Study materials
Leymus chinensis and ten common species, including five perennial grasses (Achnatherum sibiricum, Agropyron cristatum, Cleistogenes squarrosa, Koeleria cristata, Stipa grandis), three perennial forbs (Allium ramosum, Anemarrhena asphodeloides, Potentilla acaulis), one sedge (Carex korshinskyi) and one semi-shrub (Artemisia frigida), were collected in a L. chinensis community in China (116°42′ E, 43°38′ N), and then acclimated for at least two months at Nankai University (China). At the same site, soil in 0-10 cm layer was collected. In addition, L. chinensis was used as the target species, and ten common species were used as neighbor species.
Experimental design
This experiment employed a two-factor design, including soil water condition (10-15% soil water content for non-drought treatment and 5-8% soil water content for drought treatment) and neighbor richness (1-, 3- and 6-neighbor species).
On July 2, 2018, 104 plastic pots (20 cm in diameter, 21 cm in depth) were filled with the homogenized soil (4.0 kg per pot). According the species density and their relative abundances in the sampled community, six individuals of L. chinensis and six individuals of neighbors were transplanted into each pot. For 1-neighbor species treatment, six individuals from one common species were transplanted, using 10 common species as replicates for each soil water condition. For 3-neighbor species treatment, two individuals from each of three different common species were transplanted. For 6-neighbor species treatment, one individual from each of six different common species was transplanted. The treatment of 3-/6-neighbor species had 12 replicates per soil water condition, and there were the same plant assemblages under both soil water treatments. All species used except L. chinensis had the same probability in any pot within each neighbor richness treatment (supplementary data Table S1). To test the competitive effects of neighbors on L. chinensis, the monoculture of L. chinensis (18 replicates) per soil water condition was set up, with 12 individuals per pot. All individuals of the same species used in this experiment were roughly the same size.
Non-drought and drought treatments were carried out from July 15, 2018 to the end of experiment, and soil water condition per pot was monitored by an ECH2O Check except for the winter when the plants wilted. The experimental pots were randomly placed under the large-scale rain-proof shelter at Nankai University and changed the position every other week.
Data collection
Considering that the main limiting factors in the semi-arid grasslands are light and nutrients as well as water, three functional traits related to light and nutrients acquisition were measured every year following the standard protocols (Pérez-Harguindeguy et al. 2013), including plant height (hereafter Height), specific leaf area (SLA) and leaf dry matter content (LDMC). Height is the vertical distance from the upper boundary of plant photosynthetic tissue to ground level. SLA is the ratio of the leaf area and leaf dry mass, and LDMC is the ratio of the leaf dry mass and water-saturated fresh mass. For each species, these traits were measured with at least 3 replicates under each soil water condition, and mean values of each trait were used to calculate functional profiles of plant communities.
The aboveground shoot that was 2.54 cm higher than ground level were harvested on October 15, 2018, and June 25, 2019, respectively. The tiller number and aboveground shoot dry biomass per species were recorded for each pot every year. The tiller number of each species was used as species abundance to calculate functional profiles of plant communities; and the biomass of target species and neighbor species were used for the assessment of competitive effects.
Data analyses
Competitive effect index
To quantify the competitive effect of neighbors on target species L. chinensis, the corrected index of relative competition intensity (CRCI) was calculated (Oksanen et al. 2006), within each soil water condition (2), neighbor richness (3) and year (2).
CRCI = arcsin (Bmonoculture – 2Bmixture)/max(Bmonoculture, 2Bmixture)
where Bmonoculture is the mean biomass of L. chinensis in monoculture, and Bneighbors is the biomass of L. chinensis in mixture with neighbors. We doubled Bneighbors because the initial density of L. chinensis in monoculture was twice of that in mixture. The positive CRCI value indicates competition, while the negative value indicates facilitation. CRCI shows a linear relationship in wide ranges of competition and facilitation intensities, so that it can detect the linear trend of competition intensity among plants, and derive an unbiased interval estimate of plant interaction intensities (Oksanen et al. 2006).
Functional profiles of plant communities
Three functional profile indices, including CWMtrait absolute distance, CWMtrait hierarchical distance and FDis, were calculated with the package ‘FD’ in R (R Core team 2020), within each soil water condition (2), neighbor richness (3) and year (2). CWMtrait hierarchical distance reflects fitness differences and CWMtrait absolute distance reflects niche differences between neighbors and the target species for each trait (Height, SLA and LDMC), and FDis which is a multi-trait index reveals the occupancy of niche space of neighbors (vacant niche in resource use).
S: neighbor richness; NSRAi: the relative abundance of neighbor species i in a certain pot; |Ti –Ttarget| and (Ti –Ttarget): the absolute and relative value of the trait difference between neighbor species i and the target species L. chinensis, respectively.
S: neighbor richness; ai: the abundance of neighbor i; zi: distance of neighbor i to c in a certain pot; Tij: the mean value of trait j of neighbor species i.
Statistical analyses
The data of CRCI and functional profile indices were log-transformed prior to analyses to meet the normality distribution and homogeneity of variances.
To assess the difference between CRCI and zero in each pot, one-sample t-test was performed. CRCI value higher than zero indicates competition, and lower than zero indicates facilitation.
To explore the effects of soil water condition, neighbor richness and their interactions on functional profile indices (CWMtrait absolute/hierarchical distance of each trait, FDis of neighbors) and CRCI, linear mixed models (LMMs) were used for data collected in each year, with soil water condition and neighbor richness as fixed factors, and plant assemblage as the random factor. Then, the significance of differences was assessed among treatments by Tukey post hoc analysis. These analyses were conducted using the packages “nlme” and “emmeans” in R, respectively.
To quantify the relative importance of predictors (soil water condition, neighbor richness, plant assemblage, each functional profile indices) for CRCI, a variance partitioning analysis was performed for data collected in each year. Before the analysis, collinearity between variables was tested with the variance inflation factor (VIF) and correlation coefficients of pairwise predictors in both year (Menard 1995; Xu et al. 2021), and no collinearity was found because of no VIF value being higher than 5 and no correlation coefficient being higher than 0.7 (supplementary data Fig. S1). Four models were assessed. Model 1: soil water condition and neighbor richness; model 2: predictors in Model 1 and CWMtrait absolute distance of each trait; model 3: predictors in Model 2 and CWMtrait hierarchical distance of each trait; model 4: predictors in Model 3 and FDis of neighbors. Then, these four models were fitted by stepwise procedures, and the predictors which did not improve the model-fitting were removed based on the Akaike information criterion (AIC). These analyses were constructed in R, with package ‘variancePartition’ for variance partitioning analysis and ‘MuMIn’ for model selection.
To disentangle the direct and indirect causal relationships between predictors (soil water condition, neighbor richness, plant assemblage, functional profile indices) and CRCI, Shipley’ s d-sep approach (Shipley 2013) was used to construct models for data collected in each year using the ‘piecewiseSEM’ package in R, with plant assemblage as a random factor. For the case that several models were accepted, the one with the smallest AIC was selected as the final model. The direct or indirect relationships between predictors and CRCI were expressed by standardized path coefficients (Grace and Bollen 2005).