High diversity of anatomical and hydraulic strategies of dominant Amazonian savanna woody species

Amazonian savannas are isolated patches of open habitats found within an extensive matrix of tropical rain forest. The climate of Amazonia is transitioning to a state that favors savannas over forests. Still, the hydraulic functioning of savanna species in Amazonian vegetation remains unknown and unexplored. We measured 22 leaf, wood, and hydraulic traits, including embolism resistance (as P 50 ) and isotope-based water use eciency (WUE), for the seven dominant woody species (75% of biomass) of Amazonian savanna. We found wide variation in resistance to embolism (-1.66 ±0.06 MPa and -5.07±0.46 MPa) with species investing in at least three different hydraulic strategies based on anatomical adjustment: (1) Species with high G max and low water use eciency where leaf succulence may delay embolism onset; (2) Species that have low embolism resistance and high water use eciency by reinforcing the walls of longer vessel elements to prevent implosion; (3) Species with high G max and low water use eciency where short vessel elements and thick intervascular pit membranes may minimize the embolism spread across the plant. Our results suggest no unique dominant functional strategy among Amazonian savanna species illustrating the alternative ways to survive in a hot and seasonally dry environment. This study provides the rst data about hydraulic structure and functioning of woody species from Amazonian savannas. Our ndings showed wide variation in resistance to embolism and structural anatomy, suggesting no unique dominant functional strategy among Amazonian savanna species. The species diverge by investing in three different functional strategies for drought tolerance: (1) Species with high G max and low water use eciency where leaf succulence may delay embolism onset and high wood specic gravity enables more negative pressures on the xylem (Kielmeyera rubriora, Simarouba versicolor and Macairea radula); (2) Species that have low embolism resistance and high water use eciency by reinforcing the walls of longer vessel elements to prevent implosion (Norantea guianensis and Alchornea discolor); (3) Species with high G max and low water use eciency where short vessel elements and thick intervascular pit membranes may minimize the embolism spread across the plant (Maprounea guianensis and Parkia cachimboensis). some the in drought-induced embolism resistance observed locally. xylem embolism resistance in Amazonian savanna species. Our results demonstrate that species that a restrictive environment, such as Amazonian savannas, can exhibit different strategies of drought tolerance/avoidance resulting in the similar outcome of maintaining a favorable water balance that allow their persistence and co-occurrence. Distinct strategies to deal with drought permitted wide divergence regarding embolism vulnerability. The strategies that selected high xylem embolism resistance are explained by vessel element size and pit membrane thickness. However, the role of strategies related to drought avoidance should not be neglected in this ecosystem. Future climate predictions should aim to incorporate this range of strategies and could be used to test whether future climates will favor some strategies or still allow for a range of them to persist. Further studies and renement of model exercises are likely to improve our predictions on changing the species and trait composition of Amazonian savannas and also its impact of large scale ecosystem functioning.


Introduction
The Amazon Forest is the most diverse tropical forest on the planet and stands out as the main source of Neotropical biodiversity (Antonelli et al., 2018). Although the Amazon is well-known for its forests, a signi cant part of it is covered by natural grasslands and savanna-like vegetations, grouped under the broader term 'Amazonian savanna' (Pires and Prance, 1985). Amazonian savannas are isolated patches of open habitats found within the extensive matrix of Amazonian tropical forests, and possess oras distinct from that of the surrounding forest (Viana et al., 2016). Global vegetation models predict high vulnerability for Amazonian forests to drought (Huntingford et al., 2008) and even the transition to more open vegetation states (Cox et al., 2004). With deforestation and global warming causing increases in the duration of the dry season, the climate is changing to a state that favors savannas over forests (Nobre and Borma, 2009;Devecchi et al., 2020). However, the most likely scenario is that these transition processes will occur simultaneously, operating at multiple spatial and temporal scales, and interacting strongly. Thus, the knowledge about water use and drought vulnerability in savanna-like vegetations can be instrumental to understand and predict the fade of both, forests and savannas, in the future. . Importantly, taxonomic identity controls adjustments in the hydraulic structure that underlies both hydrological micro-habitat partitioning (Cosme et al., 2017) and acclimation to increasing drought (Bittencourt et al., 2020). In comparison, the functional anatomy and hydraulic functioning of savanna species in Amazonian vegetation remains unknown and unexplored.
Amazonian savannas differ from other savannas due to their high annual precipitation (~2,180 mm) and from nearby Amazonian forests due to their marked seasonality in rainfall and extreme nutrient-poor soils (Lloyd et al., 2009;2015). During periods of climatological water de cit, Amazonian savannas show a relatively higher atmospheric vapor pressure de cit (VPD) and temperature than nearby forests. Under water-restricted conditions, plants typically reduce the opening of their stomata, a strategy that signi cantly reduces water loss through transpiration, but at the same time limits CO 2 diffusion for photosynthesis (Galmés et al., 2013;Xiong et al., 2017). Thus, plants must constantly deal with the trade-off of maximizing CO 2 diffusion while minimizing water loss through transpiration (Mcadam and Brodribb, 2012). Evidence from plants growing under seasonal drought conditions, including other savannas, suggests co-variance between vessel attributes and stomatal regulation. For example, previous work relates leaf hydraulics, including turgor loss and leaf conductivity, to wood density in central Brazil savannas (e.g. Bucci et al., 2004). However, it is uncertain how those ndings can be extended to Amazonian savanna species as less than 50% of the species are shared between savanna oristic provinces in the Neotropics (Bridgewater et al., 2004). Also, the marked differences in species dominance among savanna provinces suggests that ecological, and perhaps functional, strategies that are advantageous in one region may not be so successful in the others.
A growing knowledge on how anatomical and physiological attributes coordinate to allow plants to adapt to drought shows general trends in the structure of the plants. Xeromorphic characteristics in secondary xylem include the presence of high frequency of vessels, narrower and shorter vessel elements, vasicentric tracheid, thicker ber walls, thicker intervascular pits, and well-de ned growth layers (Carlquist, 1977(Carlquist, , 2018Chave et al., 2009;Lens et al., 2011;Ziemińska et al., 2013). Fusion of wide vessel elements generally forms long vessels, and narrow vessel elements form short vessels (Tyree and Zimmermann, 2002;Nijsse, 2004;Jacobsen et al., 2012). Long vessels transport water more e ciently than short vessels because intervessel transport (ie transport through intervessel pits) largely contributes to resistance to water ow (Comstock and Sperry, 2000;Van Ieperen et al., 2000;Nijsse et al., 2001). On the other hand, short vessels are safer in case of drought, or mechanical damage, because embolisms generally cannot spread beyond the vessel in which they occur (Comstock and Sperry, 2000;Tyree and Zimmermann, 2002).
It is expected that higher frequency of short vessel elements offer greater resistance to water and air ow as a larger number of end walls must be trespassed (Hacke et al., 2006;Loepfe et al., 2007). Lateral ow mediated by the ultrastructural characteristics of the intervascular pits, especially those related to the pit membrane, can also play an important role in preventing the propagation of air bubbles along the vessel network (Choat et  . Together, those mechanisms show the potential for coordination of anatomical and hydraulic traits, but in which conditions this coordination happen is still unknown. Speci cally, we need to better understand how vessel element size and vessel diameter and density are related to the e ciency and safety of water conductivity (Zanne et al., 2010;Lens et al., 2011;Hacke, 2015;Gleason et al., 2016;Li et al., 2016) and what is the role of vessel wall thickness and the membrane of the intervascular pit in the resistance to xylem embolism (Hacke et al., 2001;Jacobsen et al., 2005).
At leaf level, observed patterns on the relationship between anatomical (e.g. cuticle thickness, leaf succulence) and hydraulic (e.g. stomatal conductance, water use e ciency) traits also suggests coordination. Different lines of evidence suggest that minimal transpiration water loss arises from leaks in two main locations: cuticle and stomata (Richardson et al., 2007;Bueno et al., 2019). The maximum stomata conductance is driven by a combination between stomata size and density (de Boer et al., 2016). The plant cuticle covers the outer cell walls of the leaf epidermis, serving to reduce water loss as well as protect against pathogens and UV damage (Schuster et al., 2017). Water restriction and high irradiance reducing isotopic discrimination and increasing the amount of 13 C incorporated in the plant's leaf tissue as stomata remain closed longer (Martinelli et al., 1998;Teixeira et al., 2018). From C (δ 13 C) isotopic ratio we can also estimate the water use e ciency (Martinelli et al., 1998;Vitória et al., 2018), which represents an important integrative trait of functions in the leaf scale. Water use e ciency can be affected by leaf succulency as water stored in the leaf tissues may delay stomata closure by supplying water to sustain hydraulic function (Rosado and De Mattos, 2007;Rosado et al., 2016). This allows the plant to reduce stomatal opening, consequently stomatal conductance (Niinemets, 2001 . Although stem characteristics, including wood density, P 50 and hydraulic conductivity, may not differ signi cantly between savanna and forest species; leaf properties, such as maximum conductivity and C, can be signi cantly higher in savanna trees compared to forest (Hao et al., 2008). This suggests that greater leaf hydraulic e ciency may be more critical than stem hydraulics in adapting to seasonality in savanna-like vegetations, but an integrated understanding on how water status control at leaf and stem level are coordinated is currently missing.
In this context, we combined studies of anatomical and hydraulic traits to understand the structure-function relationships of leaf and xylem in plants of Amazonian savannas. We measured 22 leaf, wood, and hydraulic traits of seven dominant woody species, making this the rst hydraulic, and the most comprehensive, study to date on anatomical strategies of Amazonian savanna plants. To understand trait-trait relationships among the focal species, we asked: (i) Do woody species of Amazonian savannas converge or diverge in their hydraulic strategies? Here we hypothesize that trees may adjust their wood and leaf anatomy in different ways to adapt to drought. This will result in the coexistence of several contrasting functional strategies coexisting under the same environmental conditions (Rosado et al., 2013;Dias et al., 2020).
(ii) How are hydraulic and anatomical traits related to each other? We expect some degree of integration between hydraulic and anatomical strategies between different organs, particularly between WUE cuticle thickness and maximum stomata conductance. At branch level, xylem resistance is expected to be higher in species with shorter, narrower and more frequent vessel elements and when pit membranes are thicker.
(iiWhat is the anatomical basis of leaf and branch hydraulic strategies? We hypothesized that different combination of traits at leaf and branch level could be in place to guarantee plant survival with plant anatomy and hydraulics interacting in many ways. We expect leaf succulence to be related to WUE. However, under a high transpiration demand we expect a coordination between leaf and branch hydraulic strategies that favors xylem safety. Together those strategies should be able to reduce the changes of embolism formation and spread that allow species occurrence in Amazonian savannas.

Study area
The study was conducted in a rocky Amazonian savanna community in the municipality of Nova Canaã do Norte in the state of Mato Grosso, Brazil (Figure 1), 10º53'98,7" S, 55º46'68,7" W). The soil of the study area is classi ed as Litholic Neosols that are poorly drained, dystrophic, alic, extremely acidic and sandy, and with low nutrient concentrations (Pessoa et al., 2021). The climate of the area is equatorial (Am) hot and humid, according to the Köppen classi cation, with temperatures varying from 20° C to 36° C with an annual average above 28° C (Alvares et al., 2013). Total annual rainfall can reach 2,180 mm, with two well-de ned seasons -a rainy season encompassing November, December, January, February, and March (1,180 mm) and a dry season encompassing June, July, August, and September (108 mm) -with the other months being considered transition periods (Alvares et al., 2013).

Data collection
The seven most abundant woody plant species in the study area were selected from vegetation representing 75% of the community plant biomass ( Table  1). The following morphological, anatomical, and hydraulic measurements were made for each species: leaf area, leaf succulence, leaf speci c mass, stomatal density, adaxial cuticle, maximum theoretical stomatal conductance and carbon isotopic composition (water use e ciency -WUE); theoretical hydraulic conductivity, vessel lumen area, vessel frequency, vessel element length (including the tips, Figure S1), lumen fraction, ray frequency and ray width; ber lumen, ber length, wood speci c gravity, vessel wall thickness, intervessel pits (often horizontal in tangential sections, Figure S2) and pit membrane thickness (resistance mechanism); and xylem embolism resistance and hydraulic safety margin (embolism vulnerability proxy). Trait sampling and measurements for the species are detailed below.

Leaf
Morphological measurements were made of samples from ve fully expanded leaves from three individuals of each species. Discs of 0.5 cm in diameter were removed from the median third of the leaves. The discs were hydrated for 24 h and dried with paper for subsequent determination of saturated mass (M saturated ) using a digital balance (AY220, Shimadzu), and thickness with a digital caliper (Stainless, Hardened). The hydrated discs were then placed in an oven set at 55 ºC for 72 h to obtain dry mass (M dry ). These parameters were used to calculate leaf succulence (M saturated -M dry /Areadisc) (Kluge and Ting, 1978) and leaf speci c mass (M dry /Areadisc) (Kluge and Ting, 1978). Leaf area was determined by digitally scanning the leaves used for the other morphological analyses and then measuring their area using ImageJ digital image processing system.
Leaf cross-sections were made in the middle of the leaf by freehand for measurements of leaf anatomical attributes and to observe the adaxial cuticle.
Stomatal density (mm 2 ) and stomatal pore length (µm) were determined from the analysis of images of epidermis dissociated by the Franklin method (Franklin, 1945). Theoretical maximum stomatal conductance was calculated from the relationship between stomatal pore density and size (de Boer et al., 2016), according to the following equation: where G max is maximum stomatal conductance (mm.s -1 ), D is stomatal density (mm²) and L is stomatal length (µm).

Scanning electron microscopy
Two leaves were selected for each species for investigation of stomatal morphology on the abaxial leaf face. Fragments of the median third of the leaf blade were xed in an aqueous solution of 2.5% glutaraldehyde, 4% formaldehyde, and 0.05M sodium cacodylate buffer at pH 7.2 (Karnovsky, 1965
We used a maceration method for the measurements of vessel element length (including tips) and ber length. Maceration of branch material followed Franklin (1945). Small branch fragments were removed from each sample and placed in bottles containing a macerating solution of glacial acetic acid and hydrogen peroxide (1:1). The bottles were then sealed and placed in an oven at 60 °C for 24 h or until the complete dissociation of cells. The material was then washed in distilled water, stained with 1% aqueous Safranin, and mounted on semi-permanent slides with glycerin. Imperforate tracheary elements were not observed in macerations for any of our samples, as it is common among Angiosperm species.
Quantitative analysis was performed using 12 slides per individual. All descriptions, counts and branch cellular measurements followed IAWA Committee standards (1989). Permanent and semi-permanent slides were analyzed using a light-eld light microscope (Axioplan, ZEISS, USA), with image capture via a coupled camera (Power shot A640, CANON, USA).

Transmission electron microscopy
Two branches of each species were selected for analysis of the intervascular pit membrane. Branch fragments were xed in modi ed Karnovsky solution (Karnovsky, 1965 modi ed by Da Cunha et al., 2000). Post-xed in 1% osmium tetroxide and 0.05M sodium cacodylate buffer 2 h at room temperature, and then dehydrated in an increasing acetone series and in ltrated and embedded with epoxy resin (Epon®). Ultrathin sections (80 nm) were made using an ultramicrotome (Reichert Ultracuts Leica Instruments) with a diamond knife (Diatome®), which were collected in copper grids (300 mesh) and contrasted with 1.0% alcoholic uranyl acetate, followed by 5.0% aqueous lead citrate (Reynolds, 1963). Ultrastructure analysis of pit membranes was performed using a JEM 1400 Plus JEOL transmission electron microscope at a voltage of 80 kV, with 20 observations per individual. Measurements of anatomical attributes (Table 2) were performed using Image Pro-Plus 4.0 digital image processing system. Pit membrane thickness was measured cuts made in the sapwood at three different points, later averaged to represent the sample pit membrane thickness ( Figure S4). All measurements for each species were made from three individuals with similar height and two branches between 1.5 and 2 cm diameter per individual.
Wood speci c gravity Wood speci c gravity was calculated by rst measuring the fresh volume of wood samples by displacement of a water column (Williamson and Wiemann, 2010). Samples were immersed in a beaker containing water on top of a digital balance and sample volume was converted from the weight of displaced water (e.g., 1g = 1 cm 3 ). Dry mass was obtained by drying the samples in an oven at 105 °C for 72 h. Wood speci c gravity was then calculated as: where WSG = wood speci c gravity (g.cm -3 ), Dm = dry mass, and Dv = displaced volume.
Theoretical hydraulic conductivity Theoretical hydraulic conductivity (K th ) was calculated for each sampled individual from vessel lumen area using the Hagen-Poiseuille equation: Embolism resistance and hydraulic safety (P 50 and HSM) Branches of 1.5 to 2.0 m in length were collected at dawn for assessing embolism resistance and hydraulic safety (P 50 and HSM). Ten to 15 cm long segments were cut from the base of each branch under water and were allowed to rehydrate for 12 h, keeping them covered and sealed by black plastic bags. Hydraulic measurements were then made on the distal end of each branch to ensure there were none arti cially embolized vessels in the measured sample. All samples used for hydraulic measurements were from rst or second-order branches that were 30-55 cm in length and 2-4 cm in diameter and were cut under water with a sharp blade before connecting to the apparatus to ensure that all vessels were open. We measure the water potential of the leaves after 12 hours of hydration, taken to represent the timepoint when transpiration is at its minimum and the water potential of the plant is closest to equilibrium with that of the soil. We also determined midday water potential (Ψ md ), to capture the minimum Ψ of the plant in the dry season ( Figure S5), leaf water potentials were measured using a pressure chamber (Model 1505, PMS). This measure is affected by any cuticular or stomatal transpiration and, thus, broadly captures the integrated effects of plant traits and the environment water demand on the minimum water potential a plant reaches in natural conditions. P 50 was used as an embolism resistance index, which is water potential corresponding to a 50% loss of xylem conductivity. P 50 was used to calculate the hydraulic safety margin (HSM; i.e., the difference between P 50 and Ψ md ), which is a good predictor of drought resistance (Barros et al., 2019). Xylem embolism resistance of each branch was measured using the pneumatic method in the manual measurements set-up (Pereira et al., 2016;Zhang et al., 2018). With this method, loss of hydraulic conductance is estimated from the increase in air volume inside the wood caused by the formation of an embolism, as the branch dehydrates (for details of the methods see Bittencourt et al., 2018). Branches were dehydrated using the bench dehydration method (Sperry et al., 1988). The branches were bagged for an hour to balance the water potential of the xylem with that of the leaves before each air removal measurement. The volume of air reservoir was as adjusted when a rapid drop in air discharge or values close to atmospheric pressure were detected to preserve the accuracy of the method. The reservoir volume varied between 1.305 to 2.610 depending on the species and/or individuals the volume. Water potential of one or two leaves was measured immediately after air removal. Embolism resistance is then given by increasing air removal (PAD = the percentage of air removed) with each tree xylem water potential. To calculate P 50 , we gathered data for the repetitions of two branches of the same tree and adjusted a sigmoid curve to the data where P 50 and slope are the adjusted parameters (Pammenter and Willigen, 1998).

= 100 / (1 + exp ( (Ψ -50 ). (Eq 5)
Characterization of P 50 was done for the seven dominant species of the community. The pneumatic method was applied to the branches to construct vulnerability curves for the xylem. We were not able to produce reliable vulnerability curves to Macareia radula and Alchornea discolor using the pneumatic method most likely due to an undetected leakage during the measurements in the eld. After inspecting the curves, we observed large amount of variation in the initial and nal measurements, and consequently, poor t of the curves. Both species have very tick bark and fast shrinkage requiring tting adjustments during the measurements. We believe that this may have caused air leakage and for this reason we were not con dent to use the data collected for both species for the analysis. The P 50 and HSM are presented for the remaining ve dominant species: N. guianensis, S. versicolor, P. cachimboensis, K. rubri ora and M. guianensis ( Figure S6).

Statistical analysis
To determine morphological, anatomical, and ecophysiological variables related to transpiration, water transport, and xylem vulnerability, we made pairwise tests of the relationships between attributes due to the high degree of correlation. Only attributes directly related to our hypotheses were considered for pairwise relationships.
The Shapiro-Wilk normality test was used to test the assumptions of regression and choose methods for data analysis. A correlation matrix was constructed to observe associations among the variables of this study. Correlations were performed using the non-parametric statistics to calculate Spearman's Coe cient (rho) as data and variables did not follow a normal distribution even after transformation. A graph was constructed from the correlation matrix to show the observed relationships within an intuitive model of expected relationships. The model was proposed based on leaf and wood anatomical and hydraulic attributes to explain correlated mechanisms between leaf transpiration, water transport mechanism, and resistance to embolism.
Pairwise regression models were adjusted for water use e ciency, xylem embolism vulnerability, hydraulic safety margin, and anatomical variables to evaluate the anatomical basis for observed variation in ecophysiological attributes. Adjusted coe cients of determination (R 2 ), 95% con dence intervals and p-values are reported. The adjusted regression models were corrected through a multiple comparison test (Bonferroni test). The data were transformed when necessary to meet the assumptions of regression (linearity, homoscedasticity, normality, and low leverage).
The data were scaled to standard deviation units to perform a principal component analysis (PCA), to ordinate possible plant strategies and select the variables that contribute most to species clustering. Descriptive statistical analysis, correlation matrix, regression models, and principal component analysis were performed using R © software (R Core Team, 2019).

Anatomical and hydraulic diversity
The dominant woody species in the studied Amazonian savanna have anatomical ( Figure 2) and hydraulic diversity of leaf and wood ( Table 2). Variation by a factor of two to three was observed among the seven species for most attributes, including leaf area, vessel frequency, lumen area, lumen fraction, intervascular wall thickness, ber length, P 50 , stomatal density, and G max . This variation was even greater, ve times or more, for the attributes of succulence and HSM. Exceptions to this pattern were the attributes of wood density, water use e ciency (δ 13 C), and pit membrane thickness, which varied 10-30% among species. It is worth mentioning that the relative ordering of species varied according to attributes ( Figure S7), with no species having a consistent pattern of covariation for several attributes simultaneously.

Correlation between attributes
Stomatal conductance is negatively correlated with water use e ciency of leaves, which in turn positively correlated with vessel element length (r=0.90 p≤0.01) and negatively correlated with P 50 (r=-0.64 p≤0.01). Vessel frequency is negatively correlated with vessel element size (r=-0.62 p≤0.01) and lumen area (r=-0.65 p=0.002). Vessels with smaller lumen area correlate negatively with higher lumen fraction per area (r=-0.59 p=0.004), the latter of which has a negative correlation with P 50 (r=-0.76 p≤0.01). The thickness of the intervascular pit membrane is negatively correlated with P 50 (r=-0.61 p=0.02) and positively with safety margin (r=0.61 p=0.02).

Pairwise relationships between anatomical and hydraulic attributes
Signi cant relationships were found between anatomical and hydraulic attributes of leaf and wood (Figure 4). Water use e ciency (δ 13 C) decreased with G max (r²=0.50; p≤0.01) (Fig. 4A) and with lumen fraction (r²=0.56; p=0.002) (Fig. 4B). Among the wood anatomical variables evaluated here, only vessel element length and thickness of the membrane of the intervascular pit were signi cantly related to xylem vulnerability (Figure 4). The attribute P 50 increased with vessel element length (r²=0.77; p=0.04) and decreased with the thickness of the membrane of the intervascular pit (r²=0.43; p≤0.01) ( Fig.  4C; 4D). The attribute HSM decreased with the length of the vessel element (r² = 0.69; p≤0.01) and increased with the thickness of the pit membrane (r² = 0.42; p≤0.01) (Fig. 4E, 4F).

Anatomical and hydraulic strategies
The principal component analysis (PCA) explained 73.92% of the total variation with the rst two axes ( Figure 5). The rst axis explained 54.12% of the total variation and presented species with high water use e ciency, long vessels and bers, and vessels with thicker walls, such as N. guianensis and A. discolor. The second axis explained 19.8% of the total variation and represented species with higher wood density, a greater lumen fraction, greater thickness of the membrane of the intervascular pit, higher theoretical maximum stomatal conductance (G max ) and greater leaf succulence and speci c leaf mass, such as S. versicolor, M. radula, P. cachimboensis, K. rubri ora, and M. guianensis.

Discussion
This study provides the rst data about hydraulic structure and functioning of woody species from Amazonian savannas. Our ndings showed wide variation in resistance to embolism and structural anatomy, suggesting no unique dominant functional strategy among Amazonian savanna species. The species diverge by investing in three different functional strategies for drought tolerance: (1) Species with high G max and low water use e ciency where leaf succulence may delay embolism onset and high wood speci c gravity enables more negative pressures on the xylem (Kielmeyera rubri ora, Simarouba versicolor and Macairea radula); (2) Species that have low embolism resistance and high water use e ciency by reinforcing the walls of longer vessel elements to prevent implosion (Norantea guianensis and Alchornea discolor); (3) Species with high G max and low water use e ciency where short vessel elements and thick intervascular pit membranes may minimize the embolism spread across the plant (Maprounea guianensis and Parkia cachimboensis).
Local divergence in drought-induced embolism resistance Amazonian savanna species studied here diverged widely regarding their vulnerability to embolism. We observed extremely vulnerable species, such as N. guianensis (P 50 = -1.66 MPa), coexisting with drought-resistant species, such as P. cachimboensis (-5.07 MPa), M. guianensis (-4.56 MPa), S. versicolor (-4.40 MPa), and K. rubri ora (-3.87 MPa). This variation was observed when comparing safety margins among species, which also varied widely (c.f. Table 2). Although, at global scale is observed a very clear pattern of less negative P 50 in wetter habitats and more negative P 50  Integration of structure and hydraulic functioning of plants Integrative anatomical and physiological studies are essential for understanding the diversity of drought resistance mechanisms of coexisting species in seasonal environments. Here we show that species that are less e cient in water use (e.g. K. rubri ora, M. radula, S. versicolor, P. cachimboensis, and M. guianensis) show lower degree of stomatal control (high G max ) supporting xylem functioning with leaf succulency and/or safer wood anatomical structures. Alternatively, species that are more e cient in water use (e.g. N. guianensis and A. discolor), have greater control of transpiration, thus incorporating and xing more 13 C in leaf tissues. With greater stomatal control, a lower fraction of lumen may be required to sustain the transpiration demand of these species. Thus, this control over stomatal opening may be a strategy to avoid cavitation in the xylem, as species presenting this strategy tend to have long vessel elements of greater diameter.
Vessel length and diameter directly determines water transport e ciency, and the hypothesis that there is a trade-off between safety and e ciency of the xylem has often been raised (Gleason et al., 2016). This hypothesis proposes that the xylem anatomy drives safety and e ciency but that the vessel characteristics cannot promote safety and e ciency at the same time. Our ndings support this view as most resistant species do have shorter vessel elements with smaller diameters. In this case, transport safety is ensured by increasing the thickness of vessel walls to provide su cient mechanical strength avoiding potential implosion under tension, especially if a vessel lled with water touches an embolized vessel (Hacke et al., 2001;Sperry et al., 2005). The secondary xylem of this group of species is structured by a higher frequency of short vessel elements with thick pit membranes and smaller diameters, which safely increases the fraction of lumen for water transport. Flow resistance is augmented as element vessel size decreases as water need to ow through a larger number of end walls with impacts to the plant's hydraulic conductivity (Loepfe et al., 2007). Here we show that plants with short vessel elements and higher frequency and clustering of vessels also show high lumen fraction. This may be associated with a strategy to compensate for the loss of hydraulic conductivity, since this hydraulic architecture increases the options for transport routes through the vessels.
Some controversy on the functional role of vessel grouping to xylem safety is present in the literature. Tyree & Zimmermann (2002) and Lens et al. (2011) showed that the grouping of vessels can improve embolism resistance. On the other hand, Loepfe, Martinez-Vilalta, Piñol & Mencuccini (2007) and Martínez-Vilalta et al. (2012) suggested the opposite, that is, high clustering of vessels is related to low security by increasing the probability of embolism propagation. Here, we observed greater clustering of vessels associated with embolism resistance con rming the importance of grouping to xylem safety. Again, in those species protection against embolism spread may be guaranteed by thicker intervascular pit membranes, which are related to decreased air propagation from embolized vessels to functional vessels (Lens et al., 2011;Dória et al., 2018). Our results also con rm the functional importance of the intervascular pit membrane drought-induced embolism resistance in Amazonian savanna species. Here we show that drought-induced embolism and HSM were signi cantly related to vessel element length and thickness of the membrane of the intervascular pit and that short vessel elements with thick membranes guaranteeing safe water transport for the studied species. Together, our results suggest that investigating the the role of anatomical structures to hydraulic function in a more integrative way can reconcile the different expectations about the role of individual anatomical features on hydraulic functioning. Finally, structural adjustments to allow safe or e cient water transport may not be restricted to xylem but also be related to parenchymatic tissues. A study of 800 species of trees in China found axial parenchyma to be related to theoretical hydraulic conductivity, and radial parenchyma to be related mechanical support (Zheng and Martínez-Cabrera, 2013), which may explain the increase in radial parenchyma in species with greater lumen area. Chen et al. (2020) obtained similar results for 10 species in an arid limestone habitat in China. Their results showed that resistance to embolism was positively correlated with xylem mechanical resistance indicators and negatively related to the fraction of axial parenchyma. These authors suggested that tree species with more axial parenchyma may not need high resistance to embolism to adapt to drought. The present study did not analyze xylem at the parenchymatic tissue level, but variation in the amount of axial and radial parenchyma among species was observed in the anatomical images suggests that this may be one of the mechanisms that allow vulnerable species (e.g. Norantea guianensis) to persist on Amazonian savanna's environmental conditions. Future studies relating anatomical features, tissue capacitance and xylem function may shed some light on this interesting strategy.
No single anatomical or hydraulic strategies to colonize Amazonian savannas The main novelty of the present study is that it shows no convergence of strategies for species to be dominant in Amazonian savannas, but rather there is a set of anatomical and hydraulic attributes that vary together, re ecting the speci c responses of each species to drought. Vessel element length and pit membrane thickness explained resistance to embolism while leaf succulence allow for less conservative water use strategies. Although the species of Amazonian savannas present divergent strategies, this coordination between leaf and wood is observed for the balance of water supply.
Our results demonstrate a strong degree of coordination between water transport and stomatal regulation. The observed relationships among the characteristics are probably a result of trade-offs favoring the functional coordination between leaves and wood. Decreased stomatal conductance is accompanied by a decline in leaf photosynthesis and increased water use e ciency, with greater stomatal control in the dry season (Janssen et al., 2020). Declines in hydraulic conductance and stomatal conductance with decreasing water potential of the xylem are similar (Brodribb et al., 2003), suggesting that hydraulic vulnerability of xylem and stomatal sensitivity are strongly coordinated. ). Allometry appears to be regulated at the scale of the entire plant so that water demand and supply are compatible (Brodribb, 2009). Therefore, understanding the coupling between stomatal conductance and xylem hydraulic architecture expands our mechanistic knowledge about plants in an integrated way. Here we contribute to this understanding by showing that these mechanisms can be manifested in different strategies among coexisting plants, converging on safe structures, although in different organs.

Conclusion
This study provides new insights into the different strategies that plants can exhibit in an integrated manner and represents the rst work to estimate xylem embolism resistance in Amazonian savanna species. Our results demonstrate that species that inhabit a restrictive environment, such as Amazonian savannas, can exhibit different strategies of drought tolerance/avoidance resulting in the similar outcome of maintaining a favorable water balance that allow their persistence and co-occurrence. Distinct strategies to deal with drought permitted wide divergence regarding embolism vulnerability. The strategies that selected high xylem embolism resistance are explained by vessel element size and pit membrane thickness. However, the role of strategies related to drought avoidance should not be neglected in this ecosystem. Future climate predictions should aim to incorporate this range of strategies and could be used to test whether future climates will favor some strategies or still allow for a range of them to persist. Further studies and re nement of model exercises are likely to improve our predictions on changing the species and trait composition of Amazonian savannas and also its impact of large scale ecosystem functioning.
Declarations PS collected and compiled the data alongside ALG, GVF, SP and LS. TE designed the study with MDC. PS, ALG and GVU performed the statistical analysis and PS, TE and MDC wrote the paper, with all other authors making substantially contributions to revisions.       Table S1).

Figure 5
See image above for gure legend.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download. Simioniet2021SupplementalFiles.doc