Intraspecific trait variation and covariation in a widespread tree species (Nothofagus pumilio) in southern Chile

Study species and locations

Nothofagus pumilio (Poepp. et Endl.) Krasser (Nothofagaceae) is a deciduous broadleaf light-demanding tree species that extends through a wide latitudinal and altitudinal range in the southern Andes of Chile and Argentina. Its ample distribution provides a unique opportunity to study probable mechanisms responsible for trait variation under contrasting climates while controlling for phylogenetic effects. This feature of N. pumilio has already been used in some studies that have found genetic and morphological variation (Barrera et al., 2000; Premoli, 2004; Premoli et al., 2007) across abiotic gradients, although these previous studies have not dealt with LMA and WD (but see Barrera et al., 2000), reported magnitude of variation or been framed into the functional trait paradigm. Here, we selected two environmental drivers for capturing trait variation of N. pumilio: a large-scale climatic (latitudinal) moisture difference that covers three locations, from Mediterranean to cold temperate climatic conditions; and a short-scale local influenced climatic (altitudinal) gradient that covers, at each location, a decreasing temperature trend with altitude.

The first location is in Termas de Chillán (36°54′S, 71°24′W, Fig. 1), hereafter called Termas, where treeline occurs at an altitude of 2080 m. This area belongs to the meso-Mediterranean belt with humid climatic conditions and with a predominant drought period during summer (Amigo & Ramírez, 1998). The annual precipitation is 1950 mm most of which falls as snow from May to September. The soils are derived from andesitic rocks of volcanic origin. The second location is in the Antillanca area within the Puyehue National Park (40°47′S, 72°11′W), hereafter called Antillanca, where treeline occurs at an altitude of 1350 m. This area belongs to the meso-Temperate belt with supra-Humid climatic conditions (Amigo & Ramírez, 1998). The annual precipitation is c. 4000 mm (Aguas Calientes weather station 1980–2000), most of which falls as snow from May to September. The soils are derived from andesitic rocks of volcanic origin. The third location is in Cerro Castillo National Reserve (46°04′S, 72°03′W), hereafter called Cerro Castillo, with the treeline at an altitude of 1310 m. This area belongs to the supra-Temperate belt with humid climatic conditions (Amigo & Ramírez, 1998). Mean precipitation for 1992 to 2007 was 416 mm for the five warmest months (November to March). The soil is derived from aeolian volcanic ash deposits. Soil moisture measurements (mean ± SE of volumetric water content of soil) recorded during two growing seasons were 0.0855 ± 0.0029 m³ m⁻³ for Termas, 0.193 ± 0.0034 m³ m⁻³ for Antillanca, and of 0.1659 ± 0.0030  m³ m⁻³ for Cerro Castillo (HOBO Micro station, Echo probes, F. I. Piper, unpublished).

Field sampling and traits determination

Two traits were examined: LMA (g m⁻²), which measures the leaf dry-mass investment per unit of leaf area deployed (Lambers et al., 1998; Wright et al., 2004), and WD (g cm⁻³), also referred to as wood specific gravity (Swenson & Enquist, 2007), defined here as the oven-dry mass divided by green volume (Chave et al., 2009). At each location, N. pumilio trees were studied at four altitudes, covering the last c. 300 m of altitude of the treeline ecotone (Table 1). For clarity and brevity, we hereafter refer to altitudes by their short names: CF, closed and mature lowest altitude forest; IN, intermediate stand between closed forest and timberline; TB, timberline; and TL, treeline (Table 1). Although the three treelines occur at different altitudes, we were certain that the temperature gradient with elevation was of comparable magnitude as treelines occur at analogous isotherms (treelines worldwide share a similar mean growing season temperature of 6.7°C, Körner & Paulsen, 2004), and the different elevations were similarly equidistant to each other across locations. At each altitude, we set a sampling line perpendicular to the slope and every 30 m along this line four to five trees belonging to different height classes (i.e. different ages) were selected for sampling. A total of 30–50 trees were selected at each altitude. Individuals with extensive browsing or other damage were excluded. Only unshaded individuals were sampled. Height and stem diameter at breast height (DBH, 1.35 m) were recorded on the selected trees, spanning the whole range of dimensions to enable us to estimate a wide range of potential ages. Tree ages for Termas and Cerro Castillo were estimated using exponential models fitted at the age–DBH relationship for each altitude determined from a data subset of tree cores collected at each altitude (N = 60–80) (A. Fajardo, unpublished). Tree ages for Antillanca were all directly determined (i.e. all trees were cored for age determination). Tree cores for age determination were extracted at c. 30 cm height and placed dried in labelled plastic straws. Once in the laboratory, cores were mounted in grooved wooden sticks and sanded with successively finer grades of sandpaper to reveal annual rings. Rings were read using a microscope mounted on a dendrochronometer with a Velmex sliding stage and Accurite measuring system. Cross-dating accuracy was checked using cofecha v6.06P (Holmes, 2001).

For Termas and Cerro Castillo, leaf and wood samples were collected in late February–early March 2007, once leaves had completed their development and reached their final size. For Antillanca, the collection of tissue was done in late February 2008. Sampling and measurements followed methodological protocols dictated by Cornelissen et al. (2003). In each sampled tree, we randomly collected a fully expanded, sun-exposed branch from which 15–30 leaves were selected. Stem sapwood xylem cores were extracted perpendicular to the bark at c. 50 cm height from the same tree bole using a 5.15-mm-increment bore (Haglöf, Långsele, Sweden). Water displacement measurement of WD value was not possible at the field locations, so we carefully cut the ends of each core section perpendicularly to the sides, measured its length with a calliper (10⁻⁵ m precision), avoiding pressure of the calliper blades on the wood, and calculated fresh volume by multiplying the length by the cross-sectional area. All cores were c. 10 cm long, except when saplings were considered. Tissue samples were bag-labelled and stored in a cool box for transportation. Back in the laboratory, leaves were laid flat separately and photographed with a reference square of known area using a Nikon Coolpix 5000 digital camera (Nikon Corporation, Tokyo, Japan). Total projected leaf area was calculated using the image-processing software sigmaproc (Systat Software Inc., Richmond, CA, USA). Leaves and cores were then oven-dried at 70°C to constant weight, and subsequently were weighed with a scale to a precision of 0.0001 g for determination of LMA (as oven-dry weights of the 15–30 leaves divided by their total leaf surface) and WD (as oven-dry weights of cores divided by the core fresh volume). We determined LMA and WD for a total of 470 trees. Our method to determine WD might differ from other more elaborate methods (Williamson & Wiemann, 2010); however, our procedure proved to be the best at hand to have a rapid, nondestructive sampling of trees in remote sites having few local facilities.

Statistical analysis

Wood density was normally distributed but LMA was right-skewed (Kolmogorov–Smirnov test for goodness-of-fit), and it was therefore log₁₀-transformed before analysis. First, to determine how variance of both traits was partitioned across latitude, elevation and individuals’ age, we used a nested analysis of variance (ANOVA) with random effects (Messier et al., 2010). In this analysis, each level (e.g. latitude, elevation and tree age) was considered a nested level. As tree age was determined for each individual (i.e. a continuous variable), we categorized for this sole analysis, based on tree age frequency distribution, a variable called ‘age class’ into four groups: I, trees ≤ 50 yr old; II, trees of 51–100 yr; III, trees of 101–200 yr; IV, trees > 201 yr. Thus a variable component analysis was conducted on the mixed effects nested ANOVA model using the ‘varcomp’ function, ‘ape’ and ‘nlme’ (Pinheiro et al., 2009) packages of R (R-Development, 2009). The variances computed here represent the variance around the mean for each level (e.g. sites, altitudes, age classes). We used a bootstrapping procedure to compute 95% confidence intervals (CIs) around the variance estimation.

We also computed coefficient of variation (CV) for sites and populations within sites as a complementary index to interpret intraspecific variation. Additionally, we compared the intraspecific trait variation of our study with trait variation values (SD) reported from a nonexhaustive list of other studies (Wright & Westoby, 2002; Muller-Landau, 2004; Wright et al., 2004; Chave et al., 2006; He et al., 2006; Chao et al., 2008; Ishida et al., 2008; Yang et al., 2008; Hallik et al., 2009; Souto et al., 2009) that mostly dealt with interspecific variation across broad scales.

To test whether there is any significant covariation between LMA and WD across locations, populations or individuals (within populations), we analysed our data using the Model Type II regression method (Wright et al., 2007), where the regression slope was computed as standardized major axes (SMAs). Relationship strength was quantified using the coefficient of correlation (r² and P-values) as well as CIs of both the slope and intercept. This slope-fitting technique was also considered appropriate, as there was error associated with both dependent and independent variables (i.e. the residual variance is minimized in both x and y dimensions). Additionally, as we were also interested in testing the potential differences in LMA and WD across latitude and elevation, and the relationships between both variables and tree age, we used differences in the SMA slopes (Falster & Westoby, 2005), which were tested following Warton & Weber’s (2002) method using SMA tests and routines (Falster et al., 2003). Thus, when common slopes across latitude and elevation were demonstrated (homogeneity of slopes) and calculated, differences in the y-intercept of the regression slope across latitude and elevation were tested using an ANCOVA-like procedure (ANOVA following transformation of the data to remove any correlation between, for example, LMA and tree age) (Falster et al., 2003). The statistical software (s)matr (Falster et al., 2003) was used for all the analyses. Tukey’s HSD post-hoc test was applied whenever a significant difference among locations was found.

Intraspecific trait variation

The total intraspecific variation for LMA (CV = 21.14%) was significantly higher (t = 6.233, P = 0.011) than that of WD (CV = 10.52%). The variance partitioning strongly differed depending on the scale and the trait being considered; except for the variance not explained by the factors under scrutiny (> 50%), altitude accounted for the largest total variation in LMA, whereas sites (latitude) and age classes accounted for the largest variation in WD (23 and 13%, respectively, Table 2). The ontogeny for LMA and altitude for WD accounted for the smaller total variation. When compared with other across-species studies, the intraspecific variation we found was lower, though not negligible. Intraspecific trait variation of N. pumilio was not less than 26% in LMA (except when compared with Wright et al.’s (2004) worldwide study) and 23% in WD of the interspecific trait variation across several biomes. The total variation found in the current study even reached 95% of the interspecific variation found by Hallik et al. (2009) for LMA in a set of northern hemisphere deciduous broadleaved tree species, and 42% for WD in one tropical forest location in Amazonia, Brazil (Muller-Landau, 2004). When compared with the only other study reporting trait variation at the intraspecific level, the total variation found for LMA in N. pumilio was 64% of the variation in Embothrium coccineum, another widespread tree species in the southern Andes (Souto et al., 2009) (Table 3).

The effect of soil moisture, temperature and ontogeny on LMA and WD

According to variance partitioning magnitudes and depending on the trait and the spatial scale (see the ‘Intraspecific trait variation’ section in the Results), both LMA and WD mean values varied significantly across locations (moisture); values of LMA were significantly different according to elevation (temperature), while WD differed only according to ontogeny. First, there were significant differences across locations with varying soil moisture content in the mean values of LMA (F = 4.577, P = 0.011) and WD (F = 60.903, P < 0.0001, Fig. 2), where the maximum value of each trait was found at different locations; the maximum value of LMA was in Antillanca (112.035 g m⁻², moister location) and that of WD was in Termas (0.557 g cm⁻³, drier location). Secondly, in the main, LMA increased significantly with elevation (Table 4); individuals at higher elevations had thicker and/or denser leaves than individuals at lower elevations. Finally, at each elevation, our results generally indicated that LMA remains constant across different ages of trees, while WD decreases with tree age (Table 5). WD varied negatively with tree age even at the krummholz belt (treeline in Termas) where the range of ages was notably narrower (Table 1).

Covariation between LMA and WD

Values of LMA correlated positively and significantly with WD at each location (Fig. 3, Termas, P = 0.0062; Antillanca, P = 0.0186; Cerro Castillo, P = 0.0007). The trend remained consistent across locations (Fig. 3; slopes of each location did not differ one to another, P > 0.05). When covariation patterns were scrutinized among individuals within each elevation, we did not find a significant covariation between both traits, except in Termas at the CF altitude (Table 6), where both traits covaried positively. Thus, at the within-population level (same elevation), individual trees that have similar LMA values do not tend to have similar WD values. All this suggests the existence of a similar covariation pattern between both traits relative to temperature only; that is, the environmental variation across elevation is a decreasing pattern of temperature with altitude.

Intraspecific trait variation is not negligible

Using a widely distributed tree species located in different and contrasting environmental conditions (from Mediterranean-like to humid conditions), we observed a substantial amount of intraspecific variation, particularly for LMA (CV = 21.14%). Also notable was the fact that the variation of WD was half that of LMA, suggesting the conserved character of WD, which is in agreement with previous evidence that WD is less variable within species than expected by chance (Chave et al., 2006; Swenson & Enquist, 2007). The intraspecific variation we found for N. pumilio was, for both traits, not less than 23% of the entire interspecific variation found in other cross-species studies (some of them including multiple woody species and complete biomes). Similarly, in a recent study, Messier et al. (2010), working with LMA and leaf dry matter content in tropical forests in Panama, found that the total amount of intraspecific variation was comparable to the amount of interspecific variation; thus, intraspecific variation can be important (not negligible), at least for some clades (Chave et al., 2006; Swenson & Enquist, 2007). It should be noted that in our comparative analysis with other studies reporting the magnitude of trait variation, the comparison was done in absolute terms of variation and did not consider whether the other studies were conducted under similar conditions (i.e. similar environmental conditions). As such, only one of the studies scrutinized (Souto et al., 2009) was conducted along a latitudinal gradient with one tree species (Embothrium coccineum), while the others considered biomes or precipitation gradients as the comparative pattern. With this rather simplistic comparative analysis, we have shown that, in most cases, intraspecific variation of N. pumilio represented at least a quarter of the total interspecific variation across similar spatial scales. This finding may imply two things: our tree species is unusual in that it varies in a way that other species cannot vary, since they cannot tolerate different abiotic conditions, and thus are less competitive and less persistent; and other woody species being part of community assemblies could potentially have similar and important trait variations (Hulshof & Swenson, 2010; Messier et al., 2010), which have not traditionally been accounted for. When this is the case (coexisting species having important intraspecific trait variation), it could be argued that these species may be functionally redundant; that is, their overlap may be larger than the difference between their mean trait values (Messier et al., 2010). It would be very illustrative to compute intraspecific trait variation along a species-rich gradient and to test whether this variation changes and ultimately depends on biotic interactions (Gross et al., 2009). In a similar vein, Cornwell & Ackerly (2010) found a significant relationship between abundance of species and trait values at the community scale (e.g. the higher the LMA value, the higher the abundance of those species), but not at the landscape scale. N. pumilio constitutes a forest community where the abundance may be close to 100%; does the other small fraction of woody species have radically different trait values to those exhibited by N. pumilio?

Environmental conditions and ontogeny affect LMA and WD variation differently: probable mechanisms

Although, half the variation was not accounted for by any of the drivers proposed (residuals), our results indicate that the most important environmental conditions affecting LMA and WD were elevation (temperature) and location (moisture), respectively. WD and, in particular, LMA values increased with altitude for N. pumilio (covaried similarly, see later). An increase of LMA with altitude is a common pattern found in many other species, particularly evergreens (Cordell et al., 1998; Hovenden & Vander Schoor, 2004), and it was also found for N. pumilio in an altitudinal gradient in Tierra del Fuego (Barrera et al., 2000). A higher LMA is related to thicker laminas or higher tissue density, or combinations of these (Niinemets, 1999). This may suggest the existence of more expensive-to-construct leaves at higher elevations, where to tolerate stress (e.g. low temperatures) may be more important than having high rates of photosynthesis. However, Premoli & Brewer (2007) found that under both field and common garden conditions, net photosynthesis of N. pumilio was higher in individuals of upper than of lower elevations. This finding, along with the results of the present study, suggest that LMA may be positively correlated with net photosynthesis in N. pumilio and that expected trade-offs between leaf functional traits and physiological performance (Wright et al., 2004) might not apply in this species. Even though altitude only explained 6.8% of the total variation of WD, the increasing trend with altitude may be explained in terms of a decreasing growth rate as a result of environmental stress (e.g. Termas and Cerro Castillo).

Site variation accounted for 23% of total variation in WD. The highest values were found at the location exhibiting drought (Termas), which is in accordance with other studies (e.g. Chave et al., 2006; Preston et al., 2006). Higher WD may be an adaptation to drought stress (Hacke et al., 2001) as it is associated with low stem vulnerability to cavitation (Tyree & Sperry, 1989; Preston et al., 2006), or it may be explained in terms of drought-induced growth limitations (Zhang et al., 2009). Site variation only accounted for 9% of total variation in LMA, which is why this variable was not positively related to moisture conditions. Similarly, Wright et al. (2004) emphasize that LMA and rainfall are not related in deciduous species; however, LMA does relate positively with leaf longevity which has been found to be shorter at colder locations, as a result of the shorter growing season (Wright et al., 2004). Although Wright et al.’s (2004) dealt with interspecific comparisons, the selective force of environment might act similarly within species (Cornwell & Ackerly, 2009), and thus the lower LMA in Cerro Castillo (the southernmost location) might be associated with the shorter growing season, which precludes leaves from complete development and anticipates leaf senescence.

Ontogeny affected very differently for both traits, not accounting for any variation in LMA. In general, LMA has been found to be lower in younger than in older trees (Day et al., 2001; Thomas & Winner, 2002; Zhang et al., 2009), and this is probably the result of the different light environments. Our results, however, indicate that leaves of N. pumilio do not vary their construction costs along with ontogeny. This is the opposite of what has been found for other deciduous species (Thomas & Winner, 2002). As we controlled the light environment by collecting sun-exposed leaves, our result is in line with previous studies pointing out that light exposure is the main driver of LMA variation with ontogeny (Rozendaal et al., 2006; Poorter, 2007). Regarding WD, an increase is expected when growth is restricted with ontogeny (Zhang et al., 2009). We nonetheless found the opposite: WD declined with ontogeny across two locations (for Antillanca the trend was not significant). Discarding the fact that we collected sapwood WD (see the Materials and Methods section), we have no clear explanation for this trend, but it evidently indicates that decreasing WD with ontogeny cannot be interpreted as a consequence of declining efficiency in growth, as reported, for example, by Zhang et al. (2009). One potential explanation is that denser wood in younger trees could be related to frequent water stress affecting the superficial root zone of shorter trees. In the driest site (Termas), taller trees may escape from water stress by accessing deeper sources of water. Yet another potential explanation for this pattern is that increasing vessel area with tree height may act as a compensation mechanism for increased hydraulic resistance (Becker et al., 2000). This may account for the negative relationship between WD and tree height, even when it might be somewhat counterbalanced by a reduction in vessel density.

Relationships between plant trait variation

At the interspecific level, Cornwell & Ackerly (2009) found that WD and LMA responded similarly to environmental gradients, suggesting covariation of the two traits and common tradeoffs across locations. A similar positive relationship between WD and LMA was found by Bucci et al. (2004) and Ishida et al. (2008) working in multi-species tropical forests. On the other hand, Wright et al. (2007) found that these two traits were not related to each other among seven tropical species. In particular, the highest consensus so far on the relationship between the environment and LMA and WD is that their values increase with dry conditions (Wright et al., 2004; Preston et al., 2006; Cornwell & Ackerly, 2009). We found a stronger control of temperature, rather than moisture and ontogeny, on the covariation of these traits (i.e. they covary across a temperature but not across a moisture gradient (or dry conditions)), suggesting that the effect of environmental drivers on the covariation between LMA and WD differ at the intraspecific level. Basically, we did not find any relationship between traits and ontogeny because LMA did not vary with tree age while WD did; nor was there a relationship with moisture, as WD mostly varied with it, whereas LMA’s variation with moisture was small. The fact that both traits covaried predictably with temperature represents crucial information to be considered for future incorporation into vegetation–climatic models (Wright et al., 2005).

Conclusions

Leaf mass per area and WD are part of a whole suite of interconnected traits that together shape the performance of plants (Westoby & Wright, 2006). So far, much importance has been placed on their interspecific variation. Here we quantified trait variation of both LMA and WD attributed to moisture, temperature and tree ontogeny while controlling for phylogeny. We conclude, first, that trait variation at the intraspecific level is relevant, particularly when compared with interspecific variation values reported by other studies; second, that LMA and WD varied differently depending on the environmental driver and ontogeny, for which we offered explanations on the most probable mechanisms driving these variations; and third, that LMA and WD covaried positively with temperature only. Additionally, in spite of the existence of an increasing understanding of local adaptation and acclimation to environmental conditions at the within-species level, our study, as far as we know, is the first one attempting to place this intraspecific variation into the context of community ecology and assembly processes at large scale. Intraspecific variation should be recognized and considered when interspecific trait variation studies are conducted, particularly for LMA with temperature and for WD with moisture, because environmental variation is indeed causing important trait variations but with no effects on species turnover. Widespread species forming monospecific forests, and thus having maximal abundance in the forest community, may constitute a valid exception to take into account and a key system for testing assumptions and consequences of the functional trait paradigm.