China’s South to North Water Transfer Project (SNWTP) is a world-famous hydraulic project, which aims to transfer water resource from southern part of China to the north and northwest regions. It involves water redistribution among different basins and may cause a series of changes in ecological conditions of impounded lakes and channel rivers ecosystems (Zeng et al., 2015; Guo et al., 2020; Sun et al., 2021). SNWTP consists of ERSNWTP, Central Route of South-to-North Water Transfer Project (CRSNWTP), and West Route of South-to-North Water Transfer Project (WRSNWTP). ERSNWTP pumps water from the Yangtze River in Yangzhou, Jiangsu Province, employs Lake Gaoyou, Hongze, Luoma, Nansi and Dongping for water impoundment, utilizes the Grand Canal and its parallel rivers to transfer water from south to north (http://nsbd.mwr.gov.cn). Nansi Lake was divided into upper and lower region by an artificial dam since 1960s, the water is first transported into the lower region and then is lifted and moved to upper region during water diversion. The morphometrical and limnological characteristics of these impounded lakes were detailed in Table S1 .

P. crispus is dominant and constructive species of submerged macrophyte community in the impounded lakes and channel rivers of ERSNWTP in China during winter and spring (Zhu et al., 2022). P. crispus population were sampled in five impounded lakes and Grand Canal from 10 April to 12 May 2018 when the biomass of P. crispus reached its peak in these lakes. 74 sites were surveyed in P. crispus-dominated region using site-quadrat method. The number of sampling sites in impounded lakes and Grand Canal were determined based on the distribution area of P. crispus, 3 sampling sites were selected in every Grand Canal section between each two impounded lakes ( Figure 1 ). P. crispus population were sampled using a rotatable reaping hook covering 0.2 m², three replicated quadrats at each site. At each quadrat, the collected macrophytes were rinsed in lake water and then one healthy P. crispus individual was randomly selected for subsequent functional traits measurement. Prior to plant sampling, water and sediment samples were sampled from 0.5 m below the water surface using a Schindler sampler and the top layers of sediment using a Peterson sampler, respectively. These samples were kept in a container with ice bags and taken back to the laboratory for subsequent analysis.

For all sites, water pH, dissolved oxygen (DO), conductivity (Cond), total dissolved solid (TDS), turbidity (Tur), Secchi depth (SD) and underwater photosynthetic active radiation (PAR) were recorded in situ. T, DO, pH, Cond, TDS and Tur were measured with a YSI EXO2 (Yellow Springs Instruments, USA). SD was measured by Secchi disk (30 cm diameter). Underwater PAR was recorded at intervals of 0.25 m from water surface to 1.0 m by an underwater radiation sensor (UWQ 10250) coupled with a LI-COR data logger (Li-1500, Li-Cor Company, USA). The light extinction coefficient (K) was computed based on the equation: K =(1/d) ln (I₀ /I_d ), where I_d and I₀ are PAR value at the water depth d and water surface (Krause-Jensen and Sand-Jensen, 1998).

Water samples were used to measure total nitrogen (TN), total phosphorus (TP), ammonium nitrogen (NH₄ ⁺-N), nitrate nitrogen (NO₃ ^–N), orthophosphate (PO₄ ³⁺-P), permanganate index (COD_Mn), and chlorophyll a (Chl a) in the laboratory according to the standard methods (Huang et al., 1999).

Sediment samples were used to analyze water content (S_W), organic matter content (S_O), total carbon content (S_C), total nitrogen content (S_N) and total phosphorous (S_P). S_W and S_O were determined using the methods detailed by Jin and Tu (1990). Sediment samples were air dried to constant weight and then ground into fine powder for elemental determination. S_C and S_N were measured by an element analyzer (Multi N/C 2100, Jena Instruments, German). For S_P measurement, sediment samples were digested with sulfuric acid/perchloric acid and then measured with the ammonium molybdate ascorbic acid method (Huang et al., 1999).

Eighteen functional traits of 222 P. crispus individuals sampled in five impounded lakes and Grand Canal were measured during the study period: plant height (H), stem branch (SB), internodes number (NN), internode length (NL), internode diameter (ND), relative stem length (RSL), relative leaf area (RLA), leaf width (LW), leaf area (LA), leaf number (LN), leaf thickness (LT), leaf length (LL), dry plant weight (DPW), dry leaf weight (DLW), dry stem weight (DSW), ratio of stem weight to leaf weight (SLR), leaf dry mass fraction (LMF) and stem dry mass fraction (SMF). Details of the functional traits and determination methods were described in Table 1 .

Submerged macrophytes PTNs were constructed to reveal the difference in trait associations among impounded lakes and Grand Canal. In PTN, plant traits are nodes, and trait-trait connections are edges. First, a matrix of trait-trait coefficients was computed using Pearson’s correlation coefficients (Poorter et al., 2014; Kleyer et al., 2019). To avoid spurious relationships among traits, trait-trait coefficients were assigned as 1 if significant at P<0.05 level and regarded as 0 if insignificant. Then, an adjacency matrix A = [a_i , _j ] with a_i , _j ∈ [0,1] was yielded. Thus, PTN only showed the presence and absence of correlations among traits. Next, PTNs were constructed and network parameters were calculated using the R package “igraph”. Finally, PTNs were visualized in Cytoscape 3.8.0 using the Prefuse Force Directed OpenCL layout (Shannon et al., 2003).

Three node parameters were selected to describe the properties of functional traits within PTNs. Degree (k) is calculated as the sum of the edges over all neighbors of the focal trait in the network. Plant traits with high k values can be regard as “hub traits”. Closeness (C) is the reciprocal of the mean shortest path between a focal trait and all other traits. Traits with high C values are closely associated with other traits. The betweenness (B) of a focal trait is determined as the number of shortest paths between pairs of traits that contain the focal trait. Traits that have a high k are generally considered as mediators in the PTNs ( Figure 2 ).

Four metrics were used to show the overall topology of PTNs. Edge density (ED) represents the density of the connected edges among nodes in a network, that is, the proportion of actual correlations among traits out of all possible correlations. Average path length (AL) is the mean shortest path between all traits in the network. PTNs that have higher AL indicate greater overall independence among traits. Average clustering coefficient (AC) is the mean of the clustering coefficients of all node traits in PTNs. PTNs with higher AC are less easily separated into several different modules. Modules are defined as sets of traits that present covariation among themselves. Modularity (Q) measures how well a network is divided into modules. The trait network with higher modularity may give macrophytes more flexibility to adjust its functioning to changing environments ( Figure 2 ). More details on network parameters were presented by He et al. (2020) and Kleyer et al. (2019).

To describe the importance of traits within PTNs, k and C were used to determine “the connectedness” of each trait, B was used to signify the “centrality” of each trait. Traits that have high k and high C were “hub traits”, and high B were “mediator traits”.

To describe the overall topology of PTNs, ED and AL were used to represent the “connectivity” of the PTN, AC and Q were used to characterize the “complexity” of the PTN. PTNs with low ED and high AL are “looser”, that is, they have an overall low level of covariation among traits. A high AC indicates fewer modules, and a lower PTN complexity. The Q of the PTN describes the degree of separation among modules.

One-way ANOVA with Duncan’s test (at the 0.05 significance level) was used to evaluate the difference of functional traits and also the environmental parameters among impounded lakes and Grand Canal. All data were tested for normality and homogeneity, when necessary, data were log₁₀ and sqrt transformed to meet assumptions before statistical analysis. Linear regression was used to examine the relationships between traits variations and three node parameters and four topological metrics. To explore the determining environmental factors of PTNs, Pearson’s correlation was performed to detect the significant environmental factors related to parameters of PTNs, then one or multiple variable linear regression was computed using PTN topology metrics (i.e. ED, AL, AC, Q) as the dependent variable, and the significant environmental factors detected by Pearson’s correlation as the independent variables.

Cond, pH, DO, TDS, Tur, SD, K, Chl a, TN, NO₃-N, NH₄-N, TP, PO₄-P, S_w, S_o, S_C and S_N differed significantly among impounded lakes and Grand Canal except COD_Mn and S_P ( Table S2 ).

All 18 functional traits were significantly different among impounded lakes and Grand Canal ( Table S3 ). In terms of the mean coefficient of variation (CV) of traits from high to low, the order of 18 functional traits was stem branch, dry leaf weight, relative leaf area, dry plant weight, leaf number, dry stem weight, internodes number, leaf thickness, ratio of stem weight to leaf weight, relative stem length, internodes length, plant height, leaf area, leaf dry mass fraction, stem dry mass fraction, leaf length, internode diameter, leaf width ( Table S4 ). The results showed that leaf-related traits and organ mass allocation traits were sensitive traits that varying with various environment.

The most connected traits (with higher k and/or C) were relative leaf area, dry stem weight and internodes number in Grand Canal; relative leaf area, plant height, dry plant weight, stem branch and dry leaf weight in Gaoyou Lake; dry plant weight, dry stem weight and internodes number in Hongze Lake; relative leaf area, dry stem weight and dry leaf weight in Luoma Lake; leaf thickness, dry plant weight, dry leaf weight, dry stem weight and internode diameter in lower Nansi Lake; dry plant weight, dry leaf weight, dry stem weight, internodes number and internode diameter in upper Nansi Lake; and leaf number, dry plant weight, leaf area and relative stem length in Dongping Lake ( Figures 3 , 4 ). These results indicated that seven natural P. crispus L. populations showed different properties of functional traits within PTNs among impounded lakes and Grand Canal.

The mediator traits (with high B) were relative leaf area and ratio of stem weight to leaf weight in Grand Canal; plant height, internode length and relative leaf area in Gaoyou Lake; internode diameter, dry plant weight and leaf area in Hongze Lake; relative leaf area, dry stem weight and dry leaf weight in Luoma Lake; leaf length and stem branch in lower Nansi Lake; internode diameter and leaf area in upper Nansi Lake; and leaf number and leaf area in Dongping Lake ( Figure 5 ). Such traits acted as bridges in the PTNs by connecting other traits belonging to different modules.

PTNs showed different structures among impounded lakes and Grand Canal ( Figure 5 ). The number of edges within network was highest in lower Nansi Lake and lowest in Dongping Lake ( Figure 6 ). A module was a set of plant traits more closely connected to each other but less connected with traits outside the module. In our study, different modules with different colors were observed in the PTNs among impounded lakes and Grand Canal ( Figure 6 ). Two modules were detected in PTN in lower Nansi Lake, and three modules were identified for PTNs in others impounded lakes and Grand Canal.

Traits connections in PTNs differed among impounded lakes and Grand Canal. PTNs displayed higher edge density in lower Nansi Lake (0.641) but lower in Dongping Lake (0.275) and Luoma Lake (0.314). Similarly, average clustering coefficient was higher in lower Nansi Lake (0.842) but lower in Dongping Lake (0.453) and Luoma Lake (0.610). Conversely, PTNs exhibited higher average path length and modularity in Dongping Lake (average path length = 2.046 and modularity = 0.227, respectively) and Luoma Lake (average path length = 1.830 and modularity = 0.206, respectively) but lower in lower Nansi Lake (average path length = 1.380 and modularity = 0.067, respectively).

Pearson correlation showed that average path length significantly related to TP (r=-0.84, p<0.05), PO₄ ³⁺-P (r=-0.85, p<0.05) and DO (r=0.88, p<0.05); edge density was closely correlated with TP (r=0.87, p<0.05) and DO (r=-0.86, p<0.05); average clustering coefficient significantly correlated with DO (r=-0.89, p<0.01); modularity significantly related to DO (r=0.87, p<0.05).

To examine the effects of three significant factors on PTNs average path length, a backward multiple regression analysis was computed. This regression result showed that average path length was negatively correlated to TP, but positively to DO.

To explore the relative role of the two significant factors in explaining the edge density, a backward multiple regression analysis was computed. The result indicated that edge density increases with increasing TP and with reducing DO.

To reveal the relationships between average clustering coefficient and modularity and DO, a univariate linear regression analysis was performed. The results suggested that DO increases with increasing modularity and with reducing average clustering coefficient.

The coefficient of variation of traits showed that stem branch, dry leaf weight, relative leaf area, dry plant weight, leaf number and dry stem weight were more sensitive to varying environment than others traits (Siefert et al., 2015; Yang et al., 2020; Westerband et al., 2021). Noteworthily, these sensitive traits were the hub traits of most PTNs, which indicated that traits with high variability were more likely to be the hub traits of PTNs among impounded lakes and Grand Canal overall. These results can be explained by the finding that functional traits degree displayed significant increase with increasing variation of functional traits ( Figure 7 ). Traits with higher degree plays a central role that affects the whole-plant phenotype (Kleyer et al., 2019; He et al., 2020; Li et al., 2021), thus high variability of hub traits makes the trait network more responsive to diverse environments, which may be the reason why the PTNs can be used as an effective approach to clarify functional adaptation of submerged macrophytes to changing environmental conditions.

Although the hub traits of PTNs were different among impounded lakes and Grand Canal, relative leaf area and leaf number were the robust hub traits, which was in agreement with previous study on the forest leaf trait network shifts along latitude gradients, indicating that leaf thickness and relative leaf area were hub traits of leaf trait network (Li et al., 2021). Leaf is the primary organ of submerged macrophytes for light and nutrition resources acquisition (Wright et al., 2004; Shipley et al., 2006; Donovan et al., 2011). Submerged macrophytes tend to increase leaf area, leaf number and relative leaf area to adapt to underwater light limitation (Pierce et al., 2012; Fu et al., 2018; Chmara et al., 2019; Zervas et al., 2019; Liu et al., 2021). Indeed, relative leaf area is a part of the leaf economic spectrum (LES) and closely correlated with other traits related to resource acquisition during plant life history (Price et al., 2014; Reich and Cornelissen, 2014; Díaz et al., 2016).

Moreover, organ mass allocation traits, including dry plant weight, dry stem weight and dry leaf weight, were as central as leaf traits in PTNs. The result was consistent with recent finding that stem- and leaf- traits (stem ratio and leaf ratio) significantly related to biomass allocation were the hub traits in submerged macrophyte PTNs (Yuan et al., 2023). Previous studies revealed that plant height was closely related to stem height and leaf area (Price et al., 2014; Díaz et al., 2016), and the capacity for light and nutrient resource acquisition depends on the allocation of biomass to leaves and stem (Hilbert, 1990; Barko et al., 1991). Notably, shifts in biomass allocation is an important strategy for plants to overcome environmental heterogeneity (McConnaughay and Coleman, 1998; Madsen and Cedergreen, 2002). Submerged macrophytes usually adjust organ mass allocation for optimizing resource capture and minimizing imbalance in any critical environment. (Barko and Smart, 1988; Xie et al., 2010). Thus, organ mass allocation traits as hub traits in our study mainly explained that leaf and stem mass allocation determined light and nutrient resource capture, and shifts in plant biomass allocation happen throughout the whole life of submerged macrophytes (Chambers et al., 1989; Xie et al., 2004).

Furthermore, we noted that hub traits of PTNs are relative to the included traits. A morphological trait is more likely to be a hub if many morphological traits are included in PTNs; likewise, a physiological trait is more likely to be a hub if many physiological traits are included. Traits selection could influence the hub traits of PTNs, to an extent that needs deep study in future work.

In present study, higher edge density and average clustering coefficient, whereas lower average path length and modularity of plant trait network were observed in lower Nansi Lake. The opposite trend for these network parameters was detected in Dongping Lake and Luoma Lake. Regarding to the mean CV of traits of impounded lakes and Grand Canal from high to low, the order was lower Nansi Lake, Hongze Lake and Grand Canal, Gaoyou Lake, Dongping Lake, upper Nansi Lake, Luoma Lake ( Table S4 ). This finding demonstrated that plant trait network topologies were related to the mean functional trait variations of lakes, higher mean functional variation coefficients represented tight plant trait network, lower mean functional variation coefficients indicated loose plant trait network ( Figure 8 ). These results resulted from the significantly negative relationship between mean coefficients of functional variation of lake and modularity of PTNs ( Figure 8 ). Obviously, higher mean functional variation coefficients make smaller separation of trait clusters within the network and raises the possibility of functional traits connecting to other traits.

Numerous studies proved that submerged macrophytes would alter multiple functional traits and their relationships in different environmental conditions (Fu et al., 2018; Liu et al., 2021; Wang et al., 2021). In our experiment, water total phosphorus had a significantly positive effect on the edge density and negative effect on the average path length of PTNs, which indicated that trait coordination and integration and thus efficient resource transporting was gradually strong with an increase phosphorus concentration. Contemporary study focusing on the effects of phosphorus on the submerged macrophytes trait network topologies suggested that when water TP was< 0.2 mg/L, edge density increased with increasing TP concentration. In our study, the range of water TP was 0.014 mg/L~0.055 mg/L, the mean water TP was 0.022 mg/L, far less than 0.2 mg/L. Our results were consistent with previous conclusion (Rao et al., 2022). Additionally, dissolved oxygen was another influencing factor determining PTNs topologies. On the contrary, dissolved oxygen had significantly positive effects on average path length and modularity, and negative effects on edge density and average clustering coefficient. Numerous studies demonstrated that aquatic plant tended to elongate leaf length, reduce leaf thickness and decrease overground biomass accumulation in anoxic condition (Kay et al., 2004). Hub traits held the largest connections to others traits and occupied the centre in PTNs (Kleyer et al., 2019; He et al., 2020; Li et al., 2021; Li et al., 2022; Rao et al., 2022). Therefore, an environmental factor such as dissolved oxygen selecting on leaf-related traits and organ mass allocation traits will affect other traits, thus changing the PTNs topology. Thus, both the TP and DO together determine the PTNs topologies in impounded lakes. It should be mentioned that we only studied the unilateral effects of TP and DO, the interactional impact on PTNs topologies need further evaluation.

We found that TP was one of two determined factors influencing the plant trait network of impounded lakes of ERSNWTP. ERSNWTP pumps water from the Yangtze River in Yangzhou, utilizes the Grand Canal and its parallel rivers to transfer water from south to north. For Yangtze river in Jiangsu Province, water total phosphorus concentration ranged from 0.07 to 0.10 mg/L (Chen et al., 2020), which is much higher than the impounded lakes and Grand Canal. Total phosphorus concentration of impounded lakes will increase with the enduring operation of the water diversion, which will result in a strong trait coordination and an efficient resource transporting of submerged macrophytes. Furthermore, future studies should pay more attention on long-term effects of total phosphorus change induced by water transferring on submerged macrophytes PTNs. In addition, more attention should be pay to total phosphorus reduction before submerged macrophyte restoration projects in aquatic ecosystems from PTNs perspective.