Leaf trait variations associated with habitat affinity of tropical karst tree species

Abstract Karst hills, that is, jagged topography created by dissolution of limestone and other soluble rocks, are distributed extensively in tropical forest regions, including southern parts of China. They are characterized by a sharp mosaic of water and nutrient availability, from exposed hilltops with poor soil development to valleys with occasional flooding, to which trees show species‐specific distributions. Here we report the relationship of leaf functional traits to habitat preference of tropical karst trees. We described leaf traits of 19 tropical tree species in a seasonal karst rainforest in Guangxi Province, China, 12 species in situ and 13 ex situ in a non‐karst arboretum, which served as a common garden, with six species sampled in both. We examined how the measured leaf traits differed in relation to species’ habitat affinity and evaluated trait consistency between natural habitats vs. the arboretum. Leaf mass per area (LMA) and optical traits (light absorption and reflectance characteristics between 400 and 1,050 nm) showed significant associations with each other and habitats, with hilltop species showing high values of LMA and low values of photochemical reflectance index (PRI). For the six species sampled in both the karst forest and the arboretum, LMA, leaf dry matter content, stomatal density, and vein length per area showed inconsistent within‐species variations, whereas some traits (stomatal pore index and lamina thickness) were similar between the two sites. In conclusion, trees specialized in exposed karst hilltops with little soils are characterized by thick leaves with high tissue density indicative of conservative resources use, and this trait syndrome could potentially be sensed remotely with PRI.

with geomorphic factors and climate. Karst landscapes create particularly complex habitat mosaics full of fissures, conduits, sinkholes, and steep terrains, because limestone is prone to dissolution.
Identification of traits associated with edaphic habitat specialization, such as seen in tropical karst forests, may promote a mechanistic understanding of the community assembly process and ecosystem functioning. Kraft and Ackerly (2010) demonstrated how leaf mass per area (LMA) and leaf nitrogen concentrations are associated with topography in a lowland Amazonian forest in an evolutionarily convergent manner. Tree species adapted to different habitat types within a karst landscape may exhibit trait variations reflecting adaptation to differences in supply regimes of water, nutrients, and light among these habitats. As a key axis in the leaf economic spectrum (Wright et al., 2004), LMA values exhibit negative associations with inherent growth rates and soil resource availability, of which environmental heterogeneity explains around 36% of the variation (Poorter, Niinemets, Poorter, & Wright, 2009). Long leaf lifespan, which is positively associated with LMA, is adaptive in infertile soils as it reduces the nutrient turnover rates (conservative ecological strategy: Westoby, Falster, & Moles, 2002).
Optically detectable traits (hereafter optical traits) and anatomical traits may exhibit functional adaptation specific to individual environmental stress factors. Recent advances in visible to near-infrared range (400-1,050 nm) spectroscopy allow estimation of LMA, concentrations of nitrogen and photosynthetic pigments, and water content from spectral reflectance at both leaf and canopy levels (Asner et al., 2011;Doughty, Asner, & Martin, 2010).
Photochemical Reflectance Index (PRI: Gamon, Peñuelas, & Field, 1992) in particular is significant as an indicator of the degree of protection from excess radiation with xanthophyll cycle pigments.
Plant adaptations to contrasting water regimes may be reflected in anatomical traits of vein and stomata, which may not be correlated with LMA or optical traits (Westoby et al., 2002). Vein length per area (VLA) of minor veins and stomatal traits are closely associated with transpiration and photosynthetic gas exchange capacity in angiosperms (Brodribb, McAdam, & Carins Murphy, 2017;Sack & Scoffoni, 2013). Vein and stomatal arrangement would indicate the supply and exchange capacity of water through leaf lamina.
Functional understanding of species distributions across edaphic habitats may be advanced by assessments of these multiple leaf traits (Baltzer & Thomas, 2010).
In such trait-based analyses, it is also important to recognize that many plant traits exhibit phenotypic plasticity (Bradshaw, 1965;Nicotra et al., 2010). Leaf traits germane to productivity and growth, such as LMA, photosynthetic capacity, and leaf lifespan, exhibit substantial plasticity in relation to heterogeneity of light and soil resources (e.g., Russo & Kitajima, 2016). It is extremely difficult, if not impossible, to separate such effects of phenotypic plasticity from inherent trait differences among species by sampling plants only in their natural habitats. Common garden experiments are useful in teasing apart whether an apparent association of certain species traits with their natural habitats is due to habitat filtering of species or plastic responses (Cordell, Goldstein, Mueller-Dombois, Webb, & Vitousek, 1998). For tree species, an arboretum where trees were grown on a common soil and receive adequate supply of water and nutrients may serve as a common garden in which leaf traits may be evaluated under a standardized favorable condition.
Karst forests occupy 7%-15% of the global terrestrial landmass (Fu et al., 2012;Hartmann et al. 2014). In an extensive karst forest zone in southwestern China, ranging from Yunnan to Guangxi, recent studies report affinities of tree species to various topographic positions (Guo et al., 2017;Zhang, Hu, Zhu, Luo, & Ni, 2010). Steep topography and percolation of water through the bedrock accompanied by nutrient leaching and soil erosion creates a gradient of water and nutrient availability from mountain hilltops to foothills, leading to drier and nutrient-poor conditions in hilltops and mid-slopes compared to foothills. The karst valleys are not necessarily resource-rich habitats due to prolonged inundation and soil anoxia in the rainy season (Bonacci et al., 2008;Guo et al., 2017;Schindler, 1982). Among the four topographical habitats we discussed above, foothills perhaps represent the most favorable in terms of water and nutrient availabilities for plant growth (Guo et al., 2015;Huang et al., 2014). However, potential associations of leaf traits in karst tree species to their habitat affinities are yet to be characterized.
Here, we report a comparative study of leaf traits among karst tree species growing in their natural habitats and in an arboretum in Guangxi Province, China. We selected 19 karst tree species known to show habitat affinity to four topographic positions (i.e., hilltop, midslope, foothill, and valley). Twelve were common species in Nonggang National Nature Reserve (hereafter the karst forest), and 13 were growing in an arboretum in Nanning, Guangxi. Of these, only six species affiliated to valley and foothill habitats were sampled at both sites.
We measured 13 leaf traits to test the following hypotheses: (1) species affiliated to resource-poor hilltop and mid-slope habitats express a suite of traits indicative of a more conservative leaf strategy such as high LMA, compared to those in more favorable foothill habitats, (2) values of a given trait of a species may differ between its natural habitat and the arboretum, but species means should show positive interspecific correlations between the two study sites. Inclusion of optically detectable traits in multivariate trait analysis in this study allowed us to explore a potential for spectral detection of suites of leaf functional traits indicative of ecological strategies and habitat affiliation.

| Study sites and species
We conducted this study in two sites in Guangxi Province in south China. The first was a seasonal karst rainforest (Figure 1 Table S1).
We selected 19 species mainly based on the information on tree dominance and habitat association available from the 15-ha plot, which belonged to the forest dynamics network coordinated by the Center for Tropical Forest Science (Guo et al., 2015;Huang et al., 2014). Of these 19 species, we could sample 13 in the arboretum, 12 in the karst forest, and six in both sites (Table 1). These six species were affiliated to valley and foothill habitats. Trees in the arboretum had been planted approximately 30 years earlier (except for Saurauia tristyla, see Table 1) and watered regularly during the dry season, whereas those sampled from the karst forest had been growing in their typical habitats. We sampled three to six species in each habitat affinity class (valley, foothill, mid-slope, and hilltop: see Table 1 for species codes, their habitat affinity, and taxonomy).
Species classified to the affinity class hilltop had not been planted in the arboretum and could be sampled only in the karst forest. The set of mid-slope species sampled in the karst forest was different from the mid-slope species sampled in the arboretum (Table 1).

| Sample collection
From November 2015 to January 2016, we harvested three small branches containing 5-25 sun leaves from a minimum of three mature individuals per species using a seven-meter-long telescopic pruner.
S. tristyla, sampled only in the arboretum, was an exception, as it was still at the sapling stage (<5 cm dbh). All measurements were taken from mature, fully expanded, and healthy leaves following the trait measurement protocol of Pérez-Harguindeguy et al. (2013). To maintain the water status of leaves, cut ends of branches were immediately wrapped in water-soaked tissues and sealed in airtight polythene bags.
Samples were transported from the arboretum and the karst forest in

| Leaf trait measurements
For six leaves per species, we measured lamina thickness (LT) with a digital micrometer (Mitutoyo Corporation, Kanagawa, Japan), fresh mass with a digital balance and leaf area (LA) with a LI 3050C leaf area meter (Li-COR Inc., Lincoln, Nebraska, USA). After drying to a constant mass (65°C for 72 hr), we calculated leaf dry matter content (LDMC) as dry mass divided by fresh mass and LMA as leaf dry mass divided by leaf area.
Using another set of leaves, we applied clear nail polish at six to eight locations on two leaves per tree to imprint abaxial leaf surfaces for stomatal trait measurements (Lawson, James, & Weyers, 1998).
None of our study species had adaxial stomata. Micro-photographs of these imprints were taken with a digital camera mounted to a light microscope (Leica Microsystems, Wetzler, Germany), and stomatal density (SD), guard cell length (GCL), and stomatal pore index (SPI = SD × GCL 2 ; stomatal pore area per lamina area) were measured.
We were unable to measure stomatal traits of three species (Ficus simplicissima, Ficus hispida and S. tristyla) as clear imprints could not be obtained due to trichomes.
Additional leaf samples were preserved in formalin, acetic acid, and alcohol fixative to measure VLA (Sack & Scoffoni, 2013). Three leaf pieces per tree were chemically cleared and stained before they were micro-photographed at high resolution (2,048 × 1,536 pixels).
Micro-photographs were analyzed with ImageJ v.

| Statistical analysis
All analyses were performed using R version 3. values (n = 6) for site comparisons and tree-mean trait values for examining bivariate trait correlations in the karst forest.

| Syndromes of leaf traits across karst hill habitats
The eight leaf traits of 12 tree species measured in their natural habitats exhibited multivariate trait differentiation, which was greater among habitats than within-habitats ( Figure 2 showing the result of PCA). The first principal component (PC1) explained 42.0% of the variance, which was largely contributed by LMA and leaf optical traits, separating hilltop species with high LMA, LT, and LDMC from the rest (Table 3). Foothill species were clustered at the negative end of PC1, but they were not clearly segregated from mid-slope and valley  (Table S2). But habitat affinity did not explain these among-species variations.
Overall, optical traits that contributed to PC2 were less differentiated among habitats than LMA, LDMC, and LT that contribute to PC1.
Of these Chl SPAD (Figure 3g) and PRI (χ 2 = 18.00, p < .001, Table S2) showed significant differences in relation to habitat affiliation. The Chl SPAD , which is an estimate of chlorophyll per leaf area, generally tracked the pattern of leaf thickness, except that mid-slope species growing on shaded slopes, showed high values for their relatively thin leaves. Mid-slope species also showed higher PRI values than those from exposed hilltops (Table S2), indicating that they exhibited greater light use efficiency. PRI showed significant negative correlations with LMA, LT, and LDMC (Figure 4a-c). Water band index (WBI) was positively correlated with LT and LMA (Table S3), indicating higher water contents per unit leaf area in thick leaves of hilltop species, despite their high LDMC.

| Traits syndromes in the arboretum
In the arboretum, LMA tended to be the lowest for foothill species (Figure 3b), similarly to the in situ pattern in the karst forest (Figure 3a).
There were no clear differences in LDMC among habitat classes of mid-slope, foothill, and valley in the arboretum, but LT followed the pattern similar to LMA (r = .65, p < .001, Figure 3d). In the arboretum, PRI did not correlate with LMA (r = .05, p = .67), LT (r = −.21, p = .08) or LDMC (r = .02, p = .87). Among-species variation in VLA (threefold, 3.99-12.89 mm/mm 2 ), SD (sixfold, 97-585 mm −2 ), and SPI (sixfold, 0.10-0.36) did not differ significantly among habitat affinity classes (Table S2).  Tables 2 and 3). Percentage of total variance explained by each PC axis is shown along the axis label T A B L E 3 Loadings of first two principal components and eight leaf traits in 12 karst tree species representing four habitat affinity classes in the karst forest. The PCA was conducted with tree level means. Trait abbreviations are given in Table 2 Trait the sampling sites were not significant in many traits (Table 2).

| Leaf trait consistency between the sites
Exceptions were lamina thickness, SPI, and GCL, which strongly correlated between the two sampling sites (Table 2, Fig. S2), even though stomata density was higher in the arboretum in four of five species (Figure 5c).

| DISCUSSION
Our comparative study of tropical karst tree species found that lamina thickness and LDMC differed significantly among habitat affinities, resulting in the highest LMA values for the hilltop species (Figure 3a).
At the hilltop, trees were growing on largely soil-less substrates and

| LMA as an indicator of a conservative ecological strategy
For a given leaf, LMA is a product of leaf density (dry mass per unit volume, which tightly correlates with LDMC) and lamina thickness (e.g., Kitajima & Poorter, 2010). Hence, LDMC and lamina thickness are equally important in explaining variations of LMA across a broad range of tropical tree species (e.g., Westbrook et al., 2011) as found in our study. Values of LMA generally increase with elevation (Poorter et al., 2009;Read, Moorhead, Swenson, Bailey, & Sanders, 2014). In our dataset, LMA was the lowest, not among species from the lowest elevation (the valley), but among species affiliated with the foothill, where soil volumetric water content was the highest (Fig.   S1). We could not measure the nutrient availability in the rooting zone of our study species, but it was likely that soil and water availability covaried, as in sand-or karst-dominated habitats in general (Cavender-Bares, Kitajima, & Bazzaz, 2004;Mi, Li, Chen, Xie, & Bai, 2015;Zhang et al., 2007). Collectively, high LMA among hilltop species reflected overall resource-poor conditions in hilltop habitats, suggesting LMA as an indicator of a conservative ecological strategy in karst habitats.

| Leaf optical traits in multi-trait syndromes
In the multivariate leaf trait correlations, optical traits measured in our study, including estimates of chlorophyll contents and xanthophyll cycle pigments, exhibited partially independent variation from LMA.
Our purpose was not to measure dynamic in situ response of these traits, but to compare species under a standardized protocol. Lower PRI values indicate greater quantities of xanthophyll cycle pigments which are involved in photoprotection from excess radiation (Filella et al., 1996;Gamon et al., 1997). Low PRI values among hilltop species ( Figure 4) where species were grown under fully exposed sunlight  (Asner et al., 2011). Significance of negative correlations of PRI with LMA, LDMC, and LT (Figure 4a-c, r 2 ranging from .25 to .49 Table S3) suggests a possible use of PRI as a predictor of LMA. This significance owes largely to hilltop species with high LMA, LT, and low PRI (Figure 4a-c). Spectral signatures from a wider range of 400 to 2,500 nm predict specific leaf area, the inverse of LMA, with r 2 = .79 for 42 rainforest species (Asner and Martin, 2009). Prediction of LMA with optical measurements with a relatively inexpensive spectrometer (400-1,050 nm), as used in our study, may also be useful in ecological studies.

| Leaf trait plasticity
Some of the leaf traits in our study showed substantial withinspecies differences between the two study sites, which is most likely due to plasticity, but direction and magnitude of plastic responses were inconsistent across species ( Figure 5). Among the leaf traits included in our study, stomatal pore index and lamina thickness exhibited low plasticity (Fig. S2), suggesting that these traits can be sampled as inherent traits, although Cordell et al. (1998) reported a wide degree of plasticity of lamina thickness in Metrosideros polymorpha in Hawaii. In our study, we sampled mature and fully expanded leaves from well-exposed branches of mature trees. The main difference between the karst forest and the arboretum was that the plants in the latter experienced less degrees of drought and nutrient stress due to deeper soil and frequent watering. Possibly in response to this more favorable environment in the arboretum, four of the five species showed higher SD, and three of the four showed higher VLA than at the karst forest. Limited sample size (three species each from foothill and valley), the lack of measurements of microenvironments, leaf age, ontogeny, and population sources of plants growing in the arboretum, limit the functional implications of these results.
In conclusion, leaf traits of karst tree species exhibit syndromes that apparently evolved in relation to their specialization to edaphic habitats. In particular, specialization to hilltops, where water and nutrients may be in limited supply, is associated with a conservative ecological strategy represented by high values of LMA, lamina thickness, LDMC, and low values of chlorophyll, photochemical reflectance index. Although we also expected, vein and stomatal traits associated with leaf hydraulic properties had no significant differences among contrasting types of habitat specialization. Substantial differences in values of some leaf traits between the natural habitat and the arboretum suggest that it is not often possible to infer trait values from ex situ samples. Hence, further exploration of the relationships of optical traits with other leaf functional traits will contribute to better understanding of adaptive trait syndromes in relation to habitat specialization, especially of species that specialize in difficult to access sites, such as the karst hilltops in our study.