During the first 2 months of maize growth, December 2018 and January 2019, the cumulative precipitations were 91 and 181 mm, 48 and 79% of the historic rains for these months. The precipitation in February, March, and April were 237, 194, and 125 mm, 12, 30 and 99% higher than the historic average for these months (Figure S1A). The total precipitation from December 2018 to April 2019 was 827 mm, 98% of the historic average for these periods, but the lower precipitation observed in December 2018 and January 2019 probably affected the maize yield. The monthly mean air temperature during the same period ranged from 21.8 to 24.8°C while the historic values ranged from 19.3 to 22°C, so the monthly mean air temperature was 0.9–2.8°C higher than the historic average for March and January, respectively (Figure S1B).

Soil samples were collected in each field plot at three time points: just before termination of forage grasses and weeds (0 DAT, November 20, 2018), just before maize sowing (16 DAT, December 7, 2018), and just before maize N fertilization (49 DAT, January 19, 2019) (Figure 1). Composite soil samples comprised soil from 8 sampling locations of the maize interrow of each plot to a depth of 10 cm using an Edelman auger. We measured soil water filled pore space (WFPS), pH, NH4+ and NO3- ₋content. The WFPS was calculated with the gravimetric water content determined after drying at 105°C and the soil bulk and particle densities. Soil inorganic N (NH₄ and NO₃) was extracted from fresh soil with KCl (1M) and quantified by steam distillation (32). Soil pH was measured after shaking 10 g of soil with 50 mL of 0.01M CaCl₂ for 30 min and waiting for the soil to settle.

One static chamber (30 cm diameter) was installed in each field plot to measure nitrous oxide (N₂O) and carbon dioxide (CO₂) efflux during the 2018/2019 maize growing season (34). The chamber was placed at 5 cm distance from the planted row. Frequent gas sampling (>2 sampling days per week) was undertaken after sowing and N fertilization followed by at least weekly measurements until the end of the maize season (Table S1). The sampling took place between 9:00 and 10:00 am to represent the mean daily gas flux (35). Gases were sampled 0, 15 and 30 minutes after chamber closure and stored in evacuated glass vials protected by a gas-impermeable stopper. The gases were analyzed by gas chromatography (Shimadzu GC-2014), detecting N₂O and CO₂ concentration by electron capture detector. The gas fluxes were calculated by linear interpolation of concentrations during the sampling period, according with the following equation:

where, F is the efflux (mg C-CO₂ m⁻² day⁻¹ or μg N₂O-N m⁻² day⁻¹), ΔCΔt is the variation of gases concentration while the chamber was closed (mol⁻¹h⁻¹), V is the volume of the chamber headspace (m3), A is the soil area covered by the chamber (m²), m is the molar mass (g mol⁻¹), and V_m is the molar volume of the chamber (m3 mol⁻¹).

The flux calculated by equation 1 represents the daily gas flux. To calculate the cumulative gas emission, the linear interpolation of measured fluxes on sampling events was used to estimate the flux on non-sampled events (36).

Soil DNA was extracted from 0.3 g of soil sampled from the 0-10 cm layer of four replicates in the field plots at two contrasting time points, 0 and 49 DAT. The DNA extraction was performed using Qiagen Power soil DNA extraction kit and the DNA was quantified using Qubit fluorometer (Life Technologies, Carlsbad, CA, USA), the total extracted DNA was normalized and expressed in ng g⁻¹ of soil as a proxy for total microbial abundance. The quality of the DNA was measured using NanoDrop 1000 spectrophotometer (NanoDrop Technologies, Montchanin, USA). The bacterial/archaeal 16S rDNA abundance was quantified using primers 515F (5′ -GTGCCAGCMGCCGCGGTAA-3′) and 806R (5′ -GGACTACVSGGGTATCTAAT-3′). Eight nanograms of the DNA samples were diluted in 10 μL of SYBRGreen (2X) 1 μL of the forward and reverse primer and 7 μL of MQ water (Sigma Aldrich) and the abundance of the genes were quantified using real-time PCR with PCR Detection System (Bio-Rad, USA). The thermal conditions were: 1 cycle of 3 min at 95°C for denaturation; 30 cycles of 30 s at 95°C, 45 s at 52°C and 90 s at 72^oC for annealing; and 1 cycle of 12 min at 72°C for extension. The qPCR run of DNA samples had a standard curve with serial dilution of known concentration of plasmids measured previous using the Qubit fluorometer (Life Technologies, Carlsbad, CA, USA).

The abundance of ammonia oxidizing archaea (AOA), the group of AOO more sensitive to BNI (37), was quantified by real-time qPCR using the amoA gene as reporter. The total reaction volume was 20 μL, using 10 μL of SYBERGreen (2X), 1 μL for each primer, and 2 μL (10 ng total) of DNA template. The primers combination was amoA19F (5′-ATGGTCTGGCTWAGACG-3′), amoA643R (5′-TCCCACTTWGACCARGCGGCCATCCA-3′) for AOA (38). The amplification conditions of PCR were as follows: 1 cycle of 10 min at 95°C; 40 cycles of 60 s at 95°C for denaturation, 30 s at 55°C for annealing, 45 s at 72°C for extension, and 45 s at 72°C for final extension. The melting curve was generated from 65 to 95°C measuring each 0.2°C and maintained for 1 second.

16S rRNA and ITS genes were amplified using PCR. The reaction was prepared using 2.5 μL of PCR buffer (10x), 2 μL of dNTPs mix (2.5 mM), 0.7 of Roche bovine serum albumin (20 mg mL⁻¹), 1 μL of the forward and reverse primers (515F/806R and ITS1F/ITS2R), Takara Hot Start Ex Taq polymerase (5 U μL⁻¹) and 10 ng of DNA samples. The primers were modified to include the forward and reverse adapters for Illumina Nextera and the reverse barcode contained 12-bp Golay barcode (39). The thermo cycler conditions of 16S amplification were: 1 cycle of 3 min at 95°C for denaturation; 30 cycles of 30 s at 95°C, 45 s at 52°C and 90 s at 72°C for annealing; and 1 cycle of 12 min at 72°C for extension. The set up for ITS amplification were: 1 cycle of 3 min at 95°C for denaturation; 30 cycles of 30 s at 95°C, 30 s at 51°C and 30 s at 72°C for annealing; and 1 cycle of 5 min at 72°C for extension. The products of PCR reactions were purified using SPRI bead purification (AMPure XP, Beckman Coulter Genomics, Danvers, MA) and the quality of the amplicons were checked and the average size distribution on Bioanalyzer 2100 with a DNA 7500 chip (Agilent Technologies, Santa Clara, CA). Purified amplicons of 16S and ITS were quantified using Qubit and pooled in two libraries of 16S and ITS at the same concentration (10 ng of each sample). The libraries were sequenced using v3 Illumina MiSeq at the Vincent J. Coates Genomics Sequencing Laboratory at the University of California, Berkeley.

Forward and reverse reads were aligned and paired using usearch [v10.0.240] fastq_mergepairs command (maximum diff = 3) (40). The aligned reads were quality filtered (command fastq_filter with -fastq_trunclen = 250, -fastq_maxee = 0.1), concatenated into a single fasta file, and singletons were removed (command sortbysize with minsize = 2). The resulting aligned and filtered sequences were processed using the DADA2 pipeline (41) to define amplicon sequence variants (ASVs), by first dereplicating the sequences and then denoising unique sequence pairs using the “dada” function with “err=NULL” and “selfConsist=TRUE.” Taxonomy was assigned to each ASV by a Naïve Bayes classifier using the assign_taxonomy. py command in QIIME (42) with reference to the SILVA database accessed from mothur (42). For phylogenetic inference of bacterial and archaeal ASVs, representative sequences for each bacterial OTU were aligned to a SILVA SEED sliced alignment using the PyNAST algorithm (43) and archaeal and bacterial phylogeny was inferred using FastTree (44). All sequence data were submitted to NCBI under BioProject accession number PRJNA746808.

The shoot biomass of B. brizantha, B. humidicola and weeds (fallow) grown during the fall-winter were sampled from the plots with the respective treatments on November 21st, 2018, at the end of the dry season, using 0.25 m² quadrants. The shoot biomass was dried at 65°C for 48 h and the dry masses were extrapolated to Mg ha⁻¹. The weeds from plots with both forage grasses were not sampled because visually they covered <5% of the plot area.

Maize grain was harvested from 10 m in each of two central rows in each plot on May 10th, 2019 and weighed at the field moisture content. Grain subsamples of each plot were dried at 105°C for 24 h and the maize grain yield was extrapolated to 1 ha at 13% moisture, which is the standard moisture for maize commercialization in Brazil.

The biomass produced by the forage grasses and maize not harvested as grain was left on the soil surface as mulch for the no-till system. This plant material is expected to gradually decompose and contribute to the stock of soil organic matter (SOM) over the years.

Descriptive statistics were used to obtain the frequency and distribution of data. We evaluated the effect of cropping system (maize monocrop, maize intercropped with B. brizantha and with B. humidicola) and topdress N fertilization (0 and 150 kg N ha⁻¹) on cumulative potential net nitrification rate, net N mineralization, N₂O and WFPS by analysis of variance (ANOVA) using the sp.plot function of the agricolae package (45) on the R statistical software (46). For significant factors, individual mean comparisons were performed to compare the cropping systems and the topdress N rate within each cropping system using Tukey's test (p < 0.1 or 0.05) in the same R package.

We performed non-metric dimensional scaling analysis (NMDS) to reduce data dimensionality and to observe the variance of community structure aiming to identify differences between the different treatments. Bray-Curtis distances was used for fungi and weighted-unifrac distances for bacteria. The effect of treatments on bacterial and fungi community was tested by perMANOVA test using the adonis function of R package vegan (47). We used DESeq function on DESeq2 package to identify the taxa with significant difference between pairwise comparisons (mono vs.intercrop and B. brizantha vs. B. humidicola) by calculating the log2foldchanges in relative abundances for each time point, we used the p-value adjusted for multiple comparisons by Storey's Q Value using the qvalue function of the qvalue package (48). We considered organisms that were identified at least at the order level and showed all the significant taxa in a bar plot.

The function estimate_richness from phyloseq package (49) to obtain the Shannon index used as alpha diversity. We used FUNGuild (50) to annotate and classify fungal ASVs into trophic modes and functional guilds such as saprotrophic fungi, taxa with multiple trophic modes (i.e., mixotrophic), diverse types of opportunistic fungal pathogens (e.g., Fusarium), endophytes, symbiotic organisms (i.e., Trichoderma), arbuscular mycorrhizae (AMF, i.e., Glomerales) or ectomycorrhiza (EMF). We performed ANOVA post-hoc Tukey test for alpha diversity and relative abundance of each group of fungi as described for other soil parameters.

A principal component analysis (PCA) was performed using fviz_pca_biplot function from the factoextra package (51), with the ellipses drawn using the Euclidean distance from the centroid of each treatment (interaction of cropping system and maize N topdress fertilization) with 95% confidence level. The PCAs were performed for each time point aiming to address the correlations of N cycling parameters (potential net nitrification rates, net mineralization rates and cumulative N₂O emission), soil edaphic factors and microbial groups (soil inorganic N, NH₄:NO₃ ratio, WFPS, pH, total DNA extracted from soil, the abundances of 16S rDNA and AOA, fungi and bacterial diversity, cumulative respiration, fungi trophic modes and key fungi and bacteria taxa), and plant parameters (forage grass biomass produced during the fall-winter-spring period and the maize grain yield of the subsequent growing season).

Additionally, multiple linear regression was performed to relate proxy measurements of N cycling (potential net nitrification rate, net N mineralization and N₂O cumulative emission) using the lm function from R version 3.6.2. To predict the cumulative N₂O emission the considered explanatory variables were potential net nitrification rate, net N mineralization, WFPS, pH, and inorganic N pools (NH₄ and NO₃) sampled 0, 16 and 61 DAT as well as total DNA extracted from the soil and the abundance of 16S rDNA from soil sampled 0 and 61 DAT. The explanatory variables for potential net nitrification rate were net ammonification and nitrification, WFPS, pH, inorganic N pools (NH₄ and NO₃), total DNA extracted from the soil, and the abundance of 16S rDNA and AOA genes. Prediction of net N mineralization was performed using the following explanatory variables: WFPS, pH, inorganic N pools (NH₄ and NO₃), total DNA extracted from the soil and the abundance of 16S rDNA. The relative importance of each explanatory variable was calculated by Lindeman-Merenda-Gold using booteval.relimp function in relaimpo package, the 95% confidence intervals were obtained for 1000 iterations.

The present study (2018/19) maize grain yield averaged 3.5 Mg ha⁻¹ across treatments without topdress N fertilization. N application increased maize yield by 125%, but no effect of intercropping system was observed on grain yield (Figure 2).

Despite the positive effect of topdress N fertilization on maize yield, there was no effect of N addition on biomass of forage grasses and weeds growing after the maize harvest. The highest biomass dry matter yields was observed for B. brizantha (6.6 ± 2.4 Mg ha⁻¹); B. humidicola accumulated on average 1 Mg ha⁻¹ more biomass than weeds growing under fallow plots after maize monocrop (Figure 3).

Averaged across N rates, B. humidicola, which had relatively low shoot biomass yield (Figure 3), increased the potential net nitrification rates before termination (0 DAT) by 1.86 mg NO₃-N kg⁻¹ day⁻¹ compared to fallow and B. brizantha treatments. The plots with B. humidicola and annual N fertilization resulted in higher nitrification potential (e.g., 2.22 mg NO₃-N kg⁻¹ day⁻¹) compared to plots without N fertilization (Figure 4). When the soil was sampled 16 and 49 DAT, plots intercropped with grasses had higher potential net nitrification rates regardless of the forage grass species. Intercropping with forage grasses increased potential net nitrification by 2.63 and 1.09 mg NO₃-N kg⁻¹ day⁻¹ at 16 and 49 DAT, respectively compared to maize monocrop (Figure 4).

In addition to increasing potential net nitrification, plots with B. humidicola mineralized 0.41 mg kg⁻¹ day⁻¹ more inorganic N (NH₄ and NO₃) than plots under fallow at 0 DAT. At this time point, maize topdress N fertilization increased net mineralization rates of plots by 0.71 mg kg⁻¹ day⁻¹ for plots under fallow. At 16 DAT, maize N fertilization had a positive effect only for plots under the intercropping system, increasing net mineralization by 0.36 mg kg⁻¹ day⁻¹. At the later time point (49 DAT), plots under the intercropping system had net mineralization rates 0.24 mg kg⁻¹ day⁻¹ higher than monocropping; maize N fertilization had a negative impact on mineralization of plots with B. humidicola, decreasing the mineralization rates by 0.22 mg kg⁻¹ day⁻¹ (Figure 5).

Nitrous oxide emissions were only affected by the cropping system but not by N fertilization. Plots of maize intercropped with B. brizantha and B. humidicola emitted 12.8 and 4.8 mg N₂O-N m⁻² more than that of monocropping system regardless of N fertilization; intercropped plots with B. brizantha had higher N₂O emission than B. humidicola (Figure 6). Cumulative soil respiration was higher only for plots of maize intercropped with B. brizantha, which emitted 39.3 g CO₂-C m⁻² more than the average of the other treatments (Figure 7) and had higher WFPS values than maize monocrop plots at all three sampling times (p < 0.05) (Figure 8).

The effect of cropping system and N fertilization on inorganic N was only significant at 49 DAT. At this sampling time, soil with maize and B. brizantha intercropping had higher inorganic N, averaging 3.7 mg kg⁻¹ more than plots of maize monocropping (Figure S2).

No effect of treatment was observed on total extracted DNA or the abundance of AOA genes (Figures S4–S6). The only marginally significant experimental effect was observed before forage grass termination, when B. humidicola without maize N fertilization resulted in a higher number of copies of 16S than with N fertilization (p < 0.1).

The alpha diversity (Shannon index) of the bacterial community was not influenced by cropping system or N fertilization of the previous maize crops. However, fungal alpha diversity was positively affected by B. brizantha regardless of the sampling time (Figure 9). At 0 and 49 DAT, we identified 7 and 24 bacterial, respectively, and 27 and 26 fungal taxa positively associated with the monocrop system (Figure S7) while only 4 and 1 bacterial and 3 and 6 fungi taxa positively associated with intercropping system, respectively (Figure S8). Before termination (0 DAT) B. humidicola was positively associated with Nitrospira, a nitrite oxidizing organism, while B. brizantha was positively associated with Nitrosomonadaceae mle1-7, an AOB. However, this effect did not last until the maize season (Figures S9, S10). We also observed that Myxococcaceae, positively associated with B. humidicola at 0 and 49 DAT (Figure S9), could have a significant functional role for N mineralization and it was used as a possible soil biological indicator of N mineralization for the principal component analysis.

Before forage grass termination, soil samples of the B. brizantha plots had greater fungi as well as diversity higher relative abundance of AMF and lower saprotrophs compared to other cropping systems (Figure 9 and Figure S11), however there was no effect of cropping system or maize N rate on any trophic mode for soil sampled at 49 DAT (Figure S11). In our PCA, we observed that AMF vs. EMF and endophyte vs. saprotroph opposed their relative abundances being endophytes the only trophic mode clustered with fungi diversity before the forage grasses termination (0 DAT) (Figure 10). However, at 49 DAT the AMF and endophyte clustered in a direction opposed of that saprotroph while EMF was the only group that clustered in opposition of fungal diversity (Figure 10). Although AMF clustered in opposite direction of C:N ratio at 49 DAT, the Glomerales was still positively associated with C:N ratio of the grasses even after the termination. The positive effect of B. brizantha and B. humidicola on different AMF organisms belonging to the Glomerales family persisted during the maize growing season (Figures S9, S10). One taxa of the genera Fusarium, typically classified as saprotroph organisms that could be opportunistic pathogenic or endophytic, was positively associated with intercropping system (Figure S8). Only fungal diversity clustered with grass biomass for soil sampled after the termination. The presence of soil cover was related to Trichoderma spp. regardless of the time point (Figure S8). Glomerales (AMF), Trichoderma (endophyte) and Fusarium (saprotroph) clustered with the respective trophic mode before the termination (0 DAT) but did not at 49 DAT (Figure 10). As Glomerales, Trichoderma clustered with C:N ratio in opposite direction of endophytes and AMF while Fusarium clustered in opposition to grass biomass and fungi diversity at 49 DAT (Figure 10).

PCA showed that the intercropping system impacted soil N dynamics differently during the grasses and maize growing season. Before herbicide termination, the potential net nitrification rates clustered with plant C:N ratio was associated with B. humidicola treatments and negatively correlated with grass biomass associated with B. brizantha (Figure 10). The biological indicators positively associated with potential net nitrification rates were relative abundance of saprotroph fungi and Fusarium, abundance of AOA, and relative abundance of Myxococcaceae. Net mineralization rate was associated with grass biomass and C:N ratio and clustered with relative abundance of AMF, Nitrosomonadaceae and Glomerales, bacterial diversity, and total DNA. At this time point, the cumulative N₂O emission during the maize growing season clustered with fungi diversity, abundance of 16S rDNA and relative abundance of fungal endophytes while negatively associated with soil potential net nitrification rate and abundance of AOA.

However, in soil samples taken during maize growing season, the net mineralization rate was associated with potential net nitrification rate. The N₂O emission clustered with both potential net nitrification rates and mineralization at 49 DAT, the biological indicators associated with that was fungal and bacteria diversity, and relative abundance of Myxococcaceae. Other soil parameters, such as Glomerales, Trichoderma, Nitrosomonadaceae, Nitrospira, saprotroph, AOA, total DNA, abundance of 16S rDNA and AOA were better associated with C:N ratio than biomass and negatively associated with the AMF and endophytes relative abundance.

In the multiple linear regression model, net mineralization rate was the main explanatory variable of potential net nitrification rate, accounting for 27% of the R² (Table 1). In general, abundance of AOA was negatively correlated with potential net nitrification rate (12% of the R²), while the total extracted DNA (a proxy for microbial biomass), fungal diversity and pH significantly explained potential net nitrification rates (p < 0.05) but explained < 10% of the model variance. The full model explained 77% of the variance of potential net nitrification rates. PCA showed that the effect of intercropping was different at 0 and 49 DAT, being associated with saprotrophs and AOA at 0 DAT and with fungal and bacterial diversity at 49 DAT.

Only soil WFPS was significantly and positively correlated with net N mineralization, accounting for 40% of the model R², and the multiple linear regression model explained 68% of the data variance (Table 1). The PCAs confirmed the strong correlation between soil water content and mineralization rates, except for soil sampled 16 DAT. The soil WFPS was also associated with treatments where biomass production during the fall-spring by forage grasses were higher (Figure 10).

Potential net nitrification and mineralization rates of soil sampled 49 DAT (1 day before topdress N fertilization) were positively correlated with N₂O emission and accounted for 8 and 20% of the model R², respectively (Table 2). PCA showed that cumulative N₂O was mainly correlated with the biomass produced by grasses and weeds before maize growing season and WFPS regardless of sampling time (Figure 10).

At 0 DAT we observed for that forage grass intercropping increased net N mineralization and potential nitrification rates. These changes on soil functions related to N cycling are associated with changes in saprotrophic and mycorrhizal fungi, and bacterial nitrifiers prior to maize growth.

Our results of 16S rDNA amplicon sequencing showed that B. humidicola at 0 DAT, which resulted in higher potential net nitrification rates (Figure 4), was also associated with the nitrite oxidizer, Nitrospira regardless of the maize topdress N rate. The ammonia oxidizing bacteria, Nitrosomonadaceae mle1-7, was associated to B. brizantha. Although B. humidicola was shown to exudate brachialactone that inhibits the activity of Nitrosomonas europeae (52) and Byrnes et al. (15) observed that B. humidicola could reduce the abundance of AOO and net nitrification rates, our results showed the opposite effect of forage grasses on nitrifiers organism. However, when Vázquez et al. (11) compared the same cultivar of B. humidicola used in our experiment with that of the hybrid “Mulato” of the Brachiaria genus, they found no negative effect on gross nitrification rates. Along with the effect of the forage grasses on organisms related to nitrification at the end of fall-spring season (0 DAT), B. humidicola did not reduce but rather increased the potential net nitrification rates, suggesting that forage grass might impact net nitrification rates by increasing inorganic N availability (mineralization).

Although Nitrosomonadaceae mle1-7 was associated to B. brizantha, the net nitrification rates were lower when compared with B. humidicola. At this time point (0 DAT), B. brizantha showed higher relative abundance of AMF and fungi diversity, but lower saprotrophs in comparison with other treatments. One possible mechanism of nitrification inhibition is the competition of soil fungal communities with ammonium oxidizing organisms (AOO) for substrate (17). Teutscherova et al. (12) found that grasses with high BNI potential also had higher root colonization by AMF. Thus, lower net nitrification and mineralization rates in soils with growing B. brizantha when compared to B. humidicola, may be explained by the higher inorganic N consumption by AMF since its relative abundance was higher in this treatment (Figure S11).

The association of Nitrospira and Nitrosomonadaceae mle1-7 with B. humidicola and B. brizantha, respectively, before herbicide termination did not last after the termination (0 DAT), and both forage grasses had higher potential net nitrification and mineralization rates than plots under monocrop. Fanin et al. (13) showed that after an initial N immobilization, different residues tend to have positive net N mineralization in longer term. Plots under the intercropping system showed a positive effect of intercropping on net N mineralization at 49 DAT. As net mineralization rates were increased in plots with residues of both grasses, the hypothesis of the legacy of forage grasses causing a reduction in net nitrification rates due to N immobilization or residual of BNI activity was rejected.

Using ¹⁵N pool dilution, Vázquez et al. (11) observed that genotypes associated with high BNI potential mineralized and nitrified 2.3 and 0.5 μg g⁻¹ day⁻¹ more N, respectively, than genotypes with low BNI potential. Their results suggest that the effect of B. humidicola to reduce net nitrification, AOA abundance and N₂O emissions (10, 15, 53, 54) might be due to their effect on gross mineralization/immobilization but not directly on gross nitrification rates. Higher N mineralization provides more substrate for nitrification and consequently increases the nitrification process (55, 56). The potential net nitrification rates are commonly better associated soil organic C than the abundance of AOA and AOB in soil (57), which might be a result of the impact of soil C on mineralization and consequently nitrification rate (55). In the same direction, our study showed that net mineralization rate was a good predictor of potential net nitrification rate (Table 1), supporting the idea that mineralization is an important bottleneck for soil nitrification (55, 58).

The 3-year legacy of the forage grasses to soil under maize cropping was positive on soil mineralization and nitrification rates, regardless of the species, and only B. brizantha left a positive legacy on fungi diversity either by rhizodeposition during the fall-spring or plant residues (23). Fungal and bacterial diversity was found to be also associated with residue mineralization rates (28). Our results reinforce that tropical forage grasses can select specific organism related to nitrification that affect potential net nitrification rates during the fall-spring season. However, the legacy of these grasses on N mineralization is the most important parameter influencing the potential net nitrification rate, or the potential of the soil to produce and accumulate NO₃ (Table 1). Other parameters positively related to potential net nitrification rates were total soil DNA (Table 1). Relative abundance of Myxococcaceae also clustered with potential net nitrification rate (Figure 10), this group correspond to a predatory organism that lyses and takes up nutrients from other organisms, and is associated with high microbial biomass (59) and responsive to C and N addition (60), suggesting that changes in microbial biomass known to be positively correlated with N mineralization rates (55). Thus, our results obtained in shorter incubations (7 days) support the evidence that grasses impact the bacterial community related to N cycling before the maize growing season, leaving a positive legacy to N mineralization and nitrification rates, as also observed by Vázquez et al. (11).

We found that that B. brizantha produced 5 Mg ha⁻¹ of dry mass more than the fallow plots while B. humidicola only increased the shoot biomass production by 1 Mg ha⁻¹. The biomass production of forage grasses was not affected by the N fertilization on maize. The high demand for N and fast initial growth of the cereal, shading the brachiaria plants and reducing competition, result in low N uptake by the grasses by the end of the maize growing season. Coser et al. (5) showed that only 2.08% of the ¹⁵N fertilizer applied during maize growing season was taken up by B. humidicola. In addition, N mineralization in the clay soil used in our study probably is supplying the nutrient needed for the relatively high grass biomass yields, specially of B. brizantha.

The amount of residue left in the field has been shown to be an important factor controlling soil moisture and consequently N₂O emission in tropical soils (61, 62). Thus, the biomass production by forage grasses may be the main factor driving higher soil moisture, recycling N, and adding labile C consequently increasing N₂O emission during the maize growing season. Plots with Brachiaria intercropping had the highest N₂O emissions. On the other hand, the high yielding B. brizantha resulted in an annual C input of 3.8 Mg ha⁻¹ more than the monocrop system, considering the shoot and root biomass at the top 40 cm soil layer, B. brizantha was shown to accumulate more than 100 kg ha⁻¹ of N in the biomass, but emitted only 0.4 kg ha⁻¹ of N₂O-N over the second maize harvest season of the same experiment in the field (1). Carvalho et al. (63) found that, despite of the higher greenhouse gas emission from the soil as N₂O and CH₄, the intercropping system stored 600 kg ha⁻¹ year⁻¹ of C as SOC, resulting in a negative balance of greenhouse gas emission from this cropping system: those authors estimated a net reduction of 360 kg C-equivalent ha⁻¹ year⁻¹ for the intercropping system in a tropical soil.

The stimulatory effect of plant residues on N₂O emission may be attributed to: (i) increased N supply by mineralization (64), (ii) C addition through plant residue (65), and (iii) the presence of soil cover itself (61). Canisares et al. (1) suggested also that N released shortly after grass desiccation with herbicide could increase N₂O emissions. Using a synthetic and sugarcane straw compared to bare soil, Fracetto et al. (61) observed that the soil cover was responsible for retaining soil moisture and consequently increasing N₂O emission regardless of labile C inputs and N mineralization during straw decomposition. The residues left on the field by the forage reduced evaporation an explain the higher soil moisture for soil sampled at 49 DAT (61, 62).

In our study, cumulative N₂O emissions were correlated with WFPS regardless of the time point of soil sampling, and WFPS explained 26, 65, and 33% of N₂O data variance at 0, 16 and 49 DAT, respectively (Table 2). Higher WFPS results in lower O₂ availability and consequently lower redox potential, creating a more favorable condition for denitrification than the other treatments (66). However, we did not identify strong evidence of anaerobic taxa, which suggest that soil conditions do not match with ideal conditions where denitrification is the main biogeochemical process driving N₂O emission (66).

Soil moisture is also associated with N mineralization and nitrification (55) and our results showed that WFPS was the only significant explanatory variable for net mineralization rate (Table 1). Besides WFPS, the N₂O emission was strongly associated with N mineralization rates and soil respiration as well, mainly at 49 DAT (Figure 10). Soil moisture measurements typically follow soil C content (55). However, we did not measured soil C because several years are needed for management treatments to cause significant changes on soil C content (6). The WFPS correlated well with mineralization because of the changes on soil moisture had temporal effects, while using forage grass biomass and C:N ratio as a reference for labile C input did not follow a temporal change. Thus, the higher N₂O emissions in plots with B. brizantha mulch may be attributed to a higher N and labile C pulse after forage grass termination due to its higher biomass yield than B. humidicola. The combination of higher biomass with high soil moisture at field conditions, which could not be decoupled in our experimental design, increased N mineralization, consequently supplying substrate for nitrification and denitrification (64).