The raw data were either obtained from tables or figures (extracted by using the GetData Graph Digitizer, version 2.24)¹, and a list of all contributing papers were included in Supplementary 2 (Supplementary Material). For each publication, we recorded the following data: site location (longitude and latitude) as well as herbaceous plant LMA, leaf N concentration per mass (N_mass), and leaf P concentration per mass (P_mass) as three key leaf functional traits associated with the leaf traits and growing strategies (Diaz et al., 2016). Finally, our database consists of 1,653 observations of 542 species of herbaceous plant from 316 sampling sites across China [shown in Supplementary 3 (Supplementary Material)], with longitude ranging from 74.93°E to 123.92°E, and latitude ranging from 19.10°N to 50.19°N (Figure 1). According to the dry and wet climate-region division in China² (Thornthwaite, 1948), our sampling sites are distributed in four climatic regions: arid, semi-arid, semi-humid, and humid regions in China (Figure 1).

The climate (precipitation, temperature, evapotranspiration, and solar radiation) data were derived from the database of WorldClime 2.0 Beta version 1 (June 2016), as downloaded from http://worldclim.org. They represent average monthly climate data for 1970–2000. The soil attributes [soil organic matter (SOM); total N (STN) and available N (SAN); and total P (STP)] were extracted from a Chinese dataset of soil properties, which was constructed based on a 1:1,000,000 soil map of China and 8,595 soil profiles (Shangguan et al., 2013). Due to the constraint based on data availability and accuracy, there are many measurements of leaf traits at the same soil property level especially for STP. To explore the effects of climatic factors on plant functional traits, we used the mean values from May to September, the growing season, for temperate zone where our sites are mostly located, including the mean growing season precipitation (GSP), mean growing season temperature (GST), mean growing season evapotranspiration (GSE), and mean growing season solar radiation (GSR). Abbreviations of these variables are shown in Table 1. The edaphic and climatic information of the sampling sites were extracted according to geographical location (latitude and longitude) through ArcGIS 10.1 (ESRI, Inc., Redlands, CA, United States). The frequency distributions of the leaf traits (LMA, N_mass, and P_mass) were used to ensure the normality of the distribution (Zhou et al., 2019).

The benefit for a single objective leaf trait is defined as the relative deviation from the mean of a given observation. The relative benefit for the objective leaf trait (LMA, N_mass, or P_mass) indicates the relative magnitude of the values of these leaf traits, and is calculated as follows (Bradford and Damato, 2012; Sun et al., 2018):

where i is a certain leaf trait (LMA, N_mass, or P_mass), B_i is the magnitude of the relative benefit for objective leaf trait i, x_i, x_min, and x_max are the observed value of i, minimum, and maximum values for all measured i, respectively. A three-dimensional (3D) relative benefit method was used to describe the leaf traits (relative benefits among LMA, N_mass, or P_mass) (Li et al., 2019). As shown in Supplementary Figure 1, the vertical distance (l) from the 3D point to the 1:1:1 line represents the magnitude of benefit for the three objective leaf traits, forming 3D coordinates in the LMA-N_mass-P_mass economic spectrum. The relative benefit was decomposed into the LMA (a), N_mass (c), and P_mass (b) directions, whose values could reflect a plant’s growing strategy (Supplementary Figure 1): a low relative benefit of LMA combined with high relative benefits of N_mass and P_mass implied a quick investment-return growth strategy; in contrast, a slow investment-return growth strategy was indicated by a high relative benefit of the LMA combined with low relative benefits of N_mass and P_mass.

A principal component analysis (PCA) was performed with all the potential explanatory variables pooled together, to determine the explanatory powers of variance in the leaf functional traits, soil properties, and climatic factors among different climatic regions. The packages of “FactoMineR,” “factoextra,” and “corrplot” in R software (R Core Team, 2016) were used for the PCA. A one-way analysis of variance was performed with SPSS 19.0 software (SPSS Inc., Chicago, IL, United States) to detect differences in the plant, soil, and climate variables among the four climatic regions, using the least significant difference (LSD) as post hoc analysis. Regression and correlation analyses were conducted with R Version 3.3.2 (R Core Team, 2016) and SigmaPlot 14.0 software (Systat Software, Inc., Chicago, IL, United States) in each climatic region, to explore the correlations of leaf traits with edaphic and climatic variables. Besides, the correlogram of intercorrelations among relative benefits of leaf traits, soil properties, and climate factors was graphed by “mgcv” package in software R (R Core Team, 2016).

The LMA in the arid region, with a mean value of 124.34 g m^–2, was significantly higher than the mean values of 97.64, 81.90, and 79.73 g m^–2 for the semi-arid, semi-humid, and humid regions, respectively (P < 0.05, Figure 2A). N_mass and P_mass tended to initially increase and then decrease from arid to humid regions. The average values of N_mass were 23.51, 24.82, 25.92, and 16.64 mg g^–1 for the arid, semi-arid, semi-humid, and humid regions, respectively (Figure 2B). P_mass was significantly different among the four climatic regions (P < 0.05), with mean values of 1.52, 1.70, 1.86, and 1.27 mg g^–1 for the arid, semi-arid, semi-humid, and humid regions, respectively (Figure 2C). The LMA was negatively associated with N_mass and P_mass, with the deepest slope in the arid region (Figures 2D,E), whereas N_mass was positively correlated with P_mass (Figure 2F).

The PCA results indicated that two components explained 57.4% of the total variance (Supplementary Figure 2). Specifically, the first principal component (PC1) was highly relevant to the GSP, GSE, GSR, N_mass, and P_mass; the second principal component (PC2) was highly relevant to the SOM, STN, SAN, SAP, and LMA (Supplementary Figure 2). For the soil properties, SOM, STN, STP, and SAN presented a unimodal pattern along climatic regions, and peaked in the semi-humid region, with average values of 4.77, 0.23, 0.08, and 173.65 mg kg^–1, respectively (Supplementary Table 1). For the climatic variables, from arid to humid region, the GST initially decreased, and then increased with a minimum value of 9.55°C in the semi-humid region, while there was a significant decrease in the GSR (Supplementary Table 1).

In regard to the climatic factors, GST, GSE, and GSR positively affected the relative benefit of the LMA, but negatively influenced the relative benefits of N_mass and P_mass in the arid and semi-humid regions, similar to the effect of GST in the humid region (P < 0.05, Figures 3A1–C1, A3–C3, A4–C4 and Table 2). However, the relationship patterns of the GSR with the leaf traits in the humid region were converse to those in the arid and semi-arid regions (Figures 3A4–C4). The GSP showed weak negative correlations with the relative benefits of the LMA but was positively associated with the relative benefits of N_mass and P_mass in the semi-arid and semi-humid regions (Figures 3A2–C2).

For the edaphic factors, SOM, STN, and STP were negatively related to the relative benefit of the LMA, but positively affected the relative benefits of N_mass and P_mass in the semi-arid regions (Figures 4A1–C2 and Supplementary Figure 7). Similar correlations were observed between the SAN with the relative benefits of the LMA, N_mass, and P_mass, except in the semi-humid region (Figures 4A3–C3). The number of key edaphic and climatic factors driving the leaf traits in the arid and humid regions was greater than those in the semi-arid and semi-humid regions (Figure 5).

Our study showed that the LMA in the arid region was drastically higher than those in other regions (Figure 2A), indicating that plants in hyper-arid conditions invest more photosynthate per unit leaf area. This result is in agreement with many previous studies indicating that plants inhabiting arid areas have a high LMA as manifested by thicker leaves with denser mesophyll (Fonseca et al., 2000; Niinemets, 2001; Wright et al., 2001), often combined with smaller and thicker-walled cells for reducing water loss (Wright et al., 2006). As the maximum photosynthetic rate of plants in a dry environment is relatively lower at the same level of leaf N content, the high LMA increases the photosynthetic tissue per unit area supporting photosynthesis, and allows for continued leaf function in arid conditions (Evans and Poorter, 2001). In addition, the SOM, STN, STP, and SAN in the arid region are significantly lower than those in other regions (Supplementary Table 1), and the low available water decreases the soil nutrient availability (Sardans and Penuelas, 2004) and further restricts plant nutrient absorption (Farooq et al., 2009). Consequently, survival is the primary goal for plants, which tend to select a relatively conservative investment strategy. The higher LMA in the arid region should be interpreted as an adaptive strategy of plants against extreme environmental stress, which was alleviated in other climatic regions, and the LMA was relatively smaller.

The leaf N and P values play pivotal roles in photosynthesis and protein synthesis, respectively (Niklas et al., 2005; Westoby and Wright, 2006). The N_mass and P_mass values in the humid region were dramatically lower than those in the other regions (Figures 2B,C), indicating a small carbon gain rate. This is probably because frequent strong precipitations accelerated the soil erosion and nutrient losses in the humid region, indicated by that the soil nutrients were significantly lower than those in the semi-humid region (Supplementary Table 1). Moreover, the soils were often waterlogged and anoxic, depressing the oxidative metabolic processes of the roots and microbes in the mesic lands, where the plant growth rates were greatly inhibited by excessive rainfall (Zhang S. S. et al., 2019; Zheng Q. et al., 2019).

We observed significantly negative correlations of the LMA with N_mass and P_mass (Figures 2D,E), which agrees with previous studies demonstrating that a higher LMA is often associated with lower leaf N and P concentrations, indicating slow growth rates (Santiago and Wright, 2007; Donovan et al., 2011). Nevertheless, the leaves with a high LMA and low N_mass and P_mass are more tolerant to unfavourable growing conditions. In contrast, plants would be vulnerable to damages from environmental stress if a high LMA was combined with high N_mass and P_mass, as it would accelerate energy losses via respiration (Wright et al., 2004). This might explain why negative-relationship slopes were the deepest in the arid region (Figures 2D,E).

The favourability hypothesis suggests that plants tend to choose certain optimum trait assemblages when facing particular inhabiting conditions, either through trait plasticity within a species or via species turnover in community composition (Swenson et al., 2012; Simova et al., 2015). Temperature which influences plant hydraulics, nutrient utilization, and leaf energy balance, is a central controller of leaf trait variation (Fonseca et al., 2000; Michaletz et al., 2016). In the present study, it was obvious that the orientations of GST and LMA exhibited the smallest angle (Supplementary Figure 2), indicating that the GST tended to promote an increase in the LMA, which conflicts with prior studies (Swenson et al., 2012; Simova et al., 2015; Simova et al., 2018). We propose two probable explanations for the contradictory results. First, the sampling sites in our study covered a relatively narrower GST range (from –1 to 28.63°C) than the former studies. The plant growth form was mainly herbaceous plant species in the temperate zone, unlike, e.g., boreal coniferous species, with higher LMAs in cold conditions for modulating the temperatures of leaves (Michaletz et al., 2016). Second, the soil nutrients such as SOM, STN, STP, and SAN apparently decreased with increases in GST (Supplementary Figure 4). Plants exhibited low potential for resource capture and a slow photosynthetic rate under the more unfavourable inhabiting conditions (Wright et al., 2010; Swenson et al., 2012). They can also explain why the GST negatively influences P_mass and thereby affects the photosynthesis-N relationship (Reich et al., 2009), in accordance with previous studies (Kerkhoff et al., 2005; Sun et al., 2019). In addition, the negative correlation of the GST with N_mass (Supplementary Figure 2) provides evidence for the temperature–plant physiological hypothesis, i.e., leaf N would increase with decreasing temperature, as a high N concentration could compensate for physiological inefficiency in a cold environment (Reich and Oleksyn, 2004).

In this study, we used the benefit metrics for leaf traits to determine changes in plant growth strategy by analysing their relationship with different environmental drivers. Specifically, low relative benefits of LMA, high relative benefits of N_mass and P_mass implied an acquisitive growth strategy; conversely, a conservative growth strategy. Consistent with previous studies indicating that the responses of plant physiological processes to the same factor and underlying rules may be largely divergent in face of different environments (Quan et al., 2019), we found that the relative benefit of LMA was distinctly elevated, while the relative benefits of N_mass and P_mass decreased with increasing GSR in the arid and semi-humid regions, whereas in the humid region, the leaf traits showed the opposite pattern (Figures 3A4–C4). In that regard, the GSR is strongly positively associated with the GST and GSE in the arid and semi-humid regions and displays a significantly negative relation to the GSP in the semi-humid region (Supplementary Figures 3D–F). The radiation-induced warming leads to the reduction of soil water through intensifying evaporation (Wang et al., 2009), thereby aggravating the negative influences of drought in the arid area (Quan et al., 2019). Moreover, an increase in the GSE with increasing GSR accelerates the water consumption from transpiration (Decker et al., 2013). As the GSR increases, plants are expected to adopt conservative investment-return strategies with slow growth rates and higher LMAs with smaller areas exposed to solar radiation per gram leaf, to reduce radiative heating and water loss (Wright et al., 2004; Chen et al., 2013; Moles et al., 2014).

However, for the humid region, the overall relationships of the GIR with the GST, GSP, and GSR are considerably weaker than those in the other regions (Supplementary Figure 3). Indeed, the GIR is the lowest in this region (Supplementary Table 1), implying a relatively limited light resource availability. As a vital factor directly influencing plant growth, solar radiation plays crucial roles in many plant physiological processes, such as leaf N and P accumulations (Takashima et al., 2004; Sun et al., 2019), photosynthesis (Waite and Sack, 2010), and protein synthesis (Xu, 2003). Consistent with Wright et al. (2006), who revealed that plants in higher-irradiance regions had a lower LMA and higher photosynthetic capacity than those in lower-irradiance areas, we observed that the GSR presented a negative impact on the relative benefit of LMA, but remarkably positively affected the relative benefits of N_mass and P_mass in the radiation-lacking region. This indicated that the plant growing strategy shifted from slow to fast as the GSR increased.

Unexpectedly, the GSP showed no significant effect on the relative benefit of leaf traits in the arid region (Figures 3A2–C2). The following several mechanisms might contribute to this. First, although precipitation data are relatively easier to obtain, they are not always sufficient proxies for soil water availability (Hickler et al., 2009), especially in hyper-arid grassland mostly belonging to desert steppe, where the soil water holding capacity are small. Thus, the soil available water is still limited, even if there is relatively high rainfall (Li et al., 2008; Xie, 2011). Second, it may be attributed to the inherent characteristics of plant species evolutionarily adapted to the hyper-arid conditions, with corresponding trait syndromes such as a high LMA and small stature (Wellstein et al., 2017), thereby inducing the plant community to be relatively less sensitive to rainfall (White et al., 2000). Finally, the low soil nutrient contents in this region (Supplementary Table 1) greatly restrict plant growth, which might weaken the relative importance of the GSP in mediating the leaf traits.

Soil nutrients play positive roles in multiple plant physiological activities. For example, the SOM provides energy for heterotrophic N-fixing microbes, and further facilitates plant N fixation (Reed et al., 2011; Zheng M. H. et al., 2019); the soil N can strongly affect the leaf N and plant photosynthesis (Lu et al., 2011; Yue et al., 2019); and a low soil P may lead to a lower leaf P and thus limit plant function (Han et al., 2005; Quesada et al., 2009; Sun et al., 2019). Plants invest more energy for resource acquisition in nutrient-insufficient soil (Nasto et al., 2014). Soil nutrient changes can induce species turnover, by influencing plant facilitation and competitive exclusion (Marriott et al., 2002; Guo, 2003). The regression relation between relative benefits of each leaf trait and soil nutrients indicated that the SOM, STN, and STP consistently presented significant promotion of an acquisitive growth strategy in the semi-arid region (Figure 4, Table 3, and Supplementary Figure 7). This may be ascribed to the poorer soil quality in this region (Supplementary Table 1). In contrast, the soils in the semi-humid region were relatively more fertile, providing plentiful nutrients for plant growth. In that regard, the leaf traits will not respond to the nutrient supply after the soil fertility fully meets the plant demand (Figure 5③; Bowman et al., 2003).

We found no evidence that the SOM, STN, and STP caused variations of relative benefits of leaf traits in the arid region with insufficient soil nutrients (Figure 4 and Supplementary Figure 7). This may be owing to the narrow ranges of the SOM, STN, and STP content. Moreover, drought constrains the effectiveness of soil fertility on plant growth, and thus the sensitivity of the leaf traits to nutrients in the hyper-arid area (Sardans and Penuelas, 2004; Farooq et al., 2009). Additionally, the SAN presented negative effects on the relative benefit of LMA, but positively affected the relative benefits of N_mass and P_mass, promoting an acquisitive investment strategy except in the SAN-rich semi-humid region (Figures 4A3–C3). This agrees with early studies, i.e., that the leaf traits are more related to the N supply at a global scale (Ordonez et al., 2009).

Overall, our results confirm that abiotic filtering shapes the leaf trait pattern, i.e., higher costs for resources and environmental stress lead to a slow-growing strategy with conservative leaf traits, whereas optimal conditions result in a fast tissue turnover with acquisitive leaf traits across climatic regions in China (Figure 5). This supports the favourability hypothesis, i.e., that the plant leaf traits should be governed mainly via environmental filtering in a climatically harsh temperate zone (Fischer, 1960). Furthermore, more key environmental drivers and steeper leaf-trait-environment relationship slopes were found in the arid and humid regions (Figure 5), suggesting that the degree of abiotic filtering is stronger in more stressful environments characterised by e.g., nutrient limitations or drought or over-wet conditions, in line with previous studies (Swenson et al., 2012; Simova et al., 2015). In that regard, the more stressful ecosystems are commonly inhabited by stress-tolerant species with a narrow geographical range size owing to natural selection (Thuiller et al., 2004), which respond more sensitively to environmental changes than wide-ranging species (Geng et al., 2012). Together with the finding that the environmental signal for the leaf traits would be relatively weaker in the optimal environmental conditions represented by the semi-humid region (Supplementary Figure 6), we conclude that an extreme environment elicits extreme ecological responses (Zhang F. Y. et al., 2019) and predict that frequent extreme climatic events caused by future climate change may enhance the fluctuations of the leaf traits in the terrestrial ecosystems (Jentsch and Beierkuhnlein, 2008; Stott et al., 2016).