A comparison of coffee floral traits under two different agricultural practices

Floral traits and rewards are important in mediating interactions between plants and pollinators. Agricultural management practices can affect abiotic factors known to influence floral traits; however, our understanding of the links between agricultural practices and floral trait expression is still poorly understood. Variation in floral morphological, nectar, and pollen traits of two important agricultural species, Coffea arabica and C. canephora, was assessed under different agricultural practices (sun and shade). Corolla diameter and corolla tube length were larger and pollen total nitrogen content greater in shade plantations of C. canephora than sun plantations. Corolla tube length and anther filament length were larger in shade plantations of C. arabica. No effect of agricultural practice was found on nectar volume, sugar or caffeine concentrations, or pollen production. Pollen total nitrogen content was lower in sun than shade plantations of C. canephora, but no difference was found between sun and shade for C. arabica. This study provides baseline data on the influence of agronomic practices on C. arabica and C. canephora floral traits and also helps fill a gap in knowledge about the effects of shade trees on floral traits, which can be pertinent to other agroforestry systems.

Pollination is a critical ecosystem service, with up to 90% of flowering plants requiring insects or other animals for pollination 1 and approximately 35% of the global plant-based food supply being dependent on animal-mediated pollination 2 . Floral traits and rewards, including nectar and pollen, are important in mediating interactions between plants and pollinators. Pollinators can use a combination of visual and olfactory signals from flowers to determine which patches, plants, and individual flowers to visit 1 . Floral morphology, including anther and stigma heights, can affect how effective different pollinator species are at removing pollen from anthers and depositing it on stigmas 3,4 . Despite the importance of floral traits in pollinator attraction and pollination and well-known examples of pollinator-mediated selection on floral traits 5,6 , there are a surprising number of plant species, including both wild and agricultural species, for which we have little information about variation in their floral morphology and reward chemistry, what influences this and how it affects pollinator visitation and pollination. Floral traits in horticultural crops have been influenced through breeding practices and domestication with potential consequences for pollinators [7][8][9] , but there is less evidence of how cultivation practices influence floral traits. The goal of this study was therefore to assess variation in morphological and chemical traits of flowers, nectar, and pollen of two important agricultural species, Coffea arabica and C. canephora, under different farm management strategies.
Floral traits can vary in response to environmental pressures 10,11 . For example, the application of low concentrations of nitrogen-based fertilizer can result in plants with larger flowers, which produce more nectar than plants exposed to higher concentrations of nitrogen 12 . This in turn can result in increased pollinator visitation rates to the low-nitrogen plants 12 . In a similar vein, the shading of flowering species can also affect floral traits and rewards. For example, increased solar irradiance can have a positive effect on nectar production rate of Thymus capitatus 13 . Moreover, Campanulastrum americanum plants in the sun have larger floral displays and receive seven times more pollinator visits than plants in the shade 14 . While natural variation in nutrient and light availability can affect floral traits important for pollinator visitation and seed production, agricultural management

Results
Floral shape. We found that many of the floral morphological traits (Fig. 1) were positively correlated (Table 1). All significant correlations in C. arabica shade plantations were positive (Table 1A, 1C). In contrast, there were more significant correlations among floral traits in C. canephora shade plantations than non-significant ones; and, all but one was positive (Table 1B, 1D). Among the strongest were the correlations between corolla diameter and petal length, and petal length and anther filament length; thus, as one trait in flowers of C. canephora sun increased in size, so did most of the others. The number of floral petals affected the allometric relationships of flowers. For example, corolla tube length of C. canephora was negatively correlated with petal width for flowers that had 6 petals, but the opposite was true for flowers with 5 petals. There were more significant correlations in the shaded C. canephora flowers with 5 petals than 6 (Table 1B, 1D).
Some floral morphological traits differed significantly by species and by farm type. For C. arabica, there was only a marginal main effect of farm type on reproductive floral traits (F 1,6 = 5.56; P = 0.054), a significant main effect of floral trait (F 2,550 = 616.86; P < 0.001), and a significant interaction between farm type and floral trait (F 2,550 = 12.06; P < 0.001). Similarly, there was no significant main effect of farm type on floral traits important for visual attraction (F 1,28 = 0.4; P = 0.53), but there was a significant main effect of floral trait (F 5,1375 = 6955.5; P < 0.001) and a significant interaction between farm type and floral trait (F 5,1375 = 10.5; P < 0.001). Specifically,   Table 1. Spearman rank correlation coefficients by species, farm management type (sun vs. shade), and petals (5 or 6 petals) among morphological traits. Bolded values and asterisks indicate significant correlations (*P ≤ 0.05, **P ≤ 0.01, ***P ≤ 0.001). In each sub-table, correlations for two sites are depicted, with one site above the diagonal and another site below the diagonal, as follows: (a) C. arabica flowers with 5 petals Shade above and Sun below, (b) C. canephora flowers with 5 petals Shade above and Sun below (c) C. arabica flowers with 6 petals Shade above and Sun below, (d) C. canephora flowers with 6 petals Shade above and Sun below.
Nectar standing crop, sugar concentration, and caffeine concentration. Some nectar traits differed significantly between coffee species, but farm management type had no effect on nectar reward traits. Specifically, nectar standing crop differed significantly between species (F 1,70 = 9.68; P = 0.003), with 1.3-times more nectar in flowers of C. canephora than those of C. arabica (Fig. 3). Nectar standing crop did not differ by farm type (F 1,49.3 = 0.0005; P = 0.98), and there was no interaction between species and farm type (F 1,70 = 0.28; P = 0.60). For nectar sugar concentration, we found no effects of species, farm type, or their interaction (F < 4.04; P > 0.065 in all cases). Across both species and farm types, nectar sugar concentration ranged from 12.6-25.0%. Finally, nectar caffeine concentration was 1.5-times greater for C. canephora than C. arabica (F 1,11 = 11.29; P = 0.007; Fig. 4), with no difference in caffeine concentration between farm types (F 1,10 = 0.06; P = 0.81).
Pollen production and nitrogen content. Pollen production and nitrogen content varied by species and farm management type. For pollen production, we found that C. canephora produced 1.7-times more pollen than C. arabica (F 1,15 = 62.03; P < 0.001; Fig. 5). Pollen production did not differ by farm type (F 1,13 = 0.68; P = 0.43), but there was a marginal effect of the interaction between species and farm type (F 1,15 = 4.41; P = 0.05). Even so, post-hoc analysis showed no significant difference between pollen production in sun and shade plantations of C. arabica or C. canephora (T 15 = 0.98; P = 0.76; T 12 = −2.04; P = 0.23). Although C. canephora produced more pollen per flower, its pollen had 1.16-times lower total N than C. arabica (F 1,36 = 33.89; P < 0.001; Fig. 6). There was no overall main effect of farm type on pollen N content (F 1,36 = 2.11; P = 0.16), but there was a significant interaction between species and farm type (F 1,36 = 6.40; P = 0.02; Fig. 6). Farm type modified pollen N content of the two species differently. For C. canephora, pollen from sun farms had significantly lower N content than pollen from shade farms (T 36 = 3.08; P = 0.02). However, for C. arabica, there was no significant difference in pollen N content between sun vs. shade (T 36 = −0.71; P = 0.89).

Discussion
Plants that rely on animal pollinators are dependent on their floral display to attract visitors that can effectively pollinate flowers. We assessed variation in floral morphological, nectar, and pollen traits of two important agricultural species, Coffea arabica and C. canephora, under different farm management cultivation strategies (sun and shade). Floral traits were generally positively correlated with one another within each species, with a few exceptions. Our results showed that corolla diameter was larger in shade coffee plantations of both C. arabica and C. canephora and anther filament length was longer in shade plantations of C. arabica. Corolla tube length differed in response to shade between both species, with larger tube length in sun for C. arabica and shade for C. canephora. There was no effect of farm management strategy on nectar standing crop, caffeine concentration, or sugar concentration nor was there an effect on pollen production per flower, but there was a significant difference between species with more nectar, caffeine and pollen per flower being produced in C. canephora flowers. Only pollen total nitrogen differed between farm type and species, with more nitrogen found in the pollen of flowers of C. arabica, followed by C. canephora flowers grown under shade, and then sun. Understanding the ways in which management practices impact floral traits can be especially important for agricultural systems, where variation in these traits could affect variation in pollination and, consequently, yield and profits for pollen-limited systems.
In general, our correlation analyses indicate that many of the floral traits were positively correlated in sun and shade plantations of both species. As such, flowers that are larger in one trait are generally larger overall, and management practices that might have an effect on floral morphological traits will affect these traits in a similar way. Floral traits are often positively correlated with one another in other plant systems 46,47 , suggesting that plants likely exhibit more variation in flower size than flower shape. For example, ccorrelations between related floral morphological traits, and between flower number and plant size in Erysimum mediohispanicum (Brassicaceae) have been recorded, but no correlation between corolla shape and any other trait Gomez et al. 48 . In our comparison of the effects of management practices on floral traits, we found that three out of the nine floral traits measured differed significantly between sun and shade plantations. Corolla diameter was larger in shade coffee plantations of both C. arabica and C. canephora, anther filament length was longer in shade plantations of C. arabica, and corolla tube length was larger in shade plantations of C. canephora. Studies in other floral systems have shown that larger flowers are preferred by bees compared to smaller flowers 49,50 . If this is the case in coffee systems as well, then this would suggest that bees might prefer flowers in shade plantations than sun plantations.
Differences in floral trait size between sun and shade plantations can be due to a variety of abiotic factors, including variation in soil nutrient levels 12,15 , soil moisture 11 , temperature 51 and incoming solar radiation 14 . For example, high watering regimes resulted in significantly larger calyx lengths, and stigma-anther distance of Lythrum silicara compared to medium and low watering regimes 11 . Similarly, Aquilefia coerulea plants had longer stigmas in wetter conditions, and shorter anther and stigma lengths in hotter, drier conditions 51 . Given that shade   www.nature.com/scientificreports www.nature.com/scientificreports/ plantations exhibit less microclimatic extremes than sun plantations 21,52 , it is likely that the more constant soil moisture and cooler temperatures 18 resulted in overall larger floral traits in shade.
It is surprising that although corolla tube length was positively correlated with all floral traits in C. arabica sun and the rest of the floral traits were smaller in sun than shade, that corolla tube length was still larger in sun than shade. Corolla tube length can be an important trait influencing pollinator behavior 41,53 . For example, a longer tube relates to a longer distance that must be traversed by the visitor to reach the reward -either with its body or tongue 41 , and this can in turn affect flower handling time. Bumble bees, for instance, handled lavender flowers faster than honey bees, whose tongues were slightly shorter than those of the bumble bees and the average tube length 54 . Unlike many of the floral traits measured, mean corolla tube length was noticeably (1.3 times) larger in C. canephora than C. arabica, suggesting that longer tongued bees might be more effective at handling C. canephora flowers, just as they might be for flowers under shade and sun for C. canephora and C. arabica, respectively.
Among the floral traits that we considered important for reproduction, anther filament length was the only one that differed between sun and shade plantations, and this difference was only observed in C. arabica plants. Anther filament length and style length are important structures for reproduction as they produce and receive pollen. For example, shorter styles and anther filaments, which could be closer to one-another than longer styles and anthers, can result in sexual interference 55 . There are two types of intra-floral interference, one of which involves pollen clogging, whereby self-pollen compromises female function, and the other occurs when the plant parts impede the positioning of the pollinator preventing effective pollination 55 . Differences in the relative sizes of anthers and styles of the two coffee species matched expectations based on their mating systems. Specifically, C. arabica is self-compatible, and in this species both style and anther filament lengths were similar (0.85 mm difference), and differed less in shade than sun (0.65 mm vs 1.58 mm, respectively). This similarity in anther and style lengths may result in autogamous self-pollen transfer and pollination insurance in cases where flowers do not receive outcrossed pollen. Nonetheless, C. arabica fruit production has been shown to benefit from cross-pollination 56 , so there may be some detrimental effects on pollination in shaded plants if sexual interference is occurring. There was a much larger difference in size of these reproductively important traits in C. canephora (3.2 mm), the self-incompatible species, that relies on cross-pollination for effective fruit set. In this case, the spatial separation may reduce self-pollen deposition from anthers to stigmas, but experiments are needed to test this hypothesis.
Nectar sugar concentration surprisingly did not differ across the type of farm or species. These results differ from those of Wright et al. 23 who found that C. arabica had a higher sugar concentration than C. canephora. Field measurements of nectar sugar concentration can be influenced by temperature and humidity 57,58 . Thus, it is possible that differences in environmental conditions between the two management practices drove differences in the nectar sugar concentration results (SG Prado, unpubl. res.). Alternatively, rainfall may have played a role in balancing out nectar sugar concentrations in both treatments, as many of the flowers we sampled experienced afternoon or early morning rainfall prior to sampling. Although we made sure to collect nectar samples from flowers that were angled sideways or downward, we cannot rule out the possibility that they received some www.nature.com/scientificreports www.nature.com/scientificreports/ rainwater. Increased volume, viscosity and sugar concentrations in nectar have all been shown to increase bee handling times 57,59 and handling time has in turn been linked to greater pollen transfer by bees 59 . As such, the self-incompatible C. canephora plants may be benefiting from improved pollination services compared to the self-compatible C. arabica plants.
The nectar of Coffea flowers not only contained sugars but also the alkaloid caffeine. Consistent with Wright et al. 23 , we observed a higher caffeine concentration in C. canephora than C. arabica flowers; however, for both species, the caffeine concentrations were much higher (ca. 4 and 30 times greater for C. canephora and C. arabica, respectively). Previous studies have suggested that caffeine may have a stronger effect on bee olfactory memory than sugar concentration, resulting in bees becoming more likely to prefer and return to plants with those similar caffeinated signals 23,24 . However, the caffeine concentration of C. canephora flowers in both shade and sun farms in our study exceeded prior studies, with our flowers containing mean caffeine concentrations above 1000 µM. Such concentrations have been shown to have the opposite effect on bees, diminishing a bee's ability to learn and may be a deterrent to honey bees 30 . As such, the likelihood that the caffeine in C. canephora is ensuring pollinator fidelity might be lower than for C. arabica. This would suggest that bee pollination of C. canephora might be compromised, potentially making it more dependent on abiotic pollination for seed set 60 .
Pollen production per flower did not differ between sun and shade plantations but did differ between species. Flowers of C. canephora had significantly more pollen per flower than those of C. arabica. As pollen is the male gamete of the plant, there's a trade-off experienced by the plant to maximize reproduction, while also attracting and rewarding flower visitors 1 . Therefore, producing more pollen may be one way the plant ensures sufficient pollen transferred for reproduction 61 . This would be especially important for C. canephora as it relies on animal and wind pollination for fruit set, and thus not all pollen grains produced will successfully reach conspecifics. Alternatively, C. canephora might simply require greater pollen deposition than C. arabica for successful fruit set 62 . Contrary to pollen production, pollen total N content was greater in C. arabica pollen than C. canephora, and greater in shade plantations of C. canephora than sun. Pollen N content has been shown to vary between species of many flowering plants, including Hibiscus spp. and Passiflora spp. 62 and such variation between species might explain our observed differences between C. arabica and C. canephora. Similarly, to nectar, plant pollen characteristics can differ with environmental factors, and therefore differences in environmental conditions may help explain these results. For example, high levels of phosphorus in soils of Cucurbita pepo can result in pollen that also contains higher concentrations of phosphorus 63,64 . It is therefore possible that the differences in nitrogen content are due to the different levels of nitrogen found in sun and shade plantations (e.g., nitrogen-fixing leguminous trees, slow release through leaf litter decomposition in shade and chemical fertilizers in sun). Additional research is needed to identify the ways in which different nitrogen inputs and nitrogen release times affect pollen protein content.
The comparative work conducted in this study is a necessary first step in understanding the relationship between large-scale agricultural practices and changes in floral traits. We found that corolla diameter, corolla tube length and pollen total nitrogen content were greater in shade plantations of C. canephora than sun plantations. Likewise, corolla tube length and anther filament length were larger in shade plantations of C. arabica. As larger floral displays are generally preferred by bees 49,50 and higher nitrogen content results in increased net nutritional gains, the variation in floral traits in shade plantations might benefit plant pollination and pollinators alike. This study not only helps fill a gap in knowledge about the effects of shade trees on floral traits, which can be pertinent to other agroforestry systems, but to our knowledge, it is also the first study to provide baseline data on C. arabica and C. canephora floral traits. As such, it lays a foundation upon which to formulate hypotheses to investigate causal mechanisms underlying pollinator-coffee relationships.

Methods
Study system. Study area. This study was conducted from January 2017 through April 2017 at 16 coffee plantations located in the central and western part of Puerto Rico ( Table 2). The 16 farms varied in size (0.393-31.44 ha) and agricultural practices ( Table S1). All of the C. canephora farms used also had C. arabica planted. Two of the farms were used for both C. canephora and C. arabica floral trait measurements. Four farms were in sun and five under shade for C. arabica, and five in sun and four under shade for C. canephora. All of the farms had coffee rust (Hemileia vastatrix), although C. canpehora plants were less affected than C. arabica plants. Five focal coffee plants per species were selected randomly within each of the farms, and all floral trait measurements were taken from these same plants. When possible, C. arabica var. Bourbon was sampled. All C. canephora were of the same variety -Robusta.
The land-cover in these regions is classified as lowland moist and montane wet evergreen coffee plantations 65 . Elevations in these regions ranged from 375-875 m.a.s.l., with mean annual rainfall between 1743-2428 mm and mean annual temperatures between 21.6-25.7 °C 66 . In Puerto Rico, there are two rainy seasons, a short one in April-May and a long one in September-December. Likewise, there are two dry seasons, a short one between June-August and a long one between January-March.
Coffee cultivation in Puerto Rico experienced a period of nearly 20 years of agricultural intensification 67 , starting in the late 1980s, resulting in a drastic increase in the number of sun coffee farms 68 . It is only recently that specialized shade coffee (plantations with a restored shade layer; Fig. S1) have been adopted as an alternative to strike a better balance between conservation and coffee production. These two cultivation practices (sun vs. shade) create contrasting environmental conditions, some of which are directly attributable to management practices. For example, sun coffee plantations rely less on ecological processes than shade plantations, replacing them with various agrochemicals, including fertilizers, insecticides and herbicides 69 . Moreover, the excessive use of these agrochemicals can contribute to high levels of soil erosion 70 and nutrient leaching 16 . In contrast, restoring the shade layer can convey some resilience to increasing daytime temperatures, maintain a moister and cooler microsphere than sun coffee plantations, and provide a buffer against extreme climate events, such as hurricanes 18,71 . www.nature.com/scientificreports www.nature.com/scientificreports/ These conditions can help improve plant growth and development by maintaining or improving soil fertility directly by reducing erosion 18 or indirectly through the addition of leaf litter 69 and nitrogen fixation, in the case of leguminous shade trees 16 . Conversely, there are physiological drawbacks, such as resource competition, when shade trees are planted within coffee plantations 18 . Shade vs. sun cultivation may therefore have different effects on floral traits.

Study species. Both
Coffea arabica and C. canephora are native to the African equatorial forest 72 . Coffea arabica, which is native to the Ethiopian tropical forests, can be cultivated between a range of 800-2000 m, and C. canephora, which is native to the lowland forests of the Congo river basin can be grown between <500-1500 m 42,72 . Optimal rainfall for C. arabica ranges between 1200-1800 mm, and temperatures between 18-21 °C 42 . Coffea canephora in turn, can adapt to intensive rainfalls exceeding 2000 mm and has an optimal mean temperature ranging between 22-30 °C 42 . Unlike C. arabica, C. canephora thrives under high air humidity 42 . Coffea canephora is self-incompatible and C. arabica is self-compatible, although it has been shown to experience increased yield from cross-pollination by bees 56 . Green beans of C. canephora contain more caffeine and have a higher concentration of caffeine than those of C. arabica (2.2% vs. 1.2% of dry mass, respectively) 44,45 . Similarly, leaves of C. canephora also contain more caffeine than those of C. arabica (3% vs. 1.6% of dry weight, respectively 73 ).
In Costa Rica and Mexico, the main pollinators of coffee were found to be social bees in the genera Melipona and Trigona as well as Apis mellifera 74,75 . In Puerto Rico, an island with over 35 species of bees, the main pollinator seen in coffee plantations was A. mellifera (SGP, personal observations), the only social bee on the island 76 . A Lasioglossum species and Xylocopa mordax were also observed pollinating the coffee flowers, but these sightings were rare (SGP, personal observations).
Floral shape. To study the morphological variation of C. canephora and C. arabica flowers, for each species we randomly selected ten open flowers on the five focal bushes within each farm. We collected measurements in all but two farms, resulting in a sample of 66 bushes. A total of 729 flowers were measured, 369 of which were of C. canephora (207 sun, 162 shade), and 360 of which were of C. arabica (180 sun, 180 shade). To describe floral traits important for visual attraction of pollinators, we measured the following on each flower: petal width and length, corolla diameter, corolla tube length, corolla tube diameter at opening, and counted the number of petals (Fig. 1). To describe variation in reproductive traits that can affect the ability of insects to pollinate 3,55 , we measured anther filament length, style length, and number of stigmatic lobes (Fig. 1). Measurements were taken using a Mitutoyo digital calliper to the nearest 0.01 mm (Model No. 500-196-30, Mitutoyo, Auroral, Illinois, USA).

Floral nectar sugar concentration and standing crop.
A total of 67 nectar sugar concentration readings were taken, 47 for C. canephora (38 sun, 9 shade), and 20 for C. arabica (12 sun, 8 shade). A total of 249 nectar standing crop measurements were taken, with 160 taken from C. canephora (130 sun, 30 shade) and 89 from C. arabica (50 sun, 39 shade). To measure nectar standing crop per flower, we bagged several bunches of flowers which were 1-2 days from blooming, using bridal veil fabric, to exclude floral visitors. Once the flowers bloomed, we removed the fabric, and collected nectar from 10 randomly selected flowers. We sampled nectar using 5 and 10 µL microcapillary tubes inserted into the base of the flower; we did not squeeze flowers for nectar collection but instead allowed the nectar to suck into the tubes via capillary action. Samples were taken between 9:00-14:00, during which time temperatures ranged from 23-32 °C and windspeeds ranged between 0 and 4.7 Km/h. To measure total sugar concentration, we collected approx. 20 µl of nectar from one or more flowers, as necessary, and measured concentration on an Atago 2352 Master-53T hand-held refractometer with automatic temperature compensation (Atago, Bellevue, Washington, USA), and noted the sugar concentration to the nearest 0.5%.

Species
Type www.nature.com/scientificreports www.nature.com/scientificreports/ Nectar from the standing crop measurements was used, and if more nectar was necessary to obtain the 20 µl for the sample, then nectar was extracted from additional flowers on the same coffee plant.
Floral nectar caffeine content. Using 5-54 flowers from the same coffee plants, we collected 43 nectar samples of between 20-35 µl to measure nectar caffeine content (C. arabica: 8 shade, 10 sun; C. canephora: 13 shade, 12 sun). We immediately placed the nectar samples into a cooler with ice. They were then stored in a freezer at 0 °C until they were lyophilized. Each sample was then diluted with 100 µl of methanol. Samples (5 µl) were analyzed directly by liquid chromatography-mass spectroscopy using a Dionex UltiMate 3000 LC system with separation of compounds on a Phenomenex Luna C18(2) column (150 Å~3 mm i.d., 3 μm particle size) at 400 μL min −1 and eluted using a linear gradient of 90:0:10 (t = 0 min) to 0:90:10 (t = 20-25 min), returning to 90:0:10 (t = 27-30 min). Solvents were water, methanol and 1% formic acid in acetonitrile, respectively. The column was maintained at 30 °C. Compounds were detected by MS on a Thermo Fisher Velos Pro Dual-Pressure Linear Ion Trap Mass Spectrometer. Samples were scanned, using FTMS, from m/z 194-196 corresponding to the molecular ion for caffeine (M + H = m/z 195.1) in positive mode. Peak areas were quantified against a calibration curve of an authentic caffeine standard (Sigma, Dorset, UK).
Pollen production and nitrogen content. Using 1-10 flowers per coffee plant, we collected anthers from a total of 11 plants in 4 C. arabica shade plantations, 12 plants in 4 C. arabica sun plantations, 14 plants in 4 C. canephora shade plantations, and 10 plants in 2 C. canephora sun plantations. A total of 481 flowers were used to measure pollen production per flower (C. arabica -96 shade, 120 sun; C. canephora -126 shade, 139 sun). To measure pollen production per flower, we bagged several bunches of flowers which were 1-2 days from blooming, using bridal veil fabric, to exclude floral visitors. Once the flowers bloomed, we removed the fabric, and collected the anthers from 10 randomly selected flowers, placing the anthers from each flower into separate microcentrifuge tubes. To remove the pollen from the anthers, we added 1500 µl of 70% ethanol to each microcentrifuge tube and sonicated the tubes for 5 minutes to release the pollen from the anther sacs. We then vortexed the samples for approximately 10 seconds, moving the pollen into suspension in the tube. We extracted 4 µl of the suspended solution and placed it on a hemocytometer and counted the number of coffee pollen grains under a dissecting microscope (Nikon SMZ1000) at 20X magnification. We counted 6 subsamples from each tube. We then took the mean of the subsamples and used that mean to calculate the number of pollen grains in the original 1500 µl of liquid (hereafter pollen grains per flower).
We also used some of the freshly opened, bagged flowers, to collect pollen for nitrogen (N) analysis. We removed 12-18 randomly selected flowers from 39 of our focal bushes, and using an electric toothbrush, we vibrated the flower, with the anthers placed within a microcentrifuge tube, to release pollen from the anther sacs. Pollen samples were kept in a freezer at 0 °C until processing. We added 400 µl of 200-proof ethanol to each tube and centrifuged on low RPM for 15 seconds to move the pollen to the bottom of the tube. We removed excess ethanol with a pipette and allowed any remaining ethanol to evaporate off over 24 hr. Pollen samples were then stored in the freezer at −30 °C until analysis. The 39 samples were sent to the UC Davis Analytical Laboratory (Davis, CA, USA) to determine total N using combustion with a LECO FP-528 and TruSpec CN Analyzers. Total N can be used as a proxy for crude total protein content in pollen 77  Data analysis. All statistical analyses were performed in R studio (Version 1.0.44). We used Spearman's rank nonparametric correlation analyses to assess the degree to which Coffea floral traits were related to one another using package Biotools and Hmisc 78,79 . Data were grouped by farm management types (sun/shade), species within management type, and the number of petals (5 or 6) within species. The allometric relationships of floral traits were evaluated within the context of farm management types (sun/shade), species within management type, and the number of petals (5 or 6) within species. To assess variability in floral shape of each coffee species further, we grouped floral traits into two categories: those important for attracting pollinators (petal width and length, corolla diameter, corolla tube length, corolla tube diameter at opening, and the number of petals) and those important for reproduction (anther filament length, style length, and number of stigmatic lobes). We tested whether these traits differed between sun and shade plantations of C. arabica and C. canephora using four linear mixed effect models (LMER)-one for each category of floral traits. In these models, fixed effects were: farm type (sun vs. shade) and traits measured; and random effects were flower nested within bush nested within farm. Although we conducted multiple tests, we followed the guidelines of Moran 80 and Gotelli and Ellison 81 and report unadjusted P-values.
We used a LMER to compare nectar standing crop, sugar concentration, and caffeine concentration between species and shade and sun plantations. We square-root transformed nectar standing crop and caffeine concentration to improve normality. One value for caffeine concentration was removed from analysis as it was an outlier, being 7 times greater than any of the other concentrations found for C. arabica. We also used a LMER to compare pollen production per flower (square-root transformed) and total pollen N (log-transformed) between sun and shade coffee plantations. In the models for nectar sugar concentration, nectar standing crop and pollen production per flower, fixed effects included species (C. canephora and C. arabica) and farm type (sun vs. shade), and random effects included flower nested within bush, and bush nested within farm. For nectar caffeine concentration and pollen total N, we used a similar model but only included bush nested within farm as the random effect. A post-hoc test was performed for caffeine concentration, pollen production per flower, and pollen total N, given that there were two-way interactions between coffee species and farm type. We used package lmerTest for the LMER analyses, and lsmeans for the post-hoc analyses 82,83 .