Large hydraulic safety margins protect Neotropical canopy rainforest tree species against hydraulic failure during drought

Abundant Neotropical canopy-tree species are more resistant to drought-induced branch embolism than what is currently admitted. Large hydraulic safety margins protect them from hydraulic failure under actual drought conditions. Xylem vulnerability to embolism, which is associated to survival under extreme drought conditions, is being increasingly studied in the tropics, but data on the risk of hydraulic failure for lowland Neotropical rainforest canopy-tree species, thought to be highly vulnerable, are lacking. The purpose of this study was to gain more knowledge on species drought-resistance characteristics in branches and leaves and the risk of hydraulic failure of abundant rainforest canopy-tree species during the dry season. We first assessed the range of branch xylem vulnerability to embolism using the flow-centrifuge technique on 1-m-long sun-exposed branches and evaluated hydraulic safety margins with leaf turgor loss point and midday water potential during normal- and severe-intensity dry seasons for a large set of Amazonian rainforest canopy-tree species. Tree species exhibited a broad range of embolism resistance, with the pressure threshold inducing 50% loss of branch hydraulic conductivity varying from − 1.86 to − 7.63 MPa. Conversely, we found low variability in leaf turgor loss point and dry season midday leaf water potential, and mostly large, positive hydraulic safety margins. Rainforest canopy-tree species growing under elevated mean annual precipitation can have high resistance to embolism and are more resistant than what was previously thought. Thanks to early leaf turgor loss and high embolism resistance, most species have a low risk of hydraulic failure and are well able to withstand normal and even severe dry seasons.


Introduction
Tropical rainforest ecosystems are characterized by high annual rainfall. Yet, most Amazonian rainforests are regularly exposed to seasonal droughts due to the latitudinal movements of the Intertropical Convergence Zone . Reduced soil water availability due to low precipitation during the dry season can cause a reduction in tree growth (Wagner et al. 2012), photosynthetic assimilation, and whole-tree transpiration due to stomatal closure . Increased mortality can occur with increasingly severe dry seasons (Phillips et al. 2010) and species exhibiting contrasting drought-resistance characteristics are likely to be affected differently (Anderegg et al. 2016). This indicates that drought resistance may provide an advantage for tropical trees under climate change and that droughts are likely to alter tree species' distribution as well as community composition and functioning (Esquivel-Muelbert et al. 2019). Furthermore, little is known about the trait syndromes that confer drought resistance in tropical rainforests or how they participate in shaping species distribution along macro-environmental gradients of water availability (Zhu et al. 2018).
Assessing tropical rainforest species and community resistance to drought in a context of increasingly frequent severe dry seasons (Duffy et al. 2015) requires a deep understanding of the mechanisms enabling the maintenance of physiological functions, and hence survival, during periods of low water availability (Anderegg et al. 2015). Tree species display a variety of drought tolerance and drought avoidance strategies (Delzon 2015), especially in hyper-diverse tropical rainforests . Therefore, using traits to describe these strategies may be particularly helpful in predicting whether a given species will be able to withstand future climatic conditions which could potentially alter its present bioclimatic distribution range.
Loss of hydraulic conductivity leading to xylem hydraulic failure and plant desiccation has been identified as an important mechanism involved in drought-induced tree mortality (Adams et al. 2017). Therefore, the spread of branch embolism during drought could not only have an impact on the overall fitness of individual trees but could also affect tree community functioning. Plants can, however, avoid hydraulic failure through the expression of several hydraulic traits. Plant xylem vulnerability to embolism-expressed as the xylem pressure at which 50% of branch hydraulic conductivity is lost (Ψ 50 ; MPa)-is an adaptive (Maherali et al. 2004) and mechanistic trait with strong predictive power for plant survival and distribution (Urli et al. 2013;Anderegg et al. 2015;Brodribb 2017). Embolism resistance varies with mean annual precipitation at the inter-biome scale (Maherali et al. 2004;Choat et al. 2012) and with water availability at the local scale . Species with more negative Ψ 50 values are able to tolerate more negative water potentials and sustain hydraulic conductivity and can therefore thrive in drier conditions . The xylem hydraulic safety margin (HSM; MPa), expressed as the difference between seasonal minimum xylem water potential and Ψ 50 , was found to better explain mortality during severe droughts than Ψ 50 alone or than other traits (Anderegg et al. 2016). In a metaanalysis, Choat et al. (2012) showed that trees typically operate at positive HSM, indicating that they are not being routinely exposed to high levels of branch embolism. Yet, Choat et al. (2012) reported converging, narrow HSM values (< 1 MPa) among biomes for Angiosperm species, suggesting that forests will be highly vulnerable to an increase in the frequency of severe droughts. During moderate droughts, some species have the ability to preserve leaf functioning (i.e., growth, gas exchange, leaf hydraulic conductance) thanks to a lower leaf water potential at which leaf cells lose turgor (henceforth denoted π tlp ; MPa; Bartlett et al. 2012a, b). With increasing drought intensity, leaf water potential decreases to and past π tlp , mediating stomatal closure and thereby limiting leaf transpiration and the consequential further decrease in water potential (Brodribb et al. 2003; Bartlett et al. 2016). As a consequence, π tlp has been used as an indicator for stomatal behavior  and as a proxy for stomatal closure during drought (Martin-StPaul et al. 2017;Hochberg et al. 2018).
These hydraulic traits can act in concert or as a result of tradeoffs that determine plant drought resistance, which can be attained through multiple strategies (Pivovaroff et al. 2016). It is therefore paramount to consider drought resistance strategies in terms of interactions among traits expressed in different organs. The longstanding view that Ψ 50 and π tlp are coordinated to maintain leaf functioning and productivity under drought stress (Jones and Sutherland 1991) has recently been challenged Martin-StPaul et al. 2017). Indeed, π tlp was much less variable and in most cases less negative than Ψ 50 along the global spectrum of Ψ 50 variations. The same pattern was found in another meta-analysis for~80% of tropical rainforest tree species for which sigmoidal vulnerability curves were selected ). This indicates that (i) trees generally have a positive hydraulic safety margin between stomatal closure (π tlp ) and the spread of embolism (Ψ 50 ) and that (ii) stomata must close in order to avoid the rapid spread of branch embolism and subsequent hydraulic failure.
Over the past decades, a tremendous effort has been made to improve our understanding of the mechanisms of plant response and resistance to drought in various biomes. Yet, a major knowledge gap still remains for tropical rainforest tree species ). These species, which typically experience high precipitation regimes, are believed to have low intra-biome variation and less negative Ψ 50 values (i.e., higher vulnerability to drought-induced branch xylem embolism) than those in drier forest biomes (Maherali et al. 2004;Choat et al. 2012). Unfortunately, the data available for Amazonian rainforest canopy-tree species are limited and some are subject to methodological doubt (see Torres-Ruiz et al. 2016concerning Rowland et al. 2015. Tropical rainforests also represent the woody-biome with the least negative mean π tlp values observed so far (Bartlett et al. 2012a, b;Zhu et al. 2018), reinforcing the claim that rainforest tree species cannot tolerate strong declines in water potential without closing their stomata (Fisher et al. 2006). However, π tlp showed significant interspecific variation for Amazonian rainforest canopy-tree species in French Guiana (Marechaux et al. 2015), which extends the expected range for tropical rainforest canopy tree species. Moreover, species identity was shown to be the main driver of variation in π tlp , with, respectively, low and negligible intraspecific and seasonal variability (Maréchaux et al. 2016). At present, the few studies providing a combined investigation of Ψ 50 and π tlp for tropical rainforest species concern only a limited number of species and have yielded contradictory conclusions Nolf et al. 2015;Bartlett et al. 2016;Martin-StPaul et al. 2017;Powell et al. 2017). Likewise, the absence of a general consensus on the magnitude of Ψ 50 and the risk of hydraulic failure in hyper-diverse ecosystems such as tropical rainforests highlights the need for more research, in particular for large canopy-trees, which are at higher risk of drought-induced mortality during extreme climatic events (Bennett et al. 2015).
We assessed the vulnerability of rainforest tree species to drought-induced branch xylem embolism and the risk of hydraulic failure under normal-and severeintensity dry seasons. We explored interspecific variability and covariations in branch and leaf hydraulic traits known to be important in drought response and survival. Xylem branch vulnerability to embolism, leaf water potential at turgor loss point, and midday leaf water potential were estimated in 30 co-occurring rainforest canopy-tree species in French Guiana. With this unique set of hydraulic traits, we addressed the following questions: (i) What is the range of variation in branch xylem vulnerability to embolism in abundant, co-occurring lowland tropical rainforest canopy-tree species?
(ii) Are these species exposed to a risk of hydraulic failure during normal and severe droughts? The study was conducted at the Paracou experimental site, managed by CIRAD, in plots 6, 11, and 15, each of which are 6.25 ha of undisturbed forest. All trees ≥ 10 cm in diameter within the plots have been tagged, mapped, identified to species level, and annually or biennially measured for diameter growth for about 30 years. Based on total basal area by species in this botanical inventory data, we selected the most common canopy-tree species present on plateaus or moderate slopes (terra firme) at the site. The sampling included the three main commercial species (Dicorynia guianensis, Qualea rosea, and Sextonia rubra), which account for 71.7% of the logging volume in French Guiana (Fargeon et al. 2016).
We sampled branches and leaves of co-occurring canopytree species belonging to 11 botanical families ( Table 3 in the Appendices) of which Burseraceae, Chrysobalanaceae, Fabaceae, Lecythidaceae, and Sapotaceae were the most represented in our sample. Thirty tree species were investigated, eight of which produce exudates ( Table 7 in the Appendices). We selected healthy dominant or co-dominant canopy trees. Only one understory species (Gustavia hexapetala) was selected. Accessibility for tree climbers to sunlit leaves or branches was also a selection criterion.
We measured leaf traits on 29 species and the hydraulic characteristics of the branches supporting these leaves for 25 species (Table 4 in the Appendices). The latter measurements were conducted on three to ten individuals per species ( Table 4 in the Appendices), with exceptions for seven species out of the total 25 for which we could obtain only one or two vulnerability curves. Branch hydraulics could not be obtained for four species: Licania alba and Licania heteromorpha had extremely long maximum vessel lengths and were not measured (see Vulnerability curves); we did not succeed in measuring Vouacapoua americana despite repeated attempts; and too few Sextonia rubra individuals were sampled to yield results. Leaf traits could not be obtained for two species: the only Protium sagotianum individual on which branches were sampled died; and the dense secondary vein network of Qualea rosea hampered an accurate determination of turgor loss point (Maréchaux, Bartlett et al. 2016).

Vulnerability curves
Professional tree climbers sampled 2-3-m-long, sun-exposed canopy branches with a diameter under bark ranging from 8.7 to 18.2 mm during the rainy season, i.e., between January and July 2017. While sampling the branches, the climbers attached strings to nearby fine, sun-exposed twigs so that they could later be pulled down for leaf sampling. After sampling, the branches intended for the determination of branch xylem vulnerability to embolism were immediately defoliated to prevent transpiration loss, recut under water, conditioned in wet cloth, and sealed in heavy-duty black plastic bags to halt moisture loss (Delzon et al. 2010). They were taken to the laboratory in Kourou (a 45-min drive), recut to 1.5 m in length, reconditioned in damp absorbent paper with both ends covered in plastic wrap, sealed with tape inside heavy-duty plastic bags, and immediately sent by priority transport (1-2 days) to the Caviplace phenotyping platform (Genebois, Univ. of Bordeaux, Pessac, France).
Prior to final measurements, the branches were recut under water to a length of 1 m and debarked at both ends to help limit the amount of exuded mucilage. Vulnerability to embolism was measured following the flow-centrifugation technique ) in a large Cavitron equipped with a 1-m diameter custom-built rotor (Cavi1000; DGMeca, Gradignan, France). This large rotor was designed to process species with long vessels (Lobo et al. 2018). This limits the occurrence of 'open-vessel' artifacts if the vessel length is much shorter than the diameter of the rotor . The percentage loss of conductivity (PLC) was measured for each branch in 0.5 MPa pressure steps (Cavisoft v1.5; Univ. Bordeaux, Pessac). Vulnerability curves (VCs) were obtained by plotting increasing PLC with decreasing xylem pressure. VCs were fitted with a sigmoid function (Pammenter and Vander Willigen 1998) following the Nlin procedure in SAS (SAS 9.4, SAS Institute, Cary, NC, USA) and hydraulic traits were determined. Ψ 50 (MPa), the xylem pressure inducing 50% loss of branch hydraulic conductivity, is a widely used metric representing the steepest point of the VC, meaning that even a small decrease in water potential will cause a substantial reduction in hydraulic conductivity (Maherali et al. 2004;Choat et al. 2012). a x (%MPa −1 ) is the slope of the VC at the inflexion point and is a good indicator of the speed at which embolism affects the stem (Delzon et al. 2010). Additionally, we obtained the xylem pressures inducing 12% (Ψ 12 ; MPa) and 88% (Ψ 88 ; MPa) loss of branch hydraulic conductivity. Ψ 12 corresponds to the water potential which reflects the initial air-entry producing embolism (Meinzer et al. 2009), while Ψ 88 is thought to be the xylem water potential threshold above which hydraulic damage may be irreversible, leading to high levels of tree mortality in Angiosperm tree species (Choat et al. 2012;Urli et al. 2013).
We also measured the maximum vessel length (MVL; cm; Table 7 in the Appendices) in canopy branches of similar dimensions from the same trees sampled for the VCs, following the air-injection method (Ewers and Fisher 1989). Air was injected at a pressure of 1 kPa from the distal end of the branch while the basal end was recut underwater every 2 cm. Once the first air bubbles appeared from the cut end surface, branch length was measured with a measuring tape and MVL was recorded. We excluded species with a MVL far above 1 m for a given diameter of ca. 2 cm (i.e., Licania alba and Licania heteromorpha) and selected smaller diameter branches for species with a MVL close to 1 m (i.e., Eperua falcata, Goupia glabra, and Licania membranacea).

Leaf turgor loss point
Leaf water potential at turgor loss point (π tlp , MPa) was estimated during the 2017 dry season on canopy leaves with a vapor pressure osmometer (VAPRO 5520, Wescor, Logan, UT, USA; Bartlett et al. 2012a, b). This method was previously validated and used in a study conducted on tropical rainforest trees in French Guiana (Maréchaux, Bartlett et al. 2016). After harvest, the leaves were immediately sealed in plastic bags containing moist absorbent paper, placed in a cooler, taken to the laboratory in Kourou and placed at 5°C to allow them to rehydrate overnight. The next day, one 8-mm disk per leaf was sampled with a sharp cork borer, care being taken to avoid first-and secondorder veins. The disk was immediately wrapped in tinfoil and frozen by immersion in liquid nitrogen (LN 2 ) for at least 2 min, then punctured 10-15 times with a needle and sealed in an osmometer with a standard 10-μl chamber well. The equilibrium solute concentration value C 0 (mmol kg −1 ) was recorded from the osmometer when the difference between two consecutive measurements fell below 5 mmol kg −1 . A previously established linear relationship with the osmotic potential at full turgor (π osm ) was used to convert to π osm following the van't Hoff equation relating solute concentrations to vapor pressure: where the numerator of the first term represents R × T = 2.5 L MPa mol −1 at 25°C, with R, the ideal gas constant, and T, the temperature in degrees Kelvin. The value of π tlp was then calculated from π osm (Bartlett et al. 2012a, b) as 2.4 Dry season midday leaf water potential Midday leaf water potential (Ψ md ; MPa) was measured with a pressure chamber (Model 1505D, PMS, USA) between 11:00 and 14:00 on clear sunny days with a high vapor pressure deficit (VPD = 1.32 kPa) during the 2018 dry season in early October when soil relative extractable water was low (REW = 0.21; Fig. 4 in the Appendices). The measurements were conducted on individual trees for which vulnerability curves were available and also on additional co-occurring canopy trees of the same species in order to increase sample size and reach three to five replicates per species. Additionally, we measured Ψ md on four species for which we had not obtained VCs. We followed a specific protocol to ensure reliable readings of Ψ md for laticiferous species exuding clear-exudates (as opposed to colored ones). When clear foam (latex plus air) appeared, usually from only a few distinct points on the petiole crosssection, it was rapidly blotted away with absorbent paper. This was repeated until a liquid (xylem water) appeared and covered the entire cross-section, then Ψ md was immediately recorded. Environmental conditions during these measurements were typical of an average-intensity (henceforth denoted "normal") dry season, according to 40 years of available REW data (Fig. 4 in the Appendices). We also compiled Ψ md data from the literature (Stahl et al. 2010) recorded for the same species (n = 11) at the same site during a strong-intensity (henceforth denoted "severe") dry season in early November 2008 (REW = 0.10; Fig. 4 in the Appendices; VPD = 1.51 kPa), the second driest in the past 40 years at our site. This allowed us to compare our data with those from more severe soil drought conditions.

Hydraulic safety margins
Xylem hydraulic safety margins (HSM; MPa) were calculated as the difference between either (i) dry season midday leaf water potential (HSM Ψmd-Ψx ; Meinzer et al. 2009) or (ii) leaf turgor loss point (HSM πtlp-Ψx ; Martin-StPaul et al. 2017) and the xylem pressure inducing either 12, 50, or 88% loss of branch hydraulic conductivity. Leaf safety margins were calculated as the difference between Ψ md and π tlp (SM leaf ; MPa). All safety margins were calculated for individual trees and then averaged at the species level. We report HSM Ψmd-Ψx calculated from Ψ md datasets recorded both in 2018 (normal dry season; Table 1) and in 2008 (severe dry season; Table 8 in the Appendices).

Data analysis
We used ANOVA tests to estimate differences in mean species values for embolism resistance parameters, leaf water potentials and HSM, and unpaired t tests for differences in mean Ψ 50 values between our study and other biomes. Linear models were performed to analyze trait correlations. We used a sensitivity analysis to assess the relative contributions of Ψ 50 , Ψ md , or π tlp in driving the variations in HSM. To test whether species ranking in Ψ md between 2018 and 2008 was conserved, we computed Kendall's coefficient of concordance (τ). The above-mentioned analyses were performed for all species with at least three individuals measured concurrently. Species that did not meet this criterion were simply projected onto the figures. All statistical analyses were performed with the R software.
There was a significant species effect in SM leaf measured in 2018 (F = 1.9; p = 0.02; Table 1). π tlp values were more negative than Ψ md values measured in 2018 for more than 80% of the species (Fig. 2c), with subsequent positive SM leaf ranging from − 0.58 to 1.10 MPa with an average of 0.30 ± 0.01 MPa (Table 1).  Stahl et al. 2010), showed strong interspecific variation (F = 18; p < 0.001), and was, on average, 0.62 ± 0.10 MPa (57%) more negative than Ψ md in 2018 for the species in common to both years (p < 0.001; data not shown). Consequently, the HSM Ψmd-Ψ50 calculated from the Ψ md measured in 2008 for ten species was lower Fig. 2 Relationships between Ψ 50 , the water potential at 50% loss of branch hydraulic conductivity (Ψ 50 ; MPa), and a dry season midday leaf water potential (Ψ md ; MPa), b the water potential at turgor loss point (π tlp ; MPa), and c π tlp and Ψ md , for the 30 canopy-tree species sampled in French Guiana (black n ≥ 3; gray n < 3). All error bars represent standard errors of the mean. The 1:1 line (dashed line) is represented in all panels; the 99% percentile of π tlp (− 2.12 MPa; dotted line) is represented in b; and the regression (solid line) is represented in c. Coefficients of determination (R 2 ) and significance levels (p) are shown  Table 8 in the Appendices). Mean SM leaf calculated from Ψ md measured in 2008 for 12 species ranged from − 1.09 to 0.01 MPa, with an average of − 0.43 ± 0.09 MPa. SM leaf was negative for all species but one (Symphonia sp. 1; Fig. 3a and Table 8 in the Appendices) and species ranking in SM leaf was not conserved between years (τ = 0.5; ns; Fig. 3a).

Discussion
Considerable interspecific variability in branch xylem vulnerability to embolism In this study, our sampling of 25 tropical rainforest canopy-tree species showed a four-fold range of magnitude in branch xylem vulnerability to embolism (Fig. 1, Table 1). This range encompasses 72% of the previously observed angiosperm variation in Ψ 50 at the global scale (Choat et al. 2012) and expands the range of known branch xylem vulnerability to embolism for tropical forest Angiosperm tree species (Choat et al. 2012;Nolf et al. 2015;Rowland et al. 2015;Christoffersen et al. 2016;Powell et al. 2017;Santiago et al. 2018). The limited importance of confounding factors (e.g., genetics, ontogeny, competition, habitat) indicates that the observed interspecific variability was mostly related to the species' intrinsic functional characteristics.
Mean community branch Ψ 50 (− 3.93 ± 0.31 MPa; Fig. 1 and Table 1) was much more negative than the worldwide mean for tropical rainforests; it was also more negative than the mean values published to date for terra firme canopy-tree species (− 2.31 ± 0.20 MPa; Sperry et al. 1988;Machado and Tyree 1994;Tyree et al. 1998;Choat et al. 2007;Sperry et al. 2007;Meinzer et al. 2008;Nolf et al. 2015;Rowland et al. 2015;Powell et al. 2017;Santiago et al. 2018) and was comparable to values found for drier biomes (Fig. 1). Our results therefore clearly contrast with the global and inter-biome pattern of interspecific variation in branch embolism resistance, where total precipitation is thought to be the main driver (Maherali et al. 2004;Choat et al. 2012).
The differences in branch xylem vulnerability to embolism between our study and the existing literature on tropical rainforests could be due to differences in water availability or habitat type, the species' ontogenetic stage of development, species phylogeny or methodological aspects of hydraulic measurements. However, the lower branch xylem vulnerability to embolism we found cannot be explained by differences in water availability since all our measurements were conducted on similar habitats (i.e., terra firme). Furthermore, the differences are unlikely to be due to either ontogeny or limited phylogenetic differences since we sampled branches of mature canopy trees covering a broad range of species and botanical families, as was also the case for the previously recorded Ψ 50 values. It therefore appears that methodological choices are probably the main reason for the inconsistencies that have appeared across studies. The precautions we took when sampling, combined with the use of the in situ flow-centrifuge technique, makes it unlikely that our measurements were affected by native embolism. Herein, we therefore provide robust results for an ecosystem where the scarcity of data limits our understanding of the hydraulic functioning and fate of tree species Delzon and Cochard 2014). Although it has been shown that branch xylem vulnerability to embolism varies along macro- (Maherali et al. 2004;Choat et al. 2012) and micro-environmental scales of water availability ), we still must decipher the past and present ecological processes underlying the large Table 2 Pearson's correlation coefficients (r) between hydraulic traits and species bioclimatic affiliation for the canopy-tree species sampled in French Guiana. Bivariate comparisons were applied using species means for each variable, with n varying from 18 to 26 species. r values are shown in bold type when significant (p < 0.05). Abbreviations are as follows: a x (MPa −1 ) represents the slope of the vulnerability curve; Ψ 12, Ψ 50, and Ψ 88 (MPa) represent the water potential at 12, 50, and 88% of branch embolism, respectively; Ψ md (MPa) represents midday leaf water potential measured in 2018; π tlp (MPa) represents leaf turgor loss point; HSM Ψmd-Ψ50 and HSM πtlp-Ψ50 (MPa) represent the hydraulic safety margins calculated as the difference between Ψ md or πtlp and Ψ 50 ; SMleaf represents the leaf safety margin calculated as the difference between Ψ md and π tlp interspecific variability we found in co-occurring rainforest species in the same habitat. Several approaches may be pertinent, for example, the investigation of trade-offs related to hydraulic safety vs efficiency (Santiago et al. 2018), the existence of contrasting strategies of drought resistance (Pivovaroff et al. 2016), or better knowledge of species biogeographic distribution in relation to water availability (Esquivel-Muelbert et al. 2019).
Low risk of hydraulic failure during a normal dry season Our results show that most of the studied species share a low risk of hydraulic failure during a normal dry season since both xylem hydraulic safety margins were large with regard to critical thresholds of branch embolism (i.e., Ψ 50 and Ψ 88 ). One of our major findings concerns leaf turgor loss point, for which we gathered one of the largest datasets to date for tropical rainforest tree species. The range of π tlp we found confirms previous findings (Marechaux et al. 2015) while the mean was more negative than what is currently generally accepted for tropical rainforests (Bartlett et al. 2012aZhu et al. 2018). However, π tlp was relatively constant along the full range of Ψ 50 . Our results suggest that turgor loss, which was used in this study as a proxy for stomatal closure, precedes the rapid spread of branch embolism (Brodribb et al. 2003;Creek et al. 2018), regardless of species branch xylem vulnerability to embolism. This result both supports Martin-StPaul et al. 2017) and contradicts (Christoffersen et al. 2016) patterns found in recent global meta-analyses, and it is consistent with the rather conservative behavior of rainforest species with regard to water-use (Fisher et al. 2006). Although we found that higher leaf tolerance to desiccation allows leaves to function at more negative water potentials during the dry season ( Fig. 2c; Zhu et al. 2018), our results indicate that in this ecosystem, Ψ 50 and π tlp are not linearly coupled to ensure the maintenance of leaf functioning. Therefore, lower branch xylem vulnerability to embolism may not involve maintaining gaseous exchanges to maximize carbon gain as long as possible during drought (Jones and Sutherland 1991), but rather maintaining tissue water potential (Martin-StPaul et al. 2017). This indicates that preserving perennial organs (i.e., branches) is usually favored at the expense of cheaper, renewable organs (i.e., leaves). However, the six species that would reach Ψ 12 before π tlp may be able to tolerate a certain percentage of branch hydraulic conductivity loss or this may indicate that other mechanisms are acting (e.g., at the leaf level) to avoid the initial spread of branch embolism. In order to have a more comprehensive understanding of how the functioning of branches and leaves interact, we need to  (2018) and a severe (2008) dry season. a SM leaf was calculated as the difference between dry season midday leaf water potential (Ψ md ) and the water potential at turgor loss point (π tlp ; MPa) for, respectively, 28 and 12 canopy-tree species sampled in French Guiana. Species common to both years are identified by hash signs (#). b HSM Ψmd-Ψ50 was calculated as the difference between Ψ md and the water potential at 50% loss of branch hydraulic conductivity (Ψ 50 ; MPa) for, respectively, 23 and 12 canopy-tree species sampled in French Guiana. The ranking of species between both years was not retained for SM leaf (τ = 0.5; n = 8; ns) but was retained for HSM Ψmd-Ψ50 (τ = 0.71; p < 0.01). Each bar represents the species mean for 2018 (black; n ≥ 3; gray, n < 3) and 2008 (black dashes, n > 3; gray dashes, n < 3) and all error bars represent standard errors of the mean for SM leaf measured in 2018 investigate leaf drought resistance in greater depth, notably to evaluate the potential role of the loss of leaf conductance prior to turgor loss (i.e., the loss of leaf extra-xilary hydraulic conductance; Scoffoni et al. 2017), hydraulic segmentation (Pivovaroff et al. 2014), vulnerability segmentation (Zhu et al. 2016), and cuticular conductance (Duursma et al. 2019) as potentially determinant drought-resistance features in rainforest tree species.
During our study, we also evaluated whether species exceeded xylem hydraulic safety margins (HSM Ψmd-Ψ50 ) during a normal dry season. Our results show that most of the studied species operate without developing any xylem embolism. Only five showed slightly negative HSM Ψmd-Ψ12 meaning that branches could have been subjected to some degree of embolism. However, all remained above Ψ 50 (Fig. 2 and Table 1) and far from Ψ 88 (> 1 MPa), the threshold water potential thought to cause irreversible hydraulic failure in angiosperms ( Table 6 in the Appendices; Urli et al. 2013). However, six species had relatively narrow values of HSM Ψmd-Ψ50 , close to or lower than 1 MPa. Overall, our data support the view that the hydraulic safety margins of most tropical rainforest tree species are wide during a normal dry season (Delzon and Cochard 2014).
Limited risk of hydraulic failure during a severe dry season In 2008, our site experienced the second most severe dry season in the past four decades (Aguilos et al. 2019). The SM leaf values estimated in 2008 were equal to or below zero, indicating that most species had their stomata closed ( Fig. 3a and Table 8 in the Appendices). Additional water loss and the resulting decline in water potential from π tlp to Ψ md in 2008 could therefore have been due to cuticular transpiration (Martin-StPaul et al. 2017;Choat et al. 2018;Duursma et al. 2019). In these conditions, water potential gradients between branches and leaves are strongly reduced and midday leaf water potential measured that year can therefore be considered as an estimate of the lowest branch water potential these trees may have experienced in the past four decades (i.e., minimum water potential). However, the same year, though there was a clear decrease in photosynthetic assimilation for most of the trees, it was never null, some trees were not affected , no leaf mortality was observed (Clément Stahl, personal communication), and litterfall was similar to other dry seasons . Despite considerable diversity of response to drought, this suggests that no leaves were fully embolized.
Additionally, we found no clear pattern in species SM leaf between a normal and a severe dry season (Fig. 3a), indicating that the response to drought might be non-linear and affected by threshold effects depending on the intensity and duration of the dry season. The lack of a clear pattern could also arise from the large uncertainties involved or the low sample size for some of the measured species. For the ten species in common to both years, the calculated HSM Ψmd-Ψ50 was narrower in 2008 than in 2018, but always remained positive (Figs. 3b; Fig. 7 and Table 8 in the Appendices). Unfortunately, Ψ md and therefore HSM Ψmd-Ψ50 were not available in 2008 for some species with the narrowest HSM Ψmd-Ψ50 in 2018. Yet, species' ranking was comparable between the 2 years (τ = 0.71; p < 0.01; data not shown), hinting that some species with relatively narrow HSM Ψmd-Ψ50 in 2018 could have suffered from reduced branch hydraulic conductivity during the severe 2008 dry season due to partial branch xylem embolism. However, this was not a general rule since our data support that most canopy tree species, which experience steep gradients in evaporative demand even during the rainy season, develop an embolism resistant xylem, and share a limited risk of hydraulic failure during a severe dry season. Such severe dry season droughts due to anthropogenic-induced climate change are projected to increase over the next century in the Eastern-Amazon (Duffy et al. 2015). Though it is believed that previous shifts in species composition occurred throughout the late Pleistocene and Holocene because of dramatic changes in climate and in rainfall seasonality ), a vast majority of the Amazon basin remained as intact forest (Mayle and Power 2008). At present, there is already evidence that the recruitment of drought-tolerant taxa has increased over the past three decades in Amazonia in areas affected by droughts, while mortality among taxa affiliated to wet conditions has simultaneously increased (Esquivel-Muelbert et al. 2019). However, such changes in tree communities lag behind the pace of climate change, thus raising concern about the fate of these rainforest ecosystems. A trait-based approach encompassing a mechanistic understanding of drought resistance may help decipher what facilitates such changes in species composition and may help predict how long species will be able to avoid reaching catastrophic levels of branch embolism, which still needs to be addressed by modeling approaches.

Conclusions
In this study, we report key branch and leaf hydraulic traits related to drought response and resistance for a set of 30 co-occurring Neotropical rainforest canopy-tree species. We found that (i) rainforest canopy-tree species that grow with elevated mean annual precipitation and yearly experience several months of meteorological drought and reduced soil water availability can have high resistance to embolism and are more resistant than what was previously thought and that (ii) most species have a low risk of hydraulic failure and are well able to withstand normal and even severe dry seasons thanks to early leaf turgor loss and low branch xylem vulnerability to embolism. Yet, in the forest we studied, such a severe drought has not occurred in the previous four decades and it is therefore unlikely that canopy-tree species suffered from hydraulic failure during that time, even during severe dry seasons as the one experienced in 2008. Our results therefore support the idea that, at least for the most abundant canopy-tree species, expected climate change over the next decades, at least in terms of water availability, will not directly or severely affect the survival of these species at the adult stage in terra-firme forests of the Guiana Shield. Yet, considering that the studied canopy trees are long-living organisms, they may have previously experienced severe droughts in the past beyond our records and will most likely do in a near future, while considerable uncertainties remain on their capacity to withstand repeated drought episodes. Whether species capacity to avoid hydraulic failure is influencing the slow compositional shift in the Amazon rainforest at the early development stage (Esquivel-Muelbert et al. 2019) in response to increasingly frequent and severe dry seasons is a worthy question. Moreover, little is known about the drought-resistance characteristics of the less-abundant species that compose the vast majority of Neotropical tree communities. We therefore encourage further research in this domain to better predict tree species' and forest communities' fate in a rapidly changing climate (Brodribb 2017).
Data availability statement All data corresponding to species' mean trait values are included in this published article (and its supplementary information files). All raw data generated during the current study are not publicly available at the moment because the authors of the study wish to keep them for further analysis. They will then be made public on an online repository. In the meantime, they can nonetheless be made available by the corresponding author on request.

Compliance with ethical standards
Conflicts of interest The authors declare that they have no conflict of interest.       shaded bands represent standard errors; and a 50% loss in conductivity is indicated by a horizontal dotted line for Gustavia hexapetala (red; n = 3), the most resistant species in our dataset, Chrysophyllum sanguinolentum (blue; n = 5), an intermediately-resistant species, and Lecythis poiteaui (green; n = 4), the least resistant species Fig. 7 Relationships between Ψ 50 , the water potential at 50% loss of branch hydraulic conductivity (Ψ 50 ; MPa), and midday leaf water potential measured during the severe dry season of 2008 (Ψ md ; MPa) for 10 canopy-tree species sampled in French Guiana (black n ≥ 3; gray n < 3). All error bars represent standard errors of the mean.