Quantifying Leaf Trait Covariations and Their Relationships with Plant Adaptation Strategies along an Aridity Gradient

Simple Summary Plants usually adopt different strategies to adapt to their surrounding environments. Accurately quantifying plant strategies is of great interest in trait-based ecology, in particular to understand the responses of ecological structures and processes. In the last two decades, these strategies have been described qualitatively; however, the use of quantitative methods is still lacking. In this study, we used a plant functional trait approach to discuss plant strategies along an aridity gradient. We found that eight functional traits divided into four dimensions represent four adaptation strategies: energy balance, resource acquisition, resource investment and water use efficiency. We also concluded that climate and soil together with family (vegetation succession) were the main driving forces of trait covariations. Our study provided a new perspective to understand plant functional responses to aridity gradients, which is helpful for ecological management and vegetation restoration programs in arid regions. Abstract A trait-based approach is an effective way to quantify plant adaptation strategies in response to changing environments. Single trait variations have been well depicted before; however, multi-trait covariations and their roles in shaping plant adaptation strategies along aridity gradients remain unclear. The purpose of this study was to reveal multi-trait covariation characteristics, their controls and their relevance to plant adaptation strategies. Using eight relevant plant functional traits and multivariate statistical approaches, we found the following: (1) the eight studied traits show evident covariation characteristics and could be grouped into four functional dimensions linked to plant strategies, namely energy balance, resource acquisition, resource investment and water use efficiency; (2) leaf area (LA) together with traits related to the leaf economic spectrum, including leaf nitrogen content per area (Narea), leaf nitrogen per mass (Nmass) and leaf dry mass per area (LMA), covaried along the aridity gradient (represented by the moisture index, MI) and dominated the trait–environmental change axis; (3) together, climate, soil and family can explain 50.4% of trait covariations; thus, vegetation succession along the aridity gradient cannot be neglected in trait covariations. Our findings provide novel perspectives toward a better understanding of plant adaptations to arid conditions and serve as a reference for vegetation restoration and management programs in arid regions.

Simple Summary: Plants usually adopt different strategies to adapt to their surrounding environments. Accurately quantifying plant strategies is of great interest in trait-based ecology, in particular to understand the responses of ecological structures and processes. In the last two decades, these strategies have been described qualitatively; however, the use of quantitative methods is still lacking. In this study, we used a plant functional trait approach to discuss plant strategies along an aridity gradient. We found that eight functional traits divided into four dimensions represent four adaptation strategies: energy balance, resource acquisition, resource investment and water use efficiency. We also concluded that climate and soil together with family (vegetation succession) were the main driving forces of trait covariations. Our study provided a new perspective to understand plant functional responses to aridity gradients, which is helpful for ecological management and vegetation restoration programs in arid regions.
Abstract: A trait-based approach is an effective way to quantify plant adaptation strategies in response to changing environments. Single trait variations have been well depicted before; however, multi-trait covariations and their roles in shaping plant adaptation strategies along aridity gradients remain unclear. The purpose of this study was to reveal multi-trait covariation characteristics, their controls and their relevance to plant adaptation strategies. Using eight relevant plant functional traits and multivariate statistical approaches, we found the following: (1) the eight studied traits show evident covariation characteristics and could be grouped into four functional dimensions linked to plant strategies, namely energy balance, resource acquisition, resource investment and water use efficiency; (2) leaf area (LA) together with traits related to the leaf economic spectrum, including leaf nitrogen content per area (N area ), leaf nitrogen per mass (N mass ) and leaf dry mass per area (LMA), covaried along the aridity gradient (represented by the moisture index, MI) and dominated the trait-environmental change axis; (3) together, climate, soil and family can explain 50.4% of trait covariations; thus, vegetation succession along the aridity gradient cannot be neglected in trait covariations. Our findings provide novel perspectives toward a better understanding of plant adaptations to arid conditions and serve as a reference for vegetation restoration and management programs in arid regions.

Introduction
Aridity acts as a strong environmental filter for plant survival, growth and development and therefore has considerable effects on community structure and ecosystem functions, including primary productivity and nutrient cycling [1,2]. For instance, aridity favors species with small leaves as a part of their adaptation strategy to adapt soil to water deficits and nutrient limitations [3]. Plants undergo environmental adaptation and evolve an optimal phenotype, forming a set of adaptation strategies at both the community and individual levels [4]. To date, researchers have usually ascribed the single axis of a plant's adaptation strategy to environmental changes, but strategies explained by covariation and a tradeoff in traits related to nutrient acquisition and conservation have been neglected [5,6]. These strategies can be quantified by the covariation of several plant functional traits, which are any morpho-physio-phenological traits that have an impact on plant growth, reproduction and survival [7,8]. From the perspective of trait-based ecology, more information on plant adaptation to aridity gradients is needed.
Functional trait covariation may reflect plant strategies to stressful aridity conditions. In fact, variability in environmental conditions is considered as a filter of traits [9,10]; therefore, variations in plant functional traits reflect shifts in plant adaptation strategies under changing environmental conditions. For example, changes in plant height, leaf mass per area (LMA) and leaf nitrogen content per area (N area ) are directly linked to resource acquisition and drought tolerance, which have been widely reported [8,11,12]. Meanwhile, plants in resource-poor environments prone to arid conditions are expected to favor conservative trait syndromes, such as slower growth rates, lower production and higher root density [13]. Within arid sites, C 3 plants show higher values of leaf 13 C, which reflect higher water use efficiency [14]. Thus, it has been reported that plants balance their investment in either enzymatic reactions or electronic transport in the process of photosynthesis to avoid wasting resources [15].
Several studies have successfully quantified plant adaptation to arid environments [1,14,16]; nonetheless, only a few of them have yielded results that consider the interpretation difficulty of trait covariation and their controls. Some issues remain to be resolved: (1) Traits do not vary independently but show covariation and tradeoff relationships [17].
However, from the perspective of plant functional traits, plant strategies adopted to simultaneously balance conservation and resource acquisition remain unclear. (2) Climate and soil are widely recognized to control trait variation [18][19][20]. However, the effects of climate and soil intersections have not been fully quantified, especially under arid conditions. (3) Vegetation succession along an aridity gradient resulted in the replacement of larger leaf plants by plants with small and high-efficiency leaves [21]. The contribution of vegetation distribution (family) to trait covariations is still unknown along aridity gradients.
In the past few decades, the Loess Plateau in China has implemented many largescale ecological protection and restoration projects [22,23]. The observed regional climate changes (especially in precipitation trends) has caused great effects on ecosystem composition and functions [24]. Therefore, quantifying plant adaption to regional climate changes, especially in arid areas, is a matter of urgency. In this study, taking the Loess Plateau as the study area and using multivariate statistics, we aim to (1) quantify the covariation characteristics of eight functional traits and their representative dimensions (adaptation strategies); (2) reveal trait variation along the precipitation and temperature gradients; and (3) explore the controls of eight key traits along the aridity gradient on the Loess Plateau. Hence, the aim of this study is to reveal the adaptation strategies of plants to aridity gradients throughout quantifying the covariation of eight fictional traits.

Study Area and Sampling Strategy
The Loess Plateau is a highland area located in the middle of the Yellow River basin in China (Figure 1a) and is prone to serious soil and water loss [25]. With an area of more than 620,000 km 2 , this region features a temperate continental climate with a mean annual temperature ranging from 4 • C to 14 • C and annual precipitation ranging from approximately 200 mm to 800 mm, forming three distinct zones along an aridity gradient: semihumid (moisture index, MI > 0.67), semiarid (0.55 < MI ≤ 0.67) and arid (MI ≤ 0.55). From southeast to northwest, precipitation decreases on the Loess Plateau ( Figure 1b). Thus, from south to north, the temperature (represented by a growing-season monthly mean temperature above 0 • C (mGDD 0 )) increases and is closely related to changes in elevation. In the past few decades, many ecological projects have been implemented on the Loess Plateau (Figure 1c), and, therefore, the effects of the local climate and plant adaptation to environmental changes have been widely studied.
in China (Figure 1a) and is prone to serious soil and water loss [25]. With an area of m than 620,000 km 2 , this region features a temperate continental climate with a mean ann temperature ranging from 4 °C to 14 °C and annual precipitation ranging from appro mately 200 mm to 800 mm, forming three distinct zones along an aridity gradient: sem humid (moisture index, MI > 0.67), semiarid (0.55 < MI ≤ 0.67) and arid (MI ≤ 0.55). Fr southeast to northwest, precipitation decreases on the Loess Plateau (Figure 1b). Th from south to north, the temperature (represented by a growing-season monthly me temperature above 0 °C (mGDD0)) increases and is closely related to changes in elevati In the past few decades, many ecological projects have been implemented on the Lo Plateau (Figure 1c), and, therefore, the effects of the local climate and plant adaptation environmental changes have been widely studied.
Eight morphological, chemical and photosynthetic traits were measured and sa pled from 105 species and 271 records along 22 sites in August 2016 and 2017 ( Figure  Tables 1 and A1). The 105 species were grouped into 38 families, and the main famil are described in Table A2. The sampling procedure allows one to cover all of the vege tion types and climate gradients of the Loess Plateau, which represents an ideal design reveal plant responses and functional strategies to aridity gradients. Sampling was stricted to dominant species and communities with no clear human disturbance. Th along the sampling transect, the vegetation types changed from forestland to shrubla and desert, with noticeable vegetation succession, making it ideal to study changes in e logical structure and functions from the perspective of plant functional traits.   Table A2. The sampling procedure allows one to cover all of the vegetation types and climate gradients of the Loess Plateau, which represents an ideal design to reveal plant responses and functional strategies to aridity gradients. Sampling was restricted to dominant species and communities with no clear human disturbance. Thus, along the sampling transect, the vegetation types changed from forestland to shrubland and desert, with noticeable vegetation succession, making it ideal to study changes in ecological structure and functions from the perspective of plant functional traits.

Data Descriptions of Traits, Climate and Soil
All plots were located on sunny slopes, and middle-aged plants were sampled. Eight leaf functional traits, namely leaf area (LA), leaf nitrogen content (N mass ), N area , LMA, leaf dry mass content (LDMC), maximum carboxylation rate standardized to 25 • C (V cmax25 ), maximum electron transport rate standardized to 25 • C (J max25 ) and the ratio of intercellular to ambient CO 2 concentration (c i :c a , termed χ in this study), which could capture most plant functions, were measured. All traits were measured following standardized protocols, as proposed by Cornelissen et al. [26]. LA is the projected area of a leaf (or leaflet for compound leaves), which is linked to the leaf energy balance [27]. LDMC is the oven-dry mass divided by its fresh mass. LMA is the oven-dry mass divided by fresh LA, which is the inverse of specific leaf area (SLA) and highly related to the resistance of plants to both herbivory and drought [28]. N mass is the leaf nitrogen per mass, and N area is the leaf nitrogen per area, which are two indicators of photosynthetic capacity [29]. LDMC, LMA, N mass and N area are key traits in the leaf economics spectrum (LES), which reflects a plant's investment of nutrients along the acquisition-conservation continuum [30,31].
V cmax and J max were calculated using a one-point method from the light-saturated rate of net CO 2 fixation at ambient CO 2 (A sat ) and the light-saturated rate of net CO 2 fixation at high CO 2 (A max ) [32]. V cmax and J max were standardized to 25 • C following the proposed methods of Niinemets et al. [33]. From the 272 records, 46 records for V cmax25 and J max25 were not measured, and, therefore, we used a multilayer perceptron (MLP) [34], which is a class of feedforward artificial neural networks with high simulation accuracy. In this experiment, the training and validation accuracy for V cmax25 and J max25 reached 0.55. χ was converted from leaf δ 13 C measurements after eliminating the effects of sampling year and latitude, following the method proposed by Cornwell et al. [35]. χ is a useful indicator in aridity areas because it determines the balance between carbon gain and water loss, which is closely regulated by the stomata and reflects the adaptation of plants in arid conditions. The climate of each site was defined using three climate variables, namely growing degree days above 0 • C (mGDD 0 ), monthly accumulated photosynthetically active radiation during the thermal growing season (mPAR 0 ) and the ratio of mean annual precipitation to annual equilibrium evapotranspiration (moisture index, MI). MI is a widely recommended indicator of aridity stress, which is the inverse of dry index and can reflect the plant-available water content in the soil. These bioclimatic variables were calculated from observations retrieved from meteorological stations on the Loess Plateau. Two variables representing soil conditions for optimal plant growth were selected, namely soil organic matter (SOM) and soil pH at 0~30 cm depth.

Multivariate Statistical Analysis
Principal component analysis (PCA) and redundancy analysis (RDA) were used to analyze trait covariation and to reveal the relationships between traits, considering traits as response variables and climate/soil/family variables as predictors [36,37]. PCA and RDA were performed with the vegan package in R language [38]. LA was square-root transformed to create a linear relationship. Calculated using the method of variation partitioning, a Venn diagram was used to show the contributions of the climate, soil and family, and their independent and intersection effects were explicitly considered [39]. The taxonomic rank at the family level was used in forward analysis looking at the shared evolutionary history between species and the assumed similarities in their responses to environmental changes. Thus, for the purpose of our study, it was more suitable to use the family level rather than the genera or species in the statistical analysis. Ordinary linear regression was employed to describe the variation in a single trait response to the climate gradient, and the 95% confidence intervals are shown.

Trait Covariation and Corresponding Adaptation Strategies
The PCA analysis that included all plants and sites confirmed our hypothesis that the eight traits studied are not independent but show covariation patterns ( Figure 2, Table 2). The first three axes captured 84.32% of the total variation in the eight traits. The first axis captured 41.13% of variation, which was mainly controlled by LA, N area , V cmax25 and J max25 . The second axis accounted for 32.61% of variation and was dominated by LA, V cmax25 and J max25 . The third axis accounted for 10.58% of variation and was dominated by N area and LMA. Both LA and χ changed independently and did not covary with the other traits ( Figure 2). V cmax25 and J max25 closely covaried in the first two PCA axes. N mass , N area , LMA and LMDC, four important traits of LES, also presented clear covariation characteristics, especially on the second axis (capturing 32.61% of the total variations), which show similar weights to the first axis. Table 2. Trait loading, eigenvalues and percentage of trait variation explained by the first three principal components (PCs). LA: leaf area; N mass : leaf nitrogen content; N area : leaf nitrogen per area; LMA: leaf mass per area; LDMC: leaf dry mass content; V cmax25 : maximum carboxylation rate standardized to 25 • C; J max25 : maximum electron transport rate standardized to 25 • C; χ: ratio of intercellular to ambient CO 2 concentration.   The correlation matrix ( Figure 3) confirmed our results obtained from the PCA analysis that the eight studied traits are organized into four trait dimensions. The first dimension was LA, which had a low correlation with other traits and represented plant behavior during evaporation (energy balance). The second dimension was composed of LDMC, LMA, Narea, and Nmass, which were closely related within this dimension and weakly related to traits in other dimensions. This dimension is linked to a plant's strategy for resource acquisition. Vcmax25 and Jmax25 constituted the third dimension, with a correlation coefficient reached at 0.77. In this dimension, Vcmax25 and Jmax25 varied tightly because plants should not invest more in either photosynthesis or electronic transport to maintain a lower energy cost. The fourth dimension was χ, which was found to be unrelated to the other traits. This dimension is directly linked with water use efficiency. The clear color differ- The correlation matrix ( Figure 3) confirmed our results obtained from the PCA analysis that the eight studied traits are organized into four trait dimensions. The first dimension was LA, which had a low correlation with other traits and represented plant behavior during evaporation (energy balance). The second dimension was composed of LDMC, LMA, N area , and N mass , which were closely related within this dimension and weakly related to traits in other dimensions. This dimension is linked to a plant's strategy for resource acquisition. V cmax25 and J max25 constituted the third dimension, with a correlation coefficient reached at 0.77. In this dimension, V cmax25 and J max25 varied tightly because plants should not invest more in either photosynthesis or electronic transport to maintain a lower energy cost. The fourth dimension was χ, which was found to be unrelated to the other traits. This dimension is directly linked with water use efficiency. The clear color differences of points in the histogram ( Figure 3) and scatter plots (Figure 2a,b) indicate that trait covariations were not random but presented significant variation along the precipitation gradient.

Trait Variations along the Aridity Gradient (MI Decreased)
Except LDMC, all the studied traits showed significant variation trends along the aridity gradient (decreasing MI) (Figure 4). LA increased along with an increase in MI, which is consistent with the transition of vegetation types from forests (large leaf areas) to grasslands and deserts (small leaf areas). LDMC was not sensitive to moisture changes. Plants tended to have higher LMA values when facing arid conditions because of its construction cost. Both N mass and N area increased significantly along the aridity gradient and were related to leaf construction and photosynthetic capacity. V cmax25 and J max25 were directly linked with photosynthetic capacity, exhibiting an increase along the aridity gradient (decreasing MI), because the growth period was shorter in the arid area than in the humid area, and they required more photosynthesis efficiency to recover construction cost. χ decreased along the aridity gradient, which reflected an adjustment in plant stomata in arid conditions to maintain high water use efficiency.

Trait Variations along the Aridity Gradient (MI Decreased)
Except LDMC, all the studied traits showed significant variation trends along the aridity gradient (decreasing MI) (Figure 4). LA increased along with an increase in MI, which is consistent with the transition of vegetation types from forests (large leaf areas) to grasslands and deserts (small leaf areas). LDMC was not sensitive to moisture changes. Plants tended to have higher LMA values when facing arid conditions because of its construction cost. Both Nmass and Narea increased significantly along the aridity gradient and were related to leaf construction and photosynthetic capacity. Vcmax25 and Jmax25 were directly linked with photosynthetic capacity, exhibiting an increase along the aridity gradient (decreasing MI), because the growth period was shorter in the arid area than in the humid area, and they required more photosynthesis efficiency to recover construction cost. χ decreased along the aridity gradient, which reflected an adjustment in plant stomata in arid conditions to maintain high water use efficiency.

Trait Variations along the Aridity Gradient (MI Decreased)
Except LDMC, all the studied traits showed significant variation trends along the aridity gradient (decreasing MI) (Figure 4). LA increased along with an increase in MI, which is consistent with the transition of vegetation types from forests (large leaf areas) to grasslands and deserts (small leaf areas). LDMC was not sensitive to moisture changes. Plants tended to have higher LMA values when facing arid conditions because of its construction cost. Both Nmass and Narea increased significantly along the aridity gradient and were related to leaf construction and photosynthetic capacity. Vcmax25 and Jmax25 were directly linked with photosynthetic capacity, exhibiting an increase along the aridity gradient (decreasing MI), because the growth period was shorter in the arid area than in the humid area, and they required more photosynthesis efficiency to recover construction cost. χ decreased along the aridity gradient, which reflected an adjustment in plant stomata in arid conditions to maintain high water use efficiency.  Trait variation patterns were not the same along the temperature gradient (increasing mGDD 0 ) and the aridity gradients ( Figure 5), which were mainly affected by growth temperature and latitude. LA had a pronounced decreasing trend because of the combined effects of vegetation types and latitude, indicating that large areas have a strong ability to regulate temperature. Likewise, LDMC had no significant trend toward temperature changes. N mass and N area increased with an increase in temperature; thus, N area had a higher correlation coefficient than that of N mass . V cmax25 increased with an increase in temperature, which is explained by the activity of Rubisco regulating by temperature. J max25 reflects photosynthetic electron transport ability, which was higher under warm conditions. χ decreased with an increase in temperature as a result of the tendency of plants to close their stomata to avoid water loss in arid areas. per mass; Narea: leaf nitrogen per area; LMA: leaf mass per area; LDMC: leaf dry mass content; V maximum carboxylation rate standardized to 25 °C; Jmax25: maximum electron transport rate s ardized to 25 °C; χ: ratio of intercellular to ambient CO2 concentration.
Trait variation patterns were not the same along the temperature gradient (incre mGDD0) and the aridity gradients ( Figure 5), which were mainly affected by growth perature and latitude. LA had a pronounced decreasing trend because of the comb effects of vegetation types and latitude, indicating that large areas have a strong abil regulate temperature. Likewise, LDMC had no significant trend toward temper changes. Nmass and Narea increased with an increase in temperature; thus, Narea had a h correlation coefficient than that of Nmass. Vcmax25 increased with an increase in tempera which is explained by the activity of Rubisco regulating by temperature. Jmax25 reflects tosynthetic electron transport ability, which was higher under warm conditions. creased with an increase in temperature as a result of the tendency of plants to close stomata to avoid water loss in arid areas.

Trait Covariations and Adaptation Strategies Related to Climate and Soil Variables
The highlighted four dimensions composed from the eight studied traits rem independent when filtered by climate and soil variables ( Figure 6). In fact, covariatio the eight traits could be explained by the effects of climate and soil variables rea 27.6% (Table A1), presenting distinct environmental adaptation of plants. The mos portant information was the covariations in LA along the MI and soil pH gradients, w dominated the first RDA axis (Figure 6a). Three traits in LES (Narea, Nmass and LMA changed with MI changes, but LDMC was not sensitive to the first two RDA axes second RDA axis (RDA2) was mainly dominated by the covariations in Vcmax25 and along the temperature (mGDD0) and soil pH gradient, which accounted for 5.2% third axis accounted for only 0.5% of trait covariations, which was related to χ vari

Trait Covariations and Adaptation Strategies Related to Climate and Soil Variables
The highlighted four dimensions composed from the eight studied traits remained independent when filtered by climate and soil variables ( Figure 6). In fact, covariations in the eight traits could be explained by the effects of climate and soil variables reaching 27.6% (Table A1), presenting distinct environmental adaptation of plants. The most important information was the covariations in LA along the MI and soil pH gradients, which dominated the first RDA axis (Figure 6a). Three traits in LES (N area , N mass and LMA) also changed with MI changes, but LDMC was not sensitive to the first two RDA axes. The second RDA axis (RDA2) was mainly dominated by the covariations in V cmax25 and J max25 along the temperature (mGDD 0 ) and soil pH gradient, which accounted for 5.2%. The third axis accounted for only 0.5% of trait covariations, which was related to χ variation along Biology 2021, 10, 1066 9 of 16 with MI, mGDD 0 and mPAR 0 (Figure 6b). Along with the climate and soil variables, the traits also showed covariation trends within four trait dimensions. For example, three traits in LES covaried along RDA1, and two photosynthetic traits (V cmax25 and J max25 ) covaried along RDA2. y 2021, 10, x FOR PEER REVIEW along with MI, mGDD0 and mPAR0 (Figure 6b). Along with the climate and soil var the traits also showed covariation trends within four trait dimensions. For example traits in LES covaried along RDA1, and two photosynthetic traits (Vcmax25 and Jmax25) ied along RDA2.

Controls of Trait Covariations along the Aridity Gradient
In total, 50.19% of the covariations in traits can be explained by the combinat climate, soil and family (phylogeny/vegetation succession) (Figure 7). The largest c bution comes from family, which accounted for 38.59% (Table 3). The contribution mate to the eight trait covariations reached 20.41%, and the contribution of soil acco for 18.23%. For LA, both climate and soil contributed considerably to its variation ever, the independent control of climate was small, and notable information was fo the intersections of climate, soil and family. For LDMC, the contribution of clima soil was small, and family was the main determining factor of its variation. LMA wa controlled by family, and the contribution of climate and soil was small. Compared Nmass, Narea was largely controlled by climate and was less affected by family. Vcmax Jmax25 were mainly controlled by family, less than 40% originated from the three f combined. For χ, the contributions of climate, soil and family ranged from 13.8% to The eight studied traits had different responses to environmental changes. The section of climate and soil contributed 10.39% to their covariations, while family strong intersection with climate/soil variables (16.65%) (Figure 7a), indicating that s evolved to adapt to their changing environments along the aridity gradient. On LMA, Narea and χ were sensitive to aridity changes, while the contributions of other were less than 10%, and the most notable feature was the joint effects of family that leaf dry mass content; V cmax25 : maximum carboxylation rate standardized to 25 • C; J max25 : maximum electron transport rate standardized to 25 • C; χ: ratio of intercellular to ambient CO 2 concentration. MI: moisture index; mGDD0; monthly growth temperature above 0 • C; mPAR0: monthly photosynthetically active radiation above 0 • C; pH: soil pH; SOM: soil organic matter content.

Controls of Trait Covariations along the Aridity Gradient
In total, 50.19% of the covariations in traits can be explained by the combination of climate, soil and family (phylogeny/vegetation succession) (Figure 7). The largest contribution comes from family, which accounted for 38.59% (Table 3). The contribution of climate to the eight trait covariations reached 20.41%, and the contribution of soil accounted for 18.23%. For LA, both climate and soil contributed considerably to its variation; however, the independent control of climate was small, and notable information was found at the intersections of climate, soil and family. For LDMC, the contribution of climate and soil was small, and family was the main determining factor of its variation. LMA was also controlled by family, and the contribution of climate and soil was small. Compared with N mass , N area was largely controlled by climate and was less affected by family. V cmax25 and J max25 were mainly controlled by family, less than 40% originated from the three factors combined. For χ, the contributions of climate, soil and family ranged from 13.8% to 18.0%.

Importance and Significance of Studying Trait Covariations in Arid Areas
The present study illustrates the power of using large trait datasets from multiple c existing species, spanning a broad range of climates, to reveal general patterns of tr covariations. It also shows the utility of multivariate analysis in summarizing these p terns of covariations and the research interest in variance partitioning as a means of tributing trait variability to different (and sometimes intersecting) causes of variatio However, there is a need for further extensive trait data collection, including root-relat traits and isotopic measurements, thus covering an extended range of worldwide clima  The eight studied traits had different responses to environmental changes. The intersection of climate and soil contributed 10.39% to their covariations, while family had a strong intersection with climate/soil variables (16.65%) (Figure 7a), indicating that species evolved to adapt to their changing environments along the aridity gradient. Only LA, LMA, N area and χ were sensitive to aridity changes, while the contributions of other traits were less than 10%, and the most notable feature was the joint effects of family that intersected with climate/soil variables. After the effects of climate and soil were removed, variations in LDMC, LMA, N mass , N area and V cmax25 were largely controlled by family. For LDMC, LMA, V cmax25 , J max25 and χ, more than 50% of the variation was unexplained by the combination of climate, soil and family, which may suggest an effect of local microclimate conditions, trait-trait relationships and intrinsic factors.

Importance and Significance of Studying Trait Covariations in Arid Areas
The present study illustrates the power of using large trait datasets from multiple co-existing species, spanning a broad range of climates, to reveal general patterns of trait covariations. It also shows the utility of multivariate analysis in summarizing these patterns of covariations and the research interest in variance partitioning as a means of attributing trait variability to different (and sometimes intersecting) causes of variation. However, there is a need for further extensive trait data collection, including root-related traits and isotopic measurements, thus covering an extended range of worldwide climate types.
By analyzing the covariations of eight traits along the aridity gradient, we confirmed the existence of four main dimensions of variation, which is in accordance with previous findings [40,41]. Within these dimensions, trait-trait relationships were close, but few correlations were found between them, which could mean fewer traits representing adaptation strategies, especially in the field of ecological modeling. RDA analysis presented similar dimensionality, indicating that the trait dimensions remained consistent and stable along the arid gradient. These findings have implications toward the understanding of diverse plant strategies and species diversity. LA is the first dimension and represents a plant's strategy to maintain temperature. The second dimension is composed of LES-related traits in LES, which is an efficient indicator of quick growth and resource use. The third dimension is linked with photosynthesis, which balances construction investment and cost. The fourth dimension is χ, which is an indicator of water use efficiency.
We observed that the family contribution to trait covariation, which was characterized by phylogenetic replacement along the aridity gradient, was the most important determiner of trait variation. The family contribution to the eight trait covariations reached more than 38%. For LDMC, climate and soil explain only a small fraction of the trait variation; however, large variation was caused by the family factor. Therefore, predicting plant distribution changes and vegetation sensitivity under climate conditions is crucial to community and functional ecology. If the family dynamics and trait variations in each family were quantified, the family's role and the associated mechanisms could be accurately captured [42]. This result implies that vegetation succession cannot be neglected in traitbased ecological modeling.
Apart from the family contribution to trait covariations, the considerable role of phenotypic plasticity, originating from both a genetic and environmental basis, cannot be neglected. Phenotypic plasticity is the ability of an organism or a single genotype to exhibit different phenotypes, including behavior, morphology and physiology, in different environments [43]. For instance, in this present study, the most widely distributed plant, Lespedeza davurica, had clear differences in leaf area along the aridity gradient, which was explained by its capacity to maintain more water in its leaves and to adapt to low temperatures. Phenotypic variability is defined as the tendency or the potential of an organism to exhibit different phenotypes under a wide range of environmental conditions, which offer an important adaptation capacity in response to environmental changes [44]. Additionally, during field sampling, efforts need to be oriented to the effects of plant age and microclimate, which could also have a significant effect on the expression of trait covariations.

Mechanism of Trait Covariations along the Aridity Gradient
Trait covariations occurred along the aridity gradient, and the underlying mechanisms were behind the explanation of the different plant adaptation strategies. LA is linked with the theory of energy balance [27], which means that warm and humid conditions favor plants with large leaves; in contrast, plants with small leaves are limited by the lower temperatures. Thus, variations in N mass , N area , LMA and LDMC confirm the universal nature of LES, which reflects the optimal method of balancing the construction cost and payback time of leaves [45,46]. As expected from the theory of coordination hypothesis, close covariations were seen in V cmax25 and J max25 , which shows that photosynthesis depends on both carbonylation and electron transport, as leaves seek a balance when investing in these two processes to avoid energy waste [15,47]. χ is an indicator of CO 2 drawdown from air to leaf, and it is described in Farquhar's model [48] and applied in most ecological models. Hence, variation in χ is linked with the least-cost hypothesis, which states that leaves minimize the sum of the cost of transpiration and carboxylation [49].

Challenges and Future Directions
In this study, we quantified covariations in eight important traits that captured the most information on ecosystem functions and plant adaptations, and, thus, we explored their controls. However, a more in-depth analysis is needed regarding three main aspects. First, we sampled leaf functional traits, but some traits, such as root traits (specific root length, fine root density etc.) [50] and Huber values (the ratio of the sapwood cross-section and the supporting leaf area) [51] were equally important to studying plant adaptations. Second, family was listed as the most important control, but the driving mechanism and its intersection effects with climate and soil are not fully understood. Consequently, understanding the relationships between trait covariations and family would be helpful for regional ecological restoration and management programs. Third, vegetation-based models explore the incorporation of trait variations [52,53]; however, the extent and magnitude of trait variation and, thus, their covariation characteristics have not been fully quantified. In this sense, we suggest strengthening the observation of trait covariations in the field and building consistent theoretical equations for vegetation models.
We strongly suggest considering 'functional diversity' in ecological modeling. In fact, trait-based models can represent the co-existence of multiple trait combinations, considering the fact that this diversity provides increased community resilience when facing environmental changes [54]. The challenge is to find a generally applicable method to specify the range of proper trait combinations that is consistent with the observed patterns of trait variation within sites. Ultimately, vegetation models should be able to predict, for example, and experimentally determine the relationships between species diversity and ecosystem function [55], but this potential has yet to be realized.

Conclusions
By analyzing the covariations of eight traits, we found that plants within arid regions adopted at least four strategies related to energy balance, resource acquisition, resource investment and water use efficiency, which reflect plant adaptations to environmental changes. We also confirmed that climate, soil and family contributed to trait covariations, but the control of family in trait covariations could not be neglected. Our findings linked trait covariations with multiple adaptation strategies and provide a reference for regional ecological management and species selection in arid areas.  Data Availability Statement: Data are available on request due to restrictions, e.g., privacy or ethical. The data presented in this study are available on request from the corresponding author. The data are not publicly available due to the strict management of various data and technical resources within the research teams.     Table A2. Description of families with a record number greater than 8.