Plant functional traits in alpine meadows are critical indicators of ecosystem responses to climate warming, yet dynamic changes in trait networks across warming gradients and their consequences for community functions remain poorly understood.

We conducted a 5-year multi-level warming experiment (low: +1.3°C; medium: +2.3°C; high: +3.3°C) in a Qinghai-Tibet Plateau alpine meadow, measuring 19 functional traits across three dominant species (Kobresia pygmaea, Elymus sibiricus, Oxytropis melanocalyx) to quantify warming effects on trait coordination, community functionality, and adaptive strategies.

Warming amplitude dictated trait response directions. Critically, trait responses exhibited distinct threshold dynamics: under low warming, highly plastic physiological traits shifted rapidly while stable structural and morphological traits remained largely unaffected, thereby maintaining community functional structure. However, high warming (+3.3 °C) triggered severe oxidative stress, evidenced by a 28% increase in malondialdehyde, and caused divergent species-specific adjustments in specific leaf area (SLA). At the community level, acquisitive traits increased under moderate warming but declined at high warming, coinciding with a 34% reduction in functional diversity. Notably, trait networks shifted from energy storage to structural maintenance under warming, with modularity increasing by 41% and connectivity decreasing by 29%, reflecting disrupted trait synergies. Furthermore, warming weakened inter-trait correlations between physiological and structural traits, indicating that environmental stress disrupted trait coordination and amplified trade-offs between stress tolerance and growth-related traits.

The coordination of functional traits in alpine meadow plants exhibits a nonlinear response to climate warming, where moderate warming enhances resource acquisition efficiency, while high warming alters trait coordination patterns and diminishes functional diversity. By revealing how the magnitude of warming controls the trade-off between structural maintenance and physiological performance, the understanding of the mechanism by which warming gradients alter plant ecological strategies has been advanced. These results suggest that alpine meadows may initially buffer moderate warming through trait plasticity, but are at risk of functional instability under extreme warming conditions, providing a critical threshold for predicting ecosystem transitions.

Before data analysis, the normality and homogeneity of variance were tested, and logarithmic transformation was performed on non-conforming data. Community-weighted mean trait values (CWM) and functional dispersion index (FDis) were calculated using the ‘FD’ package in R based on species relative abundance (Garnier et al., 2004; Laliberté & Legendre, 2010). Crucially, prior to computing FDis, the trait matrix was standardized (z-score transformation) using the stand.x = TRUE argument. This ensured equal contribution of all traits, whereas CWM used unstandardized data to retain biological units. Community-level indices were based on the three dominant species, capturing >85% of community abundance, which aligns with standard trait ecology protocols and robustly captures primary functional responses. To assess the effect of warming treatments on these indices, a one-way analysis of variance (ANOVA) was performed. P-values were adjusted using the Benjamini-Hochberg False Discovery Rate (FDR), followed by Dunnett’s post-hoc tests to compare warming treatments against the control.

To examine whether there are significant differences in traits among different temperature gradients, principal component analysis (PCA) was performed using the ‘vegan’ package in R, and PERMANOVA was conducted to test for significance. The first two principal components (PC1 and PC2) were selected to identify trait distribution differences. Pearson correlation analysis was used to analyze the correlations among plant traits, and the ‘linkET’ package in R was employed for visualizing the correlation heatmap. The Mantel tests for quantifying the multivariate associations between temperature/species and plant trait composition were implemented using the ‘mantel’ function in the ‘vegan’ package. To further validate the univariate effects of temperature and species on individual traits, two-way ANOVA was performed with temperature treatment and dominant species identity as fixed factors, including their interaction term. The ‘aov’ function in R was used for the analysis, and Levene’s test was conducted to verify the homogeneity of variance assumption. Similarly, FDR correction was applied to the P-values from the two-way ANOVAs.

Finally, to explore the relationships among various functional traits, we chose to construct PTNs. Four overall parameters (modularity, edge density, average path length, diameter) and four node parameters (degree, hub, betweenness, closeness) quantified trait relationships. Modularity describes separation between modules, indicating network division into functional groups. Edge density is the ratio of actual edges to maximum theoretical connections, with higher values indicating closer relationships. Average path length is the average shortest path between all nodes, while diameter is the maximum shortest distance between connected nodes; both increase with higher trait independence. Degree refers to the number of edges connected to a node, indicating centrality within the trait network (He et al., 2020; Li et al., 2021). The formula for calculating hub trait is Hub = Degree × [ / (n − 1)], where |r| is the absolute value of the correlation coefficient, and n is the number of elements in the network. A higher Hub value indicates that the trait is tightly connected and has an important regulatory role in the overall plant phenotype (Wang et al., 2022). Betweenness reflects the shortest paths passing through a node, acting as a bridge connecting functional modules (Wei et al., 2023). Closeness is the reciprocal of average distance from a node to all others, reflecting close connections across the network. All parameters were calculated using the R package igraph.

Based on the Pearson correlation coefficient (r) matrix of plant traits under different warming gradients, a threshold of |r| > 0.2 and P < 0.05 was set (Li et al., 2021) to avoid spurious correlations. In the correlation coefficient matrix, the r values that met the criteria of |r| > 0.2 and P < 0.05 were set to 1, and all others were set to 0. Thus, the correlation coefficient matrix of traits was transformed into an adjacency matrix (Kleyer et al., 2019). . This adjacency matrix was then weighted by the absolute value of the correlation coefficient (|ri,j|) between any pair of traits to reflect connection strength (Rao et al., 2021). Subsequently, the PTNs were visualized using Cytoscape 3.10.0 (Shannon et al., 2003), and the trends in node parameters within the PTNs under different temperature treatments were plotted using Origin 2022.