1. Although grazing has significant impacts on plant functional traits and community composition in grasslands, few studies have simultaneously explored how plant above- and below-ground traits and community functional composition respond to grazing.
2. Using a grazing manipulation experiment with seven levels of grazing intensity (0, 1.5, 3.0, 4.5, 6.0, 7.5, and 9 sheep ha^-1) in the Inner Mongolia grassland, we partitioned the roles of intraspecific trait variation (ITV) and species turnover underlying the grazing induced changes in above- and below-ground functional trait composition of plant communities.  Six aboveground traits (i.e. plant height, plant aboveground biomass, plant density, leaf area, specific leaf area (SLA) and leaf density) and three root traits (average root diameter, ARD; specific root length, SRL; and root tissue density, RTD) of the first-, second- and third-order roots (1^st-, 2^nd-, and 3^rd-) were measured at the plant individual level. 
3. At the community level, plant above-ground traits shifted towards grazing avoidance strategy (e.g. plant height and leaf area decreased), and below-ground traits shifted towards conservative strategy (i.e. 1^st-SRL and 2^nd-SRL decreased, 1^st-RTD and 2^nd-ARD increased) with increasing grazing intensity.  Functional tradeoffs were found between plant individual biomass and plant density, and between leaf area and leaf density under grazing.  However, community-weighted mean SRL (SRL_CWM) and ARD (ARD_CWM) of different root orders exhibited functional coordination under grazing pressure.  SLA_CWM and SRL_CWM also showed synergistic responses to grazing.
4.  The ITV plays a predominant for the changes in above- and below-ground functional trait composition at the community level.  However, changes in mean trait values among plant species with different resource use strategies were mainly triggered by species turnover.  For species with different resource use strategies, grazing exhibited a coordinated effects on ITV but an offset effect on species turnover.
5. Synthesis. Our results demonstrate that both the above- and below-ground trait composition of plant communities shifted toward conservative strategy under long-term grazing.  This study highlights the effects of ITV and species turnover that govern the grazing-induced changes in functional trait composition of plant communities, and has important implications for grazing management in semiarid grasslands.

Study site and experimental design

This study was conducted at the Inner Mongolia Grassland Ecosystem Research Station (IMGERS, 43º38'N, 116º42'E) of the Chinese Academy of Sciences, located in the Xilin River Basin of Inner Mongolia, China.  The altitude is about 1200 m above the sea level (Bai et al. 2004).  The study area has a semi-arid continental climate, with the mean annual temperature of 0.5 ºC and mean annual precipitation of 334.6 mm (1970-2010). About 60-80% of the precipitation occurs in the growing season (May to August). Loamy sand is the main zonal soil types in the experimental area (Calcic Chernozem according to ISSS Working Group RB, 1998).  In this area, the dominant species are the perennial bunchgrass Stipa grandis and the perennial rhizomatous grass Leymus chinensis.

The experimental plots were located at the Sino-German grazing experimental site and were established in June 2005 (Schönbach et al. 2011). The gazing manipulation experiment has seven levels of grazing intensity (GI): 0, 1.5, 3.0, 4.5, 6.0, 7.5 and 9.0 sheep ha^-1. Each grazing intensity level included one standard plot, given the high intrinsic spatial heterogeneity in soil and plant community composition between the plots and high management cost (Schönbach et al. 2011). Size of each standard plot was 2 ha. To achieve a minimum of six sheep per plot, the plot size was 4 ha in GI=1.5. To prevent sheep migration between plots, we isolated each plot via using fences at the beginning of the experiment.  During the period of vegetation growth from June to September, sheep were kept in each grazing plot to graze continuously.  

Field sampling

Plant functional traits of co-occurring species were measured by soil core method (Fig. S1) in August 2013 when most species were mature (Bai et al. 2008). In each grazing plot, three enclosed cages (4 m × 4 m) were randomly set up to ensure that the plants in the cages are not disturbed by grazing in 2013. In each cage, three cores (with a diameter of 20 cm and a depth of 50 cm) were set up, thus, a total of 9 soil cores were excavated for each grazing plot. The soil cores were kept 40 cm underground and 10 cm above ground to obtain most roots, given that most species are shallow root system in the Inner Mongolia grassland (Cheng et al. 2015). We measured the plant height (PH, for vegetation branch) and plant density (PD) of each species in soil cores before they were taken. After the soil cores were fully irrigated, plants were taken out completely from the plots and transported to the laboratory to obtain the leaf and root functional traits. 

Traits measurement

The plant functional traits of all plant species were measured at individual-level in a given soil core. We separated and recorded all the vegetative leaves for each plant individual, then measured the projected area of all these leaves by using Li-3100 (Li-COR, Lincoln, NE, USA). LA was the total leaf area divided by leaf number, and SLA was the ratio of the total leaf area to oven-dried mass (oven-dried at 70 ºC for 48 h to a constant mass). The soil column was immersed in water for 24 h and then carefully washed to obtain the roots of co-occurring species. On the basis of root color, texture and the connection with its shoots, we identified and separated the plant individual-specific root systems for each species.  Following the steps described in Pregitzer (2002), the whole root systems were completely classified into three orders (the most distal roots represent the first-order roots) for all species occurring in grazing plots (Cheng et al. 2015). After classification, first-, second- and third-order roots of plant individual were scanned separately by using the Epson Perfection V750 Pro (SEIKO EPSON, Nagano, Japan) with the resolution of 300 dpi. The roots were dried and weighted separately after scanning, and the root images were analysed by WinRHIZO Pro 2019 software (Regent Instruments, Quebec, Canada) to obtain root traits including root length, ARD, and volume for each root order. SRL was calculated as the total root length divided by the root dry mass, and RTD was calculated as the root dry mass divided by the root volume. Finally, the shoot and other portion of underground (rhizome, and the connection between root and shoot) of individual plant was oven-dried to obtain the biomass of different plant organs, including plant aboveground part, leaf, and roots of different orders.  In this study, each soil core can be considered as a meta-community. Each soil core contained two to eight plant species, and 16 plant species were measured in this study.