Causes of variation in leaf-level drought tolerance within an Amazonian forest

Amazonian tree communities have already been seriously impacted by extreme natural droughts, and intense droughts are predicted to increase in frequency. However, our current knowledge of Amazonian tree species’ responses to water stress remains limited, as plant trait databases include few drought tolerance traits, impeding the application and predictive power of models. Here we explored how leaf water potential at turgor loss point (πtlp), a determinant of leaf drought tolerance, varies with species life history, season, tree size and irradiance within a forest in French Guiana. First, we provided a further direct validation of a rapid method of πtlp determination based on osmometer measurements of leaf osmotic potential at full hydration for five Amazonian tree species. Next, we analysed a dataset of 131 πtlp values for a range of species, seasons, size (including saplings), and leaf exposure. We found that early-successional species had less drought-tolerant leaves than late-successional species. Species identity was the major driver of πtlp variation, whereas season, canopy tree size and leaf exposure explained little variation. Shifts in πtlp from saplings to canopy trees varied across species, and sapling leaf drought tolerance was a moderate predictor of canopy tree leaf drought tolerance. Given its low within-species variability, we propose that πtlp is a robust trait, and is useful as one index of species’ drought tolerance. We also suggest that measuring this trait would considerably advance our knowledge on leaf drought tolerance in hyperdiverse communities and would thus likely shed light on the resilience of such vulnerable species-rich ecosystem.


Introduction
Extreme natural droughts have already impacted on Amazonian ecosystem processes and services, through a decrease in tree growth and an increase in tree mortality (Phillips et al. 2009;Lewis et al. 2011;Marengo et al. 2011;Saatchi et al. 2013).In the future, such intense droughts are predicted to increase in frequency over Amazonia (Joetzjer et al. 2013;Boisier et al. 2015).Understanding and predicting the dynamics and responses to drought of Amazonian tree communities thus constitutes a crucial research challenge.We do know that plant water availability is a strong driver (iv).Consistent with hypothesis iii, some studies reported tall trees to be more vulnerable to drought than smaller trees (Van Nieuwstadt & Sheil 2005;Nepstad et al. 2007;da Costa et al. 2010;Phillips et al. 2010;Rowland et al. 2015), although others did not (Enquist & Enquist 2011;Feeley et al. 2011;Fauset et al. 2012).Disentangling the potential structural and physiological causes of these opposed trends has proven difficult (Niinemets 2010).To our knowledge, variation in π tlp has never been compared between mature trees and saplings.Canopy trees often include both shade and sun leaves, the latter being more exposed to heat light and evaporative stress and differing predictably in a number of structural and physiological traits (Sack et al. 2003;Rozendaal et al. 2006;Sack et al. 2006;Markesteijn et al. 2007).We thus predicted sun leaves would show greater drought tolerance, and thus more negative π tlp values than shade leaves within the crown of a single tree (v).

Study sites and sampling strategy
Field measurements were conducted at the Nouragues Ecological Research Station in French Guiana.It is located 120 km south of Cayenne within an undisturbed forest, ca.50 km from Cacao, the closest village (4°05' N, 52°40' W; Bongers et al. 2001).The site receives ca.3000 mm/yr rainfall, with significant seasonal and inter-annual variation due to the movement of the Inter-Tropical Convergence Zone.A long wet season lasts from December to July, often interrupted by a short dry period in March.The dry season lasts from the end of August to November with 2-3 months with precipitation <100 mm/mo (Bongers et al. 2001).Data were collected in a 12-ha permanent forest plot on claysand soil (Petit Plateau).The permanent plot was fully censused in 2012, all trees ≥10 cm diameter at breast height (DBH) were identified and their DBH measured.Within the 12-ha plot, richness is in excess of 600 tree species (C Baraloto, J Engel, P Pétronelli, unpublished results).
Measurements of leaf water potential at turgor loss point (π tlp ) were conducted in September 2012 and September 2015 during two dry seasons, and in May 2014 in the middle of the wet season (Appendix S1).The cumulative rainfall 30 days prior the ending date of each sampling session, based on logging half-hourly data at the site (Campbell Scientific SBS500, Shepshed, Leicestershire, UK), was 21 mm, 78 mm and 265 mm for the 2012 dry season, the 2015 dry season, and the 2014 wet season respectively.Ten focal species were selected that were relatively abundant in the study plot to maximize the range in π tlp (Maréchaux et al. 2015).Three to eleven canopy trees were sampled per species in each season (Table 1).Small branches were collected by climbing the trees using French spikes (Fonderies Lacoste, Excideuil, France;de Castilho et al. 2006) or using the single-rope technique (Anderson et al. 2015).For 19 of the 56 canopy trees sampled during the wet season, both sun and shade leaves could be sampled.For five of the ten focal species, saplings (individuals with a height typically less than 1.50m, thus growing in the understory) could be easily identified, and three to seven saplings were sampled per species during the wet season in May 2014 (n=24 saplings total; Table 1).

Estimating π tlp using the osmometer method
We assessed the leaf turgor loss point, π tlp , from a previously established linear relationship with the osmotic potential at full hydration (π o ), in turn directly measured with a vapour pressure osmometer (Vapro 5520, Wescor, Logan, UT; Bartlett et al. 2012a).Henceforth, we refer to this technique as the osmometer method.
Branches were collected from the selected trees and placed in opaque bags with wet paper towel, then recut under water at least two nodes distal to the original cut and allowed to rehydrate overnight covered with plastic, and branch ends underwater.Three mature leaves were collected from each rehydrated branch and stored in Whirl-Pak bags (Nasco, Fort Atkinson, WI, USA) with a wet paper towel.One disc was sampled from each leaf centrally between the midrib and margin with an 8-mm-diameter cork borer, excluding first-and second-order veins to avoid too much apoplastic dilution that would lead to less negative osmometer values (Kikuta & Richter 1992).However, for one focal species, Qualea rosea (Vochysiaceae), a dense secondary venation disallowed excluding them.The discs were wrapped in foil and frozen by immersion in liquid nitrogen (LN 2 ) for at least 2 minutes, then immediately punctured 10-15 times with sharp-tipped forceps and sealed in the osmometer chamber, using the standard 10 µL chamber well.The discs were exposed to air for less than 40 seconds for all steps between removing the leaf from the Whirl-Pak bag and sealing the disc in the osmometer.The equilibrium solute concentration value c 0 (in mmol kg -1 ) was recorded from the osmometer when the difference between consecutive 2-minute measurements fell below 5 mmol kg -1 .This value was converted to π o values from the osmometer (π osm ) using the van't Hoff equation relating solute concentration to vapour 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 Kelvin degrees.The value π osm was then converted into π tlp using the previously published regression equation relating π osm to pressure-volume curve turgor loss point values (π tlp-PV ; Bartlett, Scoffoni, Ardy, et al. 2012).This regression was developed from woody species representing a wide range of leaf structural features and habitat preferences (n=30 species, R 2 =86%, p<2.10 -12 , π osm range= [-3.03, -0.64] MPa and π tlp-PV range=[-3.00,-1.02] MPa) and was: !"# = 0.832 ×  !"# − 0.631 (2) This reference regression is based on fundamental equations describing biophysical processes and should therefore hold universally across species.However, we sought to further validate this equation for Amazonian tree species.

Validating the osmometer method for Amazonian tree species
To this end, we calculated the water potential at turgor loss point using the pressure-volume curve method (henceforth denoted the PV curve method; π tlp-PV ) and also measured π osm with the osmometer method on the same individuals from six species (n=18 individuals and 6 species).Our validation was conducted in September 2015.Species were excluded if they produce foliar latex or resin, known to create problems with the PV curve method.Branches were allowed to rehydrate overnight as above.Five to six leaves from three individuals per species were dried on a laboratory bench and repeatedly weighed (using a Ohaus Scout Pro Balance SPU 123, Parsippany, NJ, USA) and measured for water potential with a pressure chamber with a digital gauge (PMS Instrument Co., Corvallis, OR, USA).Leaf dry mass was determined after 72 h at 70°C in a drying oven.π tlp-PV was calculated from the relationships between water potential and water content (Koide et al. 2000;Sack et al. 2011).Species-level π tlp-PV was computed as the average of π tlp-PV values of the 5-6 leaves from that species.All π osm and π tlp-PV values are available (Appendix S2).
We compared the root mean squared error (RMSE) from Equation (2) computed for the published calibration dataset (n=30 species; Bartlett et al., 2012) and for the dataset including the six additional species in this study (n=6 and n=36 species).RMSE was computed as follows: where π tlp values were turgor loss point estimates with Equation (2).We identified potential outliers in the merged dataset (n=36 species) using jackknife resampling method with RMSE as statistics (R package bootstrap; Tibshirani & Leisch 2015).We also fitted the 6-species and the merged dataset and compared the 95% confidence intervals of slope and intercept of the 6-and 36-species datasets with the 30-species dataset based on 1000 bootstraps (R package boot; Canty & Ripley 2013).

Data analyses
All analyses were conducted for the wet-season π tlp values (May 2014), except for the comparison between seasons.We considered the mean of the sun and shade leaves π tlp values for individuals for which both were collected, excepted when comparing leaves sampled at different light exposures.Also, for reasons explained below, individuals from one species, Qualea rosea, were excluded from all subsequent tests.
To test for species differences in π tlp values, we used linear models (ANOVA), with species as a fixed effect, and Tukey HSD tests for pairwise species comparisons.The relative contribution of within-and across-species differences to π tlp variability was quantified by partitioning variance (Sokal & Rohlf 1987).We also tested for differences in π tlp between species grouped into early-successional, mid-successional and late-successional species according to field observations (Table 1), using an ANOVA, with successional status as a fixed effect, and Tukey HSD for post hoc pairwise comparisons.
We compared wet-season canopy tree values with dry-season values with the same osmometer method (Maréchaux et al. 2015) for six species with at least three individuals sampled during each season.We tested for a seasonal effect using a two-way ANOVA with season and species as fixed effects and using t-tests for each species.
Using DBH as a proxy for tree size, we investigated the relationships between canopy tree DBH and π tlp , and repeated this analysis accounting for the species effect (linear regressions between DBH and residuals of the ANOVA on π tlp with species as a fixed effect) to exclude variability caused by species differences in stature.We also tested differences in π tlp for four species for which we sampled both canopy trees and saplings, using a two-way ANOVA (type II) with size (categorical: sapling/canopy) and species as fixed effects.
We tested differences in π tlp between leaves collected in full sunlight at top canopy height (henceforth denoted sun leaves) and leaves collected in shaded conditions lower in the canopy (shade leaves) within individuals using a paired t-test (n=18 individuals).We validated our classification of sun and shade leaves by testing for differences in additional traits known to vary with light environments (Niinemets et al. 2015), i.e., leaf mass per area (P<10 -4 , mean of the differences=15.3g.m -2 ), and carbon isotope ratio (P<10 -3 , mean of differences=1.23 ‰; data not shown).
All statistical analyses were conducted using the R software (R Core Team 2013) and preceded by normality and homoscedasticity tests when required, using Shapiro-Wilk test, Bartlett test or Levene test (car package; Fox & Weisberg 2011).

Model validation with Amazonian tree species
The species Qualea rosea, with its dense secondary vein network was clearly identified as an outlier: for this species, the PV-curve value of π tlp value was much lower than that obtained with the osmometer method (π tlp -π tlp-PV =0.58 MPa).We decided to remove this outlier species from all subsequent analyses.The RMSEs of the 5-and 35-species datasets were both 0.18 MPa, equal to that of the previously constructed 30-species model (Bartlett et al., 2012).95% confidence intervals of the slope and the intercept of both of the 5-and 35-species regressions included the parameter estimates for the previous 30-species calibration.The new correlation between π osm and π tlp-PV , established based on the 35 species was  !"# = 0.799 ×  !"# − 0.665 (4) (P<10 -14 , R 2 =0.85,RMSE=0.18MPa, Fig. 1).We conclude that this relationship is applicable to Neotropical trees, as assumed previously (Maréchaux et al. 2015), provided the leaf secondary vein network is not too dense.
Figure 1: Validation of the osmometer method for Amazonian tree species.Linear regressions between species leaf water potentials at turgor loss point measured with the PV curve method (π tlp-PV , in MPa) and species osmotic potentials at full hydration measured with an osmometer (π osm , in MPa), for the (Bartlett et al. 2012) 30 species means (solid line) only and including the five species in this study (excluding Qualea rosea, 35 species, dashed line).The regressions were not significantly different.

Seasonal and size-related variation in π tlp within species
We found no difference between dry-and wet-season π tlp for any of the tested tree species (n=6 species; two-way ANOVA; t-tests: all p-values > 0.3, Table 1, Fig. 2).This was still the case when excluding data collected during the September 2015 sampling session which was less dry than the particularly stressed period of September 2012.The π tlp was not statistically related to DBH among canopy trees (linear regression, P=0.44, n=50 individuals).Further, DBH did not explain residual π tlp variance among canopy trees after accounting for the species effect (linear regression between DBH and the residuals of ANOVA on π tlp with species as a fixed effect, P=0.43).This result also held when excluding individuals growing in large open gaps (n=7 species and 37 individuals, P=0.48), for which a variation in size may not result in a variation in abiotic stresses due to changes of vapour pressure deficit, temperature or light intensities.
Species, plant stature (canopy vs sapling) and the interaction of these two effects had a significant effect on π tlp (two-way ANOVA, n=4 species, all P≤0.05, range of average π tlp differences between stature within species: [-0.23, -0.11] MPa).For all species, canopy trees displayed a more negative π tlp than saplings, except the earlysuccessional Pourouma sp. which showed the opposite pattern (Table 1, Fig. 2).

Validation of the osmometer method for Amazonian tree species
We applied a rapid method of leaf turgor loss point determination based on osmometer measurements of leaf osmotic potential at full hydration (π osm ).The correlation between π osm and π tlp was initially established for 15 species sampled in gardens adjacent to the campus of University of California, Los Angeles, from vegetation zones ranging from chaparral to tropical wet forest, and a further 15 species growing in natural conditions in a Chinese tropical rainforest (Bartlett, Scoffoni, Ardy, et al. 2012).These two subgroups did not differ statistically in the relationship between measurement methods, consistent with the expectation that because the relationship is biophysically based, it can be applied across vegetation zones (Bartlett, Scoffoni, Ardy, et al. 2012).Here, we included five more species from canopy tropical trees, and confirmed the strong consistency of the correlation, lending further credence to the proposal that it can be applied universally across angiosperm species.Model error remained consistently low after adding the new species.Indeed, the new correlation was not significantly different from the previously established one.We thus conclude that the original model, or its slight improvement reported in Equation ( 4), can be reliably used to infer leaflevel potential at wilting point across a wide range of angiosperm species.The rapid method offers invaluable opportunities to better inform and quantify the diversity of leaf drought tolerance in such vulnerable species-rich plant communities.The π tlp has been used as a proxy for general plant drought tolerance, because in several species sets it was notably correlated with specialization of species between wet and dry forests and biomes (Bartlett, Scoffoni, & Sack 2012) and the water potential at stomatal closure (Brodribb et al. 2003;Brodribb & Holbrook 2003;Blackman et al. 2010), even though more research is required to further elucidate the mechanistic relationships among traits which could potentially arise due to mechanistic linkage and/or to simultaneous co-selection during evolution, e.g., for performance under a given moisture availability regime.Besides, in particular for tropical tree species, the measurement of xylem resistance to cavitation is challenging because of the presence of long vessels (Martin-StPaul et al. 2014;Torres-Ruiz et al. 2014 ;Delzon 2015), and few measurements of the xylem pressure at which 50% loss of conductivity occurs (P 50 ) are available, although these trait was correlated with π tlp in several data sets (Choat et al. 2007;Blackman et al. 2010;Fu et al. 2012).Nine published P 50 values for tree genera present at our site were reported in the global database of Choat et al. (2012), including six genera collected at the Parque Nacional San Lorenzo, Panama and three genera at La Selva Research Station, Costa Rica.The range of P 50 values ([-2.70;-1.00] MPa) was comparable to the range of π tlp values found at our site (this study: [-2.52; -1.43] MPa and Maréchaux et al. 2015: [-3.15;-1.41]MPa).Since π tlp is currently the only easily measurable drought tolerance trait for tropical woody angiosperms, it is important to further explore the relationship between P 50 and π tlp for these species.

Sun leaves
One species, Qualea rosea, departed from the model's predictions.This species displayed a dense network of leaf secondary veins, and the osmometer method led to an over-estimated (less-negative) turgor loss point value.This is consistent with previous studies finding that including secondary veins in a sample leads to a greater apoplastic dilution and thus less negative osmometer measurements (Kikuta & Richter 1992).Species with very dense secondary vein networks should be carefully excluded when using this method.In the French Guiana flora, the number of species with dense secondary venation is limited: only about 2.5% of the tree genera, and <2% of tree species contain species with a dense secondary venation (J Engel, personal communication).
Variation in π tlp strongly determined by species identity and life history Species identity was the major determinant of leaf drought tolerance, and inter-specific variation in π tlp was larger than intra-specific or intra-individual variation.In contrast, season, canopy tree size or leaf light exposure did not explain much of this variability.
Species identity explained 87% of the variance of π tlp in our dataset (n=9 species, 50 individuals), a larger proportion than previously found (40 %, n=13 species, 82 individuals; Maréchaux et al., 2015).This result is in agreement with previous studies in other ecosystem types (Lenz et al., 2006;Bartlett et al., 2015).Overestimation of interspecific variation in plant traits may result from standardized protocols minimizing intraspecific variation (Cornelissen et al. 2003;Violle et al. 2012).However, this alone could not explain our pattern since our leaf sampling was not standardized with respect to light exposure or canopy tree size and was designed to maximise the entire range of the selected species.We acknowledge that we sampled species at a single site which may induce a lower intraspecific variability than if we had sampled across many sites (Hulshof & Swenson 2010;Albert et al. 2010;Siefert et al. 2015).However in a previous study, variation as assessed at two sites did not result in a greater intraspecific variation in π tlp (Maréchaux et al. 2015).
As hypothesized from a whole-plant "fast-slow" economics spectrum (Reich 2014) and reported coordination of species' drought and shade tolerance (Markesteijn, Poorter, Bongers, et al. 2011;Markesteijn, Poorter, Paz, et al. 2011), early-successional species tended to exhibit less negative π tlp values than later-stage species.This is in agreement with light-demanding species showing higher maximum stomatal conductance and assimilation and transpiration rates than shade-tolerant species under well-watered conditions, but a stronger sensitivity of these variables to soil drought, probably due to a more sensitive stomatal control on average in early-successional species (Huc et al. 1994;Bonal et al. 2000).

Absence of seasonal osmotic adjustment
We found no seasonal adjustment in π tlp for the five species with both dry and wet seasons values in our dataset.This result reinforces the view that some tree species in French Guiana are far more drought-tolerant than previously assumed for moist tropical forests, perhaps due to adaptation to past and recurring drought events in the region (Maréchaux et al. 2015).This finding runs contrary to what was expected from a global meta-analysis that reported consistent and prevalent seasonal plasticity in π tlp across 246 species from eight biomes worldwide (Bartlett et al. 2014).However, that meta-analysis also concluded that there is a wide variability in plasticity across species.The few existing studies that reported tropical moist species π tlp values across seasons provided mixed evidence for osmotic adjustment in tropical tree leaves (Wright et al. 1992;Cao 2000;Brodribb et al. 2003).
One reason explaining the absence of osmotic adjustment in our dataset could be that dry seasons rarely result in a strong tree physiological stress in French Guiana.Values of predawn and midday leaf water potentials measured during an average dry season (Stahl et al. 2011;Stahl et al. 2013) usually were less negative than π tlp values reported at the same site (Maréchaux et al. 2015): at the end of the dry season in November 2007, the range of predawn leaf water potentials were [-1.30;-0.15]MPa, 75% of which were less negative than -0.45 MPa (n=64 individuals), while the range for midday leaf water potentials were [-2.95;-0.60]MPa, 75% of which were less negative than -1.90 MPa (n=49 individuals; C Stahl personal communication).Thus, we do not exclude the possibility of osmotic adjustment under harsher droughts.
π tlp does not depend on canopy tree size, but differs for saplings In agreement with our prediction, we found that canopy tree size did not have a significant effect on π tlp , whether through a species effect or within species.This finding is interesting in light of studies that have suggested a higher vulnerability to drought of large trees compared with smaller trees (Van Nieuwstadt & Sheil 2005;Nepstad et al. 2007;da Costa et al. 2010;Phillips et al. 2010;Bennett et al. 2015), although this pattern was not observed in other studies (Enquist & Enquist 2011;Feeley et al. 2011;Fauset et al. 2012).Emergent trees are indeed exposed to greater atmospheric evaporative demand and hydraulic constraints.However, these large trees can also store larger amounts of water and thus withstand prolonged periods of stomatal closure (Goldstein et al. 1998;Meinzer et al. 2008), compete strongly for soil resources with their large root systems and access the water table deeper down (Nepstad et al. 1994;Dawson 1996;Ivanov et al. 2012; but see Meinzer et al. 1999;Markewitz et al. 2010;Stahl, Herault, et al. 2013), and accumulate more non-structural carbohydrates that may help maintain high osmotically active compound concentrations and thus attenuate the effect of drought (Sala & Hoch 2009;Woodruff & Meinzer 2011;Sala et al. 2012;O'Brien et al. 2014 ; but see Rowland et al. 2015).
Our results indicate that tall tropical trees do not have a higher leaf-level drought tolerance relative to smaller trees.This in turn suggests that large trees are not more vulnerable to drought than smaller trees.However we found evidence for a weak shift in leaf drought tolerance between saplings and canopy trees.Notably, early-and latesuccessional species showed opposed trends in their variation in π tlp between saplings and trees, although not all were significant in our dataset of limited sample size.In the late-successional species, leaves of saplings were less droughttolerant than those of canopy trees.In contrast, for species growing in gaps, leaves were more drought-tolerant at the sapling stage.Such contrasting shifts from saplings to canopy trees across species suggest that these species operate under differential ecophysiological constraints.Differential ability to cope with drought across plant sizes has been reported in relation to changes in environmental constraints (Donovan & Ehleringer 1991;Donovan & Ehleringer 1992;Cavender-Bares & Bazzaz 2000;Niinemets 2010).Soil water availability also differs between gaps and the understory: gaps are more controlled by the evaporation of soil surface while closed-canopy forest soil water is depleted by dense root water extraction below the surface (Becker et al. 1988;Marthews et al. 2008).Saplings of early-successional species, growing in gaps, with typically more superficial roots than taller individuals despite similar high-light exposure, may thus be strongly stressed even during short dry spells in the wet season (Engelbrecht et al. 2006).On the contrary late-successional saplings, growing in closed understory, may be more strongly impacted by herbivory and thus invest more in leaf defense against herbivores than in drought tolerance (Coley & Barone 1996).It would be useful to increase sample size to further analyse these trends, as the contribution of shifts in π tlp to defining the ecological niche of plant species seems to be a promising research avenue.

A weak variation in π tlp across leaf exposure within tree canopy
Sun leaves showed structural and physiological acclimation to high radiation loads and high vapour pressure deficit, with typically higher mass per area and nutrient concentration per area than shade leaves, as expected from previous studies (Sack et al. 2003;Rozendaal et al. 2006;Sack et al. 2006;Markesteijn et al. 2007;Hulshof & Swenson 2010;Weerasinghe et al. 2014).However, we found no substantial differences in absolute π tlp values between sun and shade leaves just as reported previously for temperate woody species (Sack et al. 2003).A strong acclimation in traits related to maximum flux of water and carbon through the leaf, such as leaf hydraulic conductance, vein density, and photosynthetic gas exchange may occur alongside little acclimation in traits associated with leaf drought tolerance (Sack et al. 2003;Sack & Holbrook 2006).

On measuring the community-wide leaf water potential at turgor loss point
The rapid technique of π tlp determination validated here for Amazonian tree species opens new possibilities for studying drought tolerance within and across species.This gives access to a key drought tolerance trait as part of the plant ecologist's toolkit of routinely measured functional traits (Cornelissen et al. 2003).We propose that it should now be possible to survey the π tlp in hyperdiverse communities and thus shed light on both the resilience of this ecosystem and on how the Amazonian flora has been assembled.
We found that species identity was the main driver of π tlp values with early-successional species being less drought-tolerant than later-successional species, and that season, canopy tree size and leaf exposure contributed relatively negligible variation.Yet, we explored only some of the factors that might affect inter-individual variability, species plasticity, and species differences.For instance we did not study the effect of topography (Silva et al. 2013;Liu et al. 2014) or species maximum height (Poorter et al. 2005).Intra-specific variability, whatever the drivers, may have direct consequences on sampling design and effort for community-wide measurements.However, our result confirms that π tlp could be used as a suitable species-level trait of tree drought tolerance, thus considerably simplifying sampling and determination of drought tolerance in species-rich communities (Baraloto et al. 2010).
We also provided evidence of an ontogenetic shift in π tlp that is variable across species.Particular care should thus be taken when scaling to canopy trees leaf-level drought tolerance measurements based only on saplings (Engelbrecht et al. 2007;Kursar et al. 2009).More research is needed to elucidate how such a variation across ontogeny and across species may impact species performance and vital rates across the life cycle (Adler et al. 2014) and under multiple stresses (Niinemets 2010).

Figure 2 :
Figure2: Leaf water potential at wilting point (π tlp , in MPa) across species in our dataset, with mean ± standard error for species wet-season canopy tree (black), dry-season canopy tree (dark grey) and wet-season sapling (light grey) values.* indicates significant differences between saplings and canopy trees during the wet season within species (t.test : P<0.05).

Figure 3 :
Figure3: Differences in leaf water potential at wilting point (π tlp , in MPa) across successional stages, for early-, mid-and latesuccessional canopy trees (n=11, 10 and 29 individuals respectively).Means ± standard errors are reported.Different lower case letters indicate a significant difference between stages (Tukey test: P<0.05).

Figure 4 :
Figure 4: Comparison between sun and shade leaves leaf water potential at wilting point (π tlp , in MPa) for 18 tropical canopy trees, with mean ± standard error.

Table 1 :
Mean and standard error of π tlp values in wet and dry seasons and for saplings and canopy trees.
ES: early-successional; MS: mid-successional; LS: late-successional.Same lower case letters indicate a non-significant difference between species values for canopy trees (Tukey test: P>0.05).