The nitrogen and snow addition experiment was established in the Horqin Sandy Grassland of eastern Inner Mongolia, China (42°55′N, 120°42′E) on 5 October 2009 and ran through 10 October 2014. The total soil nitrogen content in the top 0–30 cm ranged from 0.057 to 0.199 (%), and the soil bulk density ranged from approximately 1.29–1.59 g cm^-3 (Mao et al., 2012). Soil nitrogen and snow were treated as limiting factors that reflect both stressed and unstressed habitats depending on their availability. For the soil nitrogen, stressed conditions represented ambient conditions due to the characteristically low nutrient levels (represented by N₀). For the unstressed condition, we added extra 536 g (equal to 100 kg ha y^-1) of (NH₂)₂CO to each plot to represent high nutrient levels (represented by N₁). For snow, control plots received no additional snow (represented by W₀). For the high snow level, we added snow equivalent to 100 mm precipitation (weight conversion) applied during the winter (represented by W₁).

Nitrogen was added each May from 2009 to 2014, and snow was added in the winter from December 2009 to April 2010, from December 2010 to April 2011, and from December 2011 to April 2012. Snow was collected outside of the experimental plots during snowfall in winter and added to the each W₁ plots evenly. The snow added was equal to 30% of the average annual precipitation during the growing season from the last 20 years (Mao et al., 2012).

We used a randomized complete block design, consisting of 24 10 m × 10 m plots. We employed two nitrogen and two snow treatments, as well as their interactions. The four experimental treatments are as follows: (1) control (N₀W₀), (2) high nitrogen and low snow (N₁W₀), (3) high nitrogen and high snow (N₁W₁), and (4) low nitrogen and high snow (N₀W₁). Each treatment was replicated six times.

Vegetation surveys were carried out during peak biomass accumulation in August from 2010 to 2014. In order to prevent edge effects, the quadrats were set at least 1 m from the edge, with the size of each sampling plot 1 m × 1 m. Species richness, abundance and the e were measured in each plot. Aboveground biomass was clipped, dried at 65°C and weighed to the nearest 0.01 g. Root-type and cluster perennial species, like Pennisetum purpureum and Cleistogenes squarrosa, were used to calculate the number of ramets. All communities were located in a grazing-free enclosure.

Leaf samples were collected in August 2010–2014 from the same plots used to assess SLA (kg m^-2). We randomly chose three individuals of the selected plants from the six plots of each treatment for leaf functional traits analysis. A single fully expanded mature leaf was selected from three individuals of each species and used to determine leaf area and dry mass (drying for 48 h at 65°C) (Cornelissen et al., 2003; Perez-Harguindeguy et al., 2013). Selected species were determined to be dominant based on their abundance by weight within each given community. The percent abundance of dominant annual species was higher than 80% of the community among all plots.

We selected SLA at the species level to describe the total FD at the community level. We chose this trait because it is directly related to plant competitive ability with respect to different levels of nitrogen availability, and reflects trade-offs between tolerances of environmental stressors (e.g., nutrient or climatic) (Lepš et al., 2011). High SLA also indicates a fast rate of resource uptake (Wright et al., 2004; Laughlin, 2014), and thus increases in SLA will likely enhance plant-atmosphere gas exchange (Lepš et al., 2011; Koerner et al., 2016).

The total community FD, within species FD, and between-species FD were quantified for each treatment (utilizing R code by de Bello et al., 2011). This method can be used with multi and single trait approaches (Albert et al., 2010; de Bello et al., 2011). The contribution of intraspecific variability to community variability was assessed via T_IP/IC, which is the ratio of within-species (IP) variance to total community variance (IC) (Violle et al., 2012). All FD was calculated on the basis of three groups of species data; the first set of data was based on all species in each community. The second data set was based only on the annual plants that were determined as dominant in the community. The third data set was based only on all perennial species that were determined as subordinate in the community (Mao et al., 2012). A null modeling method was used to analyze the trait overlap for each community within the four different treatments.

To make sure that the observed pattern of T_IP/IC responses to different environmental variables was non-random (Mason et al., 2011; Kumordzi et al., 2015a), trait overlap (i.e., T_{IP/ICexpected}) was calculated by randomizing SLA values between all individuals in the given community. We utilized the equation T_IP/ICSES = [T_{IP/ICobserved} – mean (T_{IP/ICexpected})]/sd (T_{IP/ICexpected}) (for further details see Kumordzi et al., 2015a). We then developed a large matrix that takes into account both species relative abundance and richness within the community.

We analyzed the effect of nitrogen and snow on biomass of annual and biomass of perennial species using analysis of variance (ANOVA). We also analyzed the intraspecific FD, interspecific FD and total FD at the community scale (de Bello et al., 2011). A null model was used to analyze the significance (p < 0.05) of intraspecific FD, interspecific FD and total FD at the community scale, annual species scale and perennial species scale, and was calculated as the difference between the observed and expected values (Kumordzi et al., 2015a). We used both R statistical software (R Development Core Team, 2014) and SPSS (version 19.0; SPSS Inc., Chicago, IL, USA) to analyze the data.

We observed shifts in the trait overlap between annual species and perennial species in response to key global change drivers of the study region, of which further underscores the potential role of niche packing as a central mechanism for the coexistence of plant species (Kraft et al., 2008; Mason et al., 2011; Le Bagousse-Pinguet et al., 2015; Fajardo and Siefert, 2016). Due to land desertification and subsequent wind erosion, the available soil nitrogen in this grassland is very low (Zhou et al., 2008). As a result of this limitation, the majority of the dominant species in this grassland are annual species, which are also sensitive to changes in nitrogen availability (Bai et al., 2010; Mao et al., 2012).

Indeed, in the plant communities of our study, the dominant annual species were highly sensitive to nitrogen addition (Mao et al., 2012). We observed that increased species trait overlap within the community due to nitrogen addition was mainly caused by increases to the ITV of the dominant annual species. In this nutrition-poor habitat, nitrogen addition alleviates nutrient limitation and thus alters the role of environmental filtering as a mechanism of community assembly (Chapin et al., 1986; den Haan et al., 2016), as nitrophilic plants may show more advantages as nitrogen levels increase (Menge et al., 2015). Moreover, coexisting species are likely to exhibit more nitrophilic traits in the community after nitrogen addition, and as a result trait overlap will likely increase.

The responses of the subordinate perennial species to nitrogen addition was different than annual species (Supplementary Figure S1). It is likely that light competition increased after nitrogen addition (Hautier et al., 2009; Okubo et al., 2012). This may then increase the upper limit of the SLA range, and further increase ITV via greater trait overlap among individuals. However, biomass of the dominant perennial species increased quickly after nitrogen addition and thus may have ultimately heightened competitive exclusion of perennial species in the community. This dynamic may operate to reduce trait overlap. The trait overlap of perennial species is likely the result of the trade-off between the overlap increase caused by nitrogen addition and the overlap decrease caused by the increase in the intensity of competition. The trait overlap of perennial species increased after nitrogen addition (Le Bagousse-Pinguet et al., 2015), which indicates that environmental filtering played a more obvious role in driving increased trait overlap as compared to interspecific competition (Kumordzi et al., 2015a). These results are consistent with hypothesis i.

We did not find a detectable decrease in interspecific FD variation after nitrogen addition. This result contradicts with the prediction that nitrogen addition will impact species composition, and thus potentially increase the functional trait diversity at community scale (Weiher et al., 2011; Siefert and Ritchie, 2016) by extending the niche space of species in the community at the interspecific level (Kumordzi et al., 2015a). There may be two reasons for this result. First, the dominant species in our sandy grassland communities are annual forb species. The intraspecific variability of these species increased sharply after nitrogen addition, which may have obscured the impacts caused by interspecific FD variation. For example, in Qinghai Tibet the dominant species maximize resource use after nitrogen addition through intraspecific variability in SLA to occupy greater niche space, thus allowing more nitrogenous plants to coexist (Li et al., 2016) and leading to higher T_IP/IC in the community. Second, the species richness is very low in our study area. In species poor communities, the intraspecific variation is typically much higher than species-rich communities (Mao et al., 2012), while in comparison the interspecific variation is relatively small.

We found that snow did not change the intraspecific FD variation and trait overlap of the subordinate perennial species, which is not consistent with hypothesis ii. However the dominant annual species in the community increased their intraspecific FD variation and trait overlap after snow addition. The reason for the weak effects of snow on perennial species may be that, firstly, snow addition in this study did not directly affect the growth of perennials. The increase of snow in winter increases the thickness of snow, soil temperature and soil water supply in early spring, and thus affects plant growth and biomass allocation (Chen et al., 2008; Wipf, 2010). However, in the study area of Inner Mongolia, strong continuous winds blow accumulated snow rapidly, while the sandy soil also has poor water holding capacity (Yao et al., 2013). As a consequence, water produced after snowmelt may migrate to deep soil layers very quickly, where plant roots may not be able to fully access this resource.

Second, although snow addition slightly increased available soil available nutrients, this did not impact perennial species. This may be due to the fact that the perennial species in our study were mostly clonal plants, and that redistribution of available nutrients among individual clones could reduce the importance of small-scale spatial heterogeneity for FD and community assembly (Gundale et al., 2011). Patterns of snowfall are likely to change within the Inner Mongolian region, with these changes are less certain as compared to temperature change. One key prediction is that the amount of snowfall will increase each time, yet that the interval between events will also increase (IPCC, 2013). In this snowfall change scenarios, different functional groups of plants show different response strategies. Thus it may be the case that the response of annuals to snow changes is more pronounced in arid and semiarid regions, while perennial responses may comparatively insensitive to snow changes (Mao et al., 2012).

The change of trait overlap of annual species was more obvious after snow addition, and the variation in this pattern was similar to the changes at the community scale. This result is consistent with the ‘mass ratio hypothesis,’ which posits that the function of the ecosystem is largely determined by the traits of the dominant species or functional groups (Grime, 1998; Johnson et al., 2015). Therefore, the ITV of SLA for the dominant annual species is likely to play a more important role in determining patterns of community level FD than that of subordinate species (Lavorel and Garnier, 2002; Mao et al., 2012; Volf et al., 2016). The different variance pattern of annual species and perennial species suggests that changes to the niche space occupied by the dominant species or functional group will likely affect niche packing at the community level (Breza et al., 2012; Li et al., 2016). These contrasting responses of intraspecific FD and interspecific FD between dominant and subordinate species also suggests that species niche packaging plays an important role in maintaining the FD of plant communities (Johnson et al., 2015).

Although the effects of snow addition alone on intraspecific FD and interspecific FD of sandy grassland ecosystems were weak, there was an interaction between snow addition and nitrogen addition. This interaction remained 1 or 2 years after cessation of snow addition. Even if the interaction between snow addition and nitrogen addition are mostly driven by the effect of nitrogen alone, increasing snowfall amount may lend insight to potential variability in the effects of nitrogen on the community (Figure 4). For instance, the interaction of snow addition and nitrogen addition on intraspecific variability are significant, while the effect on interspecific FD was weak. There were temporally different responses between annual species and perennial species in N₁W₁ treatment. For instance, the interaction of nitrogen and snow had no significant effect on intraspecific FD and interspecific FD of annual species in 2010 and 2012 when both nitrogen and snow were added, yet affected the intraspecific FD and interspecific FD of perennial species. This is clearly different from the addition of nitrogen alone.

The response of intraspecific variation, T_IP/IC and biomass of the dominant annual species to nitrogen addition were related to trends at the community level, and the response of the annual species differed from that of the subordinate perennial species (Figure 5 and Supplementary Figure S2). This result supports our third hypothesis. Previous studies have shown that communities with higher intraspecific FD, potentially indicative of high phenotypic plasticity or genetic variation, may also have increased capacity to respond to environmental change through changes in traits within the same species (Grime, 2006). In this study, the biomass allocation of dominant species changed after nitrogen addition, and more nutrients were likely distributed to the leaves, which expanded the range of leaf traits and increased leaf, species and community biomass. Conversely, the effect of nitrogen addition on perennial plants was relatively weak and even negative (Mao et al., 2012). We also used standardized major axis tests to analyze the variation of annual species and perennial species after nitrogen addition. The results indicate that the resource utilization axis of the annual species shifted to the direction of nitrogen use after nitrogen addition, and that the community showed more nitrophilic functional traits (e.g., high SLA and leaf nitrogen content) as the ratio of leaf weight to total weight increased significantly (Mao et al., 2012). This effect in turn increased the biomass of the community, without significant changes in biomass allocation of perennial species after nitrogen addition.

The biomass of the community and biomass of annual species increased at nitrogen addition treatment and N₁W₁ treatment, yet interestingly the biomass of perennial species decreased significantly. This contrasting result suggests that different community components, such as dominant versus subordinate species, will likely have differential responses to environmental change drivers (Fargione and Tilman, 2006; Kumordzi et al., 2015b). Moreover, depending on a species role in the community, the response of a species to environmental change may or may not translate to detectable impacts at the community level. For example, community biomass was mainly affected by the biomass of dominant annual species, which showed the same trend with the T_IP/IC of annual plants, despite a negative trend with the changes of subordinate perennial plant biomass. This further suggests that changes in the functional traits and productivity of dominant and subordinate species may help to more accurately understand which component of the community has a greater impact on community assembly and potentially ecosystem processes.