Manipulating plant community composition to steer efficient N-cycling in intensively managed grasslands

1. Minimizing nitrogen (N) losses and increasing plant N uptake in agroecosystems is a major global challenge. Ecological concepts from (semi)natural grasslands suggest


| INTRODUC TI ON
The invention of the Haber-Bosch process, by which inert atmospheric nitrogen (N 2 ) is transformed into reactive forms, has marked a profound shift in the relationship between humans and the global N cycle. This reactive N enters the N cycle as an external input via mineral fertilizers applied to boost yields in intensive agroecosystems (Moreau et al., 2019). However, plants only take up and retain approximately 40% of fertilizer N in their biomass (Cassman et al., 2002). Most of the remaining N can be lost from the agroecosystem with dire environmental consequences. Nitrogen can be lost as nitrous oxide (N 2 O) emissions from nitrification, denitrification and related processes, contributing to global warming and ozone depletion. Alternatively, soil mineral N may leach to water bodies, resulting in harmful algal blooms and human health problems . One promising strategy to steer the biological processes that underpin soil N availability, N losses and plant N retention is by manipulating the plant community composition of the agroecosystem.
Translating concepts from ecology in natural grasslands, recent field studies in fertilized, intensive agroecosystems have shown that increasing plant species richness from monocultures up to fourspecies mixtures can increase yields (Finn et al., 2013) and plant N retention (Suter et al., 2015). In these studies, resource use complementarity was the most likely mechanism behind the biomass increases in mixtures. Therefore, plant species mixtures may mitigate N losses by reducing the availability of soil mineral N to micro-organisms due to high rates of plant N uptake (Moreau et al., 2015). Within these mixtures, the presence of specific plant functional groups may be a key determinant of plant community effects on N-cycling.
Legumes, because of their ability to fix N 2 , represent a keystone biological input of N in natural and managed ecosystems. Grown together with grasses, legumes can meet more than 80% of their own N-demand from fixation, promoting facilitation effects by transferring biologically fixed-N to grasses (Rasmussen et al., 2013), and enhancing exploitation of soil resources through spatial or temporal niche complementarity in resource uptake (Mueller et al., 2013).
Since the identity of the companion grass exerts a strong effect on the overall distribution of N in plants and soil (Pirhofer-Walzl et al., 2012), we need to identify species combinations that increase the sustainability of our agroecosystems and understand the main determinants behind such positive plant-plant interactions.
A robust approach to obtain a mechanistic understanding of how plant communities regulate N-cycling is to study relationships between plant traits and ecosystem functioning (Pommier et al., 2018).
In mesocosm experiments, grasses with leaf traits related to fast N acquisition (high specific leaf area, SLA, and low leaf dry matter content, LDMC) have been linked to reduced N losses in fertilized systems via decreases in N 2 O emissions (Abalos et al., 2018) and N leaching (De Vries & Bardgett, 2016). Below-ground, grasses with root traits associated with fast N acquisition are expected to lead to the same results, and empirical evidence to validate these theoretical claims is starting to emerge (Abalos et al., 2014(Abalos et al., , 2018Cantarel et al., 2015). It is therefore plausible that combining a legume with an acquisitive grass will result in higher plant N uptake and lower N losses. A key next step is to (a) test these findings under realistic field conditions and (b) unveil whether the functional traits of the dominant species in the community (Garnier et al., 2004) or the functional trait diversity among species (Laliberte & Legendre, 2010) determines the overall association between plant traits and N pools and fluxes.
Plant community composition can be strategically designed to regulate N losses and N retention not only via direct effects through plant nutrient acquisition but also indirectly by affecting below-ground biotic communities contributing to N-cycling.
For example, previous research has shown positive relationships between plant species richness and earthworm species number and abundance, particularly when legumes are present (Milcu et al., 2008). In turn, higher earthworm abundance may enhance N mineralization and N 2 O emissions (Lubbers et al., 2013). Legumes can also increase the abundance of soil diazotrophs (Xiao et al., 2020) rising N inputs into the soil. Leaf traits such as SLA have been linked to microbial N uptake and ecosystem N retention (De Vries & Bardgett, 2016). By producing higher root biomass, plant species richness can promote the abundance of arbuscular mycorrhizal fungi (AMF; Chung et al., 2007;Wolf et al., 2003); plants with low specific root length (SRL) generally have high levels of AMF colonization of their roots (Cortois et al., 2016). In turn, increased AMF colonization may favour plant N retention and reduce the emission of N 2 O and N leaching (Bender et al., 2016;Martínez-García et al., 2017). Specific plant traits are also linked to the abundances and activities of microbial nitrifiers (e.g. SRL; Cantarel et al., 2015) and nitrate reducers (e.g. root N uptake rate; Moreau et al., 2015), and these processes regulate in which form N is lost (Hallin et al., 2018). To date, most studies on plant community effects on N-cycling have overlooked these meaningful plant-soil interactions, despite their importance to transition into more biologically based agroecosystems (Abalos et al., 2019;Bowles et al., 2015).
The objective of this field study was to explore the potential of plant community composition (species mixtures, species identity and functional trait dominance/diversity) to regulate N-cycling in intensively managed grasslands. We hypothesized that: (a) plant species mixtures would increase plant N retention and decrease soil inorganic N concentrations and N 2 O emissions; by providing more diverse litter inputs, plant mixtures would also promote the abundance and activity rates of below-ground biotic communities; (b) the presence of a legume would increase N pools, fluxes and biological N-cycling activity, but (c) in combination with an acquisitive grass more N can be retained in the plant community with lower N losses; and (d) the relationships between plant communities and N-cycling can be explained by the dominant traits or the trait divergence of the plant community.

| Experimental design
We conducted our field study at 'Droevendaal' experimental farm, Wageningen, the Netherlands (51°59′N, 5°39′E). The soil is classified as a typic endoaquoll (Soil Survey Staff, 1999) with 75% sand, 23% silt and 2% clay. Initial physico-chemical characteristics of the 0-15 cm soil layer were pH-CaCl 2 , 5.1; soil bulk density, 1.42 Mg/m 3 ; organic matter, 2.2%; total N, 1.30 g/kg. The climate is temperate maritime (Cfb according to Köppen-Geiger), with an average temperature of 9.4°C and mean annual precipitation of 780 mm. During the years before the establishment of our experiment, the field site had been used as an arable field involving annual crops (mainly wheat and maize), managed following standard farming practices (i.e. conventional tillage, mineral fertilizers according to crop requirements, use of herbicides and pesticides).
The experiment was set up using a randomized block design in March 2016, and consisted of 60 plots (3 × 3 m) laid out in five blocks, with grass strips (Phleum pratense L.) of 3 m between plots.
The treatments covered a plant species richness gradient of one, two and four species, and two functional groups: grass (Lolium perenne L., Lp; Festuca arundinacea Schreb., Fa; Poa trivialis L., Pt) and legume (Trifolium repens L., Tr). In total, there were 12 treatments: no plants (control); monocultures of the four species; all possible combinations of two species; and a mixture with all four species (n = 60 plots). The levels of species richness from one to four species were selected based on the strong benefits for ecosystem functioning of this diversity gradient in intensive grasslands (Finn et al., 2013;Nyfeler et al., 2009), while maintaining a level of management complexity that is feasible to adopt by farmers. The selected plant species are commonly found in N-rich soils and cover a relatively wide range of functional traits, both above-and below-ground (e.g. Abalos et al., 2014;Mason et al., 2016). All plots were weeded manually two to three times per year to maintain the target species composition.
Plowing took place in March 2016 and sowing in April 2016 (establishment year), at a final seeding density of 1,500 viable seeds per m 2 divided equally among the species of each mixture (i.e. 750 seeds m −2 for each species in the two-species mixtures, and 375 seeds m −2 for each species in the four-species mixtures). During 2016 and 2017, N fertilizer (calcium ammonium nitrate) was split into three applications at a total rate of 150 kg N/ha (50 kg N/ha in May, June and September). This total N rate is realistic from a farmer's perspective, but lower than the highest levels of N application to grass monocultures in intensive grassland management (200-300 kg N/ha), and it does not impair positive grass-legume interactions (Nyfeler et al., 2009). All plots also received superphosphate (50 kg P ha −1 year −1 ) and potassium sulphate (200 kg K ha −1 year −1 ).
All measurements were conducted during the second productive year (2017), when the grassland communities were well established.

| Soil and plant N pools and fluxes
Plant N pools were quantified for above-and below-ground plant biomass. Above-ground biomass was cut in subplots (25 × 25 cm) three times over the growing season (May, July and October) and the vegetation was sorted per species. The remaining above-ground biomass was harvested with an experimental plot harvester immediately after sampling. Root biomass was sampled at the end of the growing season after the last harvest (October 2017) with a soil corer (8 cm diameter, 0-30 cm depth) and roots were separated from soil and cleaned with water using a root washing facility; a representative root subsample of the cleaned roots was collected and stored in 50% ethanol to be used for quantifying root traits (see below). Ballmilled dry plant samples were analysed for N concentration using a CN elemental analyser (LECO). For plant communities consisting of more than one species, we used species-specific above-ground N uptake (kg N/ha) per plot to quantify the net, complementarity and selection effects using the additive partitioning method (Loreau & Hector, 2001).
Flux measurements of N 2 O were taken daily during the first 4 days after fertilizer applications, every other day in the second week, two consecutive days after rainfall events and two to three times per week for the remainder of the experiment. The flux measurement protocol largely followed that of previous studies (Abalos et al., 2014(Abalos et al., , 2018Lubbers et al., 2015). Polypropylene flux chambers (20 cm diameter, 14 cm height) were placed on top of permanently installed collars (10 cm depth) in each plot for approximately 30 min.
Gas measurements were taken from the headspace with an Innova 1312 photo-acoustic infrared gas analyser (LumaSense Technologies A/S). Emission rates of N 2 O fluxes were consequently estimated as the slope of the linear regression between concentration and time.
Cumulative emissions were calculated by linear interpolation.
To determine soil mineral N, five soil cores per plot were taken with a soil auger (25 cm deep × 1.5 cm diameter) in May, June, July and October. Due to the exponential decline of root biomass with soil depth (Frank et al., 2010) and the mobility of NO − 3 , high NO − 3 concentrations in the top 25 cm soil layer are generally an indicator of high risk of N leaching (Klaus et al., 2018). Soil samples were extracted for mineral N analyses by shaking (1/10 w/v, dry weight basis) with 0.01 M CaCl 2 for 2 hr (Houba et al., 2000). Extracts were centrifuged (15 min), filtered (0.45 µm) and NH + 4 and NO − 3 concentrations were determined by colorimetry (Brann en LuebbeTrAAcs 800 Autoanalyzer, Skalar Analytical B.V. Breda).

| Soil microbiological properties, biological N-fixation, arbuscular mycorrhizal fungi and earthworms
To quantify the abundances of microbial populations involved in N-cycling by real-time quantitative PCR (qPCR), five soil samples  Table S1.
To determine the influence of carbon availability on N 2 O production and reduction by denitrification, we determined denitrification capacity as described in Abalos et al. (2018). Briefly, we supplied 10 ml of a 1 mM KNO 3 solution to airtight 1 L flasks containing two subsamples per plot (20 g each) collected in July. For one of the duplicate flasks, 10% C 2 H 2 was added to inhibit N 2 O-reductase activity and the flasks were incubated at 20°C. Nitrous oxide was quantified after 24 and 48 hr, and the ratio of N 2 O to total N 2 O + N 2 production was derived from the ratio of N 2 O production in flasks without and with C 2 H 2 .
To sample earthworm populations, a soil monolith of 20 × 20 × 20 cm was dug out and hand-sorted from every plot in October. Earthworms were counted and their intestines voided for 48 hr following the wet paper method of Dalby et al. (1996) before being weighed.
To estimate arbuscular mycorrhizal fungi (AMF) colonization in the root samples collected in October 2017, the fungal structures in roots were stained following Brundett et al. (1996). The AMF colonization percentage of the roots was estimated according to the grid line intersection method (McGonigle et al., 1990).
Biological N-fixation in the plants was assessed using the natural 15 N abundance method as percentage of N derived from the atmosphere (%Ndfa). The shoot biomass of five legume leaves per plot was sampled in October, dried, ball-milled and analysed for total N and atom% 15 N at the UC Davis Stable Isotope Facility, on an ANCA-SL Elemental Analyser coupled to a 20-20 mass spectrometer using the Dumas dry combustion method. The %Ndfa was calculated following Shearer and Kohl (1986).

| Plant traits
Plant functional traits were measured following standard pro- Leaf dry matter content (LDMC) was estimated as the ratio of dry to saturated weight. The root subsamples that were stored in alcohol were stained with neutral red (Bouma et al., 2000) and

| Calculations and statistical analysis
All statistical analysis was performed in R version 3.5.0. (R Core Team, 2018). The effects of plant species mixtures (1 vs. 2 or 4 species), species identity and functional trait dominance (CWM) and diversity (FD) on N pools, fluxes and biological drivers of N transformations were tested with linear mixed effects models, lme from package nlme (Pinheiro et al., 2016), fitted with a type III SS ANOVA. Plant species mixtures, species identity and functional trait measures were considered as explanatory variables, and block was included as a random factor. When residual variances were heterogeneous, a variance structure (varIdent, package nlme) was included in the model. Models with and without the variance structure were compared based on Akaike information criterion (AIC) and the model with significantly lower AIC was retained. The same models with monoculture data were used to test differences in functional traits at the plant species level. Since the abundances of N-cycling microbial communities were highly correlated (r = 0.72-0.99; p < 0.001), we performed a principal component analysis (PCA) on gene abundances using the package vegan (Oksanen et al., 2013), and PC1 was used in subsequent analysis. Spearman correlations were used to assess the relationships between biotic N drivers and N pools.

| Plant monocultures versus species mixtures
Plant N pools increased in the mixtures compared to the monocultures (p = 0.04; R 2 = 0.11; F = 3.34), with the four-species mixture having a 29% larger plant N pool (N uptake in above-and belowground biomass) than the average of the monocultures (Figure 1c).
This effect was mainly due to root N uptake rather than aboveground plant N uptake. Root biomass in the mixtures was higher than in the monocultures (p < 0.001; R 2 = 0.34; F = 16.57; Figure S1). The additive partitioning analysis of above-ground N uptake ( Figure S2) showed a positive net effect of mixtures (i.e. enhanced performance of plant mixtures relative to that expected from monocultures).
These results also revealed that the complementarity effect contributed more to the net effect (61%-75%) than the selection effect (i.e. the presence of a specific species). Soil NO − 3 availability (p = 0.01; R 2 = 0.14; F = 4.32) and earthworm biomass (p = 0.007; R 2 = 0.07; F = 5.51) were lower in mixtures than monocultures (Figure 1). All the other N variables were not different between plant mixtures and monocultures.

| Plant species identity and community composition
Trifolium repens presence increased N 2 O emissions by 58% (p < 0.001;

| Relationships between biotic N drivers and N pools
Earthworm biomass was related to all the measured variables, whereas AMF colonization was only related to earthworm biomass (Table 1).
The collective abundance of genes of the soil microbes involved in N-cycling (PCA axis projection 1) had a positive relationship with N 2 O emissions and with N-fixation. Plant N pools were related to all N variables except AMF, and the strongest association was with N-fixation.

| Plant functional traits in monocultures
Above-ground, L. perenne and T. repens had the lowest LDMC, T. repens had the highest LNC and lowest C:N ratio and P. trivialis had the lowest SLA values (Table 2). Belowground, P. trivialis had the highest SRL and lowest root diameter; P. trivialis and L. perenne showed significantly larger RLD values than F. arundinacea and T. repens. The legume T. repens had lowest root C:N ratio. F I G U R E 2 Effect ofTrifolium repenspresence/absence on (a) cumulative N 2 O emissions; (b) the abundance of N-cycling microbial communities; (c) plant N uptake, depicted separately for above-and below-ground components; (d) earthworm biomass; (e) soil NO − 3 concentration; (f) soil NH + 4 concentration; (g) arbuscular mycorrhizal fungi (AMF) colonization in roots. Significant differences are tested using linear models (NS: non-significant)

| Trait dominance and trait diversity effects on N-cycling
Community-weighted means (CWM) of leaf traits showed more associations and stronger effects on N pools and biotic drivers of N-cycling than functional diversity of leaf traits (Table 3)

| Plant community composition and N-cycling
We found that an agriculturally feasible level of plant species mixing (from one to two or four species) increased plant N retention by 25% and reduced soil NO − 3 availability and earthworm biomass compared to monocultures (Figure 1). The positive effect of species mixtures on plant N retention is consistent with results previously found in natural grasslands (Scherer-Lorenzen et al., 2003;Wang et al., 2018), yet our study is one of the few reporting this finding for an intensively fertilized grassland (Suter et al., 2015).
This effect could be due to enhanced complementarity (De Deyn et al., 2009), and/or due to the presence of a certain influential species or functional groups (i.e. selection effect; Hooper, 1998).
Our results suggest that both mechanisms may have taken place during the experiment. Above-ground, complementarity was the main mechanism causing enhanced N uptake in plant mixtures relative to that expected from monocultures ( Figure S2). The importance of complementarity is also supported by our finding that increased functional diversity of LNC (higher for the four-species mixtures) was associated with higher plant N pools (Table 3). The higher resource complementarity and plant N capture probably led to the lower NO − 3 availability for the four-species mixtures (Figure 1). Despite this evidence for the occurrence of resource use complementarity, the dominance of CWM over FD trait measures in explaining biotic and abiotic N processes (Table 3)  Indeed, we found that the presence of the legume T. repens increased N 2 O emissions, plant N pools, soil NO − 3 and NH + 4 concentrations, the abundance of N-cycling microbial communities and earthworm biomass at plant community level (Figure 2). Legume presence also played a minor but significant role for higher yield ( Figure S3). By establishing a symbiotic association with N-fixing bacteria (and by increasing N-fixers abundance, i.e. nifH), legumes add an internal input of biological N resulting from the mineralization of their N-rich litter (T. repens had the lowest leaf and root C:N ratios; Table 2) and root exudates. In semi-natural grasslands with low soil fertility levels, positive effects of legumes on N 2 O (Niklaus et al., 2006), soil mineral N (and associated N-leaching;Scherer-Lorenzen et al., 2003), plant N pools (De Deyn et al., 2009, earthworm biomass (Milcu et al., 2008) and some N-cycling microbes (e.g. AOB: Malchair et al., 2010) have been previously reported. Here, we show for the first-time similar effects for an intensively managed grassland receiving external inputs of N via fertilizer. Moreover, our study shows that legumes can also increase the abundance of nirS, nirK, nosZI and nosZII microbial guilds, further stimulating soil N-cycling.
Among the combinations of T. repens with grasses, L. perenne was the species leading to lower N 2 O emissions and a high plant N pool, without a negative influence on the other N-cycling drivers or pools. This was probably because this fast-growing grass had higher N uptake rates than the other grass species (Figure 3), allowing L. perenne to use the available sources of N from soil, fertilization and N-fixation more efficiently. Indeed, L. perenne was the grass species with the lowest LDMC (Table 2), indicative of an acquisitive strategy for N capture. It was also the species with the highest RLD and with high SRL, architectural root traits that determine the configuration of the root system and thereby the plant's ability to acquire N (Bardgett et al., 2014) and potentially reduce N 2 O emissions (Abalos et al., 2019). As such, higher RLD led to lower soil NO − 3 concentrations (Table 3), and because L. perenne had the highest root biomass production of all the species (Figure S1), its combination with T. repens had the highest RLD values among the combinations with the legume.
Although grasses with acquisitive leaf traits and high RLD appear to be of pivotal importance for efficient N-cycling in combination with legumes, these traits alone did not explain a significant fraction of the variation in N pools and soil N-cycling processes (Table 1; Table 3). P. trivialis, for example, also presented these set of traits, but did not result in the same positive effects when mixed with T. repens. This was probably because high root biomass production may be a critical factor as well, since L. perenne in monoculture and two-species mixtures had high biomass production, whereas P. trivialis had very low biomass (also aboveground), contributing little to the overall traits of the plant communities. The negative correlation between root biomass and the abundance of AOB (r = −0.45; p < 0.05) and AOA (r = −0.37; p < 0.05) supports the relevance of root biomass for soil mineral N capture. Hence, the lower abundance of both AOA and AOB suggests a lower soil N availability and that ammonia oxidizers are out-competed when the root biomass production is higher. Root biomass could also stimulate denitrification and N 2 O emissions, as the quantity of root exudation, which may provide a carbon source for heterotrophic microorganisms as denitrifiers, is thought to be positively correlated with root biomass (Bengtson et al., 2012;Orwin et al., 2010). However, denitrification capacity was not affected by plant community composition and N 2 O emissions were not correlated with root biomass, suggesting that root biomass alone was not a good indicator of carbon availability in our study. These results also indicate that future studies should consider other traits or factors not measured in our study to unravel the full potential of plant community composition to improve N-cycling.

| Relationships between N pools and soil N-cycling processes
Our results shed light on the relative importance of N-cycling processes as regulators of N pools in intensive agroecosystems.
Earthworm biomass was a particularly pivotal player for N-cycling, showing positive associations with all the evaluated variables (Table 1). By stimulating mineralization from plant residues and soil organic matter, earthworms can increase soil mineral N and N leaching (Dominguez et al., 2004), N 2 O emissions (Lubbers et al., 2013) and plant N uptake . As opposed to the prominent role of earthworms, AMF colonization showed a general absence of direct relationships with biological drivers of N-cycling or with the measured N pools and fluxes. By improving plant nutrient (mainly phosphorus (P) but also N and other plant growth limiting nutrients) and water acquisition, AMF can reduce soil mineral N and N leaching, N 2 O emissions (Storer et al., 2017) and increase plant N retention (Bender et al., 2016). However, since P and N availability were not limiting plant growth strongly in our experiment due to fertilizer application, the chance of a large mycorrhizal benefit through improved plant nutrition may have disappeared (Ryan & Graham, 2018).
The abundance of N-cycling microorganisms was related to the emission of N 2 O (Table 1), which was probably mediated by the presence of T. repens. As explained above, legume presence increased the availability of NH + 4 in the soil (Figure 2), resulting in higher abundance of AOA (91%) and AOB (38%), and accordingly the potential N 2 O emissions from nitrification (Prosser et al., 2019). Trifolium repens also had positive effects on the abundance of all the denitrification genes, and marginally increased the ratio of N 2 O-producers (nirK + nirS) to N 2 O-reducers (nosZI + no-sZII; Figure S4), and as expected this ratio was positively correlated with N 2 O emissions (r = 0.25; p < 0.05). Therefore, T. repens probably also increased N 2 O emissions from denitrification. Some authors argue that the N 2 O emissions induced by the growth of legumes may be estimated solely as a function of the above-and below-ground N inputs from their residue during pasture renewal (Rochette & Janzen, 2005). Here, we show that, to some extent, the positive feedback between legumes and the abundance of N-cycling microbial communities may be a hitherto overlooked mechanism contributing to the higher emissions of N 2 O when legumes are grown.

| Implications of our study and future research directions
Intensively managed grasslands are commonly renewed after 2-5 years (Nevens & Reheul, 2003;Schils et al., 2002), and therefore, the effects of any intervention must become apparent within a relatively short timeframe. Our results show that manipulating plant community composition has a profound effect on multiple facets of N-cycling already in the second productive year. However, the extent and direction of such effects and the ecological processes involved should be confirmed in longer term studies. For example, it remains unknown whether the legume effects persist when soil fertility levels increase after several years of legume use. Such long-term studies can also clarify whether legumes increase (Fornara et al., 2009) or decrease (Prommer et al., 2019) soil carbon sequestration in fertilized systems.
Legume effects on P-cycling should also be addressed, since legumes have high P requirements compared to grasses (Haling et al., 2016), but also an advantage in P acquisition in P-limited soils (Houlton et al., 2008).
In productive temperate grasslands, grass monocultures are often used to simplify management options (e.g. easier to estimate the harvesting time, more stable forage quality). Here, we show that using plant species mixtures of two or four species instead of monocultures in these systems may increase plant N pools due to complementary resource acquisition, without increasing N losses.
Our experiment also indicates that grass-legume mixtures have strong potential to optimize N-cycling in agroecosystems: they reduced soil mineral N and the potential N losses associated with this pool compared to legume monocultures, and increased plant N retention and the abundance of N-cycling microbial communities compared to grass monocultures. By generating high forage yields of high nutritive value, these mixtures can also lead to improved livestock performance (Lüscher et al., 2014). Combining legumes with grasses with low LDMC and high RLD may provide a promising opportunity to further promote rhizosphere biological interactions that regulate internal soil N-cycling, while ensuring productive and sustainable agroecosystems that retain and use N more efficiently.