Flower and leaf samples of the 10 species in this study ( Figure 1 and Table 1 ) were collected on the campus of Guangxi University, Nanning (Guangxi, China, 22°50′N 108°17′E), which has a subtropical monsoon climate with a mean annual temperature of 21.8°C and a mean annual precipitation of 1,290 mm. Three to five randomly selected individuals per species were selected for sampling. On each plant, a sun-exposed branch with leaves and flowers was cut and immediately placed in a bucket with water in the evening or early morning and transported back to the laboratory on campus.

All measurements were made on a fully expanded, healthy, sun-exposed branch with flowers and leaves for each of the 3-5 individuals sampled per species. From each leaf or flower, approximately 1-cm² sections of lamina were excised, avoiding the margin and midrib. These sections were cleared in a 1:1 solution of H₂O₂ (30%) and CH₃COOH (100%), then incubated at 70°C until all pigments had been removed. Sections were then removed from this solution and rinsed in water for 3 minutes, then the epidermises separated with forceps from the mesophyll and veins, allowing the upper and lower epidermises to be stained and mounted separately. To increase contrast, all samples were stained with Safranin O (0.5% w/v in water) for 5 min and Alcian Blue (1% w/v in 3% acetic acid) for 20 secs - 1 min, then washed in water and mounted on microscope slides.

Cross-sections of petals and leaves were made with a sliding microtome (RM225, Leica Inc., Germany) with a tissue thickness of 35 µm. Cross-sections with the same thickness were also made of peduncles and petioles. Sections were bleached for 10 min, rinsed in water, and then stained with Safranin O (0.5% w/v in water) for 5 min and with Alcian Blue (1% w/v in 3% acetic acid) for 20 secs - 1 min, rinsed, and then mounted on glass slides.

Images were taken at 5x, 10x, or 20x magnification, which had fields of view of approximately 3.99 mm², 0.89 mm², and 0.22 mm², respectively, using a compound microscope outfitted with a digital camera (DM3000, Leica Inc., Germany). Both abaxial (lower) and adaxial (upper) leaf and petal surfaces were imaged for all species to determine whether they were amphistomatous. In subsequent analyses, we used sum of abaxial and adaxial stomatal densities for comparisons. We found no stomata on petals of Catharanthus roseus and Rosa sp.

All anatomical measurements from images were made using ImageJ (Rueden et al., 2017). From images of paradermal sections, vein density (D _v) was measured as the total length of leaf or petal vascular tissue per mm² of leaf area or petal area, stomatal density (D _s) was measured by counting the number of stomata in the image and dividing by the area of the field of view, stomatal size (S _s) (comprising a pair of guard cells) was directly measured on at least five stomata per image. Partial stomata and epidermal cells were included in the density counts if visible along the top and left borders of the photomicrographs and discarded if visible along the bottom and right borders (Carins Murphy et al., 2017). Leaf and petal thicknesses were directly measured from the cross-section images.

Vessel double wall thickness (T _w) was measured on at least 10 pairs of connected vessels per image from cross-sections of peduncles and petioles. Mean hydraulically weighted vessel diameter (D _h) for each species was calculated as (Tyree and Zimmermann, 2013):

where D is the equivalent circular diameter of a vessel whose area was calculated from its long and short diameters and N is the number of vessels measured. The D _h is biased towards wider vessels that conduct the majority of water according to the Hagen–Poiseuille law. Average vessel frequency (VF) was calculated per image by dividing the total number of vessels by the image area. For each species, we calculated the theoretical hydraulic conductivity of as (Rakthai et al., 2020):

where ρ is fluid density (assumed to be 998.2 kg m^-3 at 20°C) and η is viscosity of water (1.002 x 10^–9 MPa s^-1 at 20°C).

Upon returning samples to the lab, peduncles and petioles were immediately cut into small segments and placed in 100 ml of 5% FAA fixative (90:5:5 ratio of 70% ethanol, acetic acid, formaldehyde) at room temperature (25°C) to prevent expansion or shrinkage. Longitudinal sections of the segments were made with a sliding microtome (RM225, Leica Inc., Germany) at a thickness of 2-3 mm. The sections were fixed to aluminum sample holders with an electron-conductive carbon adhesive tape (Nisshin EM Co. Ltd., Tokyo), air-dried for 12 h at room temperature, and coated with gold using a sputter coater (Cressington 108Auto) for 40 secs at 0.08 mA to get a 20-nm-thick gold layer, under an argon atmosphere. A conventional scanning electron microscope (FEI Quattro S, US) with a voltage of 2 kV was used to visualize intervessel pit parameters according to standard protocols (Jansen et al., 2009; Lens et al., 2011; Zhang et al., 2021).

ImageJ (Rueden et al., 2017) was used to determine the following intervessel pit characteristics ( Table 2 ): intervessel pit aperture surface area (A _pa), intervessel pit surface area or intervessel pit membrane surface area (A _pit), pit aperture longest diameter (D _pal), pit aperture shortest diameter (D _pas), pit aperture shape (R _pa = D _pal/D _pas), pit membrane longest diameter (D _pml), pit membrane shortest diameter (D _pms), pit shape or pit membrane shape (R _pit = D _pml/D _pms) and pit density (D _p). Mean values of these intervessel pit traits were calculated from at least 50 measurements from SEM images of various intervessel walls per individual.

Shoots with leaves and flowers were collected from at least three individuals per species at night or at predawn and transported back to the laboratory. In the lab, all shoots were recut underwater to rehydrate for at least 2 h and covered with a black plastic bag during equilibration. Initial water potentials were checked and always close to -0.1 MPa. Pressure–volume curves were constructed for each sample by repeatedly measuring the bulk water potential using a pressure chamber (0.01 MPa resolution; PMS Instruments, Albany, OR, USA) and the mass to determine the relationship between water potential and water content following standard methods (Scholander et al., 1965; Tyree and Hammel, 1972; Sack and Pasquet-Kok, 2011; Roddy et al., 2019; Jiang et al., 2022). Prior to each water potential measurement, samples were enclosed in humidified plastic bags for about 20 min to allow equilibration. The pressure chamber was kept humidified with wet paper towels to prevent evaporation during the water potential measurement. After water potential measurement, the sample was weighed on a balance (± 0.0001g, model ML204T; Mettler Toledo). At the end of measurements, samples were oven-dried at 70°C for at least 72 h before determining dry mass. Because measuring flower surface area is difficult after turgor loss, pressure–volume parameters were expressed on a dry mass basis, according to previous analyses (Roddy et al., 2019). From these pressure-violume curves, we calculated saturated water content (SWC), absolute capacitance (C_T), water potential at turgor loss point (Ψ_tlp), and osmotic potential at full turgor (Ψ_sft) ( Table 2 ).

Shoots with leaves and flowers were collected at night from at least five individuals per species, recut underwater, and rehydrated over night while covered with a black plastic bag. Leaf and flower samples were excised in the morning, including the petiole or peduncle. Immediately following excision, their cut ends were sealed with glue and the entire organ was weighed every 10 min using an electronic balance ( ± 0.0001g, model ML204T; Mettler Toledo) in a dark room. The room was equipped with an air-conditioning to control the temperature and humidity, and samples were hung in front of a large fan as they desiccated. The velocity of air flow was high enough to physically move the samples. A small temperature and humidity sensor was kept near the samples, and temperature (T) and relative humidity (RH) were recorded manually each time a sample was weighed. After ten measurements, samples were scanned to determine projected area and then oven-dried at 70°C for 72 hours before determining dry mass.

Minimum diffusive conductance (g _min) was calculated as (Bourbia et al., 2020):

where WL is the water loss rate (mmol m^-2 s^-1) calculated as the slope of mass (g) over time (s) and normalized by the projected area (m²) or dry mass (g) of each organ; P_atm is the atmospheric pressure (101.3 kPa); VPD is the vapor pressure deficit determined using the Arden Buck equation (Buck, 1981).

All statistical analyses were conducted in R (v. 4.0.3) (Core Team, 2022). Paired t-tests were used to determine differences between flowers and leaves. Differences of pit area and density among different plant lineages were tested through one-way ANOVA. We used linear regression and standard major axis (SMA) regression (R package ‘smatr’) to determine the relationships between traits (Warton et al., 2012). Principal component analysis (PCA) was carried out on centered and scaled trait data using the ‘vegan’ package. A phylogenetic tree was built using the R package ‘V.PhyloMaker’ and phylogenetic independent contrasts (PICs) were calculated using the ‘pic’ function in the R package ‘ape’ and PIC correlations tested using linear regression. All statistical tests were considered significant at P< 0.05. In order to contextualize our measurements of inter-conduit pit traits on leaves and flowers, we compiled published data reporting pit membrane surface area (A _pit) and pit density (D _p) for a diverse set of vascular plants ( Supplementary Table 1 ). We used this broad dataset to examine how pit membrane area and pit density vary among lineages and organs.

Minimum diffusive conductance (g _min) was significantly higher in flowers than in leaves, whether it was normalized by dry mass (t = 5.48, P<0.001) or by projected area (t = 4.88, P<0.001) ( Figures 2A, B and Supplementary Table 2 ). In addition, traits from pressure-volume curves were also significantly higher in flowers than in leaves (P< 0.01) ( Figures 2C–F and Supplementary Table 2 ).

SEM images were used to examine intervessel pits in peduncles and petioles ( Figure 1B ). Compared to petioles, peduncles had significantly smaller pit membrane diameters D _pms (t = -2.36, P< 0.05) and D _pml (t = -2.86, P< 0.01) and smaller pit area A _pit (t = -3.09, P< 0.05), as well as differences in pit aperture shape R _pa (t = -2.16, P< 0.05) ( Figures 2G–J and Supplementary Table 2 ). However, pit aperture diameters D _pas (t = -0.43, P > 0.05) and D _pal (t = -1.49, P > 0.05) and pit aperture area A _pa (t = -1.15, P > 0.05) were not significantly different between petioles and peduncles ( Supplementary Table 2 ). Pit membrane shape R _pit (t = 0.41, P > 0.05) and pit density D _p (t = 2.02, P > 0.05) were similar in peduncles and petioles ( Supplementary Table 2 ). Peduncles and petioles differed significantly in the size of the pit membranes despite having similar pit aperture sizes ( Figure 2 and Supplementary Table 2 ).

In leaves, g _min,mass was positively correlated with Ψ_tlp (R ² = 0.56, P = 0.013) and Ψ_sft (R ² = 0.50, P = 0.022), which remained significant after accounting for shared evolutionary history ( Figures 3A, B and Table 3 ). No similar relationships between g _min,area or g _min,mass and C _T in leaves ( Figures 3C, D ) or g _min,mass and Ψ_tlp or Ψ_sft in flowers ( Figures 3E, F ) were found. g _min,area was positively correlated with C _T in flowers (R ² = 0.50, P = 0.034, Figure 3H ). The relationships between C _T and both g _min,mass and g _min,area were significant after accounting for shared evolutionary history ( Table 3 ).

Pit density D _p was negatively correlated with pit membrane diameters D _pms (R ² = 0.68, P< 0.001) and D _pml (R ² = 0.71, P< 0.001) and with pit area A _pit (R ² = 0.61, P< 0.001), as well as with pit aperture diameters D _pas (R ² = 0.58, P< 0.001) and D _pal (R ² = 0.5, P< 0.001) and pit aperture area A _pa (R ² = 0.47, P< 0.001) ( Figure 4 ). These correlations remained significant after accounting for shared evolutionary history (all P< 0.05, Table 3 ). Comparing these data to previously published data from stems and leaves of a broad sampling of angiosperms, gymnosperms, and ferns ( Supplementary Table 1 ) showed that the negative correlation between D _p and A _pit was common ( Supplementary Figure 1 ), with gymnosperms exhibiting larger pits that occur at lower density ( Table 4 and Supplementary Figure 1 ). A _pit and D _p in flowers and leaves measured here were within the range reported previously for angiosperms and differed significantly from only gymnosperms ( Table 4 ).

In petioles, like in other species (Lens et al., 2011; Mrad et al., 2018), K _th was positively correlated with D _pas (R ² = 0.45, P = 0.004), D _pal (R ² = 0.58, P = 0.011), and A _pit (R ² = 0.74, P = 0.001), but in peduncles these relationships were not significant, mainly because of the relatively constant D _pas and D _pal ( Figures 5A-C ). These correlations were statistically similar after accounting for shared evolutionary history (P< 0.05, Table 3 ). D _pal was positively correlated with D _h (R ² = 0.39, P = 0.003) and with T _w (R ² = 0.45, P = 0.001) in both petioles and peduncles ( Figures 5D, E ). These correlations remained significant only in leaves after accounting for shared evolutionary history (P< 0.05, Table 3 ). The positive relationships between D _pal and D _s (R ² = 0.25, P = 0.005) ( Figure 5G ), A _pit and T _w (R ² = 0.49, P = 0.025) ( Figure 5F ) became non-significant in leaves after accounting for shared evolutionary history ( Table 3 ).

In peduncles, R _pa was positively correlated with D _h (R ² = 0.43, P = 0.04), K _th (R ² = 0.43, P = 0.038), SWC (R ² = 0.53, P = 0.018), Ψ_sft (R ² = 0.54, P = 0.016), Ψ_tlp (R ² = 0.50, P = 0.022), C _T (R ² = 0.43, P = 0.041), FT (R ² = 0.54, P = 0.015) ( Figures 6A-G ). However, none of these correlations remained significant after accounting for shared evolutionary history ( Table 3 ). R _pa was unrelated to any of these hydraulic traits in petioles, but R _pa was positively correlated with leaf thickness after accounting for shared evolutionary history (R ² = 0.72, P< 0.01) ( Table 3 ). Only R _pit was negatively correlated with D _v (R ² = 0.42, P = 0.041) ( Figure 6H ), even after accounting for shared evolutionary history (R ² = 0.72, P< 0.01) ( Table 3 ).

Phylogenetic independent contrast correlations (PIC) were made between all 24 traits measured in both flowers and leaves. Positive correlations of D _pms (R ² = 0.62, P = 0.012), D _pml (R ² = 0.84, P = 0.001), A _pit (R ² = 0.86, P< 0.001), D _pas (R ² = 0.91, P< 0.001), D _pal (R ² = 0.79, P = 0.001), A _pa (R ² = 0.93, P< 0.001), D _p (R ² = 0.78, P = 0.002) between flowers and leaves were found ( Figures 7A–G ), and species with larger pit membranes and pit apertures in petioles also had larger pit membranes and pit apertures in peduncles ( Supplementary Figure 2 ). Interestingly, pit membrane (R ² = 0.09, P = 0.444) and pit aperture shape (R ² = 0.14, P = 0.33) showed non-significant relationships after accounting for shared evolutionary history ( Figures 7H, I ).

There were positive correlations in traits between organs for S _s (R ² = 0.50, P = 0.048), D _v (R ² = 0.41, P = 0.047), T _w (R ² = 0.49, P = 0.024), and g _min,mass (R ² = 0.46, P = 0.031) ( Supplementary Figure 3 ), which became non-significant after accounting for shared evolutionary history ( Table 5 ). Meanwhile some traits were not correlated among organs and remained uncorrelated even after accounting for shared evolutionary history: g _min,area (R ² = 0.13, P = 0.336), SWC (R ² = 0.07, P = 477), K _th (R ² = 0.06, P = 0. 523), Ψ_sft (R ² = 0.44, P = 0. 051), C_T (R ² = 0.02, P = 0.721), D _s (R ² = 0.13, P = 0.336), and LT/FT (R ² = 0.04, P = 0.626) ( Table 5 ). Among physiological traits, only Ψ_tlp exhibited correlated evolution among flowers and leaves (R ² = 0.54, P = 0.024) ( Table 5 ).

The principal components analysis using all 24 traits revealed that the first two principal components explained 37.20% and 24.06% of the total variation, respectively. The first PC was driven by D _h, K _th, and some pit characters, including D _p, D _pms, and A _pit. The second PC was largely driven by pressure-volume parameters, including SWC, C_T, Ψ_tlp, Ψ_sft, as well as anatomical traits, including D _v, D _s, and LT. Flowers and leaves largely differed in the regions of trait space they occupied ( Figure 8A ).

Using only the pit traits in a principal components analysis revealed that the first two principal components explained 86.55% of the total variation among species and organs. The first PC (66.85%) was driven primarily by D _p versus all of the other pit traits except R_pit and R_pa . The second PC (19.70%) was driven primarily by R _pit and R _pa. There was a high level of overlap among flowers and leaves in the regions of pit trait space they occupied ( Figure 8B ).

In order for terminal organs to avoid desiccation, water loss must equal water supply, at least over diel timescales. In leaves, which maintain relatively high transpiration rates, the need to maintain water balance has resulted in coordinated evolution of key anatomical traits that influence both liquid water supply and water vapor loss, particularly leaf vein density (D _v) and stomatal density and size (Sack et al., 2003; Brodribb et al., 2013; Simonin and Roddy, 2018; Zhang et al., 2018). While similar coordination between veins and stomata has been observed in flowers (Zhang et al., 2018), flowers often have few or no stomata, meaning that other traits, such as g _min, may be more important to regulating water balance (Roddy et al., 2016; Roddy, 2019). Furthermore, in both flowers and leaves, higher water contents and hydraulic capacitance can buffer water potential declines and lengthen the time required to reach steady state transpiration when water supply equals water loss (Simonin et al., 2013; Roddy et al., 2018; Roddy et al., 2019). Because flowers can have higher g _min than leaves ( Figure 2 ), we predicted that there may be coordination between g _min and hydraulic capacitance, which would indicate that higher hydraulic capacitance can compensate for higher g _min in flowers. Across organs, g _min and hydraulic capacitance were correlated even after accounting for shared evolutionary history ( Figure 3 and Table 3 ), with flowers having both higher g _min and higher capacitance than leaves ( Figure 2 ). These patterns suggest that multiple traits and hydraulic strategies may be employed to maintain water balance among leaves and flowers.

In the absence of high hydraulic capacitance to buffer water potential declines, high g _min may cause water potentials to decline enough to initiate xylem embolism in flowers before leaves (Bourbia et al., 2020). If this were the case, we would predict that to prevent embolism in flowers, intervessel pit traits may have experienced selection to reduce embolism vulnerability. However, there were overall relatively small differences in intervessel pit traits between petioles and peduncles ( Figure 2 ). One possible explanation is that intervessel pit traits may be shielded from selection by high hydraulic capacitance in flowers that allows g _min to be high without causing water potential declines and embolism spread. Because intervessel pit traits also influence hydraulic conductance (Choat et al., 2005; Ellmore et al., 2006; Hacke et al., 2006; Lens et al., 2011; Jacobsen et al., 2016), differences in intervessel pit traits between leaves and flowers may be due to divergent selection on hydraulic conductance among leaves and flowers. However, while flowers generally have relatively low hydraulic conductance, they are not necessarily outside the range of hydraulic conductance of leaves (Roddy et al., 2016), further suggesting that the strength of selection due to hydraulic efficiency acting on pit traits may be relatively weak.

In some cases, leaves and flowers exhibited similar coordination between intervessel pit traits despite the large morphological, anatomical, and physiological differences between these organs. We found similar coordination between A _pit and A _pa in both leaves and flowers ( Figure 4 ), with larger A _pit being associated with higher theoretical hydraulic conductivity K _th in leaves ( Figure 5 ). It is worth nothing that K _th incorporates only vessel traits and not intervessel pit traits, so coordination between K _th and A _pit and A _pa suggests that variation in vessel size and density are linked to intervessel pit variation. A broad sampling of vascular plants ( Supplementary Figure 1 ) revealed strong coordination between A _pit and pit density (D _p), due largely to packing constraints similar to those elucidated for stomata on the leaf surface and mesophyll cells inside the leaf (Franks and Beerling, 2009; Théroux-Rancourt et al., 2021; Borsuk et al., 2022; Jiang et al., 2023). Compared to other vascular plants, angiosperm A _pit and D _p were closer to the theoretical packing limit, which may be important to increasing hydraulic efficiency of angiosperm xylem regardless of other xylem traits. Coordination between A _pit and D _p was also found among flower peduncles and leaf petioles, similar to other angiosperms, regardless of the fact that previously published data were taken from both stems and leaves ( Supplementary Figure 1 and Supplementary Table 1 ). Previous studies have shown that larger A _pit is associated with larger A _pa, leading to higher hydraulic conductivity (Orians et al., 2004; Wheeler et al., 2005; Lens et al., 2011; Johnson et al., 2016). Similarly, we found that A _pit was positively linked to K _th in leaves but not in flowers (noted that A _pit was positively correlated with D _h but not VF in leaves, Table 3 ), suggesting that coordination in pit traits and vessel dimensions and packing were decoupled in flowers ( Figure 5 ). Thus, despite obeying similar biophysical packing principles as vegetative organs–i.e. stomatal and vein densities (Zhang et al., 2018)–flowers can deviate in other traits that also influence hydraulic performance.

From the hydraulic efficiency perspective, plants that have larger diameter vessels will have higher hydraulic efficiency (Hargrave et al., 1994; Christenhusz et al., 2011; Liu et al., 2019), while larger conduit diameter, larger pit membrane area, and larger pit aperture area, will be expected to decrease hydraulic safety (Pittermann et al., 2010; Lens et al., 2011; Brodersen et al., 2014; Jacobsen et al., 2016). Conduit wall thickness is thought increase hydraulic safety (Hacke et al., 2001; Brodribb and Holbrook, 2005). In our results, both pit membrane and pit aperture traits were correlated with K _th and D _h in leaves, no such correlations were found in flowers ( Table 3 ). These results may indicate that leaves increase hydraulic efficiency with larger vessels and A _pit, but they may increase hydraulic safety through thicker vessel walls and more elliptical pits ( Table 3 ). On the other hand, high hydraulic capacitance of flowers prevents water potential declines may relax the selective pressure of intervessel pit traits for hydraulic efficiency and safety. In general, elliptically shaped pit apertures are associated with greater embolism resistance, with more cavitation-resistant species exhibiting narrower and more elliptical pit apertures (Lens et al., 2011; Scholz et al., 2013; Jacobsen et al., 2016). Consequently, angiosperms adapted to dry environments might have smaller conduit diameters and thicker, denser, smaller, and more elliptical pit apertures (Wheeler et al., 2005; Hacke et al., 2007; Jansen et al., 2009; Lens et al., 2011; Scholz et al., 2013). While intervessel pit traits might influence both hydraulic safety and efficiency, none of the pit traits were correlated with hydraulic traits in flowers ( Table 3 ). It is highly likely, therefore, that traits exhibiting greater differences between leaves and flowers (e.g. pressure-volume traits) may be more important to flower water balance than intervessel pit traits.

Flowers have been shown to exhibit high diversity in hydraulic traits with higher water content and higher hydraulic capacitance than leaves (Roddy et al., 2019). We found similar patterns in our data, with flowers exhibiting greater variation in g _min and pressure-volume traits than leaves ( Figure 2 ). However, flowers exhibited less variation in pit traits than leaves ( Figure 2 ), although there were some clear differences in intervessel pit traits between leaves and flowers. This was further validated by the PCA results ( Figure 8 ), in which R _pit and R _pa loaded on the same axis as the pressure-volume traits, in contrast to all other intervessel pit traits, which were orthogonal to other hydraulic traits except D _h and K _th. Further corroborating the role of R _pit and R _pa in causing the divergence in hydraulic strategies between leaves and flowers, R _pit and R _pa exhibited no correlated evolution between leaves and flowers, in contrast to all other intervessel pit traits ( Figure 7 ). Taken together, these results suggest that while most of the hydraulic differences between leaves and flowers is due to stomatal and vein anatomy and pressure-volume traits, differences in pit and pit aperture shape may also signify important differences.