Differences in N2O Fluxes and Denitrification Gene Abundance in the Wet and Dry Seasons Through Soil and Plant Residue Characteristics of Tropical Tree Crops

The conversion of forest to agricultural soils is a widespread activity in tropical systems, and its link to nitrous oxide (N2O) fluxes and nitrogen cycling gene abundance is relevant to understand environmental drivers that may interact with climate change. A current challenge to estimating N2O emissions from land use conversion is an incomplete understanding of crop-specific impacts on denitrifier communities and the N2O fluxes driven by differences in the above- and below-ground inputs with crop type. To address this knowledge gap in tree crops, we evaluated N2O fluxes and denitrification gene abundance and their relationships with soil and plant residue characteristics in citrus and eucalyptus plantations in the field and in soil incubations. We found that the accumulated N2O fluxes from soil were lower for the two agricultural field sites than those for their adjacent forest sites in dry and wet seasons. The N2O fluxes were higher in the wet season, and this seasonal difference persisted even when the soils collected from both seasons were incubated under the same moisture and temperature conditions in the lab for 30 days. Increased N2O fluxes in the wet season were accompanied by an increase in soil nirK and nosZ gene abundance, the dissolved organic carbon (DOC) concentration, and the total soil carbon (C) and nitrogen (N) content. In turn, the abundance of denitrifiers, as indicated by nirK, nirS and nosZ gene copy numbers, showed a low but significant positive correlation with soil bulk density. Our results suggest that soil moisture, leaf litter and crop residues influence the seasonal differences in both N2O fluxes and abundance of denitrifiers in citrus- and eucalyptus-cultivated soils, likely through effects on soil physicochemical characteristics. These findings highlight the overwhelming role of environmental drivers that can make investigating microbial drivers difficult in the field and open the possibility for a better understanding of N cycling processes in tropical soils based on paired field- and incubation-based experimentation.

The conversion of forest to agricultural soils is a widespread activity in tropical systems, and its link to nitrous oxide (N 2 O) fluxes and nitrogen cycling gene abundance is relevant to understand environmental drivers that may interact with climate change. A current challenge to estimating N 2 O emissions from land use conversion is an incomplete understanding of crop-specific impacts on denitrifier communities and the N 2 O fluxes driven by differences in the above-and below-ground inputs with crop type. To address this knowledge gap in tree crops, we evaluated N 2 O fluxes and denitrification gene abundance and their relationships with soil and plant residue characteristics in citrus and eucalyptus plantations in the field and in soil incubations. We found that the accumulated N 2 O fluxes from soil were lower for the two agricultural field sites than those for their adjacent forest sites in dry and wet seasons. The N 2 O fluxes were higher in the wet season, and this seasonal difference persisted even when the soils collected from both seasons were incubated under the same moisture and temperature conditions in the lab for 30 days. Increased N 2 O fluxes in the wet season were accompanied by an increase in soil nirK and nosZ gene abundance, the dissolved organic carbon (DOC) concentration, and the total soil carbon (C) and nitrogen (N) content. In turn, the abundance of denitrifiers, as indicated by nirK, nirS, and nosZ gene copy numbers, showed a low but significant positive correlation with soil bulk density. Our results suggest that soil moisture, leaf litter, and crop residues influence the seasonal differences in both N 2 O fluxes and abundance of denitrifiers in citrus-and eucalyptus-cultivated soils, likely through effects on soil physicochemical characteristics. These findings highlight the overwhelming role of environmental drivers that can make investigating microbial drivers difficult in the field and open the possibility for a better understanding of N cycling processes in tropical soils based on paired field-and incubation-based experimentation.
Keywords: tropical soil, denitrifiers, nitrous oxide, citrus, eucalyptus INTRODUCTION Many gaseous nitrogen (N) emissions from soils are a consequence of soil microbial processes (Hallin et al., 2018). In soils, nitrous oxide (N 2 O) is produced by N cycling nitrification and denitrification reactions, which are mediated by enzymes encoded by nirK, nirS, and nosZ genes (Levy-Booth et al., 2014). Denitrification in agricultural soils is a leading source of anthropogenic N 2 O, especially in humid climates (Lammel et al., 2015;Kanter et al., 2016). Atmospheric N deposition is chronically elevated and predicted to increase in tropical regions (Reay et al., 2008;IPCC, 2014) and may stimulate additional N cycling pathways in these regions. A high demand for wood material and orange exports causes eucalyptus and citrus crops to play important roles in the agricultural sector in Brazil (Neves et al., 2010;Cavararo, 2014). However, the impacts of N additions on N 2 O emissions have been studied most often in natural forests and rarely in tropical plantations (Martinson et al., 2013;Mori et al., 2013;Zhang et al., 2014). The effects of agricultural management of tree crops (e.g., citrus orchards and eucalyptus plantations) on the N 2 O balance remain unclear.
It is critical to understand the sensitivity of the microbial communities that regulate ecosystem N cycling to the soil, and the changes in both environmental and crop-specific conditions that accompany natural forest loss via conversion to agriculture. Microbe-mediated N 2 O production begins with the reduction of nitrate (NO − 3 ) to nitrite (NO − 2 ), followed by the reduction of NO − 2 to nitric oxide (NO), which is catalyzed by enzymes encoded by nirK and nirS genes (Pajares and Bohannan, 2016). The conversion of one-third of NO to N 2 O is followed by its reduction to N 2 by N 2 O reductases encoded by nosZ genes, thereby partially mitigating N 2 O release to the atmosphere (Jones et al., 2008;Hallin et al., 2018). The presence of crop residues and other leaf litter affects the rate of denitrification primarily by increasing organic C availability, which is one of the most important factors influencing denitrification (Handerson et al., 2010). While different studies have investigated the links between N 2 O fluxes and key marker genes (Lammel et al., 2015;Liu et al., 2016), few have investigated the interplay between tree crop species, their residues, and denitrifiers that lead to different N 2 O emission profiles from various types of agricultural management.
In this study, we evaluated seasonal patterns in N 2 O fluxes and denitrification gene abundance and their relationships with soil and plant residue characteristics in citrus and eucalyptus plantations. We evaluated the following hypotheses: (1) that citrus-and eucalyptus-cultivated soils emit higher amounts of N 2 O to the atmosphere than their adjacent natural forest soil counterparts and (2) that the abundance of denitrifier genes, as indicated by nirK, nirS, and nosZ gene copy numbers, would be increased in citrus-and eucalyptus-cultivated soils in comparison to that in natural forest soils. We predicted that these observed differences would be driven in part by seasonal environmental variables, the chemical characteristics of plant residues and soil physicochemical characteristics. To evaluate these expectations, we determined N 2 O soil atmosphere trace gas fluxes, nirK, nirS, and nosZ gene copy numbers, and soil physicochemical and plant residue chemical characteristics in samples taken in the dry and wet seasons in the field and using soil and plant residue incubations. The results from our study provide the first comprehensive picture of the complex determinants of N 2 O cycling in the tree crops that are widespread in the tropics.

Site Descriptions
Four different field sites were chosen in Southeastern Brazil; two sites (Citrus and Forest 1) are located in the Sorocaba municipality ( (Figure 1).
Among the four sampling sites, Forest 1 and Citrus are located on São Pedro Farm, with a total area of 431.5 ha with 333.7 ha of conventional citrus orchards. According to the records of 18 August 1965 and 29 July 1970, Forest 1 can be considered to be in an advanced regeneration stage without recorded anthropogenic disturbances within the last 52 years. The conversion from forest to citrus orchard occurred 48 years before our study. The orchards have been planted every 10 years since. On an annual basis, the citrus orchard received 856.8 kg ha −1 of NPK (15-10-15). This dose was split into three applications as follows: the first application was 55.1 kg ha −1 of N as NH 4 NO 3 , 36.7 kg ha −1 of P 2 O 5 as triple superphosphate and 55.1 kg ha −1 of K 2 O as KCl, and the second and third applications were 36.7 kg ha −1 of N as NH 4 NO 3 , 24.5 kg ha −1 of P 2 O 5 as triple superphosphate and 36.7 kg ha −1 of K 2 O as KCl. Although fertilizer recommendations were expressed in kg ha −1 , all the fertilizers were mechanically spread under the canopy, which occupied half of the field area. In contrast, the Eucalyptus site is located in a 38 ha plantation. Fertilization was performed in the furrow once during eucalyptus planting and consisted of 50 kg of N in the form of urea, 250 kg of P 2 O 5 (triple superphosphate), and 50 kg of K 2 O (potassium oxide) per hectare. Broadcast fertilization was performed 9 and 18 months after planting with 55 kg of N, 20 kg of P 2 O 5 , and 75 kg of K 2 O per hectare. The Eucalyptus site was planted 4 years prior to our study. The native 10 ha vegetation area adjacent to the eucalyptus (Forest 2) is in an advanced natural regeneration stage without any recorded anthropogenic disturbance in the last 50 years.
The regional climate is subtropical warm (humid temperatures with a warm summer according to Köppen's classification) with annual precipitation of 1,285 mm and 16.8 • C and 30.1 • C as the minimum and maximum average temperatures, respectively. The air and soil temperature varied <7 and 5 • C, respectively, among the different sampling sites during the gas and soil sample collection. The only soil in the study sites is an Udox in a hyperthermal soil temperature regime according to the classification of Soil Survey Staff (1999).

N 2 O Fluxes From the Soil
The N 2 O flux from the soil was measured using the chamberbased method (Davidson et al., 2002) in the regional dry FIGURE 1 | Sampling site locations for Citrus, Forest 1, Eucalyptus and Forest 2. The chambers for gas sample collection (red and yellow points in the diagram) were installed at each sampling site in dry and wet seasons. Soil and plant residue samples (also the yellow points in the diagram-soil and plant residue sampling cores were randomly chosen next to the yellow point chambers) were taken at each sampling site (for a total of five samples per sampling site at each season).
(July 2016) and wet (January 2017) seasons. Gas sampling was performed at all four sampling sites (Citrus, Eucalyptus, Forest 1, and Forest 2). At each site, eight cylindrical PVC bases (30 cm diameter, 22 cm height) were installed in soils under the tree canopies. All gas sampling was performed between 7:00 and 12:00 a.m. After closing the chambers, 60 mL samples were collected at 1, 10, 20, and 30 min using syringes and were transferred and stored under pressure in 20 mL evacuated penicillin flasks sealed with gas-impermeable butyl rubber septa (Bellco Glass 2048). The gas sampling was performed over 5 days at each site in each season for a total of 1,280 gas samples (i.e., 32 gas samples for each sampling site per day) in order to fit to the soil sampling and it may not represent behavior over the entire seasons. The N 2 O mole fractions were determined by gas chromatography with an electron capture detector (GC 2014 Shimadzu, Columbia, MD, USA). The system was equipped with a HayeSep TM N packed column (1.5 m, 80-100 mesh) for N 2 O separation. Helium was used as the carrier gas. Gas samples were stored at 20 • C and analyzed 1-2 weeks after collection. Measurements of atmospheric pressure, chamber height and air temperature were taken during the gas sampling to determine the total moles of air in the chamber volume. The N 2 O fluxes were calculated for each sampling event in each site and season according to equation (Jantalia et al., 2008):

Soil, Leaf Litter, and Crop Residue Sampling
Soil samples were collected from five different points at each sampling site from the upper 10 cm (topsoil) layer on the 3rd day of gas sampling in both the dry and wet seasons. At each of the 5 sampling points at the Citrus and Eucalyptus sites, crop residues (i.e., leaves, stalk, and husk) (Supplementary Figures 1A,C) were collected from a 30 cm diameter area and stored in plastic bags, and soil samples were collected using a 5 cm diameter aseptic cylindrical corer. The same sampling approach was taken at 5 sampling points in the Forest 1 and Forest 2 sites for soil and plant residues, where litter layer samples (Supplementary Figures 1B,D) were collected in the same manner as crop residues in the agricultural sites. A total of 40 soil samples were collected from the field (4 sampling sites × 5 sampling points × 2 sampling periods). The samples were transported to the laboratory on ice, stored at −20 • C, and processed within 72 h of sampling.

Soil Physicochemical and Plant Residue Chemical Analysis
The soil physicochemical analysis was performed on each of the 40 soil samples. The granulometric analysis was performed as described in Camargo et al. (2009) using the soil texture triangle from IAC (Agronomic Institute of Campinas). Soil moisture was determined by the gravimetric method as described in EMBRAPA (2009). Bulk density was measured by the volumetric flask method (EMBRAPA, 2009). For chemical analysis, a subsample of each sample was air dried, passed through a 2 mm mesh sieve, and shaken in pure water. Of this solution, 100 µL was acidified with 1.5% of 2 M HCl, followed by a purge with synthetic air for 5 min to remove the inorganic carbon. Then, the dissolved organic carbon (DOC) content was determined on a Shimadzu TOC-5000 A analyzer (Shimadzu, Columbia, MD, USA). Another soil subsample was air dried and passed through a 0.149 mm mesh for total C and N determination by dry combustion on a LECO R CN-2000 elemental analyzer (PerkinElmer, Waltham, MA, USA). Soil NH + 4 -N and NO − 3 -N were extracted by 2 M potassium chloride (KCl) and quantified using spectrophotometry as described by Norman et al. (1985) and Krom (1980), respectively. The soil pH in CaCl 2 , soil moisture, total soil porosity, and particle density were analyzed according to EMBRAPA (2009). The determination of particle density was based on the relationship between soil mass and the volume occupied by the soil particles. All analyses were performed in triplicate for each soil sample. Total C and N, the only analyses performed on plant residues (leaf litter and crop residues), was performed by dry combustion on a LECO R CN-2000 elemental analyzer (PerkinElmer, Waltham, MA, USA) after the samples were completely dried at 65 • C and ground.

Soil, Leaf Litter, and Crop Residue Incubation Experiment
To evaluate the N 2 O fluxes and denitrification gene abundance across our sites while normalizing the influence of environmental parameters (e.g., moisture and temperature variability) on the soil, a set of microcosms were prepared with the soils and plant residues collected from the field sites in the dry and wet seasons. The soil characteristics such as the water holding capacity (WHC) of the soil, soil moisture, and soil density were analyzed prior to soil incubation as mentioned above for these soil samples. The WHC was determined by transferring the soil into a metal ring with a paper filter until it was fully covered, followed by a water soaking step until water completely diffused in the ring. The weights of the soil, ring, and wet set (ring + wet soil) were previously noted for further calculations. After all of these analyses, the samples were incubated at 105 • C for 24 h and then weighed. The WHC was calculated from the ratio between the water and soil weight multiplied by 100 (results in %). For each sampling site and season, five microcosms were prepared by placing ∼300 g of soil into 500 mL sterilized Schott bottles (63.58 cm 2 basal area) covered by plant residues. The plant residues from the corresponding site were added on top of the soils to achieve the same coverage as observed in the field (visual evaluation); the forest sites had thicker litter layers than the citrus and eucalyptus site crop residue layers. The microcosms were incubated at room temperature (between 20 • and 22 • C) for 1 day at a moisture level of 45% of WHC, after which the soil moisture was adjusted to 70% of WHC by adding water and weighing. The microcosm soil moisture was maintained at this level for the remaining 30 days by weighing daily to check and adjust for moisture loss. The N 2 O fluxes were measured after the first 24 h (at 45% of WHC) and then at 1, 7, 15, and 30 days following wetting (at 70% of WHC). For each flux measurement, five gas samples were taken from the microcosm headspace using syringes at 1, 30, 60, and 90 min after closure of the microcosm, followed by the respirometric assay proposed by Pitombo et al. (2018). The N 2 O concentrations were determined by gas chromatography, and the fluxes were calculated as described above. On the last day of gas sampling, soil samples were taken from each of the 40 microcosms (4 sampling sites × 5 microcosms × 2 sampling periods) and stored at −20 • C until the DNA extraction process.

Isolation of DNA From the Soil and Quantitative Real-Time PCR Assays for the Denitrification Genes
Genomic DNA was extracted from 250 mg of soil from the field and soil incubation experiments using the Power Lyzer TM Power Soil R DNA Isolation Kit (MoBio Laboratories, Carlsbad, CA, USA) according to the manufacturer's instructions. The DNA purity and concentration were assessed spectrophotometrically (Nanodrop ND-1000, NanoDrop Technologies, Inc., Wilmington, DE, USA) to determine absorbance at the following wavelengths: 230, 260, 280, and 320 nm. All 80 DNA samples were stored at −20 • C until use. Quantitative real-time PCR (qPCR) using the nirK, nirS, and nosZ genes as biomarkers was performed to assess the abundance of the denitrifying bacterial community in each of the 40 soil samples from the sampling sites and 40 incubation experiment samples. Amplicons of Sinorhizobium meliloti (DSM 30135), Pseudomonas fluorescens (DSM 50090), and Bradyrhizobium japonicum (DSM 1755) were used as standards for the nirK, nirS, and nosZ genes, respectively. DNA standard curves were generated by dilution series of 10 2 -10 7 copies µL −1 using duplicate 10-fold dilutions of S. meliloti, P. fluorescens, and B. japonicum standard DNA.
The following primer pairs were used for qPCR of nirK, nirS, and nosZ gene fragments ( Table 1).
Each 10 µL reaction contained 5 µL of SYBR green master mix (Applied Biosystem, Foster City, CA, USA), 0.7 µL of each primer (0.3 µM for nirK and nirS and 0.5 µM for nosZ), 0.5 µL of bovine serum albumin (BSA; 0.12 mg mL −1 ) and 1-2 ng µL −1 (final concentration) of template DNA. The following conditions were chosen for nirK amplification after optimization: initial denaturation for 10 min at 95 • C and 40 cycles of 30 s at 95 • C, 30 s at 58 • C, and 40 s at 72 • C. The PCR conditions for nirS amplification were as follows: initial denaturation for 2 min at 94 • C and 50 cycles of 15 s at 95 • C, 1 min at 57 • C, and 1 min at 72 • C. For nosZ amplification, the following PCR conditions were used: initial denaturation for 10 min at 95 • C and 45 cycles of 15 s at 95 • C, 30 s at 60 • C, and 30 s at 72 • C. A negative (no template) control reaction in which soil DNA was substituted for nucleasefree water was used in qPCR assays. PCR amplification and product quantification were performed using the QuantStudio TM 3 Real Time PCR System (Applied Biosystems, Foster, CA, USA). A melting curve analysis of amplicons was conducted to confirm that the fluorescent signals originated from specific amplicons and not from primer dimers or other artifacts. Automated analyses of PCR amplicon quality (e.g., PCR baseline subtraction, Ct threshold set to the linear amplicon phase) and quantity were performed using QuantStudio R Design and Analysis Software (Applied Biosystems).

Statistical Analysis
A Tukey's HSD test was used to determine the significance of the differences in N 2 O fluxes between dry and wet seasons within the same sampling site and agricultural and adjacent forest site in the same season based on data obtained in field and incubation experiment. The statistical comparison of N 2 O fluxes was performed using Statistica v. 10.0 software (Statsoft Inc., Tulsa, OK, USA).
Pearson's correlation analysis was performed to explore the relationships among the abundances of denitrification genes (nirK, nirS, and nosZ), N 2 O fluxes, and soil physicochemical characteristics by using the function "prcomp" in the R packages "factoextra" and "FactoMineR" (version 3.3.3; R Core Team, 2017). The explicit relationships among these variables were examined by constrained ordination generated by principal component analysis (PCA). The PCA and Pearson's correlation analyses were performed based on a data frame in which all variables had the same row length. The field N 2 O flux data in these analyses were based on averages by "day" across "chamber, " resulting in eight unique flux measurements per sampling site per season. The soil incubation N 2 O flux measurements were averaged by "day" across microcosm bottles only.
Linear mixed effects models (R version 3.3.3; R Core Team, 2017) with "chamber" nested in "day" were used to test N 2 O fluxes across sites and seasons; P-values were obtained using Satterthwaite's approximation of degrees of freedom (lmerTest). Mean N 2 O fluxes were compared with Tukey HSD and were considered significantly different at P ≤ 0.05 (lsmeans). Data for Citrus and its adjacent forest (Forest 1), and Eucalyptus and its adjacent forest (Forest 2) were analyzed independently; four mixed effects models were tested ( Table 2). Model 1 used field data from Citrus and Forest 1; Model 2 used field data from Eucalyptus and Forest 2; Model 3 used lab data from Citrus and Forest 1; Model 4 used lab data from Eucalyptus and Forest 2. All models contained data from both wet and dry seasons. Analyses of Models 2-4 were done on the natural log scale; constants were added in these models to make all N 2 O flux values >0 so that data could be log-transformed. Constants' units were in mg m −2 day −1 in the field data and mg kg soil −1 day −1 in lab data. LSmeans were back transformed and the respective constant was subtracted for presentation and interpretation.
Linear mixed effects models were also conducted using total gene abundances as the response variable. These models had the same data groupings as above; replicate was used as the random effect.

Field and Soil Incubation Fluxes of N 2 O
The field N 2 O fluxes increased by 10-77% from the dry to wet season. At all sites, the field average N 2 O fluxes were lower in the dry season than in the wet season (Figure 2A). The N 2 O fluxes were lower for the two agricultural sites than those from their adjacent forest sites in both the dry and wet seasons. These differences in average N 2 O fluxes were significant (Tukey's HSD, P < 0.05) between the dry and wet seasons and between agricultural sites and their respective adjacent forest sites (Figure 2A), except for the Eucalyptus and Forest 2 sites that showed no difference between the dry and wet seasons (Figure 2A). The accumulated flux for dry season was between O flux was measured as mg -1 m -2 day -1 in the field and mg kg soil -1 day -1 in the lab.
The incubated soils demonstrated different trends in average N 2 O fluxes with season and site than those observed in the field ( Figure 2B). The incubations of soil and plant residues collected in the dry and wet seasons showed higher average N 2 O fluxes from the soils collected from the Eucalyptus, Forest 1, and Forest 2 sites in the wet season than those collected in the dry season even though soil moisture was fixed at 70% of WHC throughout all soil incubations ( Figure 2B). The Citrus site showed the opposite pattern, with higher average N 2 O flux in the dry season than in the wet season. In the Eucalyptus, Forest 1, and Forest 2 sites, the differences in average N 2 O fluxes between the soils collected in the dry and wet seasons from a given sampling site were significant (P < 0.05), while the difference was not significant for the Citrus site ( Figure 2B).
A significant interaction between site and season was observed for N 2 O flux in model 3 (P = 0.009, t-value = 2.71; lmerTest). Lsmeans determined no difference in Forest 1 Wet Season and dry season fluxes, which had the greatest mean N 2 O fluxes (11,613.3 and 26,131.7 mg N-N 2 O kg soil −1 day −1 , respectively); Citrus Wet Season had the smallest mean N 2 O flux (685.1 mg N-N 2 O kg soil −1 day −1 ); there was no difference between Citrus Dry Season mean fluxes (7,925.3 mg N-N 2 O kg soil −1 day −1 ) and any of the other groups' (P < 0.05; Table 2). Model 4 showed site and season were both significant predictors for N 2 O flux (P < 0.001, t-value = 4.29; P = 0.04, t-value = 2.10, respectively); lsmeans determined the wet season had a greater mean flux (24,604.7 mg N-N 2 O kg soil −1 day −1 ) than the dry season (2,685.7 mg N-N 2 O kg soil −1 day −1 ); and forests had a greater mean flux (13,172.3 mg N-N 2 O kg soil −1 day −1 ) than agricultural sites (4,843.6 mg N-N 2 O kg soil −1 day −1 ) (P < 0.05; Table 2).

Soil Physicochemical and Plant Residue Chemical Characteristics
The soil physical factors varied across the sites but did not change significantly with season. The range of soil texture fractions across all sites was 9-54% clay, 20-66% sand, and 12-35% silt. The soils were described as Forest 1 soil as clay, Citrus soil as clay-sandy, Forest 2 soil as sandy and Eucalyptus soil as clay. The soil pH ranged from 4.5 to 7.0 for the dry season and from 5.0 to 6.9 for the wet season. The pH was the lowest at the Eucalyptus site and the highest at the Forest 1 site in both seasons. These soil factors differed (Tukey's HSD, P < 0.05) between the Citrus and Forest 1 sites in both seasons and between the Eucalyptus and Forest 2 sites in the dry season only ( Table 3) but not between dry and wet seasons at the same site (Tukey's HSD, P > 0.05).
Soil C and N varied with site and season in distinct ways for the tree crops and forests. Total soil C, N, and DOC were higher in Forest 1 than Citrus site in both seasons ( Table 3). The same pattern was not observed for the Forest 2 and the Eucalyptus site, where differences were typically small. In general, soil C, N, and DOC increased by 121-162%, 117-181%, and 130-260%, respectively, in the wet season when the difference was significant (e.g., increase in DOC for the Eucalyptus and Forest 2 sites). The balance of the pools of C and N, expressed as the C/N ratio, was lower in the soil from the agricultural sites than from the forest sites (indicating more N) and tended to decrease further into the wet season.
The plant residue pool, which represents key matter inputs to the soil, was highly dynamic for the tree croplands but not for the forests. The plant residues (crop residues) in the Citrus and Eucalyptus sites were twice as high in C and N in the dry season compared to those in the wet seasons, while a corresponding change in the forest residues was not observed. The statistical analysis showed significant differences for total C and N of the plant residues between the Citrus and Forest 1 sites for both seasons and between Eucalyptus and Forest 2 in the dry season. The C/N of the plant residues was stable across the wet and dry seasons at all sites except for the Citrus and Eucalyptus sites, where the crop residue was higher in N in the dry season (C/N∼13.0 and ∼25, respectively) than in the wet season (C/N∼17.0 and ∼32, respectively).
The levels of the key denitrifying substrates nitrate and ammonium were the most variable in the forests, which drove the dynamics of the observed trends with site and season. Nitrate was higher in the forests than in the croplands, and the levels were significantly different (Tukey's HSD, P < 0.05) between the Citrus and Forest 1 sites in the dry and wet seasons and between the Eucalyptus and Forest 2 sites in the wet season ( Table 3). The nitrate levels did not vary significantly with season, but the ammonium levels increased (between 4 and 197%) in the wet season for all sites except Forest 2. Ammonium increased significantly in Forest 1 and decreased in Forest 2 in the wet season, which drove their corresponding differences from the adjacent tree croplands that had lower significant increase on levels of ammonium across the seasons.

Denitrification Gene Abundance
The genetic markers of the denitrifiers (nirK, nirS, and nosZ) increased in abundance from the dry to wet season in all sites, except the Citrus site. The abundance of these three denitrification genes, as determined by qPCR (expressed in 10 5 gene copies per gram of soil), indicated that the nirK and nosZ gene copy numbers were generally highest in the wet season in both the field and incubation experiments ( Table 4). These differences were for all sites and conditions (field vs. incubation), except for the Citrus site in the soil incubations, which demonstrated a higher nosZ abundance in the dry season soil incubations and little difference in nirK. Of all three gene targets, the nirS gene fragment had the lowest abundance at all sampling sites regardless of season or condition, and the differences with site and season were not significant. The gene abundance did not differ between the agricultural and adjacent forest sites in the same season. Finally, using linear mixed effects models, we found no significant effects (P > 0.05) of site and season on the total gene abundance (sum of nosZ, nirK, and nirS) in the field and lab experiments (Table S1).

Relationships Among Denitrification Gene Abundances, N 2 O Fluxes, and Soil and Plant Residue Physicochemical Factors
The relationships between nirK, nirS, and nosZ gene abundance, N 2 O fluxes, and soil physicochemical factors were examined by PCA ordination analysis separately for each agriculture-forest pair from both the field and incubation results. The field data (Figure 3) revealed distinct groups for the geographical locations of the adjacent sampling sites (Citrus vs. Forest 1 and Eucalyptus vs. Forest 2). For the Citrus and Forest 1 sites, there was no association between nirK, nirS, and nosZ gene abundance and soil and plant residue chemical factors such as DOC, total C, or total N, which were related to soil bulk density (Figure 3A). A significant positive association (P < 0.05) was found among DOC, soil and plant residue total C and N, nitrate, ammonium, pH, and N 2 O fluxes at the Citrus and Forest 1 sites. Soil moisture was positively correlated with the soil and plant residue chemical factors, as well as with N 2 O fluxes. The PCA ordination revealed a positive correlation among N 2 O fluxes, ammonium and DOC for the Eucalyptus and Forest 2 sites ( Figure 3B). The denitrification genes showed a positive correlation with bulk density and plant residue total C and N for these sampling sites.
Pearson correlations were determined between all data from each of the four sampling sites in both the field and soil incubation experiments and season (Supplementary Figure 2). A greater number of negative correlations were observed in the agricultural Citrus and Eucalyptus sites (Supplementary Figures 2A,C) than in the Forest 1 and Forest 2 sites (Supplementary Figures 2B,D). For instance, nirK and nirS gene abundance revealed a negative correlation with DOC in the Citrus and Eucalyptus sites, while in both forest sites, the same correlation was positive for the field and soil incubation experiments. The correlations common to both the field and soil incubation experiment data included those between soil total C and N and the nirK, nirS, and nosZ genes. Furthermore, in all sampling sites, soil moisture was positively correlated with N 2 O fluxes for the field measurements. In the soil incubation experiment, DOC was correlated with N 2 O fluxes at all sampling sites (Supplementary Figure 2).

DISCUSSION
Through complementary field-and incubation-based analyses, we characterized the fluxes of N 2 O from the soil and their plant residue covers to evaluate our research hypotheses. In contrast to the first hypothesis that citrus-and eucalyptus-cultivated soils emit higher amounts of N 2 O to the atmosphere than their adjacent natural forest soils counterparts, we observed higher average N 2 O fluxes from the natural forest soils than from the adjacent crop-cultivated soils (Figure 2; Table 2). Season emerged as a key driver of N 2 O fluxes and marker gene abundance for denitrifiers, which did not significantly differ with land use (Table 4; Table S1), in contrast to our second hypothesis. The factors such as soil moisture and plant residue characteristics that may drive the observed differences with season, land use, and their combinations are discussed below.
Our results showed a strong influence of season on N 2 O fluxes, with the N 2 O flux increasing in the wet season in both the field and soil incubation experiments. Field moisture was an important predictor of the N 2 O fluxes in the field and incubations (Figures 3, 4, and Table 2). Clough et al. (2017) also showed higher N 2 O fluxes in soil at high moisture levels once a lower oxygen content and, consequently, an anaerobic environment, stimulate denitrification (Van Spanning et al., 2007). Leaf litter and crop residues can contribute to increased soil moisture retention and may consequently promote N 2 O production (Kravchenko et al., 2017). Forest 1 and Forest 2 soils had thicker  litter layers than the cultivated soils, and our results are therefore consistent with the higher N 2 O fluxes from soils under thicker plant residue layers. In addition, Taketani et al. (2017) stated that soil moisture influences changes in soil microbial communities by increasing nutrient transportation during the high moisture period. Thus, plant residues can promote soil moisture retention, oxygen limitation, nutrient transport, and the corresponding microbial community composition in ways known to promote the denitrification process, a critical component of N 2 O fluxes. We discovered that the drivers of the differences in the soil N 2 O fluxes in the wet and dry seasons were not limited to the soil moisture conditions. If that had been the case, we would have seen the differences in the average N 2 O fluxes observed in the field (Figure 2A) vanish upon incubation under identical moisture conditions in the lab ( Figure 2B). Instead, we found that the soil collected from the two seasons demonstrated persistent differences in N 2 O flux dynamics in lab incubations. Therefore, this persistent and surprising legacy behavior was used to evaluate the key drivers of N 2 O fluxes based on the chemical characteristics of leaf litter and crop residues and soil physicochemical characteristics. The differences among soil texture as showed for the studied sites may be also an important factor on N 2 O production, as previous state by Harrisson-Kirk et al. (2013) and Skiba and Ball (2002). However, considering a Udox soil under a hyperthermal soil temperature regime, which C and N turnover is faster, nutrient cycling may have a more significant effect on N 2 O fluxes.
The N 2 O fluxes were increased with the carbon content due its support of denitrifying organisms, and DOC may have been particularly important. Tiedje et al. (1983) suggested that organic C is more important than oxygen status in determining the denitrifying enzyme content of habitats. Higher concentrations of organic C usually support higher populations of heterotrophs, which include denitrifiers. Positive correlations among the DOC concentration, total soil C and N 2 O fluxes were revealed in most soils in the field and soil incubation experiments. Higher DOC concentrations may drive more N 2 O fluxes as labile C increases and plays a role as an electron acceptor for denitrifying microorganisms (Harter et al., 2014). Marques et al. (2012) showed a strong and positive relationship between the DOC concentration, total soil C and soil moisture when studying Amazonian soils under different land uses. In superficial soils (0-10 cm), as used in this study, the DOC concentration is mostly influenced by the type of plant residue on the soil surface (Marques et al., 2012). Harter et al. (2014) showed a positive correlation between the DOC concentration and N 2 O fluxes in biochar-amended soil, possibly due to the high root quantity in forest fields, which supports DOC transportation throughout the soil in the dry season. This process makes the denitrification process efficient in the absence of water in soil under native forests when compared with that in agricultural fields (Marques et al., 2012).
As the substrate for denitrification, the availability and type of N in the soils and plant residues is an important factor in N 2 O fluxes. In this study, the N 2 O fluxes were positively correlated with soil total N mainly in the forest soils. This relationship has been described by previous studies, specifically for N in the nitrate form (NO − 3 ) (Carmo et al., 2005;Cerri et al., 2006). However, the site with the highest ratios of N to C (lowest soil C/N) in the soil and in the crop residue inputs was the Citrus cultivation site. That site may have had greater N availability due to the annual fertilizer application, but, in fact, it had the lowest N 2 O fluxes and a lower N 2 O emission potential in the wet than the dry season, perhaps due to the increase in C/N in the wet season crop residue. The inorganic-based fertilizers used in citrus orchards are rapidly mineralized and have the possibility to supply NO − 3 in addition to NH + 4 , which is the preferred N species for uptake by most higher plants (Britto and Kronzucker, 2002). The NH + 4 is constantly regenerated via dissimilatory nitrate reduction to ammonium (DNRA) due to the redox fluctuation conditions in tropical soils (Silver et al., 2001;Templer et al., 2008), and with a subsequent influx of O 2 , this NH + 4 becomes re-available to ammonia oxidizing communities, disfavoring the denitrification process. Higher levels of nitrate and ammonia were observed in the forests compared to the cultivated sites that received historical or annual fertilization (see section Site Descriptions), and the concentrations increased in the wet season. Therefore, the site-and season-specific trends in N 2 O flux are driven in part by changes in total N but perhaps most strongly by whether the N input is supplied through plant litter or fertilizer and how that impacts the relevant C/N ratios.
The soil pH may have also contributed to the site differences in N 2 O fluxes, as denitrification rates are pH sensitive. Several studies have established that denitrification rates tend to decrease at low soil pH values (Simek et al., 2002;Saleh-Lakha et al., 2009). Parkin et al. (1985) previously demonstrated a 2-fold decrease in the denitrification rate and a 3-fold decrease in denitrification enzyme activity when soil pH decreased from approximately pH 6.0-4.1, suggesting that prolonged exposure to a low soil pH selected a denitrifier population that was more adapted to the low pH environment. Subsequently, the N 2 O reduction by the acid-tolerant population was insignificant compared to the N 2 O production rates. Although our results did not reveal a significant correlation between the N 2 O fluxes and soil pH for the different samples, the relationship between N 2 O fluxes and soil pH in the Citrus and Forest 1 sites corroborates these previous findings.
The denitrifying gene markers tracked changes in N 2 O fluxes in some but not all cases, and their poor prediction of N 2 O dynamics may indicate that other microbial N cycling processes were also important (Martins et al., 2015). Increased N 2 O fluxes in the wet season in the field and incubation experiments were accompanied by increasing nirK gene abundances in the soil from all sampling sites. However, the gene-based measurements did not predict N 2 O dynamics in all sites, experiments, and seasons (Table S1). Although, the higher levels of nirK than nirS indicate that the N 2 O emission potential by microbial denitrification is consistent with the previous determinations for agricultural soils by Dandie et al. (2011) and Wang et al. (2017). In addition, the abundance of denitrifiers, as indicated by nirK, nirS, and nosZ gene copy numbers, neither differed between the agricultural and adjacent forest sites in the same season nor did they have a clear correlation with the chemical characteristics of the leaf litter and crop residues and soil physicochemical characteristics by site or condition (field vs. incubation). The exception was that the soil density has a low but significant positive correlation with denitrification gene abundance (Supplementary Figure 2); more dense soils may have lower porosity and higher moisture levels. Previous attempts to relate denitrifier abundance to environmental factors or N 2 O emissions have been undertaken in a range of environments with varying results (Rich and Myrold, 2004;Philippot et al., 2009;Wertz et al., 2009;Enwall et al., 2010). Soil denitrifiers have been found not to respond to the environmental properties in grassland and agricultural fields, waterlogging, and changes between ecosystems, while the structure or composition of the denitrifying microbial communities indicated differences (Philippot et al., 2009;Dandie et al., 2011;Wang et al., 2017). Therefore, we acknowledge that although surface soils have a more favorable environment for denitrifying microorganisms, the denitrification gene abundance does not necessarily relate to the real community composition. Despite this, it may be equally plausible that other N cycling pathways not represented by our three marker genes (nirK, nirS, and nosZ) are important factors in some or all of our sites. Additional pathways that produce N 2 O or drive other N cycle conversions are known (Pajares and Bohannan, 2016) although their importance to N 2 O cycling in tropical soils is unclear and variable. While beyond the scope of this study, the characterization of additional N cycling pathways may be necessary to explain the observed discrepancies between denitrification gene abundance and N 2 O fluxes across seasons and sites.

CONCLUSION
Taken together, our findings through complementary fieldand incubation-based analyses showed higher N 2 O fluxes from natural forest soils than from the adjacent citrus-and eucalyptus-cultivated soils, with denitrification gene abundance not significantly different between the natural forest and cropland soils. Based on these findings, both hypotheses of this study were rejected. Additionally, we conclude that the differences observed in the N 2 O fluxes and in some cases the nirK, nirS, and nosZ gene copy numbers are related to seasonal environmental variables and soil and plant residue characteristics, especially the DOC concentration, C and N content and density. Our results highlight the overwhelming role of environmental drivers (e.g., moisture dynamics) that can make investigating microbial drivers difficult in the field and open the possibility for a better comprehension of N cycling processes in tropical soils based on soil incubation experiments, particularly when both methods are comparatively employed as presented here.