Plant resistance to drought relies on early stomatal closure

Plant resistance to drought has long been thought to be associated with the ability to maintain transpiration and photosynthesis longer during drought, through the opening of stomata. This premise is at the root of most current framework used to assess drought impacts on land plants in vegetation models. We examined this premise by coupling a meta-analysis of functional traits of stomatal response to drought (i.e. the water potential causing stomatal closure, Ψclose) and embolism resistance (the water potential at the onset of embolism formation, Ψ12), with simulations from a soil-plant hydraulic model. We found that Ψclose and Ψ12 were equal (isometric) only for a restricted number of species, but as Ψ12 decreases, the departure from isometry increases, with stomatal closure occurring far before embolism occurs. For the most drought resistant species (Ψ12<-4.2 MPa), Ψclose was remarkably independent of embolism resistance and remained above −4.5 MPa, suggesting the existence of a restrictive boundary by which stomata closure must occur. This pattern was supported by model simulations. Indeed, coordinated decrease in both Ψclose and Ψ12 leads to unsuspected accelerated death under drought for embolism resistant species, in contradiction with observations from drought mortality experiments. Overall our results highlight that most species have similarity in stomatal behavior, and are highly conservative in terms of their water use during drought. The modelling framework presented here provides a baseline to simulate the temporal dynamic leading to mortality under drought by accounting for multiple, measurable traits.

embolism resistant species, in contradiction with observations from drought mortality experiments. 23 Overall our results highlight that most species have similarity in stomatal behavior, and are highly 24 conservative in terms of their water use during drought. The modelling framework presented here 25 provides a baseline to simulate the temporal dynamic leading to mortality under drought by accounting 26 for multiple, measurable traits. and have undoubtedly affected a panel of ecosystem services [5]. Identifying the mechanisms and traits 35 underlying drought resistance is essential to understanding and predicting the impact of widespread 36 droughts over many land areas. Experimental studies have provided empirical evidence that the failure 37 of the water transport system is tightly linked to tree dehydration and mortality in drought conditions. 38 This was confirmed by a recent study reporting that global patterns of mortality was predictable from 39 hydraulic safety margins [6]. Two key types of traits shaping the trade-off between drought resistance 40 and the maximization of carbon dioxide assimilation have been identified: hydraulic traits ensuring the 41 integrity of the hydraulic system under drought [7], and stomatal traits controlling gas exchanges at the 42 leaf surface [8]. However, efforts to model tree mortality in response to drought is still hindered by a lack 43 of understanding of how these traits interact to define physiological dysfunctions under drought stress 44 [9]. In this study, we analyzed the overall connections between these two types of traits for the full range 45 of drought resistance, with a soil-plant hydraulic model. 46 The stomata have two key functions: controlling transpiration, which supplies nutrients and 47 regulates leaf temperature, and controlling the entry of CO2 into the leaf. Stomatal closure in response 48 to water deficit is the primary limitation to photosynthesis [10], and constitutes a key cost in terms of 49 plant growth and leaf temperature under drought conditions. However, stomatal closure also limits 50 excessive decreases in water potential (quantified as a negative pressure, ψ) in the plant, thereby 51 ensuring that the water demand from the leaves does not exceed the supply capacity of the hydraulic 52 system, which would lead to embolism of the vascular system and complete dehydration of the plant. 53 These key, but opposing roles of stomata in regulating CO2 influx and H2O loss pose a dilemma that has 54 occupied scientists for centuries 1 and has led to the view that plant stomata probably operate at the 55 edge of the supply capacity of the plant's hydraulic system, to balance different cost such as productivity 56 leaf temperature regulation during drought [11,12]. 57 Conversely, maintenance of the supply capacity of the hydraulic system depends on the ability 58 of a species to sustain high negative pressure to limit embolism. Embolism resistance is usually 59 quantified by the value of ψ causing 50% embolism (Ψ50), and the rate of embolism spread per unit drop 60 in water potential (slope). From these two traits the Ψ at the onset embolism formation can be computed 61 (Ψ12, see equations 1 to 3), which give a more conservative functional limit to the hydraulic system. 62 Embolism resistance varies considerably between species and with the dryness of species habitat 63 [7,13,14]. A recent study has suggested that hydraulic systems highly resistant to embolism have 64 evolved in response to the selective pressure associated with increasing drought levels during a 65 paleoclimatic crisis [15]. Some contemporary plants have extremely drought-resistant vascular systems, 66 with Ψ50 values reaching -19 MPa [16]. 67 These findings have led to the suggestion that an efficient match between the capacity of the 68 hydraulic system to sustain water deficit (i.e. embolism resistance) and the regulation of demand by the 69 stomata is a prerequisite for the maximization of gas exchanges without dehydration [11,[17][18][19]. This 70 notion naturally leads to the hypothesis that stomatal behavior and embolism resistance have followed 71 a similar evolutionary trajectory under drought constraints, and that plants have increased their intrinsic 72 embolism resistance to allow stomata to close later during drought, thereby maximizing plant productivity 73 [8,12,20]. The coordination of stomatal and hydraulic traits and their role in shaping drought resistance 74 has yet to be addressed on a global scale. This would help to clarify the interplay between mechanisms 75 and plant traits in defining the physiological dysfunctions occurring under drought stress, whic h remains 76 one of the principal challenge faced in the modeling of tree mortality in response to drought. 77 In this study we gathered stomatal regulation traits and embolism resistance traits for different 78 species. We compiled ψ12, ψ50 and slope values derived from stem vulnerability curves published by us 79 during the past 20 years, representing 151 species from different biomes (Tables S1). Recent direct 80 observations of embolism formation by mean of X-ray tomography [21][22][23][24] confirmed the reliability of 81 these values. For those species, we performed a seach for data of water potential causing stomatal 82 closure (Ψclose). We used concurrent measurements of gas exchange and leaf water potential, from 83 which the Ψ value at 90% stomatal closure was calculated following [8,20]. Stomatal opening increases 84 with guard cell turgor pressure [25][26][27], thus, we also used leaf water potential at turgor loss (Ψtlp) as a 85 surrogate for Ψclose. We then explored the range of variation and the coordination between Ψclose and 86 Ψ50. Finally, we used a soil-plant water transport model to elucidate how different associations between 87 Ψclose and Ψ50 determine the time until hydraulic failure during drought (see Methods). We validated 88 model predictions using empirical data for time to shoots death collected in drought mortality 89 experiments[28-31] (see Methods and Appendix 1). 90

RESULTS AND DISCUSSION 91
Embolism resistance (taken as the Ψ50) ranged between -1.3 and -19 MPa (Figure 1   The relationship between Ψ12 and Ψclose did not follow the isometric line (i.e. the 1:1 line, Figure  116 1c), but had a slope lower than unity : 0.4 for species with low embolism resistance (Ψ12>-4.2) and null 117 for drought resistant species p.value=0.4). This contradicts the expected match between 118 Ψ12 and Ψclose and indicates that the departure between Ψ12 and Ψclose increase with increasing embolism 119 resistance. These results were confirmed when we used Ψ50 as an indicator of embolism resistance 120 (Figure 1 D, Table S2). The relationship between Ψclose and embolism resistance (Ψ50 , Ψ12) presented 121 a marked interruption (i.e. for Ψ12 of -4.2 or Ψ50 ca.  Table S2). Overall, these results indicate that stomatal closure always occurs 123 at much higher water potential value than the one triggering embolism and that the difference between 124 Ψclose and Ψ50 increases continuously with increasing embolism resistance. Contrasting the 125 hypothesized coordination between stomatal closure and embolism resistance, our finding supports a 126 similarity in stomatal behavior particularly among embolism resistant species, and indicates that most 127 species are highly conservative in terms of their water use during drought. 128 The similarity of stomatal behavior between species suggests that keeping stomata open is not 129 beneficial in terms of fitness, particularly for survival under drought conditions. To get more insights into 130 the relationship between stomatal function and plant resilience to drought, we developped a soil plant-131 hydraulic model (see Methods) computing the survival times under drought conditions (i.e. the time to 132 hydraulic failure) for the range of hypothetical species covering the full spectrum of embolism resistance. 133 We used three different postulates to assess stomatal behavior (Figure 2 A, see Methods and Appendix 134 2). Firstly (hypothesis 1), we assumed that stomata do not close to regulate transpiration (E) during 135 drought (i.e. stomata are maintained open whatever the soil and plant water potential). Secondly, we 136 assumed (hypothesis 2) that stomata regulate water losses to maintain plant water potential (Ψplant) 137 above the water potential resulting in the onset of embolism (i.e. Ψ12), according to the premise that the 138 integrity of the hydraulic system is functionally linked to stomatal closure in response to drought. show that high levels of embolism resistance, and thus stomatal closure at lower water potential 203 accelerated death, because of faster water potential drops. More detailed simulation results are given 204 Appendix 2. 205 206 The vascular system of terrestrial plants has evolved toward very high levels of embolism 207 resistance (Ψ50 values down to -19 MPa) allowing the colonization of dry environments 21 . Stomatal 208 closure was thought to have evolved along similar lines, to keep carbon assimilation levels for longer 209 periods, even at low xylem water potential. Different recent studies have moreover reported tight co-210 variations between stomatal closure to drought and embolism resistance, but for relatively low drought 211 resistant species [8,12,20]. Our results highlight that the range of variation of Ψclose appears much 212 reduced when seen in the light of the full range of embolism resistance. Such uncoupling between 213 stomata closure and the failure of the vascular system may be the result of selection pressures that have 214 favored survival under extreme water scarcity over growth under mild drought. 215 These findings provide a complementary view to the widely accepted framework for drought 216 response strategies based on the water to carbon trade-off that plants have to face(e.g [35] , [36]). 217 According to this framework, plant drought response strategies fall in between two extreme categories 218 called isohydric and anisohydric [8,35,37,38]. Isohydric plants close their stomata rapidly in response to 219 drought, thereby maintaining a high water potential to limit embolism, but at the risk of death due to 220 carbon starvation. Conversely, anisohydric plants keep their stomata open at low water potential, 221 maintaining carbon assimilation levels, but at the cost of damage to the water transport system through 222 embolism. This framework has been the focus of many scientific studies on drought-induced mortality in 223 recent decades and has underpinned our understanding and modeling of drought induced plant mortality 224 [35,36,39]. The fact that plants among the most drought resistant close their stomata at much higher 225 potential than embolism can occur, indicates that resisting drought may not involve achieving further gas 226 exchanges during drought conditions, but demonstrates on the contrary, that plants have to limit water 227 potential drops as confirmed by the modelling analysis (Figure 2b). 228 The relative consistency of Ψclose among plants may appear contradictory with the large 229 variations in minimum water potential reported by different studies [6,7,38]. However, this may highlight 230 the importance of accounting for the multiple traits driving the demand for water when stomata are close, 231 if we want to represent water potential decline and thus, plant dehydration. For instance, the minimum 232 conductance (i.e. when stomata are closed) or the leaf area must be important traits driving plant water 233 potential decline. The hydraulic model presented is consistent with this view. Accordingly, model 234 simulations indicated that there are two main stages defining the temporal sequence leading to plant 235 dehydration in situations of water scarcity (Figure 3). The first step is defined by the time between the 236 start of water shortage and stomatal closure. Its duration depends principally on the rate of water uptake, 237 given the relative constancy of Ψclose in plants and the competition between plants for water in community 238 ecosystems. The second stage is defined by the time between stomatal closure and plant death (100% 239 embolism). The duration of this stage depends on a set of drought resistance traits allowing plant tissues 240 to retain water under very high tension, to decrease water loss when the stomata are closed and to limit 241 the decrease in water potential during embolism through deeper rooting or the release of water for 242 internal stores. It remains to be seen how these other different traits covary with embolism resistance, 243 are coordinated and have coevolved in plants to shape the spectrum of drought adaptation strategies. 244 Overall, the model analysis presented in this study, demonstrates that multiple measurable 245 drought resistance traits can be integrated into a consistent and thermodynamically reliable formal 246 framework to define hydraulic failure. This modelling approach must be validated carefully against 247 temporal dynamics of water potential, hydraulic conductance, proper embolism data and experimental 248 and field mortality for different species. However, it constitutes an important step towards assessing the 249 consequences of drought in land plants and the effects of climate change on terrestrial ecosystem 250 functions. It could also be a powerful tool for taking multiple traits into account in breeding strategies. 251

1-Data Meta-analysis 254
In this study we compiled embolism resistance traits and stomatal regulation traits from different 255 species. We first compiled ψ12, ψ50 and slope values derived from stem vulnerability curves (i.e. the 256 curve that relates the percent loss of conductivity to the xylem water potential) published by us during 257 the past 20 years, representing 150 species from different biomes. Recent direct observations of 258 embolism formation by mean of X-ray tomography confirmed the reliability of these values [21][22][23]40]. 259 All stem vulnerability curves were fitted with a sigmoidal function [41]: 260 Where is the percent loss of embolism ψ is the xylem water potential, ψ50 is the water 261 potential causing 50% loss of plant hydraulic conductivity and a is a shape parameter related to the rate 262 of embolism spread per water potential drop. Equation 1 allows computing the 2 other parameters used 263 in this study (ψ12, Slope). A more intuitive way to represent a is to relate it to the derivative of the function 264 at the inflexion point of the VC, in other word, the slope of the linear portion of the VC: 265 = × 100 4 (1) Where slope is expressed in %.MPa -1 . From the slope and the ψ50, the water potential at the 266 onset of embolism, as defined by the xylem water pressure causing 12% loss of embolism, can also be 267 computed: 268 12 = 50 +

50
(2) We managed to reassemble ψ50, ψ12 and slope parameters for 150 species (see Table S1). We 269 did not considered root and leaf vulnerability curves, because it is still unclear what mechanism is 270 responsible for the decline in hydraulic conductance measured with classical methods on these organs 271 [42][43][44][45]. Therefore we focused our study on stem embolism that we consider being the main 272 mechanisms responsible for extreme-drought induced mortality. We therefore neglected possible 273 variations in embolism resistance among plant organs as we still don't know how general is this 274 mechanism [20,44,46]. 275 For all the species with available stem embolism resistance traits, we collected different traits 276 indicating the level of plant water deficit (Ψ) causing most stomatal closure (called Ψclose in the main 277 paper and hereafter). A first group of indicators was derived directly from gas exchange or transpiration 278 measurements along with water potential data. A second indicator of Ψclose was the bulk leaf water 279 potential causing turgor loss [27,47,48] that was derived from pressure volume curves or from osmotic 280 pressure at full turgor. 281 We searched the litterature for concurrent measurements of stomata conductance (gs) and leaf 282 (or xylem) water potential to build gs(Ψ) curves, following the approach of [8]. We then computed the 283 water potential corresponding to 90% of stomatal closure (Ψ90gs). Most of our gs(Ψ) were based on 284 diurnal dynamic of leaf water potential of gs and Ψleaf or Ψxylem measurements over a drought period. In 285 a few cases, however, we used data of concurrent measurements of water potential (leaf or xylem) and 286 transpiration assessed through gravimetric methods (i.e. mass loss on detached leaves [49] or on potted 287 plants [50] over a dehydration period obtained under constant relative humidity. To select the relevent 288 literature, we primarily used the references provided in [8]. For each species individually, we also perform 289 a google scholar search by using the key words "stomatal conductance" or "transpiration" AND "water 290 potential" or "drought" or "water deficit" and the Latin name of the given species. We obtained in total, 291 66 species for this trait. We then searched the litterature for ψ tlp values. Most of them were derived from 292 pressure volume curves [51,52], but for 10 species for which we no pressure volume curves available 293 were found, we computed ψtlp from the osmotic potential at full turgor (π0) using a linear relationship 294 between π0 and ψtlp following [53,54]. Overall, 40% of ψtlp data (48 over 101 species) came from a 295 previously published database [52], and the rest was collected from different published literature. We 296 searched these data in google scholar, for each species for which no data were available from[52], we 297 used the key words "osmotic potential" or "pressure volume curves" or "turgor loss point" AND the Latin 298 name of the species. When only π0 was measured at different time of the season only the driest date 299 (i.e the lowest value of π0) was retained. 300 We studied the statistical associations between the different traits by using R (version 3.3.1), 301 following a two-step procedure. First, we fitted a segmented regression to the scatter plot of Ψclose (or its 302 component Ψtlp or Ψgs90) versus embolism resistance (Ψ50 or Ψ12) by using the package segmented. 303 Then we identified (i) the break points in the x axis (i.e. the embolism resistance value at which there is 304 a change in the co-variation between Ψclose and embolism resistance) and the y axis intercept for this 305 break point (i.e. the global limit for Ψclose). Second, we computed the correlation value as well as the 306 linear regression between Ψclose and embolism resistance for the data on either side of the break point. 307 In addition to the results developed in the main manuscript, we provided a separate analysis per group 308 (gymnosperm and angiosperm) and per trait (ψ50, ψ12, ψgs90, ψtlp) in Table S2. All parameters used in 309 this study are given in a supplementary Excel file. 310

Model: description, simulation and validation 311
We used a simplify discrete-time-hydraulic-model (called Sur_Eau) to simulate the time until 312 hydraulic failure for the spectrum of embolism resistance reported in our database, and according to 313 different hypothesis regarding the stomatal regulation of transpiration. Sur_Eau relies on the principle of 314 the original Sperry's model [19] but has been simplified to consider only two resistances (rhizosphere 315 and plant). This simplicity makes easier its applicability with only one stem VC and avoids making 316 assumptions on hydraulic segmentation, a phenomenon that depends on mecha nisms that are still 317 controversial. 318

Description of the Sur_Eau model 319
Sur_Eau assumes that liquid water flow through the soil-plant system is exactly compensated 320 by gaseous water losses at the plant foliage surface (i.e. steady state condition) which is true at large 321 time steps (>1day) or for small plants. The assumption is also made that leaf and air temperatures are 322 roughly equal, which is reasonable in well coupled canopies. We can then write : 323 where gl is the leaf conductance to vapor water, VPD is the vapor pressure deficit at the leaf 324 surface, ψsoil is the soil water potential, ψplant is plant water potential, and is the plant leaf area specific 325 hydraulic conductance over the soil to leaf pathway. gl includes both the stomatal, cuticular and boundary 326 layer conductances of the leaf. The control of E through stomata has been treated through different 327 assumptions that are described below (Appendix 3). ksl was computed as the result of two conductances 328 in series: 329 where is the hydraulic conductance of the soil to root surface pathway and the 330 hydraulic conductance of the whole plant (i.e. from the roots to the leaves). 331 was allowed to vary only to account for loss of hydraulic conductivity caused by xylem 332 embolism [55] : 333 where is the initial (i.e pre-drought) plant hydraulic conductance, and PLC is the percent 334 loss of plant hydraulic conductance due to xylem embolism. PLC is computed at each time step by using 335 the sigmoidal function for the vulnerability curve (VC) to embolism (see equations 1 to 3). 336 The model considers the capacitive effect of xylem embolism and symplasm dehydration. The 337 water freed by air filling of the apoplastic reservoir feeds the water stream of the system and thus dampen 338 the water potential decrease [56]. Following 24 , we considered that any change in PLC is followed by a 339 proportional change of the water volume that is freed back to the system: 340 Where Wxv is the amount of water freed to the system and is the total water filled xylem 341 volume of the plant (m 3 ) and PLC is defined in Equation 1.
was computed as: 342 Where Emax is the maximum diurnal transpiration, α is the apoplasmic fraction of the plant and 343 G is the ratio of the total amount of water in the xylem conduits of a tree (i.e. apoplasmic volume) to the 344 maximum diurnal transpiration. The capacitive effect of cavitation on plant survival is mainly sensitive to 345 G factor and its range of variation. The effect of G on cavitation dynamics is discussed in 24 . We also 346 accounted for the capacitive effect of symplasm dehydration (i.e. the water released by the symplasmic 347 tissue ) by using the same formulation as for cavitation (Equation 6). Computations were based on 348 the symplasmic fraction (1-α ) of the plant and PLC was replaced by the relative water content of the 349 symplasm (Rs). Rs was computed from ψleaf by using the pressure volume curve equations (Appendix 350 3). 351 The variations of the soil and the rhizophere conductance ( soil ), as well as the mean soil water 352 potential in the root zone are computed with 57] from the 353 unsaturated hydraulic conductivity of the soil, scaled to the rhizophere according to the  formulation [58,59]. The rhizophere conductance ( soil ) can be expressed as: 355 where Ksoil is the unsaturated hydraulic conductivity of the soil at given water content ( ) or water 356 potential (see below), and B is the root density conductance factor that accounts for the length and 357 geometry of the root system. B is based on the implicit assumption of a uniform roots distribution in a 358 soil layer following the Gardner-Cowan formulation 28 . B is also called the "single root" approach [60] as 359 it is equivalent to assuming that plant water uptake occurs from a unique cylindrical root that has access 360 to a surrounding cylinder of soil : 361 where is the root length per unit area, is the mean root radius, and is the half of mean 362 distance between neighboring roots. can be evaluated from , the root length per unit soil volume. 363 ksoil decreases with decreasing soil because of the displacement of water filled pores by air, as capillary 364 forces linking water to soils particles fail with increasing tension, thus creating dry non-conductive zones 365 in the rhizosphere. The parametric formulation of 25 for the water retention curve was used in combination 366 with the equation of Mualem (1976) 26 where m, n and are empirical parameters describing the typical sigmoidal shape of the 370 function. Mualem (1976) provides the formulation for the evolution if hydraulic conductivity with soil water 371 content (Θ): 372 where Ks is the saturated hydraulic conductivity, is a parameter describing the pore structure 373 of the material (usually set to 0.5), and m is again fixed as in Equation 11. The relative extractable water 374 content (Θ) is expressed as a function of volumetric soil water content as follows: 375 where θ, θs and θr are the actual relative soil water content and the relative soil water content at 376 saturation and at wilting point respectively. θs and θr are parameters measured in laboratory or derived 377 from soil surveys using pedotransfert functions. By contrast, θ is a variable, dynamically changing with 378 changes of the soil water reserve (WR). The water reserve depends on the soil volume θs and θr. The 379 calibration and sensitivity analysis for all parameters is provided below (Appendix4). 380

Dynamic simulations 381
Dynamic simulations of the different variables were performed by the mean of the discrete time 382 model described above. Under well-watered conditions, transpiration (E) is forced at a constant value, 383 assuming a constant high vapor pressure deficit. Then, E is regulated with decreasing water potential 384 as leaves loose turgor inducing stomata closure, with a submodule based on the pressure volume curve 385 (following [32]; see Appendix 3). At each time step the soil water reserve (WR) is first computed and 386 then used to compute all other variables. WR is then computed as a result of a water balance: 387 Where E is the cumulated transpiration over the time step, is water release due to cavitation 388 and is the water release due to symplasm dehydration (Equation 7 & 8 and the following text). The 389 time step was set to 0.1 day, but increasing this value up to 0.5 or down to lower values had little 390 influence on the general pattern of our results. E was computed as follows: 391 where Emax is the maximal rate of transpiration, LA is the plant leaf area and f(ψplant) represents 392 the stomatal regulation which was set according to different hypothesis as described in the next section 393 (Appendix 2). The calibration for Emax and LA as well as all other parameters are provided in Appendix 394 4. 395

Hypothesis testing on the interrelation between ψclose and ψ50 396
We used the above described model to evaluate the role of ψclose and ψ50 in determining the 397 survival time until hydraulic failure under drought for the full range of embolism resistance reported in 398 the database. Three different hypotheses regarding how stomata regulate transpiration were tested 399 ( Figure 2A): (1) No stomatal regulation of E (ψclose=-∞) whatever the embolism resistance, (2) E 400 regulation to maintain ψplant above the water potential causing the onset of embolism (ψplant >ψ12) and 401 (3) Early stomatal regulation of E during drought, so that ψclose varied with Ψ12 until ψclose=-3 MPa, in 402 accordance with the global limit observed in empirical data (Figure 1c, main manuscript). For each of 403 these hypothesis, survival time was computed for the full range of embolism resistance encountered in 404 our database, from ψ50= -1.5 to ψ50=-15 MPa, every 1MPa step. A more detailed analysis of these 405 simulations is given in the Appendix 2. 406

Model validation: survival time during drought from drought mortality experiments 407
To validate the relationship between embolism resistance traits (ψ 50) and survival time predicted 408 by the model under different hypothesis, with used data from drought mortality experiments. We found 409 survival time data for 15 species covering a wide range of embolism resistance (ψ50 from -1.5 to -11), 410 from four different drought mortality experiments published recently [29][30][31]50]. One study was 411 performed on gymnosperm species only 17 and three other studies were performed on angiosperm 412 species [29][30][31]50]. All these experiments were made under semi-controlled conditions on potted 413 seedlings or saplings and shoot death was recorded at different time since the beginning of an 414 experimentally imposed drought. Here we used the average time needed to reach 50% death (T50) after 415 the beginning of the drought treatment as an indicator of survival time during drought. In each 416 experiment, identical soil volume and climate were used across species. However, there were 417 differences in air relative humidity and soil volume across experiments (both of them can strongly affect 418 the survival time during a water deficit episode), particularly between the gymnosperm and all 419 angiosperm experiments. This precluded the direct comparison of the survival time across the four 420 different studies. To overcome this problem, we standardized the T50 of each experiment by differences 421 in soil volume or air relative humidity (Appendix1). 422 423