost of resin production when harvested daily | 47 | 0·17 ± 0·59 |
| 4 vs. 5 | Cost of an increase in resin secretion by daily removal for 5 d | 45 | 0·11 ± 0·56 |
| Comparison | Effect | n | Mean (±s.e.) difference (mg) |
|---|---|---|---|
| 1 vs. 2 | Cost of resin secretion | 51 | –0·66 ± 0·54 |
| 1 vs. 6 | Costs of male flowers + resin gland | 51 | 0·44 ± 0·69 |
| 3 vs. 6 | Cost of manipulation (removal of the bract) | 50 | –0·53 ± 0·54 |
| 5 vs. 6 | Cost of resin production when harvested once | 45 | 0·06 ± 0·72 |
| 4 vs. 6 | Cost of resin production when harvested daily | 47 | 0·17 ± 0·59 |
| 4 vs. 5 | Cost of an increase in resin secretion by daily removal for 5 d | 45 | 0·11 ± 0·56 |
The effects tested are explained in the column ‘Effect’ and the mean ± s.e. differences between treatments are presented in the final column. These means were obtained by specifying the contrasts in the mixed-effects model including treatment as fixed factor and individual as random factor.
Blossom size estimated by the upper bract area had no marked effect on seed mass (Table 2). However, we observed a positive correlation between the relative amount of resin secreted (corrected for variation in gland area) and the average seed mass at the individual level (Fig. 5), indicating that individuals producing more resin also produced larger seeds (a similar result was obtained when seed mass was corrected by either gland area or bract area; data not shown).
Relationship between the relative amount of resin secreted and seed mass at the individual level (r = 0·40, 95 % CI = 0·16–0·60; n = 56, P = 0·001). We considered here the individual mean for both the relative amount of resin secreted (i.e. residuals of the ANCOVA between the amount of resin secreted and the gland area in the two treatments) and the seed mass.
DISCUSSION
Resin production, and honesty of reward signalling
Previous studies on variation within and among species of Dalechampia suggested that resin production depends strongly on the area of the secreting surface of the resin gland (Armbruster, 1984, 1988), although the daily rate of resin secretion can vary from species to species (Armbruster and Steiner, 1992). Our results are consistent with these observations. Furthermore, as is generally the case for nectar (Pyke, 1991; Ashman and Schoen, 1997; Ordano and Ornelas, 2005; Ornelas and Lara, 2009), resin was replenished when collected, so that the daily removal increased the total production over the life of a blossom. Note that there is no evidence that the standing crop of resin is ever reduced by resorption or volatilization (see Armbruster, 1984; Armbruster et al., 1997). We also observed that the amount of resin secreted each day increased with blossom age. Because we measured the size of the gland only on the first day of the experiment, it remains to be seen whether the increase in secretion rate resulted from the growth of the resin gland or if it was a response to the daily harvesting.
Both bract area and gland area were positively correlated with the amount of resin produced and therefore provided honest indications of the quantity of reward offered. However, the reliability of the information carried by each of these two traits differed and gland area signalled more reliably the amount of resin secreted by the blossom. The path analyses further revealed that the correlation between bract area and resin secretion was generated by a weak direct relationship between the two variables and a strong indirect relationship caused by the correlation between bract area and gland area. Note that when concluding that gland area is a better indicator of resin production than bract area, we assume that the size of both bracts and resin gland can be assessed by the bees with similar accuracy. Indeed, the quality of the information revealed by each trait will also depend on the relative size of the error pollinators make during their assessment. If bract size can be assessed more accurately, especially at long distance, than gland size, it may still represent a better signal of resin quantity for foraging bees.
We noticed that the correlation between gland area and the amount of resin secreted was particularly high, despite the difficulties of harvesting the sticky resin and the associated measurement errors. With reference to the signalling theory, this suggests that gland area should be referred to as an index, i.e. a signal whose intensity is causally related to the quantity being signalled and which cannot be faked (Maynard Smith and Harper, 1995, 2003). Bract size, on the other hand, can be considered as a signal. Although evolution of involucral bracts in Dalechampia may have been partly driven by selection for protection of the flowers and developing fruits, that bracts are usually colourful indicates that they also have evolved as signals to pollinators (Armbruster, 1993, 1996, 1997). We speculate that, in the two Dalechampia species where phenotypic selection studies have shown that pollinators base their foraging decision on the size of the bract instead of the size of the resin gland (Armbruster et al., 2005; R. Pérez-Barrales et al., University of Portsmouth, UK, unpubl. res.), bract size is more easily assessed from a distance, and often provides enough reliable information about the amount of available reward to be used efficiently by pollinators during their approach (Armbruster et al., 2005). Confirmation of this speculation could be obtained by testing the relationship between bract size, gland size and resin secretion in D. schottii and D. bidentata, in which bees preferentially visit blossoms with large resin glands rather than large bracts (Bolstad et al., 2010; Armbruster et al., 2011).
Even after accounting for variation in blossom size, we observed that plants differed in the amount of resin secreted. Because plants were grown in a nearly uniform environment, this observation suggests that there is individual, possibly genetic, variation in rates of resin secretion.
Costs of resin production
Our manipulation of resin production did not affect either seed set or seed mass, suggesting that resin production does not have reproductive costs, at least at the blossom level. As suggested above, this absence of cost may be a consequence of the high-light conditions to which plants were exposed in the greenhouse. Following the carbon/nutrient-balance hypothesis, production of carbon-rich rewards such as resin (triterpenoid – large carbon-rich molecules) may be relatively cheap in a high-light environment compared with nutrient-rich rewards. When considering the light environment in the previous studies testing for a cost of reward production, no clear pattern emerges, however. An absence of cost in high-light environments has been reported for Polemonium (open subalpine; Zimmerman and Pyke, 1988) and Prosopis (Golubov et al., 2004) but also in a low-light environment for Moussonia (subshrub in forest; Ornelas et al., 2007). Conversely, negative effects of reward production on reproductive output have been observed both in high-light environments for Blandfordia (swampy heathland; Pyke, 1991) and Clarkia tembloriensis (annual herb in open grassland; Ashman and Schoen, 1997) and Penstemon roseus (shrub in openings in cloud forest; Ornelas and Lara, 2009), and in low-light environments for Tillandsia multicaulis and T. deppeana (under-canopy epiphytes in cloud forests; Ordano and Ornelas, 2005).
Despite the relative independence of the modules in plants, it may be that the cost of reward production is not detectable at the blossom (local) level but only at the whole-plant level (see Ornelas et al., 2007). Alternatively, reward production may also be condition dependent. In this case, individuals in good condition should be able to produce larger blossom with larger gland producing more resin and eventually setting more seeds of larger size. Revealing the cost of resin production would then require manipulation of the individual condition (e.g. Cotton et al., 2004). Nevertheless, the positive correlation between the relative amount of resin secreted (i.e. corrected for difference in gland area) and the seed mass tends to support the hypothesis of a condition-dependent production of reward, because individuals secreting more resin also set larger seeds, independently of blossom size.
Limiting factors on resin production and the carbon/nutrient-balance hypothesis
If there is no reproductive cost in producing resin, what factors actually limit the size of the resin glands in Dalechampia? The simplest explanation is that large resin glands are expensive in and of themselves, but our gland-removal treatment was performed too late to eliminate the cost. Another possibility is that the size of the gland is ecologically adaptive because it has evolved to attract particular types of bees. Different species of Dalechampia and different populations of D. scandens have glands of different sizes and are pollinated by bees of different sizes. There are indications of reproductive character displacement in gland size between sympatric Dalechampia species (Armbruster, 1985; Hansen et al., 2000). This is probably driven by specialization on bees of different sizes. There may also be other ecological costs to large glands in terms of attracting herbivores or reducing the need for a visiting bee to visit other blossoms (Heinrich and Raven, 1972). Another possibility is that an excessively large gland could prevent the proper closure of the bracts at night to protect the flowers. In this case the size of the bracts would impose a mechanical constraint on the size of the gland. Interestingly, this functional integration (sensuOlson and Miller, 1958) between the two structures is illustrated by the strong genetic correlation between them (Hansen et al., 2003b). This genetic correlation may ultimately act as a constraint for the independent evolution of the gland (Hansen et al., 2003b), especially because bracts may be under stabilizing selection generated by conflicting selection from herbivory (R. Pérez-Barrales et al., University of Portsmouth, UK, unpubl. res.).
The lack of a detectable cost of resin is in retrospect not too surprising given that resin is carbon based and these plants were grown in high-light environments. If soil N, P and/or K are limiting growth and seed production, then the absence of nitrogen in resin makes this a cheaper reward than nectar under high-light conditions. Indeed, the inconsistencies in the cost of nectar in high- and low-light environments noted in the introduction may reflect variation in the degree to which nectar is rich or poor in amino acids (Baker and Baker, 1986).
Interestingly, the only clade of Dalechampia to have invaded the rain-forest understory (a very low-light environment), section Cremophyllum, has abandoned resin as a reward and evolved instead fragrance rewards, attracting male euglossine bees (Whitten et al., 1986; Armbruster et al., 1989). Although this is also a carbon-based reward, it has one-third the carbon content per molecule and is effective in tiny amounts compared with resin, therefore having very low carbon cost per flower.
Given potentially strong environmental effects on the nature of reward costs, it would be worthwhile to conduct an experiment where light environment and/or soil fertility are varied, in conjunction with varying rates of reward production, with comparisons across different types of rewards. This would lead to a better understanding of the role of environmental effects and whether they are indeed related to availability of photosynthates vs. mineral nutrients, as predicted from the carbon-nutrient balance hypothesis. If this hypothesis is applicable to pollinator attractants, we expect to detect a cost of carbon-based rewards when plants are limited by light but not when limited by nutrients and vice versa for nitrogen-rich rewards.
ACKNOWLEDGEMENTS
Our research was supported by NSF Grant DEB-0444157 to T.F.H. and DEB-0444745 to W.S.A. We thank Grete Rakvaag for plant care, Juan Francisco Ornelas for information on light environments of the plants he has studied, and Geir H. Bolstad and two anonymous reviewers for their constructive comments on a previous version of the manuscript.
