Leaf turgor loss point shapes local and regional distributions of evergreen but not deciduous tropical trees

Summary

• The effects of climate change on tropical forests will depend on how diverse tropical tree species respond to drought. Current distributions of evergreen and deciduous tree species across local and regional moisture gradients reflect their ability to tolerate drought stress, and might be explained by functional traits.

• We measured leaf water potential at turgor loss (i.e. ‘wilting point’; π_tlp), wood density (WD) and leaf mass per area (LMA) on 50 of the most abundant tree species in central Panama. We then tested their ability to explain distributions of evergreen and deciduous species within a 50 ha plot on Barro Colorado Island and across a 70 km rainfall gradient spanning the Isthmus of Panama.

• Among evergreen trees, species with lower π_tlp were associated with drier habitats, with π_tlp explaining 28% and 32% of habitat association on local and regional scales, respectively, greatly exceeding the predictive power of WD and LMA. In contrast, π_tlp did not predict habitat associations among deciduous species.

• Across spatial scales, π_tlp is a useful indicator of habitat preference for tropical tree species that retain their leaves during periods of water stress, and holds the potential to predict vegetation responses to climate change.

Introduction

Global climate change is altering precipitation patterns, resulting in drying trends and increasing frequency and severity of drought in many parts of the world (IPCC, 2015; Trenberth et al., 2011; Dai et al., 2018). Forest productivity is reduced under drier climates (e.g. Poorter et al., 2017; Taylor et al., 2017; Meakem et al., 2018) and in drought years (Hofhansl et al., 2015; Meakem et al., 2018). Drought can also cause elevated tree mortality (e.g. Zuleta et al., 2017; Klein & Hartmann, 2018), particularly among the large trees that contribute substantive fractions of total ecosystem biomass and productivity (Phillips et al., 2010; Bennet et al., 2015; Lutz et al., 2018; Meakem et al., 2018). These effects are likely to result in positive feedbacks to climate change through reduced ecosystem carbon sequestration and a shift towards lower biomass forests (Weltzin et al., 2003; McDowell et al., 2020). Modeling these dynamics is critical to reducing the large uncertainty surrounding the role of forests – particularly tropical forests – in shaping Earth’s future climate (Friedlingstein et al., 2006; Cavaleri et al., 2015). Within this broader context, an important challenge is characterizing how the distribution of phylogenetically and functionally diverse tropical tree species depends on moisture availability, against the background of the copious interactions of biotic and abiotic factors integrated over the deep history of the species and ecosystems (Condit et al., 1995; Wright, 2002; Engelbrecht et al., 2007; Brodribb et al., 2020).

To effectively predict distributions of traits or species across tropical moisture gradients, it is necessary to identify plant traits that are both strongly associated with moisture availability and can be readily measured across numerous species. Hydraulic traits – for example, hydraulic conductivity of sapwood and leaves and xylem tension at 50% or 88% loss of maximum hydraulic conductivity (P50, P88) – are mechanistic traits strongly linked to a physiological function, which have been found to be correlated with drought tolerance and species distributions across moisture gradients (e.g. Nardini & Luglio, 2014; Liu et al., 2019; Laughlin et al., 2020). However, measuring these traits is very time‐consuming, presenting a substantial barrier to their use in species‐rich tropical forests. More rapidly measurable traits, including wood density (WD) and leaf mass per area (LMA), have been collected for a broader array of species and shown to predict growth, survival and light requirements of species in diverse communities (e.g. Reich, 1995; Poorter & Bongers, 2006; Shipley et al., 2006). However, WD and LMA have limited power to predict species drought tolerance (McGregor et al., 2020) or distributions across moderate moisture gradients (Umaña et al., 2020). WD is often taken as a proxy for vulnerability to xylem embolism because it correlates with P50 within and among temperate species (Rosner, 2017; Savi et al., 2019). Studies from the tropics show that WD does not predict P50 in tropical species (Trueba et al., 2017; de Guzman et al., 2020), but relates to other hydraulic properties such as sapwood water release curves (de Guzman et al., 2020).

One trait that can be measured efficiently and is strongly correlated with drought tolerance and spatial distributions with respect to water availability is turgor loss point (π_tlp) (Tyree et al., 1999; Bartlett et al., 2012a; Farrel et al., 2017). Species whose leaves lose turgor (i.e. wilt) at more negative water potentials are generally more drought‐tolerant (e.g. Baltzer et al., 2008; Bartlett et al., 2016a; Kunert, 2020; McGregor et al., 2020). The link between π_tlp and drought tolerance has been observed across species differing in local distributions with respect to topographic habitats of varying water availability (Bartlett et al., 2016b; McFadden et al., 2019), across wet‐vs‐dry climate communities (Baltzer et al., 2008; Blackman et al., 2019; Medeiros et al., 2019), and across biomes (Bartlett et al., 2012b, 2016b; Zhu et al., 2018). A tendency for higher π_tlp in wetter habitats has also been observed across a climatic gradient in northeast Spain (Rosas et al., 2019). Within the tropics, π_tlp varies with respect to moisture on local scales (Bartlett et al., 2016b; McFadden et al., 2019) and is lower in species tolerant of seasonally or perennial dry habitats (Baltzer et al., 2008; Medeiros et al., 2019).

The association of traits with climatic distribution is complicated by the fact that trees can follow a wide spectrum of strategies for coping with seasonal water shortages or drought. The endpoints of this spectrum are adaptations to either avoid or tolerate dehydration under water stress (Grubb, 1998; Tyree et al., 2003). To avoid dehydration, trees need to establish access to water, reduce water loss or concentrate their growth in the period in which water is most available. Typical adaptations are deep rooting, early stomatal closure, low cuticular conductance, water storage in plant organs, osmotic adjustments and/or leaf shedding (Westoby et al., 2002). Such adaptations occur with increasing frequency as moisture availability decreases; for instance, the fraction of deciduous species and individuals increases with dry season intensity, particularly in the larger size classes (Condit et al., 2000a; Meakem et al., 2018; Alberton et al., 2019). In contrast, to tolerate dehydration, trees need physiological adaptations allowing them to maintain water transport, gas exchange and cell survival under water stress (i.e. at a low water potential). This can be achieved by increasing the ability of cells to remain functional at low water availability and water potential. The sensitivity to embolism represents an important constraint on tree functioning and survival during drought. Because the leaf is a bottleneck in the water transport pathway within the soil–plant–atmosphere continuum (Sack et al., 2005), leaf hydraulic and drought tolerance traits are expected to be important indicators of species’ ability to tolerate desiccation. In general, however, it remains poorly understood how drought strategy (i.e. avoidance vs tolerance) mediates the distribution of species traits across moisture gradients.

Leaf phenology, which is linked to the drought strategies described above, would logically interact with traits including π_tlp, WD and LMA to shape species’ abilities to cope with drought. Tropical tree species exhibit adiversity of leaf phenological strategies ranging from drought‐deciduous (complete or partial deciduousness during the driest time of the year) to brevi‐deciduous and evergreen strategies (Reich, 1995; Powers & Tiffin, 2010; Park et al., 2019). Drought‐deciduous tree species lose their leaves at the driest time of the year as a dehydration avoidance strategy, and therefore there is little reason to expect that variation in π_tlp, WD and LMA will be related to species distribution along a regional moisture gradient. Brevi‐deciduous species are completely deciduous for brief periods of c. 1‐3 wk, but this deciduous period is not aligned with the driest time of the year. Thus, for brevi‐deciduous and evergreen species, which both maintain leaves at the driest time of the year, it is logical to expect that these species would require greater desiccation tolerance adaptations in drier environments, whereas drought‐deciduous species would not. This remains to be tested.

Here, we test the abilities of a mechanistic leaf drought tolerance trait, π_tlp, and the more widely measured plant functional traits of LMA and WD, to predict species distribution with respect to seasonal moisture variability in natural forest communities spanning local and regional moisture gradients in Panama. We hypothesized that across both local (within a 50 ha plot) and regional scales, evergreen and brevi‐deciduous species with more negative π_tlp, higher WD and higher LMA would be associated with drier environments, but that there would be at most a weak relationship for deciduous species; and π_tlp would be a much better predictor of species distributions across moisture gradients than LMA and WD.

Materials and Methods

Study sites and tree censuses

The study region features a rainfall gradient across the Isthmus of Panama from the wetter Caribbean side (3100–3800 mm annual precipitation) to the drier Pacific side (1800 mm; Engelbrecht et al., 2007). Barro Colorado Island (BCI), our focal study site, is located in the center of the gradient, and receives on average 2630 mm of rainfall per year. The proportion of deciduous species among canopy tree species increases from 14% on the wetter end to 41% on the drier end, with BCI intermediate at 28% (Condit et al., 2000a). The region has a tropical moist climate with a distinct dry season, averaging 131 d in length on BCI, between January and April (Paton, 2013). The vegetation is species‐rich tropical moist lowland forest, with 312 tree and shrub species in the 50 ha plot on BCI (Katabuchi et al., 2017), and 867 species in total in 72 plots in the region (including the BCI 50 ha plot; Condit et al., 2013).

A 50 ha long‐term monitoring plot was established in 1980 on the BCI plateau (Hubbell & Foster, 1983) and is now part of the Forest Global Earth Observatory network of large‐scale forest dynamics plots (ForestGEO; Anderson‐Teixeira et al., 2015). Regular tree censuses have recorded the diameter and location of all trees larger than 1 cm in diameter at 5 yr intervals, following the ForestGEO census protocol (Condit, 1998). This analysis uses data from the 2015 census, which were retrieved from the ForestGEO data portal (http://ctfs.si.edu/datarequest/), where it is available upon request.

Elevation varies by 38 m within the plot, leading to distinct small‐scale habitat heterogeneity and associated variation in water availability (Fig. 1a; Condit et al., 2000b; Harms et al., 2001; Kupers et al., 2019). Habitat types within the plot may be categorized into old secondary forest (c. 2 ha; Condit, 1998), high plateau, low plateau, slope, streamside and swamp areas (see Fig. 3a). For the purposes of analyses, we divided the plot into 1250 20×20 m quadrats, most of which could be assigned to one of those habitat categories (except 66 quadrates that were classified as mixed habitat). Quadrats with a slope less than 7° are designated plateau, and are further distinguished as high or low plateau based on elevation above or below 152 m (Harms et al., 2001). The habitat categories high and low plateau are well‐drained upland soils (Dietrich et al., 1996; Condit, 1998). Quadrats with a slope of more than 7° are classified as slope habitat; they are characterized by higher dry season soil moisture than the plateau areas (Kupers et al., 2019). Highest water availability is found along the stream, which contains water well into the dry season. The c. 1.5 ha swamp area is waterlogged during much of the wet season.

Fig. 1.

Map of the 50 ha ForestGEO plot on Barro Colorado Island showing the distribution of (a) dry season soil water potential of a regular dry year (from Kupers et al., 2019); (b) evergreen trees of varying π_tlp – point size is scaled to tree diameter and color indicates π_tlp; and (c) deciduous trees of varying π_tlp (note: Dalbergia retusa does not occur in the plot). Trees for which π_tlp data are not available are not plotted.

Fig. 3.

Habitat associations of evergreen and deciduous tree species to π_tlp classes within the Barro Colorado Island 50 ha ForestGEO plot. (a) Map of the plot showing the habitat classifications of Condit (1998). (b,c) Association plots representing the residuals of a contingency table assuming independence of π_tlp class and habitat for (b) evergreen and (c) deciduous species. The height of each bar is proportional to its signed contribution to the Pearson’s χ² and bar width is proportional to the square root of the expected counts corresponding to the cell, so the area of each box is proportional to the difference between observed and expected frequencies. Horizontal dotted lines indicate no difference between observed and expected frequencies. The absolute size of the residuals is indicated by the coloration of the boxes: blue for positive, red for negative, very colorful for large residuals (> 4), less colorful for medium residuals (< 4 and > 2), light gray for small residuals (< 2). Under the assumption of independence and normality of residuals, 2 and 4 roughly correspond to a significant test with P < 0.05 and P < 0.0001 respectively. The P‐value given at the bottom of the legend corresponds to the overall chi‐square test’s P‐value.

Regional analyses were based on the BCI 50 ha plot in combination with tree inventories of 71 smaller plots (0.5–6 ha) in Central Panama (Condit et al., 2013). These plots span the rainfall gradient from an estimated 1756 mm of rainfall per year at Cerro La Torre (8.91182N, −79.5868W) to an estimated 4148 mm of rainfall per year at Esperanza (9.37642N, −79.3598W). There is also considerable variation in dry season moisture availability along this gradient, with a dry season water deficit of −579 mm in the southern region and only −237 mm in the northern region (Condit et al., 2013). Small‐plot inventories followed the same protocol as at the 50 ha long‐term tree census at BCI. For all plots, full census information (diameter and identity of every individual in multiple censuses) and climate data are available through the Smithsonian ForestGEO data portal (http://ctfs.si.edu/Public/Datasets/PanamaTreePlots/PanamaPlotInfo.php).

Species were assigned to deciduousness classes based on a combination of systematic surveys (Condit et al., 2000b) and expert knowledge, as detailed in Meakem et al. (2018). For analysis, we separated species into two groups, which we refer to as ‘evergreen’ and ‘deciduous’, while recognizing that each encompasses substantial variation in leaf phenology. We grouped brevi‐deciduous species with evergreen species (see Supporting Information Table S1) because they are foliated and transpiring during the peak of the dry season (Kunert et al., 2010). Our deciduous class thus is specifically a drought‐deciduous class, including semi‐deciduous and facultatively deciduous species, as well as considerable variation in the duration of deciduousness. In total, we had 34 evergreen species (of which seven were brevi‐deciduous), and 16 deciduous species.

Leaf sampling

We sampled leaves from 50 tree species, targeting dominant canopy species (Table S1), between April and July 2018. Specifically, we sampled 40 species represented by at least 10 individuals in the BCI 50 ha plot, prioritizing based on their biomass within the plot (estimated using allometries of Chave et al., 2014). Together, these comprised 60% of the total plot biomass. We also sampled several additional species that were very abundant or had drastic changes in their abundance since the first inventory in 1982 in the BCI 50 ha plot (e.g. Hybanthus prunifolius, Piper cordulatum; see Katabuchi et al., 2017), or that were ecologically important in the larger region (e.g. Dalbergia retusa). We sampled three adult individuals per species on BCI (in the 25 ha plot, which is located 100 m from the 50 ha plot, as no destructive sampling is allowed in the 50 ha plot) or in Gamboa in the case of one species that is rare on BCI (Dalbergia retusa). We collected one fully sun‐exposed branch per tree. For canopy species we chose mature trees reaching the upper canopy; for understory species we chose sun‐exposed individuals in canopy gaps. Canopy branches were sampled with a crossbow by shooting a bolt with an attached fishing line over a branch, and then using a rope chainsaw. Directly after cutting the branches from the trees, we placed the branches in humid and opaque plastic bags. The samples were transported to the laboratory as quickly as possible (<2 h between sampling and laboratory processing) and recut underwater at least two nodes distal to the original cut. The recut samples were placed in buckets with water, covered with plastic bags again and rehydrated overnight (c. 12 h or longer). We tested that rehydration time was sufficient by measuring the leaf water potential of rehydrated branches with a Scholander pressure chamber (for a subset of 25 species).

We measured WD (g cm⁻³) of a 2 to 4 cm long segment of each cut branch. We specifically sampled the lowest (most proximal) part of each branch, to ensure that all samples were fully lignified. The segment was placed in water overnight to allow full hydration. The next day, the bark was removed and the xylem was split in the center to remove the pith. Both parts of the xylem were submerged in a container with water put on a scale to measure the displaced volume of the water. The xylem was then dried in an oven at 60°C for at least 48 h and weighed. We used three leaves of different sizes (large, medium and small) from each collected branch to estimate LMA (g m⁻²). All leaves were scanned and then dried in the oven at 60°C for at least 48 h and weighed. Leaf area was derived from scanned images using an R script by G. Arellano.

Osmometric determination of the turgor loss point

We determined π_tlp based on measurements of the sap osmotic potential of leaves (Bartlett et al., 2012a). The osmotic potential was measured with a vapor pressure osmometer (VAPRO 5520; Wescor, Logan, UT, USA). On the day following sample collection, we removed two mature, fully expanded leaves per branch from the rehydrated branches. Two sample disks were cut from the leaf centrally between the midrib and margin with a 4 mm diameter cork borer. Disks were tightly wrapped in aluminum foil and submerged in liquid nitrogen for at least 2 min (Bartlett et al., 2012a). We used the standard 10 μl chamber well of the osmometer. Immediately before putting the disks into the chamber of the osmometer, the disks were punctured with a dissection needle about 10–15 times to improve evaporation through the cuticle and to reduce equilibration time (Kikuta & Richter, 1992). The osmometer was set in the autorepeat mode and measurements were recorded until the equilibrium was indicated by an increase between measurements of < 0.01 MPa (about five osmometer readings). π_tlp was calculated from the osmotic potential given by the osmometer (π_osm) using the biophysically based calibration equation established by Bartlett et al., (2012a):

$${\pi }_{\mathit{tlp}}=0.832{\pi }_{\mathit{osm}}^{‐0.631}$$ (Eqn 1)

This biophysically based calibration equation has been verified for a variety of tropical tree species by Maréchaux et al. (2016).

Statistical analysis

We calculated the mean and standard deviation of each trait for each species, and used the mean traits in interspecific analyses. To evaluate how traits were related across species, we calculated Pearson correlations for each pair of traits. For evergreen and deciduous species groups separately, we analyzed how species mean WD, LMA and π_tlp related to both dry season soil water potential (Fig. 1a; Kuper et al., 2019) and to habitat types (see Fig. 3a; Harms et al., 2001) on the BCI 50 ha plot. For the former analysis, we used a map of mean dry season soil water potential in a regular dry year (Kupers et al., 2019; Fig. 1a), assigning each quadrat in the plot into categories of low, intermediate and high soil water potential, each spanning exactly one‐third of the total observed range. For the analysis of habitat categories, we adopted the habitat map of Harms et al. (2001), excluding the young forest (c. 2 ha of the entire plot; Condit, 1998) and mixed habitat (i.e. transition between any two other categories). We calculated the proportion of trees in a given soil water potential grouping or habitat as

$$proportioninhabitat=\frac{N_{\text{habitat}}}{N_{50ha}}\times 100$$ (Eqn 2)

where N _habitat is the number of trees in a given soil water potential grouping or habitat and N _50ha is the total number of trees across all habitats analyzed. We performed a chi‐square test to evaluate whether the observed proportion in habitat differed significantly from a null expectation representing independence of the trait of interest from habitat type, using the ‘assoc’ function from the R package vcd (Meyer et al., 2006). In this analysis, we grouped species based on their trait values into three classes of equal breadth. To test for sensitivity to the groupings, we repeated the analysis with two and five classes. To account for spatial nonindependence of different trees, we also applied the more conservative torus translation test to the habitat type analysis (Harms et al., 2001). Furthermore, we used a linear regression (function ‘lm’ from the R package stats) to test for correlations between traits (WD, LMA and π_tlp) and species‐specific proportion in habitat.

To test for moisture association on the regional scale, we adopted a moisture association index (MAI) from Condit et al. (2013), which uses a hierarchical Gaussian logistic regression to model tree distributions across the 72 sample sites included in our regional analysis. The model includes estimated dry season moisture availability and seven soil factors. We define MAI as the species‐specific first‐order response parameter to standardized dry season moisture in this model, where a negative value means a negative species response to more moisture availability and a positive value a positive response to higher moisture availability. We used linear regression (function ‘lm’ from the R package stats) to quantify the association between the different traits (WD, LMA and π_tlp) and MAI, analyzing evergreen and deciduous species separately. Our use of traits measured at one site for the regional analysis implicitly assumes that these trait values are indicative of species strategies more generally, notwithstanding probable regional intraspecific trait variation (Osazuwa‐Peters et al., 2017).

All statistical analyses were conducted using the R software (R Core Team 2019). Data are archived in Zenodo (https://doi.org/10.5281/zenodo.4431676).

Results

Inter‐ and intraspecific variation in traits

Mean species turgor loss point (π_tlp) ranged from −1.34 to −2.05 MPa in evergreen species and from −1.25 to −2.30 MPa in deciduous species (Table S1). Intraspecific variation of π_tlp was low, with the coefficient of variation (CV) below 5% in 33 of 50 species, and CV greater than 10% in only three species. Mean species LMA ranged from 39.0 to 153.4 g m⁻² in evergreen species and from 31.7 to 110.6 g m⁻² in deciduous species. Intraspecific variation of LMA was high, with CVs > 10% for 25 species. Mean species WD ranged from 0.228 to 0.894 g cm⁻³ in evergreen species and from 0.224 to 0.622 g cm⁻³ in deciduous species. Twelve species had < 5% intraspecific variation in WD, while 22 had variation > 10%. Relationships among these traits were weak, with Pearson correlations of 0.23 (P = 0.1088) for π_tlp and LMA, –0.037 (P = 0.7998) for π_tlp and WD, and 0.14 (P = 0.3432) for LMA and WD (Fig. S1).

Local scale

Species mean π_tlp was associated with soil moisture characteristics and habitat distributions within the BCI 50 ha plot in evergreen species but not in deciduous species (Figs 1, 2, 3, 4). These patterns, which are visibly apparent in our maps of trees by π_tlp (Fig. 1), were statistically confirmed by associations of evergreen species with soil water potential in a regular dry season year and habitat types (Figs 2, 3). Evergreen species with lower π_tlp (−2.28 to −1.84 MPa) were positively associated with lower mean dry season soil water potential and negatively associated with higher soil moisture potential (Fig. 2a). Evergreen species with higher π_tlp (−1.59 to −1.33 MPa) showed the opposite pattern. Deciduous species with lower π_tlp (−2.28 to −1.84 MPa) were overrepresented in areas with lowest dry season soil water potential and underrepresented in habitats with intermediate soil water potential, but there were otherwise no significant associations of π_tlp‐based species groupings to soil moisture potential classes (Fig. 2b). These trends were consistent when species were divided into different numbers of quartiles (division in median shown in Fig. S2).

Fig. 2.

Soil moisture associations of evergreen and deciduous tree species to π_tlp classes within the Barro Colorado Island 50 ha ForestGEO plot. Association plots representing the residuals of a contingency table assuming independence of π_tlp class and mean dry season soil water potential (SWP) for (a) evergreen and (b) deciduous species. The height of each bar height is proportional to its signed contribution to the Pearson’s χ² and width is proportional to the square root of the expected counts corresponding to the cell, so the area of each box is proportional to the difference between observed and expected frequencies. Horizontal dotted lines indicate no difference between observed and expected frequencies. The absolute size of the residuals is indicated by the coloration of the boxes: blue for positive, red for negative, very colorful for large residuals (> 4), less colorful for medium residuals (< 4 and > 2), light gray for small residuals (< 2). Under the assumption of independence and normality of residuals, 2 and 4 roughly correspond to a significant test with P < 0.05 and P < 0.0001 respectively. The P‐value given at the bottom of the legend corresponds to the overall chi‐square test’s P‐value.

Fig. 4.

Relationships between the proportion of trees of each species growing in high plateau and π_tlp. The linear regression line is shown when the relationship is significant (P < 0.05).

Analyzed in relation to habitat types, evergreen species with lower π_tlp were overrepresented in the drier high plateau and underrepresented in the wetter slope, whereas those with higher π_tlp showed the opposite pattern (Figs 1b, 3b). In contrast, deciduous species showed no clear relationship between π_tlp and affinity with these habitats (Figs 1c, 3c). There were a number of significant associations based on π_tlp in relation to the swamp habitat, which covers a very small area. Although this area is waterlogged during much of the wet season, it has the most negative surface soil water potential of any habitat in the dry season (Fig. 1a). Evergreen species with more negative π_tlp (≤ −1.82 MPa) were overrepresented in the swamp, consistent with their overrepresentation in other dry habitats (Fig. 3b). However, evergreen species with less negative π_tlp (≥ −1.59 MPa) were also overrepresented in the swamp, as were deciduous species with more negative π_tlp.

On a species level, there was a significant negative relationship between the proportion of individuals in high plateau and species mean π_tlp for evergreen species (Fig. 4 left panel; R ² = 0.28, P = 0.004), contrasting with no relationship for deciduous species (Fig. 4 right panel; R ² < 0.001, P = 0.914). The torus translation test, which is conservative in identification of species–habitat associations, identified significant associations to the high plateau or slope for only nine species (Fig. S3), yielding insufficient statistical power for a formal test. However, species with significant positive associations to the high plateau or negative associations to the slope (n = 2) had more negative π_tlp (≤ −1.82 MPa) than those with negative associations to the high plateau or positive associations to the slope (n = 7; π_tlp ≥ −1.74 MPa).

WD was not a significant predictor of the proportion in high plateau for either evergreen (R ² = 0.008, P = 0.647) or deciduous species (R ² = 0.001, P = 0.897). However, there was a significant positive association of evergreen species with low WD (< 0.483 g cm⁻³) to the slope, swamp and stream areas and a negative association of species with intermediate WD (= 0.483 g cm⁻³ WD ≤ 0.621 g cm⁻³) to the high plateau and the low plateau areas (Figs S4–S6). Deciduous species showed no relationship of WD to habitat association. LMA of evergreen species was negatively correlated with the species‐specific proportion in high plateau (R ² = 0.20, P = 0.0181) – the opposite pattern of what would be explained based on physiology – whereas there was no significant relationship for deciduous species (R ² < 0.001, P = 0.993; Figs S7–S9). Evergreen species with high LMA were negatively associated with high and low plateau areas, but low LMA was significantly associated with the plateau areas. A positive association was present for species with high LMA and slope, stream and swamp areas; however, low LMA species were negatively associated with those areas.

Regional scale

On the regional scale, π_tlp was positively correlated with MAI (R ² = 0.32, P = 0.003) among evergreen species (Fig. 5), indicating that low‐π_tlp species are associated with drier habitats. In contrast, π_tlp and MAI were not significantly correlated for deciduous species (Fig. 5; R ² = 0.009, P = 0.715). There was no significant correlation of MAI with WD or LMA (see Figs S10, S11) for either evergreen (WD: R ² = 0.005, P = 0.746; LMA: R ² = 0.003, P = 0.779) or deciduous tree species (WD: R ² = 0.065, P = 0.324; LMA: R ² = 0.039, P = 0.447).

Fig. 5.

Regional‐scale correlation between species’ moisture association index (MAI) and π_tlp for evergreen and deciduous species. MAI describes a species’ distribution across the Isthmus of Panama with respect to dry season moisture (positive values indicate a positive association with moisture), as modeled by Condit et al. (2013).

Discussion

Our results demonstrate that on both local and regional scales, low turgor loss point (π_tlp) is associated with the ability of evergreen species to survive and persist in habitats with greater seasonal moisture stress. Evergreen (or brevi‐deciduous) species showed clear local‐scale habitat preference and regional distribution across a moisture gradient depending on their π_tlp, indicating that leaf desiccation tolerance significantly affects a species’ ability to establish and survive in periodically water‐limited environments. In contrast, π_tlp does not explain the soil moisture and habitat association of deciduous species, suggesting that a species’ ability to shed leaves under drought conditions neutralizes the benefits of low π_tlp. The success of π_tlp in predicting local‐ and regional‐scale species distributions contrasts with the failure of WD and LMA, which showed no consistent, physiologically congruent relationships with species distributions on local or regional scales (Figs S4–S11). We therefore conclude from this extensive data collection that π_tlp is a useful trait for predicting species’ drought tolerance across diverse tropical species.

Turgor loss point drives habitat association and species distribution

We show that π_tlp is a physiological parameter influencing ecological drought tolerance of diverse tropical tree species, including on local (Figs 1, 2, 3, 4) and regional (Fig. 5) scales. π_tlp can be understood as a leaf parameter that scales with whole plant–water relationships, defining a species’ requirements of certain minimum soil water availability (Bartlett et al., 2012b). In other words, π_tlp is describing the permanent wilting point (Bartlett et al., 2012b), which refers to the soil water potential or plant‐available soil water content when trees start wilting irreversibly (Veihmeyer & Hendrickson, 1928). The fact that π_tlp is directly related to wilting explains the association of later wilting species to drier habitats, where they can survive in niches with naturally occurring water limitations and outcompete less drought‐adapted species. Therefore, species adapted to drier forest habitats can be expected to either have the ability to shed their leaves or maintain turgor under more negative water potentials than tree species adapted to moist habitats (Poorter & Markesteijn, 2008). Accordingly, tree species in seasonally dry forests are generally able to maintain leaf turgor at lower water potential (more negative πt_lp) than those in less seasonal moist forests. This indicates that dry‐season intensity is one of the main drivers of tree distributions across environmental gradients. Consequently, dry‐season performance, especially hydraulic resistance against dry season water stress among evergreen tree species, is a crucial factor for population dynamics and species distribution patterns in a community context (Engelbrecht et al., 2000, 2007). On a regional scale, π_tlp explained 35% of the variation in evergreen species association to moisture across a gradient of dry season moisture (Condit et al., 2013). This is consistent with the finding that more negative π_tlp was positively associated with more seasonal climates across the Thai‐Malay peninsula (Baltzer et al., 2008), but in this case expands to a larger number of species.

Avoidance or persistence

Unlike evergreen tree species, deciduous tree species exhibited no clear relationship between their habitat association and π _tlp. π _tlp of deciduous species explained neither habitat association and soil moisture requirements in the 50 ha plot on BCI nor species distributions across the regional moisture gradient. We explain the lack of correlation between π _tlp and species moisture associations by the simple fact that deciduous trees are avoiding the water‐limited time of the year by shedding leaves and being dormant. Deciduousness is a well‐known strategy of canopy trees to avoid drought stress in tropical forests (Poorter & Markesteijn, 2008). Thus, deciduous trees are relatively insensitive to water limitation during the dry season, as their water demands are minimal while deciduous (Kunert et al., 2010). This means that deciduous species characterized with a less negative π_tlp can survive potentially in very dry habitats like the high plateau areas. However, those species were more abundant in the slope areas (Condit et al., 2013), probably due to higher soil fertility in slopes compared to the high plateau. In general, seasonal drought favors the development of deciduous trees when sufficient soil nutrients are available, whereas infertile soils favor the long‐lived evergreen leaves to preserve limited nutrients (Givnish, 2002). In a meta‐analysis, Bartlett et al. (2012b) showed that π_tlp is a good indicator of a species’ drought tolerance if other factors do not become equally or more important than soil moisture availability. Nutrient limitation would be such a case, where nutrient use efficiency is more important for the trees than water availability. Tree species distributions in Panama depend not only on dry‐season moisture but also on phosphorus availability (Condit et al., 2013). For example, the deciduous species Cavanillesia platanifolia (π_tlp = −1.71 MPa) is restricted to slope areas with sufficient phosphorus availability. Thus, deciduousness and the dry season avoidance strategy comes at the cost that the seasonal production of new leaves requires a minimum of soil fertility.

Diverging π _tlp on temporary waterlogged swamps

A very interesting aspect of our results was that species in the most and least negative π_tlp classes were positively associated with the swamp area, whereas those with intermediate π_tlp were negatively associated (Fig. 3). Two adaptations might explain this phenomenon. First, the physiological response of trees to waterlogging is similar to drought response in that waterlogging also causes stomatal closure (Kreuzwieser & Rennenberg, 2014). Anaerobic root conditions during waterlogging cause a reduction in root conductance, especially under conditions with a high evaporative demand, and wilting can occur (Bradford & Hsiao, 1982). Physiological responses to waterlogging also include a drastic increase in the concentration of soluble carbohydrates in leaves (Ferner et al., 2012), which would directly affect their osmotic potential and thus turgor loss point. Indeed, comparing two congeneric Mora species in Guyana, one occupying a well‐drained habitat and the other a seasonally waterlogged habitat, the species associated with seasonally waterlogged soils had a more negative π_tlp than the species growing in a well‐drained habitat and regularly drought‐stressed habitat (ter Steege, 1994). The second explanation might be that trees growing on seasonally waterlogged sites have shallow rooting systems (Fan et al., 2017). In seasonal climates, this can potentially cause a shift in seasonal stress situations from seasonally waterlogged to seasonal soil water limitation. Whereas the trees try to avoid anoxic conditions during the wet season by developing shallow rooting systems, their shallow roots cannot reach the moisture as the swamp dries out during the dry seasons. Indeed, the topsoil layer in the swamp area on BCI dries out relatively fast and reaches even more negative soil water potentials than the plateau areas during the dry season (Fig. 1a; Kupers et al., 2019).

Conclusions

We deliver evidence that in tropical forests a tree species’ osmotic adaptation to a certain niche with a given moisture availability is expressed in the species‐specific leaf water potential at turgor loss point (π_tlp). This applies for nutrient‐preserving evergreen tree species but not deciduous species, which avoid dehydration via leaf loss but face nutrient limitations in constructing new leaves. Further research will be required to understand how seasonal moisture limitation and nutrient availability interactively constrain π_tlp and leaf phenology and shape species distributions. In the meantime, we suggest that π_tlp can be parameterized for the application in forest ecosystem models to predict future changes of species composition and distribution patterns under climate change.

Author contributions

NK, KAT, LS and SJD designed the research. NK and JZ collected the trait data in Panama. RP, RCC and SPH led collection of tree census data. LS, HCML, SJW and SMM contributed methods and expertise. Data analyses were performed by VH and NK. NK, KAT and LS interpreted the results. NK and KAT wrote the first draft of the manuscript, and all authors contributed to revisions.

Supporting information

Fig. S1 Correlation plot indicating the relationships between π_tlp, LMA and WD.

Fig. S2 Habitat associations of evergreen and deciduous tree species by π_tlp within the Barro Colorado Island 50 ha ForestGEO plot.

Fig. S3 Linear relationship between the proportion of trees of each species growing in high plateau and π_tlp, highlighting significant Torus associations (Harms et al., 2004).

Fig. S4 Map of the 50 ha ForestGEO plot on Barro Colorado Island showing the distribution of WD within the evergreen trees, and WD within the deciduous trees.

Fig. S5 Habitat associations of evergreen and deciduous tree species by WD within the Barro Colorado Island 50 ha ForestGEO plot.

Fig. S6 Linear relationship between the proportion of trees growing in high plateau and WD, highlighting significant Torus associations (Harms et al., 2004).

Fig. S7 Map of the 50 ha ForestGEO plot on Barro Colorado Island showing the distribution of LMA within the evergreen trees, and within the deciduous trees.

Fig. S8 Habitat associations of evergreen and deciduous tree species by LMA within the Barro Colorado Island 50 ha ForestGEO plot.

Fig. S9 Linear relationship between the proportion of trees growing in high plateau and LMA, highlighting significant Torus associations (Harms et al., 2004).

Fig. S10 Moisture association index (MAI of Condit et al., 2013) as a function of wood density (WD).

Fig. S11 Moisture association index (MAI of Condit et al., 2013) as a function of leaf mass per area (LMA).

Table S1 Summary of observed mean turgor loss point (π_tlp), leaf mass area (LMA), wood density (WD) and leaf phenology. Standard deviation is given for all mean values.

Please note: Wiley Blackwell are not responsible for the content or functionality of any Supporting Information supplied by the authors. Any queries (other than missing material) should be directed to the New Phytologist Central Office.