Lake Lugano is separated into a permanently stratified northern basin and a monomictic southern basin. This study focuses on the 95 m-deep southern basin. Water samples and incubation water were collected with a 5L Niskin bottle at the Figino Station (45.95°N, 8.90°E) during a sampling campaign on November 5, 2013. Profiles of dissolved O₂, temperature, salinity, and pH were collected by a conductivity, temperature, and depth sensor (CTD). O₂ profiles were calibrated by Winkler titration. Water from the Namibian Upwelling zone was collected by hydrocast with a 10L-Niskin bottle rosette at station 89 (20.65°S, 10.95°E) on January 28, 2014 during the NamUFil cruise of the R/V Meteor.

Water samples for NH4+ and NO3− concentrations, as well as NO3− isotope measurements were immediately filtered through 0.22 μm-pore sterivex filters (Millipore) and then frozen within 2 h of sampling. NH4+ concentrations were measured fluorometrically (Holmes et al., 1999). NO2− concentrations were determined by converting NO2− present in 10 ml of sample water to N₂O by azide reduction (McIlvin and Altabet, 2005) and then quantifying the amount of N₂O in each sample by gas chromatography-isotope ratio mass spectrometry (GC-IRMS, see below). NO2− concentration standards were prepared in 10 ml of distilled water and in lake-water or seawater, and were analyzed by GC-IRMS along with the samples. For NO3− concentration and isotopic measurements, sulfamic acid was used to remove NO2− prior to analysis (Granger and Sigman, 2009). NO3− concentrations were measured by reduction to NO with Vanadium (III) and chemiluminescence detection (Braman and Hendrix, 1989). Nitrate N and O isotope measurements of duplicate samples were performed by conversion of NO3− to N₂O using the denitrifier method (Sigman et al., 2001; Casciotti et al., 2002) and subsequent purification and analysis of this N₂O with a modified purge-and-trap gas bench GC-IRMS (Thermo Finnigan DeltaV Plus) system. Isotopic calibration was performed by concurrent analysis of NO3− isotope standards USGS 32, USGS 34, and USGS 35 (Casciotti et al., 2008). N and O isotopic data are reported on the permil (‰) scale referenced to air N₂ and Vienna Standard Mean Ocean Water (VSMOW), respectively (δ¹⁵N = ([¹⁵N]/[¹⁴N])_sample / [¹⁵N]/[¹⁴N]_{air_N2} − 1) × 1000‰ and δ¹⁸O = ([¹⁸O]/[¹⁶O]_sample / [¹⁸O]/[¹⁶O]_VSMOW − 1) × 1000‰).

Samples for N₂O concentration and isotope analyses were taken by overfilling 160 ml glass sample bottles twice from the bottom through a plastic hose connected to the Niskin outlet. The Lake Lugano N₂O samples were preserved by adding 100 μl of saturated mercuric chloride solution (HgCl₂) after a headspace was added by pipetting 1 ml of water off the top of each bottle. Each bottle was then sealed with a butyl rubber septum (VWR, 5483369) and aluminum crimps (CS Chromatographie, 300219). The marine samples were preserved by adding 5 ml of 10 M sodium hydroxide (NaOH) to the bottom of each bottle with a syringe (Mengis et al., 1997), pipetting 1 ml of water off the top for headspace, sealing with butyl septa and aluminum crimps, and then shaking vigorously to distribute the NaOH. Lake Lugano N₂O samples were analyzed within 1 week of collection. Marine samples were analyzed within 3 months of collection. The total N₂O in each sample was purged with carrier helium directly into a customized purge-and-trap system (McIlvin and Casciotti, 2010) and analyzed by continuous-flow GC-IRMS. Duplicate N₂O samples at each depth were collected for the Lake Lugano profile and one sample from each depth was analyzed for the Namibian Upwelling profile. N₂O isotope ratios were referenced to N₂O injected from a reference N₂O tank (≥99.9986%, Messer) calibrated on the Tokyo Institute of Technology scale (Mohn et al., 2012) for bulk and site-specific isotopic composition by J. Mohn (EMPA, Switzerland). Ratios of m/z 45/44, 46/44, and 31/30 signals were converted to δ¹⁵N-N₂O (referenced to N_{2_AIR}), δ¹⁸O-N₂O (referenced to Vienna Standard Mean Ocean Water), and site-specific δ¹⁵N^α and δ¹⁵N^β-N₂O according to Frame and Casciotti (2010), with an additional two-point correction (Mohn et al., 2014) using measurements of two isotopic mixtures of N₂O in synthetic air (CA-06261 and 53504; kindly provided by J. Mohn). N₂O concentrations were calculated by converting the N₂O sample peak areas measured by GC-IRMS to N₂O standards prepared by converting NO3− to a known quantity of N₂O by the denitrifier method (McIlvin and Casciotti, 2010). At each depth, temperature and salinity data were used to calculate the N₂O concentrations at equilibrium with the atmosphere according to Weiss and Price (1980), based on atmospheric partial pressures reported by NOAA ESRL Global Monitoring Division (http://esrl.noaa.gov/gmd/). The saturation disequilibrium (ΔN₂O) was calculated as the difference between the measured N₂O concentration and the atmospheric equilibrium concentration, with positive values corresponding to oversaturation.

A list of all incubation treatments is provided in Table 1. Water for the Lake Lugano incubations that was collected at 17 m depth was poured into opaque 10 L HDPE canisters (Huber, 15.0250.03), stored in the dark for ~5 h during transport back to the laboratory, and then amended with ¹⁵N-labeled incubation reagents. Water for the seawater incubations was drawn from 200 m depth and immediately mixed with the ¹⁵N-labeled substrates inside 3.4 L LDPE drinking water containers (Campmor, 81027). A dilution (1:2) of 30% hydrochloric acid (Fluka TraceSelect, 96208) with milliQ water was added to water for reduced-pH incubations of Lake Lugano water and mixed immediately before NH4+ and NO2− substrates were added. Tracer ¹⁵NH₄Cl (98.5%) and Na¹⁵NO₂ (99.2%) purchased from Cambridge Isotope Laboratories (NLM-658 and NLM-467) were paired, respectively, with NaNO₂ and NH₄Cl with natural abundance (NA) isotopic compositions, so that each incubation received either 1 μM ¹⁵NH4+ + 1 μM NA NO2− or 1 μM NA NH4+ + 1 μM ¹⁵NO2−. For each set of experimental conditions (Table 1), 10 acid-washed 160 ml glass incubation bottles (Wheaton, 223748) were rinsed with milliQ and lake or sea water, and then filled with 120 ml of incubation water and closed with gray fluorobutyl PTFE-lined septa (National Scientific, C4020-36AP) and aluminum crimps. Headspace O₂ concentrations of the reduced-O₂ incubations were adjusted by displacement with either high purity (99.999%) helium for the Lake Lugano incubations, or N₂ for the Namibian Upwelling incubations. O₂ in the headspace was quantified using a gas chromatograph with an electron-capture detector (SRI 8610C), and O₂ concentrations were calculated according to Weiss and Price (1980). During the Lake Lugano incubations, one bottle was sacrificed at the beginning of the incubation, and three bottles each were sacrificed after ~5, 17, or 30 h. During the Namibian Upwelling incubations, one bottle for each set of experimental conditions was sacrificed at the beginning of each incubation, and three bottles were sacrificed after ~12 and ~24 h. Bottles were incubated in the dark at 20–21°C for both experiments.

Immediately before an incubation bottle was sacrificed, 40 ml of liquid was withdrawn by syringe for colorimetric pH measurements with phenol red (Robert-Baldo et al., 1985) or frozen at −80°C and then stored at −20°C prior to measurements of concentration and dissolved N isotope composition. Seawater incubation samples were also filtered through polycarbonate membrane filters (Whatman nuclepore) with 0.22 μm pores before freezing. NH4+ and NO3− concentration measurements were made as described above. The initial ¹⁵N atom fraction (¹⁵F = [¹⁵N]/[¹⁵N + ¹⁴N]) of NH4+ was calculated at the beginning of the incubation using the added tracer concentration and the ambient concentration of NH4+ before tracer addition, and assuming that the δ¹⁵N of ambient NH4+ was 0‰. The ¹⁵F of NH4+ was not measured during the incubations. The ¹⁵F of NO2− during the incubations was measured using the azide reduction/GC-IRMS method (McIlvin and Altabet, 2005), and the ¹⁵F of NO3− was measured using the denitrifier method (Sigman et al., 2001) after removal of NO2− with sulfamic acid. NOx−-derived N₂O with a ¹⁵N composition <15% was analyzed using the IRMS manufacturer's resistor (3 × 10¹⁰ Ω) and capacitor pairing on the m/z 45 Faraday cup. For more ¹⁵N-enriched N₂O, the resistance on the m/z 45 cup was reduced to 3 × 10⁸. Internal isotope standards for ¹⁵F_NO2− and ¹⁵F_NO3− were prepared in triplicate by mixing NA NaNO₂ or KNO₃ of known δ¹⁵N values with either 99.2% Na¹⁵NO₂ or 98.5% Na¹⁵NO₃ (Cambridge Isotope Laboratories, NLM-157). The total N₂O in each incubation bottle was analyzed for concentration and m/z ion ratios 45/44 and 46/44 as described for natural abundance stable isotope measurements of water column N₂O samples. Additional information about converting IRMS measurements to ⁴⁵N₂O and ⁴⁶N₂O production rates is provided in the Supplementary Material S.1.

Water from the incubation depth was filtered through a single 0.22 μm-pore size 47 mm polycarbonate Nuclepore filter (Whatman). The volume of water filtered from the Namibian Upwelling was 4 L and from the Lake Lugano was 0.3 L. Filters were frozen immediately and stored at ≤ −80°C until extraction. DNA from either environment was extracted using the FastDNA spin kit for soil (MP Biomedicals).

The polymerase chain reaction (PCR) was used to amplify the V4 region of the 16S ribosomal RNA (rRNA) gene of prokaryotes using universal 16S rRNA V4 primers F515 (5′- GTGCCAGCMGCCGCGGTAA -3′) and R806 (5′-GGACTACHVGGGTWTCTAAT-3′; Caporaso et al., 2011). Forward and reverse primers were barcoded and appended with Illumina-specific adapters (Kozich et al., 2013). PCR amplifications were carried out using BioReagents™ exACTGene™ Complete PCR Kit and Core Reagent Sets (Thermo Fisher Scientific) for 30 cycles (Caporaso et al., 2012). Agarose gel electrophoresis was used to separate PCR products of the correct size that were then band-excised and recovered using a QIAquick gel extraction kit (Qiagen). For each library, triplicate PCR products were combined and quantified with a Qubit fluorometric assay (ThermoFisher Scientific) and pooled at equimolar ratios. The final pool was analyzed on an Agilent 2100 Bioanalyzer System (Agilent Technologies) using a High Sensitivity DNA chip to determine the average size of the amplicon pool. Quantitative PCR was performed on the pool using a Biorad IQ5 real time thermocycler (Bio-Rad Laboratories) and Illumina quantification standards (KAPA Biosystems). The samples were sequenced on an Illumina MiSeq (http://www.illumina.com/systems/miseq.ilmn) using 201 nucleotide paired-end multiplex sequencing with 5% PhiX spiked into the run. The library was sequenced on a single flow cell using V2 sequencing reagents, generating paired reads of ~400 bp, with ~150 bp overlap between forward and reverse reads.

Raw 16S rRNA sequence data were initially processed in Basespace using Illumina's metagenomic pipeline (https://basespace.illumina.com/home/index). Merging the paired reads and further analyses were carried out using Axiome with installed PANDAseq and the Quantitative Insights Into Microbial Ecology (QIIME v.1.8.0) software pipeline, including taxaplot for overall bacterial diversity, and calculations of Chao1 to estimate taxon richness and rarefaction curves to calculate species richness for a given number of individual sequences sampled (Caporaso et al., 2010; Masella et al., 2012; Lynch et al., 2013). On PANDAseq, the minimum overlap length was set at the threshold of 0.9. The read length maximum for all sequences was 253 bp. Since the V4 region of the 16S rRNA gene is conserved, reads were removed from further analysis if at least one of the following criteria was met: reads were ≥4 bp shorter than the maximum mentioned above, the number of ambiguous bases was ≥1, homopolymers with >4 bp were present, or sequences did not match any sequences in the database by more than 97% based on the percent coverage in BLAST. The clustering of the sequences into operational taxonomic units (OTUs) was initially performed using UCLUST (Edgar, 2010; Edgar et al., 2011) with a cutoff value of 97% sequence identity. The taxonomic identity of a representative sequence from each cluster was classified using the RDP classifier (Wang et al., 2007) and Greengenes datafiles compiled in October 2012 (downloaded from: http://greengenes.lbl.gov/), which includes chimera screening based on 16S rRNA gene records from GenBank (DeSantis et al., 2006). The confidence threshold was set at the default cutoff value of 80% in order to retrieve potential sequences from the nitrifying taxa Thaumarchaeota and Nitrosomonadaceae.

Phylogenetic trees were constructed using MEGA 7 (Kumar et al., 2016). The 16S rRNA gene sequences of representative AOA and AOB were downloaded from GenBank (Benson et al., 2015) for all phylogenetic analyses. Gene sequences were aligned using the MUSCLE algorithm in MEGA 7 and manually inspected. The final alignments consisted of 253 characters, comprising 481 and 108 taxa for sequences related to AOA and AOB, respectively. Phylogenetic reconstruction was implemented using Maximum Parsimony (MP) and Maximum Likelihood (ML). MP was implemented with complete deletion if gaps were present, and tree-bisection-reconnection (TBR) utilized for tree generation. The resulting trees were obtained via random stepwise addition of sequences, at MP search level 1, with 25 initial trees generated from a heuristic search. ML was implemented using Tamura-Nei model of nucleotide substitution rates, with tree inference based on Nearest-Neighbor-Interchange (NNI). Statistical support for MP and ML trees were obtained from 1000 bootstrap replicates under the same initial settings (only bootstrap values >50% are reported). Pairwise base comparisons of OTUs and their closest relatives were determined using BLAST (Altschul et al., 1990) and reported as % identity values.

The abundances (frequencies) of ammonia oxidizer OTUs in the Namibian Upwelling and Lake Lugano samples were counted using a program written in Python (v.2.7) (https://www.python.org/download/releases/2.7) to search for each OTU that was verified by the above phylogenetic analyses (Lau et al., 2015). The Shannon-Weiner Index and Pielou Evenness were calculated using the BiodiversityR package (Kindt and Coe, 2005) in R version 3.2.2 (http://www.R-project.org/). All partial 16S rRNA gene sequence data are available through the European Bioinformatics Institute (EBI) with project accession number PRJEB11492. The nucleotide sequences for the partial 16S rRNA genes have been deposited in EBI under individual accession numbers (LN908279-LN908785).

The profiles of the N₂O concentration and isotopic composition were useful as qualitative indicators of the depths of rapid nitrification in Lake Lugano. Although the depth of the water used for the Lake Lugano experiments (17 m) was shallower than the N₂O concentration maximum, its coincidence with a clear minimum in the δ¹⁵N-N₂O profile (Figure 2F) suggests that there was rapid in situ N₂O production there. However, the absence of a corresponding extremum in the SP profile (Figure 2H) at this depth suggests that the δ¹⁵N-N₂O minimum probably reflects a minimum in the δ¹⁵N of the precursor N molecule, rather than a change in the mechanism of N₂O formation at this depth. While there is also a minimum in the δ¹⁵N-NO3− profile at the incubation depth, the δ¹⁸O-NO3− profile indicates that NO3− was probably not the precursor of N₂O at this depth. More precisely, at this depth, the δ¹⁸O-N₂O was 44‰ higher than the δ¹⁸O-NO3−, whereas the combination of a 42‰ branching isotope effect (Casciotti et al., 2007; Frame et al., 2014) and a −22‰ kinetic isotope effect (Granger et al., 2006) associated with NO3− reduction to N₂O, should have produced N₂O with a δ¹⁸O that was at most only 20‰ higher than the δ¹⁸O-NO3−.

The similarity of the δ¹⁵N-N₂O and δ¹⁵N-NO3− profiles suggests that both compounds are derived from a shared pool of relatively low-δ¹⁵N precursor N, which could be either NH4+ or NO2−. The δ¹⁵N-NO2− in the top 15 m ranged between −29‰ and −27‰ (Figure 2L). The NH₃ oxidized to NO2− at these depths could have been relatively depleted in ¹⁵N because of rapid remineralization of isotopically lighter organic N. The 20‰ equilibrium isotope effect between NH4+ and NH₃ (Hermes et al., 1985) and/or expression of the isotope effect(s) associated with ammonia oxidation (Casciotti et al., 2003) may have also contributed to production of ¹⁵N-depleted N₂O by ammonia oxidizers at this depth.

N₂O production in the deeper water of this basin has been linked to a NH₂OH-decomposition pathway, largely by the SP value of the N₂O, which approaches a value of ~34‰ in the oxic water between 30 and 70 m (Wenk et al., 2016 and Figure 2H). This particular N₂O formation mechanism happens during ammonia oxidation in aerobic conditions (Sutka et al., 2006; Frame and Casciotti, 2010; Santoro et al., 2011). Interestingly, such a high SP value is also observed in N₂O that is formed abiotically by either the hybrid reaction of NH₂OH with HNO₂/NO2− or by the oxidation of NH₂OH (Figure 1, pathways 1 and 2; Heil et al., 2014). Thus the production of high SP (~34‰) N₂O often observed among AOB and AOA cultures may not distinguish a pathway involving only NH₂OH from a hybrid pathway where N₂O is formed via the reduction of HNO₂/NO2− by NH₂OH. To our knowledge, no data has been reported on the SP of N₂O produced by the enzyme-catalyzed reaction of NH₂OH and HNO₂ described by Hooper (1968), but it seems reasonable to assume that it may also be ~34‰. Thus, while the SP of the N₂O present in the shallower depths of Lake Lugano increases steadily between 5 and 50 m (Figure 2H), suggesting a NH₂OH-dependent N₂O formation mechanism, we cannot use SP alone to distinguish N₂O produced by the reaction between NH₂OH and NO2− from N₂O produced by NH₂OH autoxidation or disproportionation.

The high concentration of N₂O that had accumulated at the depth of the Namibian Upwelling incubation had a relatively low SP (Figure 3H), suggesting that the source of this N₂O was either NOx− reduction by denitrification or NO2− reduction by nitrifier denitrification (Toyoda et al., 2005; Frame and Casciotti, 2010). However, in our incubations we observed hybrid N₂O formation rather than denitrification or nitrifier denitrification. The absence of denitrification and nitrifier denitrification during the incubations is unsurprising, given the relatively high O₂ concentrations (20, 50, or 220 μM), all of which were well above thresholds that limit transcription of norB in denitrifiers (Dalsgaard et al., 2014) and initiation of nitrifier denitrification by AOB (Frame and Casciotti, 2010). The lower O₂ concentrations tested during these incubations were similar to the in situ O₂ concentration at this depth (56.8 μM), suggesting that if denitrification or nitrifier denitrification had occurred in this water mass, it was not happening at the time and location where we sampled it. Frame et al. (2014) have argued that in this region of the South Atlantic, transport and mixing of continental shelf water that is O₂-depleted and contains N₂O produced by anaerobic or suboxic processes with relatively O₂-rich offshore water, can produce water that contains relatively high O₂ concentrations and also N₂O with isotopic signatures that are characteristic of low-O₂ processes like denitrification or nitrifier denitrification.

The difference in ammonia oxidation rates (R_{amm_ox}) calculated during incubations with ¹⁵NH4+ vs. those with ¹⁵NO2− is probably the result of dilution and loss of the added ¹⁵NH4+ due to rapid NH4+ regeneration and uptake, which would both tend to reduce our estimate of R_{amm_ox} during the ¹⁵NH4+ incubations. It also suggests that the rate reduction that we observed in the reduced-pH ¹⁵NH4+ incubations does not necessarily reflect an actual reduction in the rate of ammonia oxidation, but instead a more rapid reduction in the ¹⁵F_NH4+ over time as compared to the control-pH incubations.

In both Lake Lugano and the Namibian Upwelling, zero-order reaction kinetics were assumed for ammonia oxidation, rather than first-order or Michaelis-Menten kinetics. This was probably a reasonable assumption in the Namibian Upwelling experiments, because AOA have an extremely high affinity for NH₃/NH4+, with half-saturation constants (K_m) on the order of 100 nM (Martens-Habbena et al., 2009; Horak et al., 2013). In contrast, AOB have a lower affinity for NH4+ than marine assemblages of AOA (Horak et al., 2013; Newell et al., 2013). The lowest reported K_m among cultivated AOB representatives is 6 μM, and the typical range for cultivated AOB is 0.05–14 mM (Knowles et al., 1965; Keener and Arp, 1993; Martens-Habbena et al., 2009; Jiang and Bakken, 1999b). NH4+ concentrations during the incubations remained well above the K_m of AOA but below the K_m reported for AOB. Since both AOB and AOA were present in the Lake Lugano incubations, we used the results of the ¹⁵NO2− incubations to model first-order rate constants for ammonia oxidation that were 0.46 ± 0.04 M⁻¹ day⁻¹ for the untreated-O₂—untreated-pH incubations, 0.45 ± 0.04 M⁻¹ day⁻¹ for the reduced-O₂—untreated pH incubations, 0.47 ± 0.04 M⁻¹ day⁻¹ for the untreated-O₂—reduced-pH incubations, and 0.42 ± 0.04 M⁻¹ day⁻¹ for the reduced-O₂—reduced-pH incubations. Using the observed concentration of NH4+ (0.92 μM) at 17 m in Lake Lugano and assuming that this concentration is in steady-state, the actual ammonia oxidation rate may be slightly lower than what we calculated using the zero-order reaction model.

Total N₂O yields measured in Lake Lugano (3.64 × 10⁻⁵ to 21.0 × 10⁻⁵ mol ¹⁵N-N₂O / mol ¹⁵N-NOx−) were comparable to those observed by Yoshida et al. (1989) in the western North Pacific, also using ¹⁵N tracer techniques (8 to 54 × 10⁻⁵). They were at the low end of the range observed during growth of batch cultures of the AOB Nitrosomonas marina (10 to 60 × 10⁻⁵) at similar O₂ concentrations (Frame and Casciotti, 2010) and were lower than those observed for batch cultures of the AOA N. maritimus (60 to 100 × 10⁻⁵) and N. viennensis (140 to 180 × 10⁻⁵) in media buffered to pH 7.5 (Stieglmeier et al., 2014b). The N₂O formed during the Lake Lugano incubations was largely derived from intermediates or products of the ammonia oxidation reactions, and relatively little N was incorporated from exogenous NO2−, as indicated by the higher rates of ¹⁵N-N₂O formation during the ¹⁵NH4+ incubations than during the ¹⁵NO2− incubations (Figures 5A–D).

During all of the ¹⁵NH4+ incubations, the measured ratio of ⁴⁶N₂O:⁴⁵N₂O production (0.38–0.67) was lower than the expected ratio produced by random pairing of N derived from NH4+ with the isotope ratio ¹⁵F_NH4+0 (expected ⁴⁶N₂O/⁴⁵N₂O = (¹⁵F_NH4+0)² / (2 × ¹⁵F_NH4+0 × (1−¹⁵F_NH4+0)) = 1.0 for the untreated-O₂—untreated-pH and reduced-O₂—untreated-pH incubations and 0.82 for the untreated-O₂—reduced-pH and reduced-O₂—reduced-pH incubations). Interestingly, Jung et al. (2014) also report relatively high ⁴⁵N₂O production compared to ⁴⁶N₂O production during tracer incubations of AOB and soil AOA cultures in the presence of 99% ¹⁵NH4+ and excess N.A. NO2−. Because they were working with laboratory cultures, their experiments started with almost no background ¹⁴NH4+ and contained no ammonium regenerating processes that would increase ¹⁴NH4+ over the course of the incubations, allowing them to attribute the ⁴⁵N₂O production to a reaction between ¹⁵NH4+-derived N and NO2−-derived N.

In the present study, incubations with ¹⁵NH4+ at the reduced pH produced lower ratios of ⁴⁶N₂O:⁴⁵N₂O than incubations at the untreated pH. Three factors may account for the difference: (1) the ¹⁵F_{NH4+_0} was lower among the reduced-pH incubations (0.62) than it was among incubations at the untreated pH (0.67), (2) there may have been differences in the regeneration rates of NH4+ among experimental treatments, which would have progressively reduced the ¹⁵F_NH4+ at different rates in the two different pH treatments, and (3) an increased contribution of N from an unlabeled N pool (i.e., exogenous NO2−) enhanced ⁴⁵N₂O production. Although we lack knowledge of the evolution of ¹⁵F_NH4+ during our experiments that would allow us to rule out factors (1) and (2), we can be certain that the more rapid production of ⁴⁵N₂O during the reduced-pH ¹⁵NO2− incubations was at least partly the result of additional N-incorporation from exogenous NO2−, given the production of ⁴⁵N₂O in the ¹⁵NO2−-amended experiment (Figure 5D), and that this argues in favor of factor (3) discussed above.

Nitrifier denitrification is unlikely to have contributed to N₂O production during the Lake Lugano experiments. During the ¹⁵NO2− incubations with untreated-O₂—reduced-pH and reduced-O₂—reduced-pH, the formation of ⁴⁵N₂O and not ⁴⁶N₂O (Figures 5C,D) suggests that nitrifier denitrification was not important. Although cultured representatives of Nitrosospira, the most abundant AOB genus in the Lake Lugano, are known to contain norB homologs (Garbeva et al., 2007), whose enzyme products reduce NO to N₂O during nitrifier denitrification reactions (Figure 1, yellow box; Schmidt et al., 2004; Kozlowski et al., 2014), the incubation conditions, such as the relatively high O₂ concentrations (70 and 290 μM), were unlikely to have stimulated nitrifier denitrification. Rather, a hybrid N₂O formation mechanism (Figure 1, pathway 2) that combines one N derived from exogenous NO2− with one N derived from a different, unlabeled N pool, explains the formation of ⁴⁵N₂O in the absence of ⁴⁶N₂O formation during the ¹⁵NO2− incubations. Stieglmeier et al. (2014b) also observed hybrid N₂O formation by AOA cultures, and have suggested that this other pool of N is NH₂OH. NH₂OH is known to form N₂O in the presence of NO2−, both enzymatically (Hooper, 1968) and abiotically (Döring and Gehlen, 1961). Since both NO2− and NH₂OH form in the periplasm (Hollocher et al., 1982), a reaction between these two compounds during the ¹⁵NH4+ incubations is also consistent with the relatively high rates of both ⁴⁵N₂O and ⁴⁶N₂O production (Figures 5A,B). However, without knowledge of ¹⁵F_NH4+ over the course of the incubations, we cannot rule out formation of some ¹⁵N-N₂O through NH₂OH autoxidation or disproportionation (Figure 1, pathway 1). A third possibility is that two intracellular NO2− molecules react with each other to form N₂O via nitrifier denitrification in a system where mixing between endogenous (i.e., periplasmic) NO2− and exogenous NO2− occurs so slowly that the two NO2− pools have distinct isotopic compositions. In the ¹⁵NH4+ incubations, nitrifier denitrification of the relatively ¹⁵N-enriched periplasmic NO2− would produce more ⁴⁵N₂O and ⁴⁶N₂O than we would predict based on the measured isotopic composition of the total NO2−. In the ¹⁵NO2− incubations, nitrifier denitrification of periplasmic NO2− would mainly produce ⁴⁴N₂O, and only a small influx of exogenous ¹⁵NO2− would be reduced to ⁴⁵N₂O after mixing with unlabeled periplasmic NO2− (as we observed during the reduced-pH incubations; Figure 5D). In this way, it would be possible for nitrifier denitrification to produce ⁴⁵N₂O and not ⁴⁶N₂O during the ¹⁵NO2− incubations. We believe that this pathway is unlikely given the relatively high O₂ concentrations in our incubations, but without more detailed knowledge of the size of the periplasmic NO2− pool maintained by the ammonia-oxidizing cells, and the rate at which this pool exchanges with the external NO2− pool, we cannot completely exclude the possibility of nitrifier denitrification.

Reducing the pH during the Lake Lugano incubations increased ¹⁵N₂O production in both the ¹⁵NH4+ experiments and the ¹⁵NO2− experiments. There may be several reasons for this, including structural (e.g., the outer cell membrane may exchange HNO₂/NO2− more rapidly between the periplasm and the outer environment at a lower pH), enzymatic (e.g., a shift toward the optimal pH of the N₂O-producing enzymes in the periplasm), transcriptional (regulation of the genes encoding the enzymes involved in N₂O production may be pH-sensitive), and chemical (due to acceleration in the rates of abiotic reactions that produce N₂O from precursor molecules made by AOB). As discussed below, the most likely explanations are a shift toward the optimal pH of enzymes that catalyze N₂O production and/or the involvement of a non-biological catalyst that accelerates the abiotic reactions that form N₂O.

The majority of ammonia oxidizers in Lake Lugano are Gram-negative bacteria, which means that they have a periplasmic space that is bounded by inner and outer cell membranes. In other Gram-negative species such as Escherichia coli, the pH of the periplasm rapidly changes to reflect that of the external environment (Wilks and Slonczewski, 2007). To our knowledge, it is unknown whether AOB regulate the pH of their periplasm or not, but decreases in the periplasmic pH are likely to enhance the rates of the N₂O-forming reactions of NH₂OH and/or NO2−. Furthermore, if there are differences in the rate at which HNO₂ vs. NO2− cross the outer cell membrane, as suggested by Hollocher et al. (1982), then a pH shift could also alter the rate at which ¹⁵N from the tracer ¹⁵NO2− enters the periplasm and the rate at which the NO2− formed during ammonia oxidation is expelled from the periplasm into the outer environment. This effect would not necessarily change the actual rate of N₂O production, just our ability to observe it with ¹⁵N tracers. However, if this occurs, it is unlikely to be the dominant effect, because we observed increased ¹⁵N-N₂O production during the reduced-pH incubations with ¹⁵NH4+, and in this case, more rapid exchange of NO2−/HNO₂ across the outer cell membrane in the reduced-pH incubations would dilute the periplasmic concentration of ¹⁵NO2−, and therefore decrease the rate of ¹⁵N-N₂O production relative to total N₂O production.

The results of the ¹⁵NH4+-incubations are consistent with an acceleration of the periplasmic reactions that form N₂O. Formation of ⁴⁶N₂O, which is composed only of ¹⁵NH4+-derived N (and should therefore be relatively independent of any influx of external NO2− into the periplasm), was also faster at the reduced pH (Figure 5A). At the enzyme level, Hooper (1968) observed acceleration of N₂O formation with decreasing pH: AOB enzyme extracts converted NH₂OH + HNO₂ to N₂O with an optimum pH of 5.75, via a reaction whose rate increased steadily as the pH dropped from 7.5 to 6. Assuming that a similar reaction also occurs in intact AOB cells, the reaction rate increase observed by Hooper (1968) was large enough to explain the pH effect observed during the Lake Lugano incubations. Decreases in pH also have effects at the level of transcription and protein expression. For example expression of enzymes involved in handling nitrogen oxides in AOB, increases as the pH of the growth medium decreases from 8.2 to 7.2 (Beaumont et al., 2004a). The mechanism for this appears to depend on a transcriptional regulator whose repression is reversed by the presence of NO2− at lower pH values (Beaumont et al., 2004a).

Without some form of catalysis, the rate constants reported for the abiotic hybrid N₂O formation reaction between NH₂OH and NO2− are too small for these reactions to have contributed significantly to N₂O production during our incubations. In particular, the set of reactions thought to produce N₂O that contains one NH₂OH-derived N and one HNO₂-derived N, has a second-order dependence on [HNO₂] (Döring and Gehlen, 1961; Bonner et al., 1983; Schreiber et al., 2012):

(Park and Lee, 1988)

(Grätzelet et al., 1970)

(Döring and Gehlen, 1961; Casado et al., 1983)

This rate-limiting step (i.e., Equation 10, the formation of NO by disproportionation of HNO₂) becomes important at pH values <4.5 (Hooper, 1968), but it is not fast enough to explain ⁴⁵N₂O production during the incubations, given the low rate constant for HNO₂ disproportionation and the low [HNO₂] during the incubations (1.8 × 10⁻¹¹ M at the untreated pH and 4.0 × 10⁻¹¹ M at the reduced pH). Thus a role for a catalyst, whether enzymatic or non-biological, is indicated.

It is important to note that the NO reacting in Equation (11) may be formed through mechanisms other than the rate-limiting abiotic HNO₂ disproportionation step. As mentioned earlier, a number of processes in ammonia oxidizers release NO. For example, nitrite reductases can convert NO2− to NO. In AOB, NO is an intermediate in the catalytic cycle of hydroxylamine oxido-reductase (HAO) (Cabail and Pacheco, 2003) and may be released from NH₂OH in HAO enzyme preparations (Hooper and Nason, 1965; Ritchie and Nicholas, 1972; Hooper and Terry, 1979). In AOA, NO is needed to oxidize NH₂OH to NO2− (Kozlowski et al., 2016). Abiotic reactions between NH₂OH and NO observed by Bonner et al. (1978) (pH 7.8, anaerobic conditions), produced N₂O that was ~75% composed of equal proportions of NH₂OH-derived N and NO-derived N, and ~25% composed of only NO-derived N. If this reaction occurs during ammonia oxidation, then depending on whether NO is derived from NH₂OH, NO2−, or both, these reactions could produce N₂O that is entirely derived from NH₂OH, or some mixture of hybrid N₂O and N₂O that is entirely derived from NO2−.

Chemodenitrification, the reduction of NO3− or NO2− coupled to oxidation of ferrous iron (Fe²⁺) (Buresh and Moraghan, 1976; Rakshit et al., 2008; Picardal, 2012) was an unlikely source of N₂O during the Lake Lugano incubations. In the top 20 m of the lake, concentrations of metals involved in chemodenitrification (Fe and possibly manganese, Mn) were less than the 1.7 μM detection limit when measured by induction coupled plasma optical emission spectrometry (J. Tischer, U. Basel, unpublished data). Furthermore, the ¹⁵F_NO2− values during incubations with ¹⁵NO2− (Figure 4B) were high enough that if chemodenitrification of NO2− to N₂O had been significant, the observed production of ⁴⁵N₂O (Figure 5C) would have been accompanied by detectable ⁴⁶N₂O production (Figure 5D), and it was not. The ¹⁵F_NO3− values were lower than ¹⁵F_NO2− values (Figure 4C), so that mixed reduction of both NO2− and NO3− by Fe²⁺ could explain detectable ⁴⁵N₂O production in the absence of detectable ⁴⁶N₂O production. However, NO3− reduction by Fe²⁺ is much slower than oxidation by NO2−, except when Cu²⁺ > 1.6 μM (Buresh and Moraghan, 1976; Picardal, 2012). We did not measure Cu concentrations in Lake Lugano, but the total dissolved Cu measured in Lake Greifen, a similarly eutrophic lake in northeastern Switzerland, were much lower than this (0.5–2.8 × 10⁻⁸ M; Xue and Sigg, 1993).

Although trace metal concentrations were low in Lake Lugano, metal ions could have played a role in accelerating the hybrid N₂O reaction. Harper et al. (2015) have reported that Cu²⁺ can drive abiotic hybrid N₂O formation rates in activated sludge that are faster than the biologically-catalyzed reactions. Furthermore, NH₂OH disproportionation and oxidation reactions may be driven by the presence of copper and iron ions (Anderson, 1964; Alluisetti et al., 2004).

The relationship between the changes in δ¹⁵N-N₂O and δ¹⁸O-N₂O during the Namibian Upwelling incubations were too small to convert to N₂O production rates, but they still contain information about the mechanism of N₂O formation. In particular, in light of the high degree of ¹⁵N labeling of the NH4+ pool (¹⁵F_{NH4+_0} = 0.94) that was achieved during the ¹⁵NH4+ incubations, random pairing of NH4+-derived N atoms cannot explain the large increase in δ¹⁵N-N₂O relative to δ¹⁸O-N₂O. In the Supplementary Material S.2 we demonstrate that the change in δ¹⁵N-N₂O relative to the change in δ¹⁸O-N₂O observed during these incubations is not consistent with the formation of N₂O composed only of N derived from NH4+. Specifically, N₂O produced with a binomial distribution of ¹⁵N and ¹⁴N derived from NH4+ with this ¹⁵F_{NH4+_0} would produce a much larger increase in δ¹⁸O-N₂O (i.e., more ⁴⁶N₂O) relative to the increase in δ¹⁵N-N₂O (⁴⁵N₂O) than what was observed. In contrast, hybrid N₂O formation (for example, by reaction of highly ¹⁵N-labeled NH₂OH with unlabeled NO or NO2−) would produce a much larger increase in δ¹⁵N-N₂O with almost no change in δ¹⁸O-N₂O (Figure S3).

If hybrid N₂O formation produced ⁴⁵N₂O during the ¹⁵NH4+ incubations, from which N pool was the ¹⁴N atom derived? ¹⁵N was more rapidly incorporated into N₂O during the ¹⁵NH4+ incubations than during the ¹⁵NO2− incubations, as indicated by the larger increase in δ¹⁵N-N₂O (⁴⁵N₂O) during the ¹⁵NH4+ incubations (Table 2). If both NH4+ and exogenous NO2− had contributed equally to N₂O formation, then we would expect approximately equal increases in δ¹⁵N-N₂O when either ¹⁵NH4+ or ¹⁵NO2− was added (¹⁵F_{NO2−_0} = 0.95). The fact that the δ¹⁵N-N₂O increased by less during the ¹⁵NO2− incubations than during the ¹⁵NH4+ incubations suggests that exogenous NO2− could not have been the sole source of ¹⁴N to ⁴⁵N₂O produced during the ¹⁵NH4+ incubations. Possibly, an intracellular pool of unlabeled NO2− was held over from before the ¹⁵NH4+ incubation started. If there was enough of this holdover NO2− to dilute any endogenous ¹⁵NO2− produced by ¹⁵NH4+ oxidation, then reactions between ¹⁵NH₂OH with the holdover ¹⁴NO2− would produce an increase in δ¹⁵N-N₂O (⁴⁵N₂O) without increasing δ¹⁸O-N₂O (⁴⁶N₂O).

Previous studies in aquatic systems have reported some variation in the importance of NH4+-derived N vs. NO2−-derived N to N₂O production. Similar results to ours are reported for incubations of suboxic Black Sea water, where Westley et al. (2006) also found production of ¹⁵N-N₂O with the addition of ¹⁵NH4+ but not ¹⁵NO2−. In the Eastern Tropical South Pacific, above the OMZ (O₂ ≥ 10 μM), the rates of NH4+-derived N incorporation into N₂O (0.01–0.02 nM/day) were similar to what was observed in Lake Lugano, although the yield (as defined in this paper) was substantially higher (80 × 10⁻⁵; Ji et al., 2015). In the North Pacific Gyre, Wilson et al. (2014) observed no changes in δ¹⁵N-N₂O during incubations with either 1 μM NA NH4+ + 1 μM ¹⁵NO2− or 1 μM ¹⁵NH4+ + 1 μM NA NO2−. Critically, however, when they reduced the NA NH4+ addition to 100 nM and added 1 μM ¹⁵NO2−, the δ¹⁵N-N₂O increased significantly over the course of the incubation. This suggests that the rate of ammonia oxidation influences the degree to which N derived from exogenous NO2− can be incorporated into N₂O, with higher rates of ammonia oxidation perhaps flooding the intracellular NO2− pool and preventing N derived from exogenous NO2− from being incorporated into N₂O.

It is difficult at this point to determine whether the same hybrid N₂O reaction mechanism(s) can explain the results of both the Lake Lugano incubations, which were numerically dominated by AOB, and the Namibian Upwelling incubations, which were dominated by AOA. To date, it is not known whether AOA enzymes also catalyze the reaction between NH₂OH and NO2− (Figure 1, pathway 2) as observed for AOB by Hooper (1968). The original AOB periplasmic enzyme complex purified in that study included HAO as well as other enzyme components. No homologs of hao have been identified in AOA genomes, though alternatives have been proposed (Stahl and de la Torre, 2012). Furthermore, if a periplasmic reservoir of NO2− plays a role in the incorporation of NO2−-derived N into N₂O, as we hypothesize here, then differences in the NO2− permeability of the outer membranes/cell walls of AOB vs. AOA could contribute to differences in the degree to which N derived from NH4+ vs. exogenous NO2− contribute to N₂O formation. Unlike AOB, almost all archaea tested have only a single cell membrane bounding the cytoplasm (Albers and Meyer, 2011). Rather than having an outer membrane, AOA have an S-layer protein cell wall separating a pseudo-periplasm from the surrounding environment (Stieglmeier et al., 2014a). Model predictions of AMO protein structure in the soil AOA Candidatus Nitrosotalea devanterra suggest that the membrane-bound enzyme faces outward into the pseudoperiplasm (Lehtovirta-Morley et al., 2016). For future reference during ¹⁵N tracer studies of N₂O production, it would be helpful to confirm that NH₂OH and NO2− both form in the pseudoperiplasm of AOA, and investigate what controls the rates at which exogenous NO2− enters, and periplasmic NO2− exits, this compartment.

During both the Lake Lugano and Namibian Upwelling incubations, more ¹⁵N-N₂O was produced at the reduced-O₂ concentrations (O₂ = 70 and 20 μM, respectively) than at the untreated-O₂ concentrations (O₂ = 290 and 220 μM, respectively). It is well known that N₂O yields by AOB increase during growth at suboxic O₂ concentrations (Goreau et al., 1980), and previous work on the mechanisms causing this increase implicated induction of the nitrifier denitrification pathway at very low O₂ concentrations (e.g., Frame and Casciotti, 2010). Reducing O₂ from 290 to 70 μM in the Lake Lugano incubations nearly tripled the yield of ⁴⁵N₂O and ⁴⁶N₂O during the ¹⁵NH4+ incubations. However, the reason for this increase was probably not increased nitrifier denitrification, since there was no ⁴⁶N₂O production during any of the incubations with ¹⁵NO2− (Figure 5C).

The mechanism conferring this O₂ sensitivity may not necessarily involve direct regulation of enzyme activity. In particular, NO removal by O₂ is a possible abiotic NO-sink that would become more important at higher O₂ and NO concentrations:

(Awad and Stanbury, 1993)

Once NO₂ is formed, in aqueous solutions it tends to react with water to form NO3− and NO2− (Park and Lee, 1988), or react with NO to form N₂O₃ (Grätzel et al., 1970). Martens-Habbena et al. (2015) measured NO concentrations of ~50–80 nM during oxic incubations of N. maritimus with 10 μM NH4+. If similar NO concentrations were produced during incubations in the present study, liquid-phase reactions between NO and O₂ may deplete NO concentrations significantly, with the rate of depletion increasing in proportion to [O₂] as well as [NO]². Thus, abiotic reaction with O₂ may compete for NO with N₂O-forming reactions that also consume NO, particularly when incubation O₂ concentrations are high.

Like O₂, NO tends to partition into the gas phase over the aqueous phase (Schwartz and White, 1981). If it is NO (aq), rather than NO2−, that participates in biological hybrid N₂O formation, an implication is that inclusion of a headspace during incubations of ammonia oxidizers suspended in water will reduce aqueous NO concentrations, and therefore slow down liquid-phase NO-dependent reactions (such as the reaction of NO with NH₂OH to form N₂O). Differences in aqueous NO concentrations might contribute to the discrepancy in the literature over whether reduced-O₂ growth conditions increase the yields of N₂O produced by AOA (e.g., Löscher et al., 2012; Stieglmeier et al., 2014b), particularly if there is variation in aeration procedures and ratios of headspace to liquid volumes.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.