a= (Cornelissen et al., 2003) ; ab= (Gong & Gao, 2019) ; ac= (Santamaría, 2002) ; ad= (Freschet et al., 2017) ; b= (Meng et al., 2015) ; c= (Dostálek et al., 2020) ; d= (Poorter et al., 2009) ; e= (Lavorel & Garnier, 2002) ; f= (Jung et al., 2010) ; g= (Loughnan & Gilbert, 2017) ; h= (Violle et al., 2009) ; i= (Butterfield & Callaway, 2013) ; j= (Rooney & Kalff, 2000) ; k= (Wehn et al., 2017); l= (Fu et al., 2018) ; m= (Bittebiere & Mony, 2015) ; n= (Gray & Brady, 2016) ; o= (Moles et al., 2009) ; p= (Tao et al., 2016) ; q= (Wang et al., 2016) ; r= (Martinez‐Almoyna et al., 2020) ; s= (Hutchings et al., 1997) ; t= (Rusch et al., 2011) ; u= (Louâpre et al., 2012) ; v= (Slade & Hutchings, 1987b) ; w= (Slade & Hutchings, 1987a) ; x= (Bittebiere et al., 2019) ; y= (Colom & Baucom, 2020).

Nutrients and phytoplankton quantity assay

For the measurements of water nutrient and phytoplankton concentrations (as depicted by the amount of chlorophyll a per mL of water), water samples were filtered through GF/F filters (0.7µm, Whatman) to remove coarse and fine particles and phytoplanktonic organisms within 24 hours of sampling. All samples (filtered water and sediments, and filters) were stored at -20 °C for several weeks before chemical analyses in metropolitan France.
Chlorophyll a pigments were quantified from GF/F Whatman filters using the Unesco method (Vohra, 1966) with a spectrophotometer, and allowed the quantification of phytoplankton per volume of water in each pond. N–NH4+,N–NO3- and P-PO43- concentrations of the water samples were determined using colorimetric methods with a sequential analyzer (SmartChem200, AMSAlliance) (Grasshoff et al., 1999). In sediment samples, organic carbon to nitrogen ratio (C:N) was measured using the “capsule method” (Brodie et al., 2011). An aliquot of 5 mg ± 10 % of homogeneous sediment sample was acidified into 100 µL of 2M HCl in silver capsules to eliminate carbonates. Capsules liquids were evaporated on a 65°C hot plate for 12 hours, before capsules were oven-dried at 80 °C for two days. Organic carbon and nitrogen total contents were then measured with an elementary analyzer (FlashEA 1112 NC Analyzers®, Thermo Fisher Scientific, Waltham, Massachusetts, USA). The bioavailable P concentration in sediments was determined following the protocol of (Ni et al., 2016). Available P was extracted from 50 mg of dry sediment samples with 5 ml of NaOH (1 M). Then, the supernatant was collected, its pH was stabilized with HCl (3.5 M), and the extracted P that has been converted into orthophosphate was quantified using the molybdate/ascorbic acid blue method (Murphy & Riley, 1962); results were reported in mg of available P per g of dry sediment.

Data analysis

We tested for correlations between all pairs of traits using Spearman rank correlations to detect possible co-variations among traits. We did not detect significant strong correlations between any pair of traits (Spearman’s rho < 0.7, Dormann et al., 2013). Thus, each trait was indicative for a part of global strategy of growth.
Abiotic and biotic characterization of ponds
First, to characterize the abiotic conditions within ponds, the mean of daily (day and night) water temperatures was calculated from November 2019 to November 2020 for each pond (Douce et al., in prep). Abiotic variables that have been measured at quadrat scales have been averaged at pond scale. Depending on the analysis, we performed several reductions of abiotic variables. Water nutrients dimensions (N-NH4+, N-NO3-, P-PO43-) were reduced to the first axis of a Principal Component Analysis (PCA), in which data were centered and standardized by standard deviation (Supporting information 2), hereafter named ‘PCAwater’. PCA Axis 1 (42.06%) was negatively related to the concentration of N-NH4+and N-NO3-. Similar reduction of dimensions was done for sediment nutrients (C:N ratio, bioavailable phosphorus) (Supporting information 3), hereafter named ‘PCAsediments’. PCA Axis 1 (58.59%) was negatively related to the C:N ratio and positively related to the bioavailable P concentration.
Second, to characterize the biotic conditions within ponds, we calculated two non-correlated indices: the macrophyte species richness, and the Pielou eveness. Biotic dimensions describing species abundance in each pond were reduced to the first-three axes of a Factorial Correspondence Analysis (FCA) (Supporting information 4). The first axis (33.81%) was respectively positively and negatively driven by the abundances of R. moseleyi and C. antarctica . The second axis (24.99%) was positively correlated with the abundance of R. pseudotrullifolius and negatively with J.scheuchzerioïdes . The third axis (16.99%) was related to the abundances of L. australis (positively) and R. biternatus(negatively). In parallel, to assess the functional characteristics of pond communities, we calculated the overall functional dispersion (then referred as FDis) with seven traits (height, SLA, LDMC, internode length, specific internode mass, maximum root length and specific root mass), using the fdisp function of package FD (Laliberté & Legendre, 2010).
Variance partitioning across nested scales
We used traits measurements from 2017 to 2020 to partition trait variances across four nested scales. A variance component analysis was performed using the Parvart procedure (package Cati), for each of the seven studied traits, across temporal, spatial, and phylogenetic (including inter- and intraspecific variations) scales introduced in this order in the model: time (year), space (site), species, and within-species. For example, the part of variation of the species scale represents the variance of the species means around the mean of their site. To that aim, the mean value of each species is calculated, then the variance of these species means are calculated around the site mean to which they belong. Thus, the intra-site variation is the sum of the species and within-species variations, that is the trait means variation between species in each site, and between individuals in each species in each site. Trait data were previously normalized (log or square-root transformations). The significance of the variance components was assessed by building 95% confidence intervals (CI) through a bootstrapping procedure. We randomly selected 100 individuals out of the 1430 of our dataset with replacement (Messier et al., 2010), and calculated the trait variance partitioning between the four nested scales. We repeated this procedure 500 times to ultimately calculate a 95% CI for each variance component. To get rid of the imperfect nesting of species within sites, we ran alternative partitioning models by deleting species scale (Supporting information 5).
Trait responses to biotic and abiotic filters
We aimed to determine which traits and categories of traits were involved in their biotic and/or abiotic resistance. On the one hand, we considered univariate responses and tested for the effects of abiotic and biotic variables measured or calculated at pond scale, on trait values measured in November 2020 and averaged at pond scale, on all ramets of all species. In the full models, we included as explanatory variables: species identity (categorical variable assessing the species to which ramets belong to), macrophyte species richness, Pielou eveness, community FDis, the first three axes of the FCA performed on macrophyte species abundances (Supporting information 4), and the phytoplankton concentration for biotic variables; water depth, pH, electric conductivity, dissolved oxygen, mean temperature, pond area, and the first axis of the two PCA performed on water nutrients and sediment nutrients (Supporting information 2 and 3) for abiotic variables. Trait data were previously normalized (log or square-root transformations) when needed, and standardized. We applied a stepwise model reduction based on the AIC criteria (MuMIn package, Bartoń, 2013) on the full model to select the most parsimonious model and we performed an ANOVA test (type II).
On the other hand, similar procedure was applied to test for the effects of the above-listed abiotic and biotic variables on traits grouped into three categories (multivariate responses): aerial (height, SLA, and LDMC), clonal (internode length, and specific internode mass), and root traits (maximum root length, and specific root mass). Trait data were previously normalized (log or square-root transformations) when needed and standardized. We selected variables by a backward elimination on the full model (deleting the non-significant effects), and then performed a MANOVA test (type II, Pillai statistic) based on the reduced model.
Traits and abiotic variable effects on species performance
Ramet total biomass (i.e. performance) and trait values recorded in November 2020 were averaged per species at the pond level (n= 97), and standardized separately for each species to get rid of variance due to species identity. We tested for the direct and indirect effects of the seven traits on individual performance using a Structural Equation Modeling (SEM) procedure (Grace et al., 2010). All possible relationships between traits were included based on the literature (Supporting information 6). We performed a path analysis with the lavaan package (Rosseel, 2012), and reduced the full model by variable selection based on AIC values.
All statistical analyses were carried out with R 4.0.3.