Twenty-seven mesocosms (185 cm Ø, 90 cm depth) were sunk into the ground outside the greenhouse facility at Radboud University (Nijmegen, the Netherlands). All mesocosms were filled with 20 cm (135 L) of carefully homogenized clay (originating from Lalleweer, 53∘16′ N, 6∘59′ E; n=9), peaty clay (originating from De Deelen, 53∘01′ N, 5∘55′ E; n=9) or peat (originating from Ilperveld, 52∘27′ N, 4∘56′ E; n=9), after which they received a layer of 50 cm of Nijmegen tap water (NH4+ < 0.03 mg L-1; NO3-: 16.40 mg L-1; PO43- < 0.03 mg L-1; pH: 7.7; total inorganic carbon (TIC): 30 mg C L-1). Sediment characteristics are displayed in Table 1, expressed per unit volume to enable comparison among sediment types with respect to nutrient exchange and plant nutrient availability. In all mesocosms, crossed transparent carbon fiber plates were used to create four fully isolated quarters. We did not include non-vegetated treatments because (1) our focus was on complete ecosystems in constructed and natural wetlands, i.e., including sediment and vegetation; (2) bare sediments always show spontaneous vegetation development if light and nutrient conditions suffice (see Sect. 2.2); (3) continuous plant removal would lead to significant sediment disturbance; and (4) dark conditions would affect sediment biogeochemistry. Mesocosms were randomly assigned to low (L), medium (M) or high (H) nutrient-loading treatment (n=3 for all). To create these, treatment solutions were added three times a week to the surface water to enable loading rates of 0.43, 21.4 and 85.7 mg P m-2 d-1 (added as NaH2PO4 ⚫ H2O and atmospheric deposition of 0.1 kg P ha-1 yr-1) (Furnas, 2003) and 1.3, 62 and 249 mg N m-2 d-1 (added as NH4NO3 and atmospheric N deposition of 35 kg N ha-1 yr-1 in this part of the Netherlands) (RIVM, 2016). In the “Results” and the “Discussion” sections, treatments will be referred to as 0.43 (low), 21.4 (medium) and 85.7 (high) mg P m-2 d-1, according to their respective P loading.

In July 2013, environmentally relevant densities (based on personal field observations) (De Lyon and Roelofs, 1986) of Ceratophyllum demersum (5.03 ± 0.24 g DW m-2 (DW is dry weight); rigid hornwort, submerged macrophyte), Chara hispida (8.66 ± 0.69 g DW m-2; bristly stonewort, submerged macroalga) and Myriophyllum spicatum (5.31 ± 0.60 g DW m-2; Eurasian water milfoil, submerged macrophyte) were planted randomly in each of three quarters of every mesocosm to establish themselves. In April 2014, patches of Azolla filiculoides (28.39 ± 0.88 g DW m-2; water fern, floating macrophyte) were added to the water layer of the remaining quarter. Apart from these four introduced species, other species colonized the quarters, including Zanichellia spp. and floating algae. As C. hispida was completely outcompeted by spontaneously developing vegetation, the quarters with this species were excluded from the rest of this study. During the experimental period, 20 % of the total plant biomass (for rooted macrophytes aboveground biomass only) was harvested when vegetation reached 100 % cover to avoid space limitation. During the final harvest, biomass of all present species was harvested separately and dried (48 h at 60 ∘C), after which it was weighed, ground and homogenized.

Surface water samples were collected every week between May and October 2014, whereas pore water samples were collected anaerobically every month using ceramic soil moisture samplers (SMS Rhizon, Eijkelkamp, Giesbeek, the Netherlands). The pH of water samples was measured between 12:00 and 14:00 (UTC + 2) using a combined Ag / AgCl electrode (Orion, Thermo Fisher Scientific, Waltham, MA, USA) with a TIM840 pH meter (Radiometer Analytical, Lyon, France). TIC of water samples was measured using an infrared gas analyzer (IRGA; ABB Analytical, Frankfurt, Germany). Concentrations of PO43-, NO3- and NH4+ in the surface water and pore water were measured colorimetrically on an Auto-Analyzer III system (Bran and Luebbe, Norderstedt, Germany) by using ammonium molybdate (Henriksen, 1965), hydrazine sulfate (Kamphake et al., 1967) and salicylate (Grasshoff and Johannsen, 1972), respectively. Concentrations of total P were measured by inductively coupled plasma optical emission spectrometry (ICP-OES; IRIS Intrepid II, Thermo Fisher Scientific, Franklin, MA, USA).

Sediment samples were collected at the end of the experiment, and subsequently volume-weighted and dried for 48 h at 60 ∘C to determine bulk density. Dry sediment samples were heated for 4 h at 550 ∘C and reweighed to determine organic matter content. Furthermore, 200 mg of dry sediment was digested in a microwave oven (MLS-1200 Mega, Milestone Inc., Sorisole, Italy) with 4 mL 65 % HNO3 and 1 mL 30 % H2O2, after which digestates were analyzed and concentrations of total Al, Fe, Ca and P in sediments were determined by ICP-OES (see above). Plant-available P was determined by extraction according to Olsen et al. (1954), whereas an NaCl extraction was performed to determine exchangeable N ions (NO3- + NH4+) as described in Tomassen et al. (2004). Total P concentrations in plants were determined by digestion of 200 mg of dry plant material and analyzed as described above. Furthermore, 3 mg of dry plant material was combusted to determine C and N content using an elemental analyzer (Carlo Erba NA 1500, Thermo Fisher Scientific, Waltham, MA, USA).

For both N and P, nutrient budgets were calculated to determine the distribution among biomass, sediment and other components. Cumulative biomass production and the nutrient content of submerged or floating macrophytes (target species and others) were used to calculate plant uptake rates of N and P. Furthermore, nutrient changes in surface water and pore water were calculated from changes in N (NO3- and NH4+) and total P concentrations (end minus start). After subtracting the N and P uptake of plants and water components from the external loading, we assume that the remainder is either stored in the sediment or, in the case of N, lost through coupled nitrification–denitrification (Wetzel, 2001).

All analyses were performed using the software program R (version 3.2.1; R Development Core Team, 2015). The effects were considered significant if P<0.05. In order to meet the assumption that residuals fit a normal distribution and homogeneity of variance, we transformed sediment characteristics, N (NO3- and NH4+) and P concentrations in surface water, biomass production rates, N : P ratios in macrophytes, N and P budgets, and N and P sequestration rates (response variables) by log(response variable) or log(response variable + 1) in the case of the lowest value of a variable being below 1. Linear mixed models were used to test the main effects and the interactions of treatments on sediment characteristics, biomass production rates, the ratios between N and P, and nutrient budgets with the mesocosm number as a random effect by using R package nlme. The main effects (including nutrient loading, sediment type, plant species and time) and interactions of treatments on N (NO3- and NH4+) and P concentrations in surface water were also tested by linear mixed models. Tukey tests were used to find differences between treatments by using R package multcomp. We analyzed the influence of nutrient loadings on P and N sequestration (uptake plus adsorption to plants) rates using linear and logistic regression models with the summary function. All graphs were plotted using R package ggplot2.

Over time, surface water P and N (NH4++NO3-) concentrations increased (Figs. 1 and 2; X2=3.44, P < 0.05 and X2=23.63, P < 0.001 for P and N, respectively), especially towards the end of the growing season. There were significant interactions between time and plant species (X2=10.18, P < 0.01) for surface water P and between time and nutrient loadings (X2=8.92, P < 0.05) for surface water N. When macrophytes were growing on peat or peaty clay sediments, P concentrations in the surface water increased with increasing external P loading (X2=99.80, P < 0.001 and X2=59.40, P < 0.001 for peat and peaty clay sediments, respectively).

Porewater nutrient concentrations depended on sediment type. Peat sediments had the highest P concentrations in the pore water, whereas the lowest were found in clay sediments (X2=20.20, P < 0.001; 4.65 ± 0.15 and 0.71 ± 0.05 mg L-1 for peat and clay, respectively), even though total P and Olsen P concentrations were much higher in clay than in the other two sediments (Table 1). In addition, mesocosms filled with peat sediments had higher N concentrations in the pore water than those with peaty clay and clay (X2=7.13, P < 0.05; data not shown). Surface water and porewater together never contained more than 12 % of total P and N added to the system at P loadings ≥ 21.4 mg P m-2 d-1 (Figs. 4 and 5).

Due to their high biomass production rates, A. filiculoides and M. spicatum could be harvested weekly and biweekly, respectively. A. filiculoides had the highest biomass production rates of all three macrophyte species (X2=55.45, P < 0.001), whereas C. demersum grew best on peaty clay sediments (X2=10.67, P < 0.01) but almost disappeared when growing on clay and peat sediments due to competition with algae and other non-target species (Fig. 3). Biomass production rates of A. filiculoides were significantly higher at high nutrient loading than at low nutrient loading (X2=11.39, P < 0.01), whereas no effect of nutrient loading was found for the other macrophytes. In quarters with C. demersum, there was a higher production rate of non-target species than in quarters with A. filiculoides and M. spicatum (X2=6.28, P < 0.05). A. filiculoides showed high N : P ratios (> 11 g g-1) when grown at ≤ 21.4 mg P m-2 d-1 (P < 0.001), whereas all other species generally showed N : P ratios ranging from 4 to 8 g g-1, without an effect of sediment type (Table 2).

A. filiculoides and M. spicatum accumulated much more P than C. demersum (X2=23.66, P < 0.001; Fig. 4). At a P loading of 0.43 mg m-2 d-1, around 100 % of added P and N were accumulated by the targeted macrophytes (Figs. 4 and 5). For the quarters with A. filiculoides or M. spicatum, around 20–40 and 10–20 % of the P added was taken up by target species at P loadings of 21.4 and 85.7 mg m-2 d-1, respectively, regardless of sediment types. C. demersum never took up more than 20 % of the P added at these loadings. Still, at a loading of 85.7 mg P m-2 d-1, removal rates by macrophytes were significantly higher than at 0.43 mg P m-2 d-1 (X2=7.22, P < 0.05; Fig. 4). The average P sequestration rates by A. filiculoides and M. spicatum were 3 to 9 mg m-2 d-1 at P loadings ≤ 21.4 mg m-2 d-1. At a high P loading of 85.7 mg m-2 d-1, the average P removal rates by A. filiculoides and M. spicatum were 16 to 20 and 6 to 14 mg m-2 d-1, respectively. In quarters with C. demersum, more P was taken up by other, spontaneously developing species than in quarters with A. filiculoides and M. spicatum (X2=6.89, P < 0.05). A. filiculoides and M. spicatum sequestrated much more N than C. demersum and the final biomass of A. filiculoides had the highest N content (including N2 fixed) among all macrophyte species (X2=10.28, P < 0.01; Fig. 5). At high N loadings, less than 21 % of added N was removed by the targeted macrophytes. In addition, C. demersum had higher P and N uptake rates in mesocosms with peaty clay compared to mesocosms with clay (X2=10.50, P < 0.01; X2=10.43, P < 0.01).

For C. demersum, nutrient sequestration rates increased linearly with increased nutrient loading, while for M. spicatum there was a logistic response to external nutrient loading (Fig. 6). A. filiculoides showed linearly increasing P sequestration rates upon increased P loading and a logistic response to external N loading.

At a P loading of 0.43 mg m-2 d-1, sediments were sources of P, whereas sediments became P sinks at P loading ≥ 21.4 mg m-2 d-1 (Fig. 4). On average, 50 to 80 % and 70 to 90 % of P added accumulated in sediments at medium and high nutrient loadings, respectively (Fig. 4). In quarters with C. demersum, more P accumulated in the sediment than in quarters with A. filiculoides (X2=11.25, P < 0.01). At medium and high N loads, 45 to 90 % and 80 to 90 %, respectively, was either taken up by the sediment or lost to the atmosphere through coupled nitrification–denitrification (Wetzel, 2001).

With average biomass production rates of 3.4 and 1.0 g DW m-2 d-1, respectively, A. filiculoides and M. spicatum showed the highest growth rates, regardless of sediment type and nutrient loading, and therefore have the best potential for being used to remove nutrients in constructed wetlands. Due to their high growth rates, these species could be harvested biweekly or even weekly. C. demersum, on the other hand, appeared to be less suitable, since this species was easily outcompeted for light by other species, such as floating algae and Zanichellia spp. P was removed most efficiently by A. filiculoides, followed by M. spicatum and C. demersum. Although a high P load (85.7 mg m-2 d-1) resulted in increased uptake rates of 6 to 14 and even 16 to 20 mg P m-2 d-1 for M. spicatum and A. filiculoides, respectively, these rates were not sufficient to efficiently filter all added P from the system.

Different response types between species to external nutrient loading most likely resulted from differences in main nutrient sources and nutrient limitation (Fig. 6). For rooted M. spicatum, plants mainly rely on sediment uptake (Best and Mantai, 1978; Barko and Smart, 1980; Carignan and Kalff, 1980), whereas for non-rooted A. filiculoides and C. demersum, water is the main nutrient source (Denny, 1987; Mjelde and Faafeng, 1997). Our results indicate that at a low nutrient loading, M. spicatum and A. filiculoides performed equally well for P removal, whereas at loads ≥ 22 mg P m-2 d-1, A. filiculoides removes P more efficiently (Fig. 6a). In addition, the effective thresholds for P purification (100 % removal) of C. demersum, A. filiculoides and M. spicatum are 1.9, 4.8 and 6.8 mg P m-2 d-1, respectively (Fig. 6a). Threshold values for complete N removal are 8.6 and 31.4 mg N m-2 d-1 for C. demersum and M. spicatum, respectively (Fig. 6b). A. filiculoides, on the other hand, hardly ever becomes N limited due to its symbiosis with a diazotrophic microbial community (Handley and Raven, 1992). Under low external P loadings, A. filiculoides therefore displayed very high N : P ratios, indicating P limitation at P loadings ≤ 21.4 mg P m-2 d-1. C. demersum, on the other hand, having no access to sediment or atmospheric N, probably showed N limitation in these systems, as indicated by their low N : P ratios. For all species, N : P ratios decreased with increasing P load.

Due to regular harvesting of A. filiculoides and M. spicatum, P and N were removed at rates of around 3 to 9 mg P m-2 d-1 and 31 mg N m-2 d-1 at loadings of 0.43 mg P m-2 d-1 and 1.3 mg N m-2 d-1. These results are comparable to those found by Van Kempen (2013), who found uptake rates of 3.7 mg P m-2 d-1 (13.4 kg ha-1 yr-1) and 13.7 mg N m-2 d-1 (50 kg ha-1 yr-1) in summer and 4.8 mg P m-2 d-1 (17.5 kg ha-1 yr-1) and 69.3 mg N m-2 d-1 (253 kg ha-1 yr-1) in the early fall for A. filiculoides grown in N-free water with 2.38 mg L-1 PO4. For M. spicatum, our results are in the same range as those reported by Smith and Adams (1986) and the N uptake rates of 0.05–1.26 g N m-2 d-1 by Myriophyllum aquaticum reported by Nuttall (1985). As low O2 concentrations, induced by the coverage of floating macrophytes or dense growth of submerged macrophytes, can mobilize P from the sediment, A. filiculoides and M. spicatum did not only take up all P being discharged into the system by both their roots and shoots but additionally took up mobilized P (Wetzel, 2001).

Since the uptake of nutrients by aquatic macrophytes depends on their biomass production and thus on macrophyte photosynthesis, these systems would only function optimally during the growing season (Wetzel, 2001). Under low external loading, sediments will take up most of the P during winter. Since submerged plants have N and P accumulation rates that are higher than the low nutrient loading, they heavily rely on the uptake of nutrients from the sediment. Thus, the nutrients stored in the sediment in winter can be mobilized and taken up by macrophytes in summer, creating an efficient and sustainable constructed wetland for water polishing in temperate climates. Furthermore, predicted climate change will lead to higher temperatures and thus longer growing seasons in temperate regions, indicating that these systems may be operational longer and longer every year.

When P loading in the treatment water increases, uptake rates of A. filiculoides double or even triple to rates of 7.87 or 17.64 mg P m-2 d-1. The highest value is lower than the results of Reddy and DeBusk (1985), who reported P uptake rates of 43 ± 15 mg P m-2 d-1 by A. filiculoides grown in an N-free, 3 mg L-1 P medium which, however, had much higher PO43- concentrations in the surface water than our high nutrient-loading treatment. P uptake rates of A. filiculoides in this study are similar to, or even lower than, the results of Brix (1994), who reported P uptake rates of 8–41 mg P m-2 d-1 by emergent macrophytes. The main advantage of using floating macrophytes instead of emergent macrophytes is, however, that they can be harvested multiple times a year and that they take up nutrients from both the water layer and the sediment. Although plants could not take up all P at medium or high external P loadings, overall surface water quality remained around or below 0.37 mg L-1 when clay sediments were used for the construction of the wetland. At the end of the growing season, however, plant uptake decreased and P availability in surface waters above peaty clay and peat sediments increased strongly to concentrations around 1.86 and 2.23 mg P L-1, respectively, indicating not only inactivity of aquatic macrophytes but probably also P saturation of sediments. Due to the 7–8 times higher Fe and Al contents (22.6 vs. 2.6–3.3 g L-1 FW (fresh weight) and 11.9 vs. 1.5–1.8 g L-1 FW for Fe and Al, respectively) of clay sediments, P was most probably immobilized more efficiently by clay (Reddy and DeLaune, 2008), which resulted in lower P concentrations in surface water above clay sediments in our study.

More than 98 % of added N was removed from the surface water during the run of the experiment. As nutrient loading increased, the amount of added N that was removed by plant uptake decreased. Harvested biomass of target plants contained 31 mg N m-2 d-1 for M. spicatum, whereas in the quarters with C. demersum, non-target macrophytes or algae sequestrated most N. Although it can be estimated that N2-fixation rates by Azolla grown in an N-free medium were in the range of 1.4–2.7 kg N ha-1 d-1 (Reddy and DeBusk, 1985), in our study we added N to the surface water which may affect N2 fixation. Therefore, it was difficult to calculate N removal rates for A. filiculoides, as the unknown N2 fixation rates lead to an overestimation of N uptake rates by A. filiculoides. N that was not taken up by plants but was still removed from the water layer most likely ended up in the sediment or was released to the atmosphere by coupled nitrification–denitrification (Wetzel, 2001). On average, inorganic N (NH4+ + NO3-) concentrations in the surface water were below 0.11 mg L-1 with external loadings ≤ 62 mg N m-2 d-1 and around 0.28 mg L-1 when receiving 249 mg N m-2 d-1. At the end of the growing season, dissolved N concentrations increased under high nutrient loading, similar to P concentrations. This increase may result from a combination of reduced plant uptake, nutrient leaching from senescing plants and reduced denitrification rates as a result of lower temperatures. Due to the different available pathways for nitrogen removal from the sediment, sediment saturation of N seems unlikely.

We showed that in macrophyte-dominated CWS, both the submerged and the floating macrophytes we tested are able to remove most of the added nutrients at low P and N loadings, whereas at higher nutrient loadings, floating or submerged macrophytes could only remove 20–45 and 10–25 % of the external P loads for 21.4 and 85.7 mg P m-2 d-1, respectively. For water management, regular mowing of fast growing aquatic macrophytes, such as A. filiculoides or M. spicatum allows the complete removal of added nutrients at relatively low nutrient loading (≤ 4.8 or ≤ 6.8 mg P m-2 d-1, respectively). Although A. filiculoides still extracted P and competed with sediment adsorption at higher P loads (≥ 21.4 mg P m-2 d-1), most external P ended up in the sediment, eventually resulting in saturated sediments and thus leading to an increase in water nutrient levels under a continued nutrient input. While aquatic macrophytes are able to remove this P from the sediments by either creating anaerobic conditions to trigger high P mobilization (Smolders et al., 2006) or through both root and shoot uptake, the external load will have to be reduced for this process to occur efficiently. Consequently, at these higher P and N loads, the macrophyte stage can only be used as an additional polishing step after a major part of the nutrients have been removed by other methods of water treatment.