Water balance is crucial for the growth and flowering of plants. However, the mechanisms by which flowers maintain water balance are poorly understood across different angiosperm branches. Here, we investigated 30 floral hydraulic and economics traits in 24 species from ANA grade, magnoliids, monocots, and eudicots. We found that basal angiosperms had richer petal stomatal density, higher pedicel hydraulic diameter, and flower mass per area, but lower pedicel vessel wall reinforcement, and epidermal cell thickness, compared to monocots and eudicots. This indicates that basal angiosperms maintain water balance with high water supply and consumption, while monocots and eudicots maintain water balance more conservatively. We also observed significant trade-offs and coordination among different floral traits. Specifically, pedicel theoretical hydraulic conductivity was positively correlated with petal stomatal density, flower water potential at turgor loss point, and maximum vessel diameter, but was negatively correlated with flower construction cost, vessel density, and pedicel vessel wall reinforcement. Floral traits associated with reproduction, such as floral longevity and size, were strongly linked with its physiological and anatomical traits. Our results systematically reveal the variation in flower economics and hydraulic traits from different angiosperm branches, deepening the understanding of flower water use strategies among these plant taxa.

Study site and species

This study was conducted at Xishuangbanna Tropical Botanical Garden (XTBG), which is situated in Yunnan Province, China, at coordinates 21°55′N and 101°15′E, with elevation of 570 m. The XTBG is located at the northern boundary of the Southeast subtropical zone and experiences a tropical monsoon climate. This region receives an average of 1859 hours of sunshine annually, with a mean annual temperature of 22.7°C and rainfall of 1447 mm. The area exhibits a distinct dry season from November to April and a rainy seasons from May to October.

We selected 24 species from the ANA grade, magnoliids, monocots, and eudicots for the study. These species were chosen from 17 families to ensure a high phylogenetic diversity (Fig. 1 and Table S1). All plants were cultivated outdoors under well-watered conditions at the XTBG. The selected species are widely distributed in southern Yunnan, and they provided a sufficient number of samples to measure their floral traits. Floral trait data for the 12 herbaceous plants are the same as shown by Ke et al. (2023), which we conducted at the same site but addressing significantly different scientific questions. All other data were newly collected for this study.

Floral longevity and nutrient concentrations

We randomly marked 10–20 flower buds for each species and measured floral longevity (FL; days) from the moment flowers opened until the corolla fell off or became wilted and discolored (Roddy et al., 2021).

To determine carbon and nitrogen contents, healthy and fresh flowers were first oven-dried at 70 ℃ for at least 48 hours, ground into powder, and passed through a 60-mesh sieve. The flower carbon (C_flower; g kg⁻¹) and nitrogen (N_flower; g kg⁻¹) concentrations per mass were determined by a C-N elemental analyzer (Vario MAX CN, Elementar Analysensysteme GmbH, Hanau, Germany). The C/N ratio of flowers (C/N_flower) was then calculated. Flower construction cost (CC_flower; g kg⁻¹) was estimated as (5.39 C_flower ‒ 1191)/1000 (Vertregt and Penning De Vries, 1987).

Petal anatomy and morphology 

We prepared petal cross-sections using the paraffin embedding method and captured images using a light microscope (Leica Microsystems Ltd., Leica DM2500, Wetzlar, Germany). For each species, six flowers were selected for measurements. The petal thickness (PT; μm), adaxial epidermis thickness (Ada_petal; μm), and abaxial epidermis thickness (Aba_petal; μm) were measured using the ImageJ software (National Institutes of Health, Bethesda, MD, USA).

We utilized a flatbed scanner (Epson Perfection V850 Pro) to determine the floral area (FA; cm²). Subsequently, we measured the fresh weight (FW_flower; g) of recently opened flowers. These flowers were then immersed in water for several hours, and the saturation weight (SW_flower; g) was recorded. Finally, the flowers were dried at 70 ℃ for 48 hours for a constant dry weight (DW_flower; g). Flower mass per area (FMA; g m⁻²) was calculated as DW_flower/FA, flower dry matter content (FMDC; g⁻¹) was calculated as DW_flower/SW_flower, and the relative water content of flowers (RWC_flower; %) was calculated as (FW_flower ‒ DW_flower)/(SW_flower ‒ DW_flower).

We used the flatbed scanner at 3200 dpi and the light microscope to measure petal veins and stomatal density. Pictures of petal stomata from 3‒6 individuals of each species were collected, with 5‒10 fields of view taken for each section. The vein density (D_{vein, petal}; mm. mm⁻²) was the total vein length per surface area, and the stomatal density (SD_petal; no. mm⁻²) was the total number of stomata per surface area. The epidermal cell size (ECS_petal; μm²) was the field of view area divided by the number of epidermal cells on that field of view.

Pedicel xylem anatomy and hydraulic efficiency

To examine xylem anatomy and hydraulic efficiency, we gathered three to six pedicels from various individuals for each species. We prepared cross-sections of the pedicels using the paraffin-embedded method, and then each part of the section was photographed with the light microscope. Finally, we utilized the ImageJ to measure the cross-sectional area of the pedicel, the total number of vessels, and the long and short axes of vessel lumens.  The pedicel hydraulically weighted vessel diameter (D_{h, pedicel}; μm) was calculated according to Poorter et al. (2010).

Mean diameters of the ten largest and smallest vessels for each pedicel were defined as the maximum (D_{max, pedicel}; μm) and minimum (D_{min, pedicel}; μm) vessel diameter, respectively. Pedicel vessel density (VD_pedicel; no. mm⁻²) was calculated as the number of vessels per unit pedicel cross-section area. Pedicel vessel lumen fraction (VLF_pedicel; %) was defined as the total vessel lumen area divided by the total cross-sectional area of the pedicel. Pedicel vessel wall reinforcement (VWR_pedicel) was calculated as the square of the ratio of pedicel vessel wall thickness (VWT_pedicel; μm) to vessel diameter (D_{i, pedicel}).

Measurement of water loss and drought tolerance in flowers

To examine the time required for drying saturated flowers to 70% relative water content (T_{70, flower}; h), we soaked freshly collected flowers in water for 4-8 hours and then recorded the SW_flower. In order to maintain constant conditions, the flowers were placed in a thermostatic incubator set at 25 ℃ and a relative humidity of 65%. We weighed the flowers every hour until no further change in weight was observed. Finally, the flowers were dried at 70 ℃ for 48 hours to obtain the dry weight (DW_flower). T_{70, flower} was calculated by analyzing the relationship between the relative water content and the time interval of each measurement, which is considered the point at which physiological damage occurs (Lawlor and Cornic, 2002; Hao et al., 2010; Zhang et al., 2015).

We used a Model 5600 VAPOR pressure osmometer (ELITechGroup Inc, Logan, USA) to measure the flower osmotic potential (ψ_{osm, flower}). Samples from the center of the midrib and margin were collected using a perforator. The flower discs, wrapped in aluminum foil, were immersed in liquid nitrogen, and placed in the osmolality chamber. The osmolality was measured repeatedly until equilibrium was reached (with a difference of less than three mmol kg^-1 between two measurements; Bartlett et al., 2012). We recorded the equilibrium osmolality in mmol kg⁻¹ and converted it to osmotic potential in MPa by multiplying it by -0.002437 m³ MPa mol⁻¹ using the Van't Hoff relationship (Laughlin et al., 2020). Flower water potential at the turgor loss point (ψ_{tlp, flower}; MPa) was calculated using the equation provided by Bartlett et al. (2012)

Statistical analysis

We first calculated the average value of all floral traits in each species. Before analysis, we performed a log₁₀-transformation on all data to improve the normality and homoscedasticity. We used independent-samples t-tests to analyze the differences in floral traits among basal angiosperms (ANA grade and magnoliids) and derived groups (monocots and eudicots) by the t-test function in the 'stats' package. Since most selected ANA grade and monocots are herbaceous plants, while most magnoliids and eudicots are woody, we considered the influence of growth forms on floral trait divergence. We utilized two-way ANOVA to test whether each floral trait differed among major evolutionary clades and growth forms and evolutionary clades of varying growth forms. To quantify the relationship between paired floral traits, we used the corr.test function in the 'psych' package. We used phylogenetically independent contrasts (PIC) correlations to test for co-evolution of paired floral traits (Felsenstein, 1985), which was performed using the pic function in the 'picante' package (Kembel et al., 2010). We constructed phylogenetic relationships for the selected species using the phylo.maker function in the 'V.PhyloMaker2' package (Jin and Qian, 2022). To test whether the relationships between floral traits differed among evolutionary clades, we conducted the standardized major axis analysis (SMA) using the sma function of the 'smatr' package (Warton et al., 2012). We also performed a principal component analysis (PCA) by the prcomp function in the 'stats' package to determine the position of basal angiosperms with monocots and eudicots in the multivariate floral trait space (Oksanen et al., 2019). We used the permutation multivariate analysis of variance (PERMANOVA) to assess whether plants of different evolutionary clades and different growth forms occupy different positions in the multivariate trait space (number of permuted datasets = 999) by the adonis function in the 'vegan' package (Anderson, 2001). All analyses were performed in R v.4.2.3 (R Core Team, 2023).