The study site is located at the Station de Biologie des Laurentides (SBL) of the University de Montréal in St. Hippolyte, Quebec (45°59′N; 74°00′O). The SBL lies within the SM-yellow birch (Betula alleghaniensis Britton) domain of the lower Laurentians, near its northern limit and at the transition with the boreal forest (Saucier et al., 2009). Sugar maple is found concurrently with RM, yellow birch, poplars (Populus spp.), balsam fir (Abies balsamea (L.) Mill.), white spruce (Picea glauca [Moench] Voss) and eastern white cedar (Thuja occidentalis L.). Due to the mosaic of tree species that formed the site, it was possible to find SM and RM seedlings growing concomitantly along an increasing gradient of soil acidity conditioned by increasing proportions of coniferous tree species.

Plots that were used for this experiment are the same as those previously studied by Collin et al. (submitted). Four plots (50 × 50 m) were delineated under each of the following forest canopies: (1) hardwood stands of SM and birch species; (2) mixed hardwood-conifer stands with SM, Betula spp. and conifers; and (3) conifer-dominated stands with SM seedlings and saplings (3 sites × 3 species composition × 4 repetitions = 36 plots). In each of these plots, SM and RM seedlings were present. Foliage from five SM and RM seedlings (between 20 and 30 cm in height) were collected within each plot for subsequent chemical analysis. Total height and diameter at the ground level were measured on all SM and RM seedlings that were sampled. We did not measure the age of SM and RM seedlings, but we assume that SM seedlings were a little bit older than RM seedlings, on average, due to their greater shade-tolerance and slower growth rate. However, nutrient concentrations in plant parts are mostly affected by their physiological age (Marschner, 2011). Therefore, comparisons of nutrient concentrations and ratios should ideally be done between plants of similar growth stages and not necessarily of similar ages. In this respect, the bias that was incurred by comparing SM and RM of similar heights and growth stages, despite possible differences in ages, is likely small.

The sandy loam soils at St. Hippolyte are Orthic Humo-Ferric Podzols (Soil Classification Working Group, 1998) developed from well-drained rocky glacial tills composed of a mixture of rocks from the small underlying anorthosite pluton (deep-seated intrusive igneous rock) of the Morin Complex (Doig, 1991) and of more felsic rocks (e.g., mangerite, syenite, charnockite) in the surroundings of the anorthosite pluton (Bélanger et al., 2012). The forest floor is a moder humus form with a thickness of 5–10 cm. Details on soil sampling, analysis and chemical properties are provided in Collin et al. (submitted). We simply confirm here that conifer-dominated stands have a significantly lower forest floor (i.e., FH horizons) pH (4.05 ± 0.07) compared to forest floors supporting mixed hardwood-conifer stands (4.29 ± 0.11) and hardwoods (4.39 ± 0.09). On the whole, nothing indicates that variations in forest floor pH are due to intrinsic site factors (e.g., parent material and topographic position, slope or aspect) other than forest type. For example, mineral soil texture is similar under each of the forest canopies and composed of about 2–3% of clay, 38–42% of silt and 55–60% of sand.

Leaf samples were oven-dried for 72 h at 65°C upon arrival in the laboratory. They were then pulverized using a ball mill prior to total C and N determination by high-temperature combustion and thermal conductivity detection (EA 1108 CHNS-O Analyzer, Thermo Fisons, Waltham, MA, USA). A 0.2 g subsample was also digested in 2 ml of 15 N HNO₃ for 4 h at 100°C in glass test tubes. Base cation concentrations (Ca, Mg, and K) of the digests were determined using atomic adsorption/emission (Model AA-1475, Varian, Palo Alta, CA, USA), whereas P concentrations were determined colorimetrically using the molybdenum blue method (Technicon Autoanalyzer, Technicon Instruments Corporation, Tarrytown, NY, USA).

Computing ilr implies that the compositional data are numerically constrained to a compositional data space. Given that not all elements of the leaf are typically measured, it is therefore a necessity to first compute a filling value (Fv), which replaces the elements that were not measured. This Fv was calculated as the difference between the unit of measurement (i.e., 100%) and the sum of all nutrients that had been measured. The compositional data matrix was then constrained as described by Aitchison (1986), i.e., the closure operation computes the constant sum of components and closes it to some whole such as 100%. Once the composition data space was closed, the ilr-transformation was applied (Egozcue et al., 2003). This ilr system of balances is first defined by a nested, binary contrasts that are implemented during the sequential binary partition (SBP) steps, in which components are hierarchically arranged into functional sub-compositions (Parent S. E. et al., 2012). This hierarchical arrangement of balances needs to be constructed based on prior knowledge or hypotheses as a means to produce interpretable results (Parent S. E. et al., 2012). In this study, the SBP (Table 1) was designed based on prior knowledge of nutrient interactions in higher plants (Marschner, 2011). We first contrasted all measured nutrients (i.e., N, P, K, Ca, and Mg) with Fv, i.e., [Fv∣Elements]. We then partitioned measured nutrients to reflect the charge balance in plant cells (Marschner, 2011). Thus, N and P were contrasted with K, Ca and Mg, i.e., the two-part amalgamated ratio [Mg,Ca,K∣P,N]. Sub-compositions were further divided into an anionic Redfield [P∣N] balance (Loladze and Elser, 2011) and cationic [Mg,Ca∣K] and [Mg∣Ca] balances (Marschner, 2011).

The ilr balances defined during the SBP steps (Table 1) were then computed according to the equation by Egozcue and Pawlowsky-Glahn (2005): ilrj=nj+nj−nj++ nj−lng(cj+)g(cj−) where, in the jth row of the SBP, nj+ and nj− are the numbers of components in the plus (+) and minus (−) groups, and g(cj+) and g(cj−) are the geometric means of components in the “+” and “−” group, respectively. Balances are generally noted [−1 group | +1 group], with denominator on the left because log ratios become more negative as values at denominator increase, hence leaning to the left as in algebra. For example, the ilr equation of the [Mg|Ca] partition is 12lnCaMg. If Ca concentration increases, then the ilr coordinate will also increase in value. In addition, an increase in Ca concentration will change the [Mg|Ca] and [Mg,Ca|K] balances, without affecting the [P|N] balance. This technique produces D−1 balances in a D-part composition, avoiding matrix singularity and redundancy. The transformation also assures that ilr balances are orthogonal to each other, thus linearly independent, making this technique totally appropriate for conducting multivariate and multiple regression analyses.

Because growth rates of maple seedlings were not measured in this study, we could not directly assess their health and vigor in relation to foliar nutrients. Therefore, we compared foliar data obtained from the literature to foliar data from SM seedlings at St. Hippolyte to assess their nutritional status. First, foliar nutrient concentrations that were measured in this study were compared to foliar nutrient concentrations from a small set of recent literature reports that included data on SM and RM seedlings that were growing on similar acidic soils (forest floor pH ranging from 3.4 to 5.5). The data were used to construct ranges of foliar N, P, K, Ca, and Mg concentrations. Second, we attempted to develop specific ranges for foliar ilr balances in healthy and declining maple seedlings from the literature. However, data on foliar nutrients for SM and RM seedlings are sparse. Many studies do not report all foliar nutrients that are required to compute ilr balances, nor do they relate foliar nutrients to seedling performance. Foliar nutrient concentrations of SM trees are more widely documented and relationships to tree fitness are often provided. A substantial SM data sets comprised of trees growing on soils (forest floors) with pH values ranging from 3.14 to 7.04 was assembled, which allowed us to construct ranges in foliar ilr balances for healthy and declining SM stands. This data set was later used as a proxy for the nutritional status of SM seedlings that were growing under three different forest types (hardwood, mixed hardwood-conifer and conifer-dominated forests) at St. Hippolyte. Conversely, foliar nutrient data for RM are sparse for both trees and seedlings and, in turn, it was not possible to compute foliar ilr balances for RM.

Data were analyzed using R software (version 3.0.0, R Core Team, 2013). The ilr-transformations were performed with the functions acomp and ilr of the compositions package (van den Boogaart et al., 2013). The function acomp is used for the closure operation, and the function ilr is used on this closed data space to perform the ilr transformation equation described above. To assess whether forest types have specific foliar nutrition signatures, linear discriminant analysis combined with a posteriori classification was performed with the function lda available in the MASS package (Venables and Ripley, 2002). Foliar ilr balances were employed as the response variables, and the multivariate homogeneity of variance-covariance matrices (permutation-based, multivariate extension of Levene's test) between species was tested using the betadisper function in the vegan package (Oksanen et al., 2013). Multivariate regressions between foliar signatures as the response variables (i.e., all balances considered) and forest floor pH as the explanatory variable were computed using data from all forest types to determine if foliar balances of SM or RM seedlings are controlled by forest species composition. Simple linear regressions between each foliar ilr balance and forest floor pH were then performed to quantify the relative contributions of each balance to the main multivariable SM and RM regressions if the latter were found significant. One-way analysis of variance and Tukey's HSD (honest significant difference) post-hoc tests were also used to separate means of forest types or species based on their respective mean foliar nutrient concentrations and balances. Testing nutrient concentrations aided our interpretation of nutrient balance results. Foliar nutrient concentrations that were used in the analysis were occasionally log-transformed to meet the assumptions of normality and homoscedasticity of the residuals.

Concentrations of foliar N, P, K, Ca, and Mg of SM and RM seedlings among the three forest types at St. Hippolyte are presented in Table 2. Sugar maple seedlings that were sampled in hardwood stands have significantly higher foliar Ca and Mg concentrations than seedlings in conifer-dominated stands. Foliar K concentrations of RM seedlings in hardwood stands are significantly higher than seedlings in conifer-dominated stands. Also, RM seedlings in mixed hardwood-conifer stands have significantly higher foliar N concentrations than seedlings in conifer-dominated stands. Finally, RM seedlings have significantly lower foliar N and higher foliar Mg concentrations than SM seedlings (Table 2).

Foliar nutrient concentrations in SM and RM seedlings that were measured in St. Hippolyte are generally within the range of concentrations that have been reported in the recent literature (Figure 1). Only foliar P concentrations of both SM and RM seedlings are at the lower limit of literature ranges.

Foliar ilr balances of SM and RM seedlings at St. Hippolyte were computed for each forest type (Table 3). Considering the influence of forest type on soil chemical properties (notably forest floor pH, see Collin et al., submitted), it was necessary to analyze the data using a mixed-effects model to separate the plot and forest type effects on the foliar balances of SM and RM. Analysis revealed that SM seedlings have significantly higher foliar [Mg,Ca,K|P,N], [Mg,Ca|K], and [Mg|Ca] balances, and a significantly lower foliar [Fv|Elements] balance compared to RM (Figure 2). It is important to note that these balances do not behave in the same manner as traditional ratios. A decreasing foliar ilr balance means that either the left part of the balance (the numerator of the ratio) is increasing or that the right part (the denominator of the ratio) is decreasing. Linear discriminant analysis suggests that the foliar [Mg|Ca] balance, followed by the [Mg,Ca,K|P,N] balance, can discriminate most RM seedlings from SM seedlings (Figure 3). The resulting classification matrix indicated 91.6% and 85% correct classifications of SM and RM seedlings, respectively, using foliar balances. This indicates that foliage of SM and RM seedlings possess distinct nutritional signatures.

Foliar [Mg,Ca,K|P,N] and [Mg,Ca|K] balances of SM seedlings rise with increasing proportions of conifers (and thus, increasing forest floor acidity) (Figure 4). These balances are significantly higher for seedlings in conifer-dominated stands than for seedlings in hardwood stands. Conversely, the foliar [Fv|Elements] balance of seedlings in conifer-dominated stands is significantly lower than that for seedlings in hardwood stands. No difference was found among forest types in the foliar balances that were computed for RM. A high degree of variability was found for the balances of RM seedlings in hardwood stands (Table 3, Figure 4).

A significant linear relationship was found between foliar signatures (i.e., all balances considered) of SM seedlings and forest floor pH which varied with forest types (adjusted R² = 0.262, P < 0.01, Table 4). However, this relationship is only due to variations in foliar [Mg,Ca|K] balances (adjusted R² = 0.494, P < 0.01). There is no relationship between foliar signatures of RM seedlings and forest floor pH (adjusted R² = −0.056, P = 0.83, Table 4). This is in agreement with the observation that foliar balances of RM seedlings do not vary significantly among forest types.

Foliar ilr balances for healthy and declining SM stands that were computed from literature foliar nutrient data are presented in Table 5. The ranges of those foliar balances are represented in Figure 4, together with foliar ilr balances of SM seedlings at St. Hippolyte. Comparisons between data sets can be used to partially assess SM seedling fitness. On the one hand, foliar [Mg,Ca,K|P,N] and [Mg,Ca|K] balances of declining SM stands are higher than foliar balances of healthy stands. On the other hand, foliar [Fv|Elements] balances are lower for declining SM stands than healthy stands. Among these three balances diverging between healthy and declining stands, foliar [Fv|Elements] of SM seedlings in mixed hardwood-conifer and conifer-dominated stands at St. Hippolyte fall within the upper limit of declining stands, whereas foliar [Fv|Elements] of SM seedlings in hardwood stands fall within the range of healthy stands. Foliar [Mg,Ca|K] and [Mg,Ca,K|P,N] balances of SM seedlings in hardwood and mixed hardwood-conifer stands fall within the upper limit of the ranges of healthy stands, whereas the balances of seedlings in conifer-dominated stands fall within or lean toward the lower limit of declining stands for [Mg,Ca|K] and [Mg,Ca,K|P,N], respectively.

Increasing proportions of conifers and soil acidity resulted in modifications of foliar ilr balances only for SM seedlings (Figure 4). A significant decline in the foliar [Fv|Elements] balance with an increasing proportion of conifers indicates a general decrease in nutrients or an increase in Fv. The Fv is composed of elements that are not often measured, particularly in forest studies, viz., C, H, O, other macronutrients (e.g., S) and micronutrients (e.g., Cu, Zn, B). An increase in foliar Al and Mn concentrations in SM seedlings under an increasing proportion of conifers may be one factor explaining Fv variation. Increases in foliar Al and Mn have been recorded in SM due to decreasing soil pH (Hallett et al., 2006; Kogelmann and Sharpe, 2006; Schaberg et al., 2006; Long et al., 2009; Park and Yanai, 2009; Beauregard et al., 2010). High ionic activity of Al and Mn in acidic soil solution is common and this tends to increase their uptake by tree roots at the expense of Ca and Mg, including SM (Cronan and Grigal, 1995; Likens et al., 1998). A significant increase in the foliar [Mg,Ca,K|P,N] balance can be due to a decrease in Mg, Ca or K concentrations, or an increase in P and N concentrations. However, our data suggest that the latter explanation is compromised, as foliar N and P concentrations of SM seedlings did not vary between forest types (Table 2). Thus, the significantly higher foliar [Mg,Ca|K] balance of SM seedlings in conifer-dominated stands compared to SM seedlings in hardwood stands, together with stability of the foliar [Mg|Ca] balance between forest types, could mean that Ca and Mg are deficient in conifer-dominated stands compared to hardwood stands. The foliar [Mg,Ca|K] balance was also the main factor explaining foliar ilr balance discrimination among forest types (Table 4). Differences in foliar balances, therefore, are clearly related to decreasing soil pH and most probably increased Al and Mn mobility/availability. The latter are both conditioned by increasing proportions of conifers. In turn, there is an antagonistic effect on Ca and Mg uptake by SM seedlings that subsequently translates into foliar imbalances. These results support the first part of our hypothesis that nutrient imbalances in SM foliage increase due to increasing soil acidity as conditioned by increasing proportions of conifers.

It is well known that SM trees are particularly sensitive to low Ca and Mg availability compared to other tree species that are growing on acidic soils (Kobe et al., 2002; Duchesne and Ouimet, 2009; Long et al., 2009). Most declining SM stands in southern Quebec were attributed to low soil Ca and Mg availability, in part, due to imbalances in Al and Mn that were induced by atmospheric acid deposition. Imbalances could also be attributed to exports of Ca and Mg in harvested biomass in excess of natural inputs, resulting in low foliar Ca and Mg (Duchesne et al., 2002; Ouimet et al., 2006; Duchesne and Ouimet, 2009). The health and vigor of SM trees is constrained by the low Ca and Mg availability of acidic soils, and accompanied by foliar Ca and Mg concentrations below deficiency thresholds (Drohan et al., 2002; Bailey et al., 2004; Houle et al., 2007; St. Clair et al., 2008; Long et al., 2009). Foliar Mg concentrations in declining SM stands are often very low and correlated with a decline in canopy health (Horsley et al., 2000). The benefits of liming declining stands of SM on rates of photosynthesis or growth were demonstrated through field studies (Liu et al., 1997; Duchesne et al., 2003; Moore and Ouimet, 2006; Schaberg et al., 2006; Moore et al., 2014). Calcium and Mg deficiencies of SM seedlings in conifer-dominated stands compared to hardwood stands at St. Hippolyte suggests that conifer-dominated boreal forests could limit the ability of SM to migrate northward to find a more adequate climate in a rapidly warming world. Our results only reflect the best-case scenario as the SM seedlings that were measured were those capable of naturally growing even under more stressful conditions that were found in conifer-dominated stands. Many conifer-dominated stands, for example, were not selected for the study because SM seedlings were not present. Similarly, the only foliar balances of SM seedlings that were significantly different among forest types at St. Hippolyte, i.e., [Fv|Elements], [Mg,Ca,K|P,N], [Mg,Ca|K] (Table 3, Figure 2), were also those differing between healthy and declining SM stands found in the literature (Table 4, Figure 4). Because the data linking yield, fitness and foliar nutrition are unfortunately limited to SM trees, not to SM seedlings, it is likely more appropriate to refer to samples in conifer-dominated stands as poorly acclimating seedlings rather than declining seedlings.

Foliar balances of RM seedlings at St. Hippolyte do not show any significant difference among forest types (Figure 4), nor are foliar balances related to forest floor pH (Table 5). These results indicate a greater capacity of RM seedlings to maintain their foliar balances within a certain range of soil acidity (i.e., forest floor pH on the study site ranges from 3.73 to 4.94) than SM seedlings. Greater variation in foliar balances of RM seedlings in the hardwood stands may be the reason that we failed to detect significant differences among forest types. However, the large variation may also reflect higher nutritional plasticity in RM to contrasting environmental conditions. In the past few decades, RM populations in eastern North America have been expanding more rapidly than those of other tree species (Alderman et al., 2005; Fei and Steiner, 2007). Expansion of RM was related to factors as diverse as fire regime, climate, biotic interactions and physiological characteristics. For example, high recruitment capacity that is associated with high seed banking (and, therefore, fast colonization capacity) was proposed as an explanation for its current expansion (Warren et al., 2004; Lambers and Clark, 2005). Lower energy and resource requirements for biomass production and maintenance were also suggested to explain its competitive success (Nagel et al., 2002). As a whole, RM is characterized as a competitive generalist possessing functional traits belonging to both early and late-successional tree species (Abrams, 1998). It is able to thrive on sites with greatly contrasting soil conditions due to its lower water and nutrient requirements for growth and survival compared to many other hardwood species in North America. Our results support the second part of our hypothesis that the lower nutrient requirements of RM seedlings allow the species to adapt nutritionally to a wide range of acid soils compared to SM seedlings. This lower nutrient requirement is another possible factor explaining the recent expansion of RM relative to SM on acidic and nutrient-poor soils.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.