Global gradients in intraspecific variation in vegetative and floral traits are partially associated with climate and species richness

Aim: Intraspecific trait variation (ITV) within natural plant communities can be large, influencing local ecological processes and dynamics. Here, we shed light on how ITV in vegetative and floral traits responds to large-scale abiotic and biotic gradients (i.e., climate and species richness). Specifically, we tested whether associations of ITV with temperature, precipitation and species richness were consistent with any of four hypotheses relating to stress tolerance and competition. Furthermore, we estimated the degree of correlation between ITV in vegetative and floral traits and how they vary along the gradients.


| INTRODUC TI ON
Knowledge of plant functional traits has advanced our ability to understand and predict species coexistence, community assembly and plant responses to environmental factors (Díaz et al., 2016;Keddy, 1992;Shipley, 2009;Weiher & Keddy, 1999;Westoby, 1999). This progress has been mostly built on approaches using mean trait values per species, without considering trait variability within species (Funk et al., 2017;Shipley et al., 2016;Violle et al., 2012). Accounting for intraspecific trait variation (ITV) has the potential to foster the understanding of ecological processes and dynamics (e.g., Albert et al., 2010;Andrade et al., 2014;Bolnick et al., 2003Bolnick et al., , 2011Carlucci, Debastiani, Pillar, & Duarte, 2015;Jung et al., 2014;Kuppler, Höfers, Wiesmann, & Junker, 2016;Spasojevic, Turner, & Myers, 2016) and is considered an important step for achieving a higher generality and predictability in community ecology (Shipley et al., 2016). At a global scale, we have a good overview of trait variation among plant species along gradients, but not within species. Kattge et al. (2011) explored intraspecific variation in species across different locations, and in a global meta-analysis Siefert et al. (2015) focused on the relative extent of ITV compared with interspecific variation at the community level. Here, we build on these findings by focusing on the absolute extent of ITV within species/populations and its global variation along biotic and abiotic gradients.
The consequences of ITV are multiple. It provides the basis for natural selection and evolution (Liu et al., 2019), it is linked to responses to environmental change (Bergholz et al., 2017;Ridley, 2003), and it boosts above-and below-ground animal diversity (Barbour et al., 2019). Intraspecific trait variation arises from a combination of genetic variation, developmental instability (i.e., the inability of an individual to produce a specific phenotype in given environmental conditions) and phenotypic plasticity owing to environmental change across time, including their interaction (Albert, Grassein, Schurr, Vieilledent, & Violle, 2011;Bradshaw, 1965;Stearns, 1989;Willmore & Hallgrimsson, 2005) and is affected by abiotic and biotic factors such as climate and species interactions (Hart, Schreiber, Levine, & Coulson, 2016;Valladares, Gianoli, & Gómez, 2007). Predicting the responses of the absolute degree of ITV to these factors is a major challenge (Barbour et al., 2019;Bergholz et al., 2017;Kumordzi et al., 2019). However, owing to the importance of ITV for the capability of plants to coexist in species-rich communities and to adapt to new climatic conditions (Banitz, 2019;Hart et al., 2016;Junker, Lechleitner, Kuppler, & Ohler, 2019), the description of global patterns in the distribution of absolute ITV is highly relevant.
To explain the relationship between ITV and climate, two opposing hypotheses have been proposed. The stress-reduced variability hypothesis states that ITV decreases with extreme abiotic conditions that generate stress (Janzen, 1967;Klopfer & MacArthur, 1961). Extreme abiotic conditions have the potential to act as an environmental filter and/or strong selective agent, causing trait convergence within species and thus reducing ITV by decreasing phenotypic and genetic variation (Caruso et al., 2017;Hulshof et al., 2013;Valladares et al., 2007Valladares et al., , 2014. In contrast, the stress-induced variability hypothesis (Helsen et al., 2017;Janzen, 1967;Klopfer & MacArthur, 1961) posits that abiotic stress increases ITV. In stressful conditions, phenotypic and genetic variation may increase owing to developmental instability and higher rates of recombination and mutation, in addition to competition avoidance when resources become less abundant (Hoffmann & Merilä, 1999;Valladares et al., 2014).
Here, we use the term "stressful conditions" to refer to environmental conditions that limit the ability of plants to convert energy into biomass, such as cold or aridity (Grime, 1977;Maestre, Callaway, Valladares, & Lortie, 2009). Studies focusing on single/few species have found species-specific relationships between ITV and climate (Albert et al., 2010;Helsen et al., 2017;Niinemets, Keenan, & Hallik, 2015), but we are still lacking the broad picture of the relationship between ITV and stress gradients. Similar opposing hypotheses have been proposed for relationships between ITV and species richness. Increasing species richness might increase interspecific competition for resources, and as a consequence, ITV may be reduced to relax it (Bastias et al., 2017;Pauw, 2013;Violle et al., 2012), whereas in areas with low species richness and dominating intraspecific competition, members of a species may occupy a larger trait space (Freschet, Bellingham, Lyver, Bonner, & Wardle, 2013;Kumordzi et al., 2019;Silvertown, 2004). This is congruent with the niche packing hypothesis, which states that an increasing number of species leads to stronger interspecific competition and increased niche density (MacArthur & Levins, 1967;Ricklefs & O'Rourke, 1975). Thus, in species-rich communities with high niche density, ITV should decrease (MacArthur & Levins, 1967;Violle et al., 2012). A contrasting hypothesis states that ITV may instead increase with species richness (Le Bagousse-Pinguet, Bello, Vandewalle, Leps, & Sykes, 2014;Clark, 2010). In favour of this hypothesis, it has been demonstrated that ITV increases with increasing species richness (Le Bagousse-Pinguet et al., 2014;increased variation hypothesis). This has been suggested to allow plants to avoid inter-and intraspecific competition by occupying a larger niche (Clark, 2010), which may lead to increasing ITV (Helsen et al., 2017;Le Bagousse-Pinguet et al., 2014). Studies published so far support either one or neither of these hypotheses (Bastias et al., 2017;Helsen et al., 2017;Kumordzi et al., 2015;Le Bagousse-Pinguet et al., 2014;Siefert et al., 2015).
Given that most studies of ITV-climate or ITV-species richness relationships have focused on a limited number of species and/or K E Y W O R D S community ecology, flower trait, functional diversity, functional trait, leaf trait, macroecology, precipitation gradient, temperature gradient, within-species variation geographical area (but see study by Siefert et al., 2015, who high-lighted global patterns in the proportion of the functional diversity of a community allocable to ITV), we lack a broad perspective on patterns of the absolute extent of ITV across large abiotic and biotic gradients, hindering general tests of the competing hypotheses regarding the effects of both stress and competition.
The aim of our study was to evaluate the relationship between the absolute extent of within-population ITV in vegetative and floral traits and abiotic/biotic gradients. Therefore, we collected geo-referenced data on ITV [coefficients of variation (CVs) of one species at one given location] of vegetative and floral traits for herbaceous and woody species from databases and from published and unpublished studies. We analysed herbaceous and woody species separately, because they represent two clearly distinct groups in the global spectrum of plant form and function (Díaz et al., 2016). These two groups can differ in their trait-trait and trait-climate correlations, which makes it necessary to investigate the groups separately along largescale gradients (Šímová et al., 2018). Trait data were combined with climatic data and regional native-species richness extracted from global models to explore three aspects of ITV. The first aspect was the relationship between variation in ITV and global heterogeneity in climate and species richness. Figure 1 shows expected patterns that would support the stress-reduced, stress-induced, niche packing and increased variation hypotheses. We expected ITV in plant traits known to respond to environmental stress (e.g., many vegetative traits; Fonseca, Overton, Collins, & Westoby, 2000;Grime, 1977;Pierce, Brusa, Vagge, & Cerabolini, 2013) to show stronger associations with climate than others, such as floral traits, that are thought to be driven by pollinator-mediated selection (Caruso, Eisen, Martin, & Sletvold, 2019). Second, we assessed the species specificity of the relationships between ITV and climate and species richness. Third, we assessed the across-trait correlation of ITV. Specifically, we hypothesized that ITV in floral and vegetative traits is correlated within but not across these two trait groups (Armbruster, Stilio, Tuxill, Flores, & Velásquez Runk, 1999;Berg, 1960), because different organs experience specific types of selection pressures related to their function; for example, resource uptake or reproduction (Junker & Larue-Kontić, 2018;Karban, 2015;Pélabon, Armbruster, & Hansen, 2011;Pélabon, Osler, Diekmann, & Graae, 2013).

F I G U R E 1
Predictions based on the four hypotheses regarding the relationships between intraspecific trait variation (ITV) and (a, c) large-scale climate gradients and (b, d) species richness gradients. (a) The stress-reduced hypothesis is supported if ITV shows an unimodal distribution along a climatic gradient (i.e., mean annual temperature and precipitation), with lowest ITV at the extremes (cold/hot, dry/wet) of the climate variable distribution that indicate high abiotic stress, or if ITV shows a linear relationship (here shown for high stress at the lower end of the climatic gradient, i.e., cold or dry climate) with opposing responses of ITV at the gradient extremes. (c) The stress-induced hypothesis is supported if ITV peaks at the extremes of the climate variable distribution (i.e., high abiotic stress) and is lowest in moderate climates (i.e., low abiotic stress) or if ITV shows a linear relationship (here shown for high stress at the lower end of the climatic gradient, i.e., cold or dry climate) with opposing responses of ITV at the gradient extremes. Both hypotheses may vary regarding the nature or length of the gradient studied. (b) For the increased variation hypothesis, a linear relationship between ITV and species richness is expected, with the highest ITV at highest species richness and lowest ITV at lowest richness. (d) For the niche packing hypothesis, ITV should show a linear relationship with species richness, with the lowest ITV at highest species richness and highest ITV at the lowest species richness intraspecific trait variation  For each species-location combination, we randomly drew five individuals (which was the lowest number in the dataset) from all individuals sampled for this species-location combination 1,000 times and calculated the CV for each drawing. The average of all drawings was used as the CV in subsequent analysis (Bastias et al., 2017).
Estimates of the rarefaction analysis are unbiased, because species ranks for the rarefied CVs were the same as for non-rarefied CV values in the complete dataset and in a subset containing only raw CV values with n > 9 (Supporting Information Appendices S5.2 and 5.3). Additionally, we explored effects of the number of individuals sampled per species per location on the non-rarefied CV. We used two different approaches: a resampling approach and a visual (i.e.,

| Relationship between ITV and large-scale climate and species richness gradients
To test the relationship between ITV and climatic factors, we cal- the linear and quadratic terms should be included (Crawley, 2009).
The determination coefficient for the final model was calculated as R 2 marginal , which is the relative contribution of all fixed factors using the rsquared function in the piecewiseSEM package (Lefcheck, 2016).

| Species-specific associations
To examine species-specific associations of ITV with climate or with species richness, we ran linear mixed-effect models, with ITV species/ location as the response variable, with MAT, MAP or species richness as fixed explanatory variables (including linear and quadratic terms), and with species identity as a random effect, allowing for a random intercept and random linear and quadratic (x + x 2 ) slope using the with a reduced model without random effects using the likelihood ratio test. Furthermore, the coefficients of determination for fixed (R 2 marginal ) and fixed and random effects (R 2 conditional ) were calculated using the r.squaredGLMM function (MuMIn package; Barton, 2018).

| Covariation in ITV among traits
To explore whether ITV covaries across traits (vegetative and floral), we first calculated the mean CV of all ITV species/locations values for each plant species (i.e., if one species was sampled at multiple locations) separately for all traits. For most species, vegetative and floral traits were measured at different locations or the sample size for measurements at the same location was small. Therefore, we calculated the mean CV for each species across locations to increase the number of traits that could be included; inflorescence diameter was excluded owing to its small sample size (n = 65). Afterwards, to identify gradients in the covariation patterns, we performed a principal components analysis (PCA) using the dudi.pca function (ade4 package; Dray & Dufour, 2007). Given that PCA requires a complete dataset with no missing values, missing values were imputed using the joint modelling approach implemented in the Amelia function (Amelia package ;Honaker, King, & Blackwell, 2011). This approach provided good estimates for missing values in datasets similar to ours (Dray & Josse, 2015). Additionally, for species with measurements of different traits at the same location, the Pearson's r was calculated for CVs of each trait combination (e.g., LDMC-flower height, LDMC-SLA or flower height-flower diameter) without calculating the mean CV first.

| RE SULTS
We found that the coefficient of variation of plant traits (ITV species/location ) varied across two to three orders of magnitude.
Although ITV varied among traits, there were no consistent differences in ITV between herbaceous and woody species (Figure 3; Supporting Information Appendix S9).

| Relationship between ITV and large-scale climate and species richness gradients
The ITV location (i.e., mean rarefied CV of one location) of single The strength of the significant correlations between species richness and ITV location were in the same range as correlations between MAT/MAP and ITV location : mean ± SD R 2 Species richness = .07 ± .16, R 2 MAT = .05 ± .08 and R 2 MAP = .03 ± .06. Associations of species richness with ITV location were negative for SLA in both groups.
Additionally, ITV location in nectar tube depth in herbaceous species was positively correlated with species richness, whereas in woody species ITV location in leaf thickness showed a concave relationship with a peak at intermediate species richness (Figure 4). We did not F I G U R E 3 Violin plot showing the absolute extent of intraspecific trait variation (ITV) of herbaceous (grey) and woody (white) plant species measured as the coefficient of variation (CV). In total, we included 18,401 measurements of 6,768 species-location combinations (herbaceous = 3,035; woody = 3,733) across all traits. The area of the violin represents the density of points at this CV value. The vertical dots denote the mean. Margins of vegetative traits are grey, and floral traits are black. The y axis is logarithmically scaled. Abbreviations: leaf C/N, ratio of leaf carbon to nitrogen content; leaf C/N/P, leaf carbon/nitrogen/ phosphorus content; LDMC, leaf dry matter content; SLA dry, specific leaf area (dry matter content) find any associations between ITV location in leaf chemical traits and species richness.

| Species-specific associations
At the species level, ITV species/location was more strongly explained by differences between species than MAT/MAP/species richness

| Covariation in ITV among traits
The PCA revealed several gradients of among-species trait covariation ( Figure 6). The first PCA axis reflected a gradient from high to low ITV in morphological and chemical leaf traits and nectar tube width. The second axis reflected mostly variation in floral traits (but also in plant height). For the pairwise correlations between ITV species/location of different traits, we also found no correlations between vegetative and floral ITV, except that flower height and stamen length increased with plant height and flower diameter with leaf area (Supporting Information Appendix S10). In addition, covariation F I G U R E 4 Correlation between intraspecific trait variation (ITV location ) and climate and species richness. The R 2 marginal values (quadratic mixed models) and schematic visualization of the fitted relationships are given for each trait and separated between herbaceous (dashed lines) and woody (dotted lines) species. Grey squares indicate that no model was fitted either because of an insufficient number of locations or because of a highly skewed distribution of locations. Black asterisks denote p < .05 and grey asterisks p = .05-.08 for mixed models. The yellow-red gradient in the left panel represents R 2 marginal values. The background colour gradients in graphs with fitted relationships show sampled gradients for MAT, MAP or species richness. The variable length of each gradient is attributable to a different combination of sample locations for each trait. Coefficients of fitted mixed models are shown in the Supporting Information (Appendices S11 and S12). Abbreviation: LDMC, leaf dry matter content; leaf C, leaf carbon content; leaf C/N, leaf ratio carbon:nitrogen; leaf N, leaf nitrogen content; leaf P, leaf phosphorus content; MAT/MAP, mean annual temperature/precipitation (Karger et al., 2017a(Karger et al., , 2017b; SLA dry, specific leaf area (dry mass); Species richness, native regional plant species richness extracted from Ellis et al. (2012) [Colour figure can be viewed at wileyonlinelibrary.com] between vegetative traits was more prominent than covariation between floral traits (Supporting Information Appendix S10).

| D ISCUSS I ON
Our findings show that ITV in certain plant traits is associated with large-scale environmental and biotic factors, which might reflect how plants cope with stressful abiotic and biotic conditions. We could show that the absolute extent of ITV in several vegetative and floral traits was associated, depending on growth form, with largescale gradients of temperature, precipitation and/or species richness, with a strong species-specific component. We found equally strong relationships between ITV and climate and species richness in both woody and herbaceous species, and in both vegetative and floral traits. Relationships were mostly present in traits with wellknown responses to climate, such as SLA or LDMC (e.g., Jung et al., 2014;Wright et al., 2004), and for traits related to competition, such as plant height or SLA (Kunstler et al., 2016). Below, we discuss our findings in the context of ecological importance, such as plant stress response, of ITV and implications for trait-based research.

| Relationship between ITV and large-scale climate gradients
Depending on the trait, growth form (woody/herbaceous) and climatic factor, our results supported the stress-reduced variability hypothesis, the stress-induced variability hypothesis or neither of the two. For leaf morphological traits and both growth forms, ITV location in LDMC was decreasing with decreasing temperature (minimum MAT −4 °C), which is consistent with the stress-reduced variability hypothesis. This means that in cold climates, LDMC values are both smaller and less variable, which might optimize leaf lifespan, photosynthetic rate and leaf temperature (Michaletz et al., 2016) and might result in a small range of possible optimized phenotypes, which F I G U R E 5 Species-specific responses between intraspecific trait variation (ITV) and climate and species richness separated between herbaceous (grey) and woody (black) species. Each graph shows the fitted random intercept and slope for each species from linear mixedeffect models. Each model contained ITV species as a dependent variable, the linear and quadratic term of climate variables or species richness as a fixed factor, and species as a random factor, allowing for a random intercept and random quadratic slope (for details, see Materials and Methods). Analyses were conducted for only a subset of traits with multiple locations per species. Asterisks indicate the significance of the random effect. These are exemplary results for a subset of traits; for all traits and full results of the linear mixed-effect models, see the Supporting Information (Appendix S8). The y axis is log 10 (x + 1) scaled. Abbreviations: LDMC, leaf dry matter content; SLA, specific leaf area (dry mass). **p < .01, ***p < .001, ns = non-significant. The stress-induced variability hypothesis was supported by high ITV location at low mean annual precipitation (MAP; i.e., water stress) for LDMC and SLA in woody species, and for SLA, LDMC and leaf C in herbaceous species. This is in agreement with previous studies focusing on fewer species, which found induced variability in SLA at low levels of precipitation (Helsen et al., 2017), although opposing patterns have also been reported (Lemke et al., 2015). Several mechanisms might potentially explain the increasing ITV. First, it can result from increased genetic variation in stressful conditions (Hoffmann & Merilä, 1999;Huang, Zhao, Zhao, Li, & Pan, 2016). Second, increased ITV location might be attributable to reduced canalization in development (Valladares et al., 2014) and thus increasing development instability (Hoffmann & Woods, 2001;Pertoldi, Kristensen, Andersen, & Loeschcke, 2006;Polak, 2003). Third, increased ITV location might result from local variation in microclimatic conditions, because water availability (which is associated with MAP) can be proportionally more variable across microsites when precipitation is low, leading to greater plasticity or, in certain conditions, local genetic differentiation (Gianoli, 2004;Hodge, 2006). Consistent with previous studies (Anderegg, 2015;Jung et al., 2014;Liancourt et al., 2015), our results suggest that greater ITV in plant populations experiencing low precipitation might increase plant-population persistence.
In general, we found that the ITV location of traits relevant to stress responses was correlated with specific climate conditions as if ITV is correlated with the evolutionary potential of species (Liu et al., 2019;Ridley, 2003). This suggests that community responses inferred from the relative extent of ITV might be modified, potentially in an opposing direction, by the absolute extent.

| Relationship between ITV and large-scale species-richness gradients
Associations between ITV and species richness were trait specific.
However, for most traits, no relationship was found at the location level. This agrees with previous studies that found ITV to be relatively invariant along species richness gradients (Bastias et al., 2017;Siefert et al., 2015). For woody and herbaceous species, a negative correlation between ITV in SLA and species richness can be viewed as support for the niche packing hypothesis. However, for ITV in leaf thickness (in woody species) the relationship with species richness was quadratic (i.e., low ITV at both ends of the gradient), which fits neither the niche packing hypothesis nor the increased variation hypothesis (Bastias et al., 2017;Clark, 2010;Violle et al., 2012).
Given that both hypotheses focus on community species richness, the absence of clear effects might be explained by the use of hypothetical/modelled regional species richness instead of local species richness at each location. However, the quadratic relationship could also be a result of the two hypotheses not being mutually exclusive.
In areas with low species richness, ITV might become large, because species experience less interspecific competition and can potentially inhabit a broader range of microhabitats, including suboptimal conditions. At high richness, ITV might also be high because the effect of avoiding inter-and intraspecific competition is stronger than the constraints imposed by available microhabitats (Clark, 2010).

| Species-specific associations
Across species, most ITV-climate associations were idiosyncratic, not showing consistent support for any of the proposed hypotheses. This is consistent with previous studies that have highlighted both the idiosyncratic nature of species responses to environmental variation and strong discrepancies between general patterns of trait variation along gradients among and within species (Ackerly, Knight, Weiss, Barton, & Starmer, 2002;Cornwell & Ackerly, 2009;Körner, 2003). However, this was not true for all traits. For example, species-specific relationships between leaf area and MAT resembled the interspecific decreasing relationship (smaller CV at higher MAT; Albert et al., 2010;Körner, 2003 (Banitz, 2019;Hart et al., 2016). Thus, depending on its structure, ITV might affect ecological processes across larger spatial scales (Armbruster & Schwaegerle, 1996) despite the increasing importance of species turnover compared with ITV (Albert et al., 2010;Siefert et al., 2015). However, it is also important to keep in mind that we might miss important structuring variables or that structure cannot be seen in a single-trait approach.  (Armbruster, 2017;Fenster et al., 2004;Gómez et al., 2008;Junker et al., 2013;Kuppler et al., 2016;Waser, Chittka, Price, Williams, & Ollerton, 1996). For specialized species, floral traits should be less variable within species and largely independent of the environment, whereas in generalists traits may be more sensitive to variation in environmental and climatic conditions, in a similar manner to vegetative traits (Armbruster et al., 1999;Galen, 2000;Junker et al., 2017).

| Floral ITV and among-traits covariation in ITV
Vegetative and floral ITV mostly separated out along the first two PCA axes, indicating that covariation between vegetative and floral ITV was weaker than covariation within each trait group (see also Kuppler et al., 2016). Also, covariation was stronger in vegetative than in floral ITV. However, somewhat surprisingly (Berg, 1960), ITV in nectar tube width was correlated with ITV in leaf traits, and ITV in plant height was correlated with ITV in several floral traits.
These observations suggest that covariation in vegetative and floral ITV can depend on function and developmental origin of those traits (Armbruster et al., 1999;Armbruster & Wege, 2019). In general, our results are consistent with previous studies that found limited support that multiple traits can be highly variable simultaneously (Ames, Anderson, & Wright, 2016;Wright, Ames, & Mitchell, 2016). This opens the question of how among-trait covariation in ITV might limit phenotypic expressions of plants in variable biotic and environmental conditions and how this affects the potential adaptation of plants to changes in these conditions (Dwyer & Laughlin, 2017;Westoby & Wright, 2006).

| Caveats
Despite our large dataset on vegetative and floral ITV, there were some constraints limiting the generality of our results. Although, to our knowledge, this is the first study to include floral ITV across large spatial scales, the number of floral traits in the dataset is still limited, which might induce a sampling bias and limit the comparability among traits. Geographically, the availability of floral trait data was largely restricted to Europe and to Central and North America, and vegetative trait data were underrepresented in some regions (e.g., Africa and Asia). Additionally, in most locations, different numbers of species were sampled, and sampling was often incomplete; this decreased the precision of the ITV estimate at the location level. Thus, there are trade-offs between sampling more individuals per species, more species at one location or at more locations. Thus, differences in sampling strategies might change the relative contribution of ITV to the overall trait variability (Albert, 2015). Additionally, the precision of ITV might vary with the number of individuals sampled per species and per location, potentially resulting under-or overestimation of ITV. This random variation can induce noise in subsequent analysis, masking patterns of interest, while a systematic ITV increase or decrease with sample size may also induce error. However, if within-species and withinlocation variation in ITV is smaller than differences between species or along gradients, errors in large-scale patterns should be minimal. In our full dataset, we did not detect systematic variation with commonly used sample sizes to estimate intraspecific variation, and there was no indication that CVs of different samples sizes within species were larger than across species (Supporting Information Appendix S4). Therefore, the unavoidable variation in such large datasets might have induced noise in our analysis that masked patterns or impacted the strength of the detected relationships (despite our use of the rarefied CV, which may affect absolute values but should not affect relative changes or generated patterns that did not exist), making our analyses conservative. Lastly, as discussed in other studies (Albert et al., 2011;Siefert et al., 2015), the use of standardized sampling protocols for plant traits (Pérez-Harguindeguy et al., 2013) is likely to affect ITV (Borgy et al., 2017), which, together with the above-mentioned point, makes sampling of ITV complex and comparisons not straightforward.

| CON CLUS ION
In summary, the associations of ITV with large-scale climate and species-richness gradients were strongest for traits related to plant stress and competition, whereas other traits mostly varied independently of these gradients. Depending on the traits considered, measurements of ITV either increased or decreased with climatic stress and species richness, suggesting that both factors, across a range of spatial scales, can constrain or enhance intraspecific variation in specific plant traits (e.g., Auger & Shipley, 2013). This might, in turn, help plant populations to cope with stressful conditions (e.g., Jung et al., 2014).
Associations between climate and ITV differed between species, indicating that general patterns might not be present. Thus, when exploring plant responses to stressful conditions and environmental change across spatial and biological scales, a consideration of ITV can improve, but also impede, our understanding of how plants cope with such conditions.

ACK N OWLED G M ENTS
The study would not have been possible without the work of the Additionally, we thank Franziska Schrodt and anonymous reviewers for constructive and thoughtful suggestions on earlier versions of this paper. Open Access funding was provided by Ulm University under the DEAL-agreement.

DATA ACC E S S I B I L I T Y
The data were extracted from openly available sources in the TRY and BIEN databases at www.try-db.org and http://bien.nceas.ucsb. edu/bien/ under the reference numbers given in the Supporting