Study sites and experimental design
In 2014, we selected six sites arrayed across the east-west extent of the grassland biome in northern China (Table S1). These six sites encompass the three major grassland types in China (i.e. meadow steppe, typical steppe and desert steppe) and vary in mean annual precipitation (MAP), mean annual temperature (MAT), plant species composition and edaphic properties. We extracted climatic variables (e.g., MAT, MAP, and potential evapotranspiration (PET, mm)) for each site from the global Worldclim data set with a resolution of 0.0083° × 0.0083° (http://www.worldclim.org). We defined the aridity index (AI) of each site as 1 - MAP/PET (Delgado-Baquerizo et al. 2013). Among the six sites, aridity increases from east (SMR) to west (UDR) (Table S1). Soil texture also varies across the six sites from sandy to clay loams, as well as other soil characteristics (Table S1). All six sites had not been grazed by domestic herbivores for the 3 years prior to the 2015 drought. The dominant species are Stipa baicalensis and Leymus chinensis in the meadow steppe, S. baicalensis and L. chinensis in the typical steppe, and S. breviflora and Peganum harmala in the desert steppe.
We established experimental drought infrastructure at each site which remained in place from 2015 to 2018. Using a randomized complete block design, we established twelve 6×6 m plots per site (6 drought; 6 control) in a topographically uniform area. Plots were located at least 2 m from each other and were hydrologically isolated by trenching the perimeter to a depth of 1 m and installing 6-mm-thick plastic barriers to prevent lateral water flow. We constructed large rainout shelters to block 66% of ambient growing season precipitation from May to August in each year (Figure S1). These rainout shelter roofs were built with transparent polyethylene panels covering 66% of the surface and were supported by a light scaffolding structure. To allow for air flow below the panels and minimize any potential greenhouse effect, we installed shelters 2 m above the ground (Delgado-Baquerizo et al. 2017). Similar experimental infrastructure has been used in previous experiments with minimal effects on light environment (permitting nearly 90% transmission) (Yahdjian and Sala 2002; Wilcox et al. 2015). For control plots, we established similar scaffolding structures but did not install polyethylene panels.
Sampling and measurements
Each plot contained a 4×4 m sampling plot, which we divided into four 1×1 quadrats. In August 2017, we randomly selected one of these quadrats and further divided it into four 50×50 cm sub-quadrats. We designated two diagonal sub-quadrats for destructive measurements of plant biomass, and the other two for surveys of plant traits (Luo et al. 2019).
We estimated aboveground net primary productivity (ANPP) in the two designated sub-quadrats by clipping plant material at the ground level during peak growth. We sorted biomass by species before oven-drying (48 hr at 65 °C) and weighing to estimate species-specific ANPP. We calculated species relative abundance as the species’ percent contribution towards total ANPP.
In the other two diagonal sub-quadrats, we measured plant traits of the most abundant species (i.e., cumulatively representing at least 90% of the total ANPP). First, we estimated plant height for three individuals of each species per plot. We then collected the youngest, fully expanded leaf from the same individuals (Perez-Harguindeguy et al. 2016) and oven-dried these leaves at 105°C for 30 min to stop enzymatic activity before drying at 65°C until constant weight. We estimated foliar soluble sugar (SS) and starch content spectrophotometrically (ultraviolet-visible spectrophotometer 723 S, Yoke Precision Instruments Co., Ltd, Shanghai, China) at 620 nm using the Sulfuric acid-Anthrone method (Li et al. 2008b). We used the sum of starch and SS concentrations as an estimate of total NSC concentration for each species. The concentrations of SS and total NSC have been widely used as indicators of NSC (Martinez-Vilalta et al. 2016).
In August 2018, we sampled leaves of the same species to determine leaf chlorophyll content. We stored samples at -20°C until foliar chlorophyll was extracted using spectra-analyzed grade N, N dimethylformamide. We measured absorbance at 663, 646 and 480 nm using a Shimadzu UV-1700 spectrophotometer (Wellburn 1994; Chen et al. 2013) and total chlorophyll content was estimated according to Porra et al. (1989).
Data analysis
We used R statistical programming (R version i386 3.6.1) to run all data analyses described below. First, we used Shapiro-Wilk and Levene’s tests to confirm the normality and heteroscedasticity of all trait data. Based on this confirmation, we used the original non-transformed data in all statistical analyses. We calculated community weighted mean (CWM) traits (i.e. plant height, chlorophyll content, SS concentrations, starch concentrations and total NSC concentrations) weighted by each individual species’ relative contribution to biomass in each plot (Khasanova et al. 2013; Griffin-Nolan et al. 2018).
To determine the spatial relationships between aridity and NSC concentrations, we regressed AI against CWM of SS and total NSC concentrations using linear or curvilinear (quadratic) equations. Based on both explained variation and Akaike’s Information Criterion (AIC) values, we found that the relationships between AI and both SSCWM and total NSCCWM concentrations were best described by a second-order polynomial, with lowest levels at a site with intermediate aridity. To determine trait-trait relationships, we used mixed linear models to relate SSCWM and total NSCCWM concentrations to heightCWM and chlorophyllCWM concentration along the sampled aridity gradient as well as on each side of the aridity gradient (i.e., moist vs. dry regions) with blocks nested within site as random effect. Here, the two sites with intermediate aridity were included as both moist and dry regions. To determine the effect of experimental drought on both SSCWM and total NSCCWM concentrations, we ran a mixed model analysis of variance with drought treatment and site as fixed factors and block as a random factor. As interactive effects of drought treatment and site were all significant (P < 0.05), mixed-effect models were applied separately for each site with drought treatment as fixed factor and block as a random factor. Additionally, the total effects of experimental drought on SSCWM and total NSCCWM concentrations were analyzed with mixed models with drought treatment as a fixed factor and site and block as random factors.
Shifts in plant SSCWM and total NSCCWM can be attributed to both species turnover (CTurn) and intraspecific variation (CIntra); thus, we isolated their relative contributions using the following equation (Jung et al. 2014; Luo et al. 2018): CTurn = NDr* - NCt and CIntra = NDr - NDr*, where NCt and NDr are SSCWM and total NSCCWM concentrations in the control and drought plots, respectively, calculated from relative biomass and SS and total NSC concentrations of each species measured in their respective plot. NDr* is SSCWM and total NSCCWM concentrations in the drought plots, recalculated using a species’ relative biomass in the drought plots, but the SS and total NSC concentrations measured in the control plots.