Whole-plant trait spectra of North American woody plant species reflect fundamental ecological strategies

The adaptation of plant species to their biotic and abiotic environment is manifested in their traits. Suites of correlated functional traits may reflect fundamental tradeoffs and general plant strategies and hence represent trait spectra along which plant species can vary according to their respective strategies. However, the functional interpretation of these trait spectra requires the inspection of their relation to plant performance. We employed principle coordinate analysis (PCoA) to quantify fundamental whole-plant trait spectra based on 23 traits for 305 North American woody species that span boreal to subtropical climates. We related the major axes of PCoA to five measures of plant performance (i.e., growth rate, and tolerance to drought, shade, water-logging and fire) for all species and separately for gymnosperms and angiosperms. Across all species a unified gymnosperm-angiosperm trait spectrum (wood density, seed mass, rooting habit) is identified, which is correlated with drought tolerance. Apart from this, leaf type and specific leaf area (SLA) strongly separate gymnosperms from angiosperms. For gymnosperms, one trait spectrum emerges (seed mass, rooting habit), which is positively correlated with drought tolerance and inversely with shade tolerance, reflecting a tradeoff between these two strategies due to opposing trait characteristics. Angiosperms are functionally more diverse. The trait spectra related to drought tolerance and shade tolerance are decoupled and three distinct strategies emerge: high drought tolerance (low SLA, dense wood, heavy seeds, taproot), high shade tolerance (high SLA, shallow roots, high toxicity, opposite arranged leaves), and fast growth/stress intolerance (large maximum heights, soft wood, light seeds, high seed spread rate). In summary, our approach reveals that complex suits of traits and potential tradeoffs underlie fundamental performance strategies in forests. Studies relying on small sets of plant traits may not be able to reveal such underlying strategies.


INTRODUCTION
Plant functional traits reflect adaptations to the environment and affect ecosystem functioning.They are thus the key inferring plant strategies and for predicting ecosystem properties (Cornelissen et al. 2003, McGill et al. 2006, Lavorel et al. 2007).Plant strategies are manifested in a suit of whole-plant performances including e.g., reproduction, growth and survival, each governed by a particular set of morphological, anatomical or physiological traits.Characterizing plant strategies therefore requires the knowledge of many traits.However, we typically lack the complete information on multiple important traits, because plant trait datasets are often biased towards a few traits that are easily measured (e.g., characteristics of leaves, seed weight, plant maximum height), while other traits may only be available for a small subset of plant species (e.g., potential allelopathy, bark thickness) (Kattge et al. 2011).
Functional traits are often grouped as sets of co-varying traits that reflect evolutionary or biophysical tradeoffs and hence jointly represent an axis of trait variation (trait spectrum) along which plant species can vary according to their respective strategies (Reich et al. 2003, Lavorel et al. 2007).Such trait spectra are typically identified with dimension reduction methods that extract lower-dimensional information (orthogonal principle components) from multidimensional trait data sets (Grime et al. 1997, Wright et al. 2007).Two prominent examples are the leaf and the wood economic spectrum (Wright et al. 2004, Chave et al. 2009) each of which described by a characteristic set of co-varying traits.The two spectra appear to be orthogonal to one another (Baraloto et al. 2010).This implies that the number and types of traits that vary independently of each other (i.e., not necessarily the total number of traits) determine unique strategy axes.However, proving the existence of trait spectra per se does not necessarily tell us anything about their relevance for whole-plant performance and for vital rates such as growth, stress tolerance or reproduction.Thus additional information on plant performance must be considered to fully understand their implications.This is challenging because comparative field data on species-specific performances controlled for the biotic and abiotic environment are scarce.Hence, studies relating traits to performances are typically conducted in controlled settings, often with short-lived herbaceous species (Poorter andVan der Werf 1998, Useche andShipley 2010).Such 'rates and traits' studies for trees are typically restricted to growth and mortality and conducted on well-studied small forest plots with homogeneous environmental conditions (Ru ¨ger et al. 2012).Biome-scale studies are lacking.Moreover, there are hardly any studies focusing on more holistic performance measures such as niche preferences or stress tolerances-integrating growth, survival and recruitment-despite the fact that these are critical for understanding vegetation composition and dynamics under global change.
In forest communities, light and water availability are critical drivers of temporal and spatial processes such as succession and zonation (Smith and Huston 1989, Pacala et al. 1996, Poorter and Markesteijn 2008).Species-specific shade and drought tolerances are typically inferred as ordinal indices ('scores') from occurrences and vitality along gradients of light and water availability and are viewed as suitable measures of performance.Niinemets and Valladares (2006) found an inverse relationship between shade and drought tolerance indices among woody species of different plant functional types for all continents that aligns with Smith and Huston's (1989) cost-benefit tradeoff model.This model predicts (1) no viable strategies should exist under both low light and low water due to conflicts between allocation to roots versus shoots, (2) under abundant light and water, the highest growth rates are expected in species deemed intolerant of both stresses, and (3) a tradeoff between high growth under favorable conditions versus low growth under more stressful conditions (Craine 2009).The first two predictions suggest a tight tradeoff between shade and drought tolerance, implying a highly constrained trait space (i.e., limited niche differentiation) for traits underlying these tolerance indices.Independence between these two tolerance indices would indicate a more flexible trait space (Sack 2004).Trait-based analyses, however, have yielded ambiguous results.For instance, Hallik et al. (2009) identified leaf traits underlying the inverse relationship between shade and drought tolerance of temperate tree species, while Markesteijn and Poorter (2009) found different trait spectra associated with shade and drought tolerance of tropical tree seedlings based on leaf, stem and root traits.
Disturbance events that lead to significant biomass loss can interact with stress tolerance strategies to affect whole-plant performance.This implies the existence of other strategies related to the ability to tolerate or recover from disturbances such as fires, wind-throw, or snow-break.For instance, Loehle (2000) described a tree strategy scheme with four independent axes (shade tolerance, tree height, seed dispersal, capacity to resprout) that was incorporated into a fitnessbased model to predict species richness under different disturbance regimes in North American forests.According to Bond and Midgley (2001), three of these four axes align with the traits and tradeoffs defined in Westoby's (1998) LHS scheme (L ¼ specific leaf area [SLA] relates to shade tolerance; H ¼ maximum tree height; S ¼ seed mass relates to seed dispersal).Loehle's fourth axis (capacity to resprout) reflects adaptations to disturbance (Pausas and Lavorel 2003).
In summary, the linkage between plant functional traits and tolerance indices which reflect whole-plant performance as a function of growth, reproduction and survival along environmental and disturbance gradients has yet to be quantified.This quantification requires species-specific information for a potentially large number of traits (Grime et al. 1997, Diaz et al. 2004, Wright et al. 2007) that reflect adaptations to key environmental factors (Cornelissen et al. 2003, Lavorel et al. 2007) and influence wholeplant performance (Violle et al. 2007) ideally for a large number of species.
In this study, we developed a database that contains information on 23 traits for 305 North American woody species that span boreal to subtropical climates.We conducted principle coordinate analyses (PCoA) to identify dominant trait spectra, and we evaluated how these spectra are related to whole-plant performance indices.In doing so, we addressed three sets of questions: (1) What are the dominant trait spectra, what are the traits underlying each spectrum, and how do they relate to whole-plant performance measures, i.e., growth rate and tolerance to shade, drought, water-logging, and fire?(2) Do the relationships between trait spectra and performances differ between major phylogenetic clades, such as gymnosperms versus angiosperms?(3) To what degree do these trait spectra agree with existing plant strategy schemes proposed by Westoby (1998) and Loehle (2000)?

Selection of species and traits
The primary literature and various existing databases (USDA, NRCS, National Plant Data Team 2007, Kattge et al. 2011) were mined for data on functional traits for the 305 North American woody species (103 gymnosperms, 202 angiosperms) identified by the US Forest Service's Forest Inventory and Analysis (FIA) program (Miles et al. 2001) (see Appendix A: Table A1 for a complete list).The species span a range of habitat types (semi-arid woodlands, mesic temperate, boreal, to sub-tropical forests), and are classified mostly as trees (only 16 shrubs are included, of which 7 are chaparral species).The 305 species are phylogenetically dispersed across 79 genera, 36 families, 16 orders, and the two major clades.
To identify traits that most likely reflect adaptation to key environmental factors and that are likely to be related to whole-plant performance we used the following criteria: (1) traits are causally related to fitness components, i.e., growth, reproduction and survival (Violle et al. 2007), (2) they reflect adaptation to constraints by water and light, (3) they are related to the ability to resist or recover from disturbances, and/or adaptation to competition stress (Lavorel et al. 2007), (4) they reflect relevant functions (storage, defense, resource acquisition, dispersal) carried out by different plant organs (leaf, stem, root, seed) and ( 5) they are quantifiable for many species spanning a range of resource, climatic, and disturbance gradients.To minimize trait redundancy, we avoided selection of multiple, correlated traits related to the same tradeoff; e.g., only one trait (SLA) was chosen from the leaf economics spectrum (Wright et al. 2004).We focused on 23 traits that fulfilled these criteria (Table 1 and Appendix A: Table A2 for a detailed description).
Abbreviations used in Figs.1-4.à For details regarding how trait scales were adapted see Appendix B.
§ Data were provided via the TRY initiative (Kattge et al. 2011), which includes data from different sources given in brackets.
v www.esajournals.org'dispersal mode', which we included despite that it was only available for 54% of the species (see Appendix A: Table A3).The species-trait matrix that we compiled was 85% complete (15% missing data).The fraction of missing information was further reduced to 4% by replacing missing continuous and ordinal traits with the mean values of the respective genera.The 23 traits included nine nominal traits (e.g., leaf type, root habit), seven ordinal traits (e.g., bark surface, seed spread rate, nitrogen fixation), and seven continuous traits (e.g., SLA, seed mass); see Table 1 for the complete list.Because we are interested in broad patterns across species, variability within a species was not incorporated into the analysis.

Selection of performance measures
We selected five species-specific performance measures: growth rate (USDA, NRCS, National Plant Data Team 2007), shade, drought, waterlogging tolerances (Niinemets and Valladares 2006 ), and fire tolerance (USDA, NRCS, National Plant Data Team 2007).These indices represent whole-plant behavior that affects plant performance along resource or disturbance gradients and are expected to result from the coordination of multiple functional traits (Reich et al. 2003, Violle et al. 2007, Webb et al. 2010).Shade, drought, and water-logging tolerance are ordinal variables ranging from 1 (intolerant) to 5 (tolerant).These species-specific tolerance indices are based on (1) site characteristics (e.g., annual precipitation and duration of the dry period for drought tolerance) representative for each species' range, (2) the physiological potential that a species can survive long periods of exposure to the associated stress (e.g., 50% of foliage damage for drought tolerance), but to a certain degree also on expert knowledge about morphological and life history traits (Niinemets and Valladares 2006).To reduce the risk of circularity we validated the shade and drought tolerance indices.To validate the shade tolerance classifications of Niienemets and Valladares (2006), we correlated these with the shade tolerance estimates of Lichstein et al. (2010), which are only derived from individual growth versus lightlevel measurements.Both shade tolerances were significantly correlated (r ¼ 0.56, P , 0.001, see Appendix A: Fig. A1), and thus, we used one of the indices (Niinemets and Valladares 2006).To validate the drought tolerance classifications, we correlated these with species-specific scores of humidity (annual P minus potential evapotranspiration derived from species' geographic distribution; see Appendix A: Fig. A2) and found a significant correlation (r ¼ 0.40, P , 0.001).Fire tolerance is also an ordinal variable that ranges from 0 (intolerant) to 3 (high tolerant) and describes the relative ability of a species to resprout, regrow, or reestablish from seeds after a fire.Growth rate is represented on an ordinal scale from 1 (slow) to 3 (fast) and describes the growth potential of a species relative to other species after successful establishment.The imputation of missing values for the performance measures was done in the same manner as for the functional traits (see Appendix A: Table A3).

Data analysis
The data analysis was performed in two steps.First, we quantified trait spectra (sets of co-varying traits) as composite variables by extracting dominant axes of trait variation running a principle coordinate analysis (PCoA) (Gower 1971).We chose PCoA over principal components analysis (PCA) or correspondence analysis (CA) because it is flexible in the choice of distance measures, which was more appropriate for our analysis given the different data types (i.e., nominal, continuous, ordinal, multi-choice nominal); Euclidean (PCA) or chi-square (CA) distances are not appropriate in our setting.PCoA provides a Euclidean representation (i.e., a graphical representation in a Cartesian coordinate system) of the distance relationships between species based on their trait values.Thus, the major PCoA axes extracted from our data represent dominant variation in the traits, i.e., the axes are explained by certain sets of traits (trait spectra).To understand the attributes of the respective trait spectra, we correlated the PCoA axes with the trait values.Second, we explored relationships between the main trait spectra (i.e., major PCoA axes) and the plant performance indices.
PCoA was conducted on a general extension of Gower's distance coefficient (Gower 1971) developed by Pavoine et al. (2009), which accounts for different trait data types by assigning appropriate metrics for the specific data types, has Euclidean properties, and accounts for missing values.Due to missing data (4%), the distance matrix was unbalanced, and thus, the PCoA yielded some negative eigenvalues.We did not correct for these because the largest negative eigenvalues were smaller in absolute magnitude than any of the first three positive eigenvalues of interest, i.e., their interpretation was still meaningful following the rules provided by Legendre and Legendre (1998).
To account for differences between gymnosperms and angiosperms, we repeated the analysis three times: once for all 305 species and 23 traits followed by two analyses that considered gymnosperms and angiosperms separately.The gymnosperm analysis was based on 19 traits (conduit type/arrangement, leaf composition, leaf margin, and nitrogen fixation were excluded because they did not vary within this group).The angiosperm analysis was based on 22 traits (dispersal mode was excluded because of missing values for 46%).If necessary traits were square-root-or log-transformed to approximate normality and to reduce the influence of extremely high values.All continuous-valued traits (original or transformed) were standardized to z-scores by subtracting the mean trait value across species from each species-specific value and by dividing this difference by the standard deviation across species.
To assess the relative contribution of each trait to the trait spectra, we computed linear correlation coefficients (Pearson's r) between all traits and the first three PCoA axes of the species-trait matrix, representing the main trait spectra.To explore the relationship between the main trait spectra and the plant performance indices, we computed linear correlations between the first three PCoA axes and the growth rate and tolerance indices.All statistical analyses were conducted in R (R Development Core Team 2010).

RESULTS
For each of the three analyses (all species, gymnosperms, angiosperms), the first three PCoA axes explain a significant amount of variation (total ;40%) in trait values (Table 2).Each additional axis explains ,6 % of the total variation, and we do not include these axes in subsequent analyses.

All 305 woody species
When considering all 305 species, the first axis explains 26% of trait variation (Table 2; Fig. 1).This axis is mainly correlated with leaf traits (leaf type, SLA, leaf margin), resprouting capacity and conduit type (vessel vs tracheids) (Fig. 1A and B, Table 2) and separates angiosperms and gymnosperms.Water-logging tolerance, fire tolerance, and growth rate are the performance indices that are most strongly correlated with this axis (Fig. 1B, Table 2).The second axis explains 9% of trait variation, and is mainly driven by traits representing responses to water availability and reproduction (wood density, root habit, seed mass).Drought tolerance and growth rate are correlated with this axis, but in opposite directions.Drought tolerant, slow growing species are located at the positive end of this axis and are characterized by a growth form intermediate between trees and shrubs (see Appendix A: Table A2 for definition) with heavy seeds, slow seed spread rate, slow vegetative spread rate, dense wood, and a taproot (Fig. 1B, Table 2).Conversely, drought intolerant, fast growing species are located at the negative end of the axis.They exhibit a tree-like growth form and support the opposite trait configuration.The third axis explains 6.9% of the trait variation and is mainly driven by leaf composition, potential for allelopathy and bark surface; it is not significantly correlated with any of the performance measures (Table 2).

Gymnosperms
For the gymnosperm-only analysis, the first axis explains 22% of trait variation and separates species in the Cupressaceae family, characterized by opposite arranged evergreen scale-like leaves, from Pinaceae, Taxaceae, and Taxodiaceae with spirally arranged evergreen needles (Fig. 2A and  B, Table 2).Growth rate and drought tolerance indices are correlated with the first axis (Fig. 2B, Table 2).The second axis explains 13% of trait variation and is driven by traits related to water acquisition and reproduction.Drought tolerance and shade tolerance are inversely correlated with this axis (Fig. 2B, Table 2), reflecting the negative relationship between these tolerance strategies.The drought tolerant and shade intolerant species located at the positive end of the second axis exhibit tree/shrub growth forms, animal-dispersed seeds, heavy seeds, a taproot, low seed spread rate, low vegetative spread rates, and small maximum heights (Fig. 2A and B, Table 2).Drought intolerant and shade tolerant species are located at the negative end of the second axis.They exhibit the opposite trait configuration and a tree-like growth form.The third axis explains 11% of the trait variation, and growth rate is (not significant).The explained variance per axis is given in %; in case of no data entry the trait did not occur or did not vary in the specific group or was excluded from the analysis (see Appendix A: Table A3 and Material and Methods).
v www.esajournals.orgcorrelated with this axis (Fig. 2C and D, Table 2).Fast growing species are located at the positive end of this axis and tend to have alternatearranged leaves, a taproot, thick bark, high resprouting capacity, and low fire resistance, compared to the opposite trait configuration for slow growing species at the negative end of this axis.Several genera such as Pinus, Larix, and Abies cover the full range of traits spanned by the third axis.

Angiosperms
In the angiosperm-only analysis, the first axis explains 14% of trait variation and is driven by traits related to water availability and reproduction (Fig. 3A and B and Table 2).This axis is positively correlated with drought tolerance and negatively with growth rate and water-logging tolerance.The second axis explains 12% of the trait variation and is driven by leaf composition and potential for allelopathy; this axis is not notably correlated with any performance measure used (Fig. 3A and B, Table 2).The third axis explains 9% of trait variation and is driven by maximum height and leaf arrangement.This axis is correlated with shade tolerance (Fig. 3C and D, Table 2).
The first and third axes reveal differences between species possessing three different strategies forming a plant strategy triangle with respect to combinations of shade, drought, water-logging tolerance and growth rate (Fig. 3C and D).The corners of this triangle are depicted by (1) species intolerant to shade and drought but with high growth rate and high water-logging tolerance, (2) species tolerant to shade but intolerant to drought and waterlogging and no correlation with growth rate, (3) species tolerant to drought but intolerant to shade and water-logging and exhibiting low growth rate (Fig. 3C and D, Table 2).The first strategy is represented by species of the genera Populus and Betula, which are characterized by low wood density, light seeds, large maximum heights, deciduous leaves, tree-like growth form, short lifespan, low toxicity, high seed spread,  1 for abbreviations and Table 2 for respective correlation coefficients); the lengths of the arrows are proportional to their correlation coefficient, and they point in the direction of most rapid change; nominal traits were dummy coded before correlation.
v www.esajournals.orghigh vegetative spread rate, and dissected leaf margins (Fig. 3C and D, Table 2).The second strategy is mainly represented by species of the genera Acer, Aesculus and Cercocarpus, which are characterized by high SLA, opposite-arranged leaves, shallow roots, small maximum heights, diffuse-porous wood, high toxicity, thin bark, and a shrub-like growth form.The third strategy type is mainly represented by species of the genus Quercus, which is characterized by low SLA, dense wood, heavy seeds, taproot, thick bark with a rough surface, long lifespan, ringporous wood, and alternately arranged leaves (Fig. 3C and D, Table 2).

Comparison of gymnosperms and angiosperms
Here we compare the PCoA axes obtained from the gymnosperm-and angiosperm-only analyses that are most strongly correlated with shade and drought tolerance (Fig. 4).The axes that are primarily related to drought tolerance (second axis for gymnosperms, first axis for angiosperms, see Table 2) are explained by the same set of traits, which are correlated with each axis in the same direction.That is, regardless of clade association, drought tolerant species tend to have tree/shrub like growth form with a taproot, dense wood, high seed mass, and both low vegetative and low seed spread rates (Fig. 4A, upper right corner); drought intolerant species are characterized by the opposite trait configuration (Fig. 4A, lower left corner).Conversely, the set of traits that are correlated with the axes related to v www.esajournals.orgshade tolerance (second axis for gymnosperms, third axis for angiosperms) differ between clades (Fig. 4B).These differences occur in two aspects: (1) different traits are correlated with shade tolerance (e.g., leaf arrangement in angiosperms vs. seed mass in gymnosperms), or (2) the same traits are correlated with shade tolerance, but in the opposite direction (e.g., maximum height increased with shade tolerance in gymnosperms but decreased in angiosperms).These clade-level differences in the trait versus shade tolerance associations explain the lack of significant trait correlations with shade tolerance when all species are analyzed together.

DISCUSSION
In this study we quantified major whole-plant spectra for morphological, anatomical and demographic traits for North American woody species and analyzed their relationship to whole-plant performance measures, i.e., growth rate and tolerance to shade, drought, waterlogging and fire (see Material and Methods for definition).The results of these analyses confirm that these holistic whole-plant performance measures reflect integrated processes of growth, reproduction and survival involving multiple traits (Reich et al. 2003, Violle et al. 2007, Webb et al. 2010).Furthermore, they support that the major whole-plant trait spectra reflect adaptations to key environmental drivers in temperate forests corroborating the prevalence of fundamental functional tradeoffs defining fundamental plant strategies (Smith and Huston 1989, Pacala et al. 1996, Poorter and Markesteijn 2008) but with substantial differences between gymnosperms and angiosperms.However, we also identified trait spectra which are not related to any of the performance measures used or which reflect clearly the differences between the major clades suggesting that there are also other factors (e.g., evolutionary history) explaining major trait variation.

Whole-plant trait spectra reflect fundamental strategies and differences between basal phylogenetic groups
When all 305 species are considered together, the difference between gymnosperms and angiosperms is captured by the first major PCoA axis.This strong phylogenetic signal essentially reflects the two dominant plant functional types that differ in leaf traits such as SLA, leaf type (evergreen needle-leaved versus deciduous broad-leaved), conduit type (tracheids versus vessels) and resprouting capacity.The strong correlation between SLA and the first major axis highlights its importance as a lineage separating trait in addition to reflecting ecological strategies (Diaz et al. 2004).We also found a weak but significant correlation between growth rate and the first major axis.Overall, this supports the notion that SLA could be used as a weak proxy for growth rate in adult trees (Wright et al. 2010).The correlation of water-logging and fire tolerance with the first axis supports the difference in functionality between the two major clades and indicates phylogenetic conservatism.
In contrast to the first axis, the traits associated with the second major axis are consistent for both clades and are positively correlated with drought tolerance and negatively with growth rate.In our study, the drought tolerant species have a lower growth rate and are characterized by a tree/shrub like growth form with high seed mass, high wood density and a taproot.High seed mass enables the rapid development of a taproot, which allows seedlings to escape dry surface soil conditions and enhances survival rates (Leishman and Westoby 1994).High wood density tends to be associated with low minimum leaf water potentials, deep rooting ability (Brodribb andFeild 2000, Bucci et al. 2004) and increases resistance to drought-induced xylem embolism (Hacke et al. 2001).Thus a complex spectrum of traits involving demographic traits (seed mass), anatomical (wood density) and morphological (taproot) is governing drought tolerance.The fact that the same set of traits governs drought tolerance in the otherwise contrasting gymnoand angiosperms emphasizes their ecological relevance.Our results also suggest that water availability is a key driver of tree growth in North American forests and that the growth rate is low in species adapted to drought.The high wood density associated with drought tolerance could be one indirect factor leading to reduced growth rate-a relationship frequently reported for tropical tree species (Muller-Landau 2004).
Fundamental relationships between drought, shade tolerance, and growth rate are reflected by whole-plant trait spectra within gymnosperms and angiosperms Shade and drought tolerance are inversely related, but the strength of this relationship differs between gymnosperms (strong correlation) and angiosperms (weak correlation) (Niinemets and Valladares 2006).Our results reveal that different trait spectra underlie the shade versus drought tolerance relationship for gymnosperms and angiosperms, reflecting different trait tradeoffs between these two major clades.
In the gymnosperms, the negative correlation between shade and drought tolerance was reflected by one trait spectrum (large maximum height and shallow roots in shade tolerant species versus small maximum height and taproots in drought tolerant species).This reflects a tradeoff between allocation to roots versus shoots (Fig. 2B).Traits reflecting growth rate are unrelated to the trait spectrum reflecting shade/ drought tolerance.This could reflect a true independence, or it may be masking a non-linear, hump-shaped relationship between growth rate and the drought-shade tolerance axis (Smith andHuston 1989, Craine 2009) because the linear methods used here are not suited to identify such non-linearity.However, the very nature of the trait spectrum reflecting high growth rates suggests independent strategies.For example, gymnosperms with high growth rate tend to be trees with large maximum heights and with thick bark that possess the ability to resprout, suggesting a relationship to disturbance strategies, which are expected to be independent of drought/shade tolerance strategies (Loehle 2000).
In the angiosperms, two independent trait spectra imply a difference between the three strategies with respect to shade, drought, and water-logging tolerance (Fig. 3C and D) suggesting different tradeoffs.Shade and drought intolerant angiosperms (e.g., species in Betula and Populus) are tolerant to water-logging and represent a resource use strategy suited to quickly exploit suitable habitats (tall stature, small wind dispersed seeds, high vegetative spread rates) at the cost of protection and maintenance structures (soft wood)-a typical pioneer trait association.In agreement with Smith and Huston (1989), these species tend to have higher growth rates compared to species that are tolerant to either shade or drought stress.Furthermore, our study agrees with Niinemets and Valladares (2006) in that the deciduous broad-leaved habit is a feature of shade and drought intolerance in North American forests, while evergreen broad-leaved habit tends to be a feature of species able to tolerate these stresses.Shade tolerant species (e.g., species of Acer) attain a relatively small stature, and light interception is enhanced by oppositely arranged leaves with a high SLA; these species also support shallow roots and produce toxic defense chemicals.High SLA is typical of winter-deciduous, shade tolerant trees growing in the understory (Lusk and Warton 2007), while the production of toxic defense chemicals might enhance their resistance to herbivores, making such species strong competitors (Kitajima 1994).Drought tolerant angiosperms (e.g., evergreen species of Quercus) exhibit a conservative resource use strategy with trait associations aligning with those reported by Markesteijn and Poorter (2009): slow nutrient turnover and long residence times (low SLA), high investment in protection and survival structures (dense wood, thick bark, heavy seeds), combined with features favored under low water availability (taproot and ring-porous wood).

Whole-plant trait spectra independent of growth rate, shade, and drought tolerance
In the angiosperm-only analysis, leaf composition and allelopathy co-vary (composite leaves paired with high potential for allelopathy versus simple leaves paired with low potential for allelopathy) and explain the second major axis.This axis is strongly determined by phylogeny as it separates species in the genera Fraxinus (Oleaceae), Carya, Juglans (Juglandaceae), and Aesculus (Sapindaceae) with composite leaves from those with single leaves (Fig. 3A and B); this axis is also independent of growth rate and drought or shade tolerance (Table 2).The fact that leaf composition is not related to shade tolerance has been observed in different deciduous woody floras (Stowe and Brown 1981, Niinemets 1998, Malhado et al. 2010).
Species with composite leaves tend to have low branching costs that allow rapid vertical growth during favorable light conditions, which is equally relevant for shade intolerant early successionals (Givnish 1978) and shade tolerant late successionals (Niinemets 1998).Stowe and Brown (1981) and Malhado et al. (2010) showed that leaf composition was related to climatic variables (e.g., spring and summer temperatures and variation in rainfall and water deficits), in such a way that seasonal drought favors composite leaves.Adaptations to episodic drought events are expected to differ from adaptations to persistent drought (Craine 2009), which may explain why we found leaf composition to be independent to trait associations reflecting high drought tolerance.High allelopathic interference more commonly occurs in stressful environments (Blanco 2007), such as under extremes in water and temperature and rapid successional changes.Thus, the co-variation of allelopathy and composite leaves seems to be an indirect relationship that merely emerges from phylogenetic relatedness.

Significance of the results in terms of data and methodological limitations
The results reported in this study could be influenced by the uncertainty caused by intraspecific variability, the choice of traits and by assignment errors in categorical traits and performance indices and the methods used.Our approach to filling-in missing species-level trait data with genus means had little influence on the results (data not shown).
The influence of intra-specific variability is likely to be negligible because we quantified tradeoffs based on 23 traits, with many traits describing morphological features (e.g., leaf arrangement) that are expected to be largely fixed for a given species; however, some traits (e.g., SLA) are likely to varying within a species (Ogle et al. 2012).However, Albert et al. (2010) found that the PCoA solution based on continuous traits exhibiting considerable intra-specific variability remains stable irrespective of whether an analysis was conducted at the species, population, or individual level.Furthermore, in our analysis, species are distributed across large environmental gradients, which is likely to lead to greater inter-species compared to intra-species trait variability (Kattge et al. 2011).
The incorporation of other potentially important traits, such as leaf area, mycorrhiza-associations, fine root diameter, serotiny or twig thickness might have the potential to detect novel tradeoffs (e.g., tradeoffs related to fire tolerance which we could not identify within the major clades), sharpen or slightly modify tradeoffs and strategies found.However, weak correlation between performance measures and the first three major axes might also depend on the resolution and information the indices are based on.Thus, fire tolerance and growth rate should rather be viewed as coarse approximations because they are primarily based on field observations, expert knowledge and estimates from the literature and not on precise measurements or experiments.However, drought tolerance and shade tolerance values used are reliable because they either are directly based on specific plant survival and site condition measurements or correlated well with species-specific values based on such measurements (see Material and Methods for details).Thus, the weak correlation of shade tolerance with the third axis analyzing the angiosperms might rather be caused by the complex interactions of functional traits.Moreover, we found a significant strong correlation of shade tolerance with the fifth PCoA axis (data not shown) suggesting that there are several different trait solutions for being shade tolerant (Valladares and Niinemets 2008).
Generally, our analysis is meant to be exploratory and aims to reveal the most important trait spectra in the first part and explores in the second part whether they reflect whole-plant performances describing fundamental ecological strategies or not.In this way it allows the exploration of novel trait spectra (e.g., the second axis in the angiosperm-only analysis) and hypothesis about underlying factors (e.g., adaptation to seasonal drought or phylogenetic constraints), which could be tested in a second step using appropriate designs (e.g., permutations and null-models).However, constrained analyses, e.g., distance-based Redundancy Analysis revealing the trait spectra which are best explained by the performance measures, yielded nearly identical results (analyses not shown), and thus underpins that adaptation to light and water availability are indeed important factors explaining major trait variation in North American woody species.

Phylogenetic signal
An explicit quantification of the phylogentic signal is challenging in this study because we imputed missing species-specific traits with genus means, which could artificially inflate the phylogenetic signal.The analysis involving the complete species pool suggests a trivial phylogenetic signal related to differences in trait strategies between gymnosperms and angiosperms.Thus, performing the separate analyses for these two major clades resulted in a coarse phylogenetic correction (Diaz et al. 2004).Comparison of these three analyses enabled us to identify important trait-performance relationships that differed between these two groups, representing potential adaptations that arose early in the evolution of these two major clades.Niinemets and Valladares (2006) found a significant phylogenetic signal in shade and drought tolerances, which pointed to trait conservatism operating between species within genera.Thus, the trait associations uncovered in this study are potentially not phylogenetically independent at, for example, the genus level.For instance, trait spectra related to drought and shade tolerance often grouped species by genera (Fig. 3); conversely, species are widely spread along the trait spectrum reflecting growth rate, and their position appears to be independent of their genus affiliation (Fig. 2).

Whole-plant trait spectra support plant strategies schemes
Comparing our whole-plant trait spectra with strategy axes of existing plant strategy schemes we found consensus but also insufficiency.For example, the widely used LHS scheme of Westoby (1998) was not sufficient to describe the main axes of trait variation of temperate woody species.The LHS scheme was moderately useful for understanding the trait spectra of angiosperms; for example, SLA (L), maximum tree height (H), and seed mass (S) were independent and contribute to complex spectra reflecting adaptation to shade and drought stress.However, among gymnosperms, SLA was irrelevant for describing inverse adaptations to drought and shade.The differential importance of SLA reflects the contrasting relationship between SLA and leaf life span for evergreen needle-leaved (little variation in SLA and large variation in leaf life span) versus deciduous broad-leaved species (large variation in SLA and little variation in life span, which reflects growing season length).The differential importance of maximum tree height with respect to plant performance indices might reflect different strategies that are controlled by different tradeoffs (Falster and Westoby 2005).In angiosperms, tall stature and fast growth correlate with a stress intolerant strategy, which might trade off with lower productivity (e.g., small stature and slow growth) when shade or drought tolerance increases.Conversely, in gymnosperms, the tall stature of shade tolerant species might be the result of competition for light.
Resprouting and leaf composition might represent additional dimensions reflecting responses to disturbance or periodic stresses that are not captured by the LHS scheme.Thus, the extension to four axes that includes resprouting capacity (Loehle 2000) seems justified for North American forests, but this may still be insufficient.The inclusion of additional traits related to adaptations dealing with disturbance and/or reflecting competitive strength would likely improve upon these existing schemata.

Conclusion
Our study shows that major whole-plant trait spectra of North American woody species are related to performance indices of growth and tolerance to shade, drought and water-logging that reflect whole-plant strategies with respect to growth, reproduction, and survival along environmental gradients in forests (light, water and disturbance).The whole-plant trait spectra related to performance measures are compound of morphological, anatomical and demographic traits and interact with each other, corroborating the assumption of fundamental functional tradeoffs between stress tolerances and growth.Ideal measures of plant performance would include direct observations of vital rates in response to environmental drivers, and a growing number of studies use repeated forest inventories to estimate these rates and relate them to local trait databases (Poorter et al. 2008, Martı ´nez Vilalta et al. 2010, Poorter et al. 2010, Wright et al. 2010).However, this has not been accomplished for a complete continental flora encompassing a wide range of long-lived species and contrasting environments.In this sense, our approach represents a macro-ecological complement to the growing field of the 'rates and traits' research.Moreover, it might be an appropriate way to associate the principles of fundamental conceptual strategy schemes (Grime 1977, Smith and Huston 1989, Grubb 1998, Craine 2009) with information on traits to refine current trait-based schemes and to identify underlying tradeoffs.

SUPPLEMENTAL MATERIAL APPENDIX A
Table A1.List of taxa names used for the analyses and accepted names after name-checking.Notes: All analyses and gap-filling (see Material and Methods) were based on the original list of taxa names for which all the traits were collected.In order to identify authority, synonyms and accepted names for the used taxa names we checked them against The Plant List (2010).After name-checking we found two taxa which were synonymous to two other taxa of the list.
Resolved accepted names are given with authority; for unresolved names the original name is in parentheses with no authority attached to it.
à Accepted names are taken from following sources: USDA, NRCS, National Plant Data Team ( 2007) and Burns and Honkala (1990).
Table A2.Description of traits used for ordination analyses with their main ecological function/proxy.Niinemets and Valladares (2006) and measures of humidity (annual P minus potential evapotranspiration [Willmott and Matsuura 2007]) derived from geographic distribution maps for 247 North American woody species (U.S. Geological Survey 1999) on 0.5 degree resolution.To obtain species-specific measures reflecting species' drought tolerance we used the lower limit (5th quantile) of the humidity measures covering a species range.We used quantiles instead of extreme values (i.e., minimum and maximum values) to minimize the effect of outliers caused by potential mismatches intersecting species range maps with climate.Documentation of the final trait scale for leaf margin, C:N ratio, life span, leaf type and standardization of SLA For leaf margin we extended the original scale (Adams et al. 2008) of 3 levels (1 ¼ entire, 0.5 ¼ toothed and/or entire, 0 ¼ toothed) to 4 levels (0 ¼ entire, 1 ¼ toothed and/or entire, 2 ¼ toothed, 3 ¼ lobed) to account for species with a pronounced lobed leaf margin.For the C:N ratio we changed the original class based scale (low , 23, medium ¼ 23-59, high .59) by taking class means instead of class borders (low ¼ 15, medium ¼ 40, high ¼ 65) to account for a realistic upper limit.For life span we combined classed based data (USDA, NRCS, National Plant Data Team 2007) and continuous data (Wirth and Lichstein 2009) as follows: (1) in cases of multiple entries per species continuous data were given priority and (2) first class (short , 100 years), second class (moderate ¼ 100-250 years) and third class (long .250 years) were converted to 80, 175 and 300 years, respectively.The two traits leaf type with 3 levels (needle-leaved, scale-like, broadleaved) and leaf deciduousness with 3 levels (evergreen, deciduous, evergreen/deciduous) were combined to one nominal trait with 6 levels (called leaf type) to reduce the strong separating effects of the traits with a low number of levels.SLA was a standardized species specific estimate based on a comprehensive meta-analysis for North America which accounts for phylogeny and intra-specific variability (Ogle et al. 2012).

Fig. 1 .
Fig. 1.PCoA ordination plot showing distances among 305 North American woody species based on 23 traits for the first two axes (A) with a histogram (a) showing the first 30 eigenvalues.In (B) significant correlations (p , 0.001) of both traits and performance measures (capitals) with the first two PCoA axes are represented as arrows (see Table1for abbreviations and Table2for respective correlation coefficients); the lengths of the arrows are proportional to their correlation coefficient, and they point in the direction of most rapid change; nominal traits were dummy coded before correlation.

Fig. 2 .
Fig.2.PCoA ordination plot showing distances among 103 North American woody gymnosperm species based on19 traits for the first two axes (A) and for the first and third axis (C) with histograms (a and c) of the first 30 eigenvalues, respectively.The dispersion of important genera (containing many species or largely explain the axes) are shown as ellipses using standard deviation of the point scores with a confidence limit of 0.7 while the lines connect the species to the genus centroid, respectively.In (B) and (D) significant correlations (p , 0.01) of both traits and performances (capitals) with the respective PCoA axes are represented as arrows, see Fig.1for detailed description.

Fig. 3 .
Fig. 3. PCoA ordination plot showing distances among 202 North American woody angiosperm species based on 22 traits for the first two axes (A) and for the first and third axis (C) with histograms of the first 30 eigenvalues (a and c), respectively.The dispersion of important genera are shown as ellipses, see Fig. 2 for detailed description.In (B) and (D) significant correlations (p , 0.01) of both traits and performance measures (capitals) with the respective PCoA axes are represented as arrows, see Fig. 1 for detailed description.

Fig. 4 .
Fig. 4. Comparison of trait spectra reflecting drought (A) and shade tolerance (B) between gymnosperms and angiosperms.Pearson's correlation (r) between traits and those PCoA axes that correlated best with drought tolerance ¼ ''drought axes'' (A) and best with shade tolerance ¼ ''shade axes'' (B) in gymnosperm and angiosperm only analysis, respectively (see Table2for correlation coefficients and explained variance).If the respective trait spectra of the two clades are similar (i.e., the same traits vary in the same manner) the Pearsons's r values show a linear arrangement along the 1 to 1 line.

Fig. A2 .
Fig.A2.Relationship between drought tolerance scores ofNiinemets and Valladares (2006) and measures of humidity (annual P minus potential evapotranspiration[Willmott and Matsuura 2007]) derived from geographic distribution maps for 247 North American woody species (U.S. Geological Survey 1999) on 0.5 degree resolution.To obtain species-specific measures reflecting species' drought tolerance we used the lower limit (5th quantile) of the humidity measures covering a species range.We used quantiles instead of extreme values (i.e., minimum and maximum values) to minimize the effect of outliers caused by potential mismatches intersecting species range maps with climate.

Table 1 .
Traits with their respective units or categorical levels and performance measures compiled including their main sources.

Table 2 .
Pearson's correlation coefficient (r)between traits and performance indices versus the first three PCoA axes (A1, A2, A3) obtained by analyzing all species, only gymnosperms and only angiosperms.

Table A3 .
Traits and performances filled for 305 species, 103 gymnosperms and 202 angiosperms, respectively before and after imputing missing values.Note: In case of no data entry the trait did not occur or did not vary in the specific group.