Introduction

Microbial denitrification converts inorganic N (NO3− → NO2−) to gaseous N forms (NO→N2O→N2), accounting for 26-40% of N outputs from natural terrestrial ecosystems worldwide (Houlton and Bai, 2009; Ciais et al., 2013). Such gaseous N losses directly modulate Earth’s climate system via the production of the potent greenhouse gas N2O, and, in N-limited ecosystems, denitrification has potential to restrict the availability of this nutrient for vegetation CO2 uptake and carbon (C) sequestration. Denitrification is arguably the most poorly understood process in the terrestrial N cycle, posing a fundamental challenge to the fields of ecosystem- and microbial- ecology alike (Williams et al., 1992; Beckman and Koppenol, 1996; Davidson and Seitzinger, 2006).

Two basic problems have limited our understanding of terrestrial denitrification. The first is systemic to ecology if not all of science—quantifying the relationship between pattern and scale (Levin, 1992). Soil microbial processes are highly dynamic and variable in space and time; yet they control many different aspects of local ecological process and global biogeochemical cycles (Falkowski et al., 2008). In the case of denitrification, microbial ecologists and biogeochemists have pointed to variations in moisture, C and N availability controlling soil denitrification rates within and among global ecosystems (Bai et al., 2012; Groffman, 2012). The second challenge centers on the composition of the air: N2 is the dominant product of terrestrial denitrifiers, but there are challenges with measuring N2 from natural terrestrial denitrification directly. The Earth’s atmosphere is ~78% N2 by volume, thus posing a fundamental ‘signal to noise’ barrier to measuring denitrification fluxes in the field. Studies using 15N tracers offer short-term snapshots of denitrification by tracking 15N-enriched NO3− into 15N2 and 15N2O; however, this approach is limited in naturally low N environments and is highly sensitive to the O2 levels used during course of incubation experiments (Kulkarni et al., 2014). Extrapolating results from local soil experiments to entire ecosystems thereby represents a perennial challenge in terrestrial denitrification research (Davidson and Seitzinger, 2006).

Here, we combine measures of natural N and oxygen (O) stable isotope composition—which can reflect broader integrated patterns in ecosystem N-cycling (Robinson, 2001)—with quantitative real-time qPCR to advance our understanding of terrestrial denitrification from gene to ecosystem scales (Figure 1). As modern genetic tools have become mainstream, researchers have begun to explore the relationships between microbial community structure and ecosystem functioning (for example, Torsvik and Øvreås, 2002; Frey et al., 2004). Microbial diversity has been measured alongside such important steps in the N cycle as mineralization and nitrification, providing novel insights to the biogeochemical dynamics of ecosystems (Bardgett et al., 1999; Calderon et al., 2001; Jackson et al., 2003). Denitrification studies have similarly identified many microorganisms and enzymes involved in NO3− reduction (Knowles, 1982; Hochstein et al., 1988; Zumft, 1997). The nir (Nitrite Reductase encoding) genes are of particular interest, because they mark the crucial first gas-formation step in the process (Hochstein et al., 1988)—the point at which denitrification becomes a gaseous N loss from terrestrial environments.

Figure 1
figure 1

Detecting terrestrial denitrification (Concept). Microbial genes (small scale) can be considered alongside natural stable isotopic signature patterns in soil (large scale) for denitrification at the nitrite (NO2−) reduction step. This first gas-formation step marks the point at which denitrification becomes a terrestrial N output.

Full size image

We use qPCR techniques to quantify the relative abundance of nitrite reductase-encoding genes nirS and nirK (Hochstein et al., 1988; Zumft, 1997). These genes are functionally equivalent but structurally different—nirS encoding for heme (cytochrome cd1)-containing enzyme NirS and nirK encoding for copper-containing enzyme NirK or Cu-NIR (Hochstein et al., 1988; Zumft, 1997). Generally, denitrifiers possess either nirS or nirK rather than both (Heylen et al., 2006); a recent study found 10 rare organisms that contradicted this general observation, although the authors did not examine whether both genes were functionally active (Graf et al., 2014). We focus primarily on nirS given its widespread abundance across soil types, as supported by previous field and cultured strain studies (Braker et al., 1998; Throback et al., 2004). Furthermore, PCR primers for nirS capture most of the known diversity of organisms possessing the gene, whereas, due to nirK sequence divergence and taxonomic diversity, PCR primer coverage for nirK is more limited (Helen et al., 2016). Nir genes have been linked to denitrification rates in soil incubation experiments across different environmental conditions (Smith and Tiedje, 1979; Patra et al., 2005; Attard et al., 2011). However, whether an organism possesses such genes does not necessarily indicate whether they will be expressed and active at a given point in time, with little field-based inquiry into denitrifier gene-abundance vs function relationships.

Natural stable isotope composition offers a complementary, non-disruptive and integrative tool for investigating soil-denitrification across different spatial and temporal scales (Robinson, 2001). Denitrifiers preferentially consume the lighter, more abundant 14N (99.7% of all N) isotope at the expense of the heavier isotope 15N (0.3% of all N), thereby enriching the 15N/14N of NO3− substrates vs gaseous N products (Mariotti et al., 1981). Over time, such isotopic discrimination (‘fractionation’) elevates the 15N/14N of soil nitrate pools relative to the air (Houlton et al., 2006), and ultimately, ecosystem to global-scale isotopic patterns in the terrestrial biosphere (Houlton and Bai, 2009; Mnich and Houlton, 2015). Other microbial processes could affect isotope signatures in field: nitrification has the potential to lower the 15N/14N of NO3−; and heterotrophic NO3− consumption and other microbial processes (for example, in Dissimilatory Nitrate Reduction to Ammonium (DNRA)) could elevate the 15N/14N of NO3− substrates similar to denitrification (Fang et al., 2015). Denitrifiers have been shown to increase the δ15N and δ18O of NO3− within a small range of characteristic slopes (Lehmann et al., 2003; Houlton et al., 2006; Granger et al., 2008), allowing for dual-isotopic evaluation of denitrification’s effect on ecosystem NO3− availability (Fang et al., 2015). Natural stable isotope budgets have implied utility of forecasting the effect of the terrestrial N cycle on global climate change (Houlton et al., 2015; Zhu and Riley, 2015); however, the importance of microbial-scale denitrification in driving larger 15N/14N patterns in ecosystems remains unclear. A particularly open question involves the extent to which denitrification expresses itself on N pools, given the potential for microsite consumption to eliminate the intrinsic isotope effect of denitrifiers on NO3− isotopes from local to ecosystem scales (for example, Houlton et al., 2006).

In this study, we ask whether the abundances of soil denitrifier genes are linked to natural stable isotope variations, and we ask whether these cross-scale data can clarify the role of microbial denitrification in regulating patterns of terrestrial NO3− availability. Specifically, we combine qPCR-derived nirS gene abundances with N and O isotopes of NO3− to investigate soil denitrification across redwood forest, oak woodland, chaparral and desert biomes. We complement these field observations with a series of in-situ incubation experiments, further isolating the temporal mechanisms behind the cross-system relationships examined. We use these approaches to test two overarching hypotheses.

The first hypothesis is that there is a significant positive correlation between nirS abundance and microbial consumption of lighter N and O isotopes. If supported, this hypothesis would point to a significant link between microbial denitrification at molecular scales and patterns of ecosystem isotopic pools at larger ones. In the absence of such a significant correlation, we would infer that there is no measurable relationship between the abundance of denitrifying genes and isotopic expression. This alternative would point to other processes driving natural isotope abundance patterns.

The second hypothesis is that there is a significant negative correlation among nirS/nitrate isotope composition patterns and soil NO3− concentrations. This finding would imply that denitrification plays an active role in controlling patterns of NO3− availability across ecosystems. However, if denitrification microbial gene abundances and isotopes increase only when NO3− is abundant, then we would infer the opposite—that denitrification increases in concert with NO3− abundance. If the gene abundances and isotopes do not correlate with NO3− levels, we would infer no directional effect either way.

Materials and methods

Study sites

We examined our overarching hypothesis and set of inferences across a matrix of terrestrial sites that occur over a short geographic domain in California (Table 1). Our diverse biome sites range only slightly in latitude (39o 22' N to 34o 47' N) yet represent substantial changes in water availability (Mean Annual Precipitation, MAP, ranged from 1103 to 292 mm across sites) and temperature (−3.9 to 12.0 °C minimum, 33.0–37.8 °C maximum) (Table 1).

Table 1 Description of field sites
Full size table

Sites fell under a mesic to dry Mediterranean climate, characterized by cool, moist winters and warm, dry summers. Two coastal redwood sites were identified at the Caspar Creek Watershed in Jackson Demonstration State Forest, located near Fort Bragg, California. Four oak woodland and two chaparral sites were selected within the Coast Range of the Mayacamas Mountain foothills, in the University of California’s Research and Extension Center. These sites have been exposed to various watershed-scale treatments, including clearcutting (1989 in one redwood site), burning (2003 in two of the oak woodland and 2004 in one of the chaparral sites), prescribed grazing (oak woodland), combinations of burning and grazing (oak woodland), and no disturbance (redwood, oak woodland, chaparral). Total dissolved nitrogen inputs from rainfall and dry deposition at these sites during the time of sampling did not differ significantly, although NO3− deposition was slightly higher in oak woodland and chaparral, dissolved organic N deposition more dominant in the redwood sites (Mnich and Houlton, 2015). The oak woodland and chaparral grazed sites may also receive NO3− inputs from grazer wastes; also, these sites are closer to farmland than the redwood sites and might receive NO3− inputs from fertilizers.

The desert site was located in the Sweeney Granite Mountain Desert Research and Extension Center in the Mojave National Preserve. Owing to patch-scale heterogeneity of the desert site, we divided our sampling scheme into two different categories: ‘under shrub/creosote’ samples were collected directly under the canopy of the dominant creosote bush, Larrea tridentata, and ‘interstitial’ samples were unvegetated. Any NO3− inputs would have to be wind-blown dry deposition, or else fixed from the atmosphere by soil microorganisms, as there was little rainfall at this site.

Field sampling approach

Surface soils (0–15 cm depth, at least five replicates per site) were sampled at all sites over a range of wet (winter) and dry (summer) seasons. This approach allowed us to examine denitrification in the organic soils where C and N concentrations were highest and heterotrophic and rhizosphere activity is most concentrated (Lennon, 2015). For homogenization, soil was sieved to the fine-earth fraction (<2 mm), and roots/organic debris were removed with tweezers. To minimize DNA or isotope cross-contamination, rubber gloves were worn and sampling tools (that is, shovels and trowels) were cleaned with ethanol between collections. Homogenized, fine-fraction, field-moist samples were mixed in half-gallon, resealable polyethylene bags, and subsamples were kept in 60 ml Falcon tubes and immediately put on ice in the field.

Week-long incubation experiments were conducted in-situ at redwood (n=10), chaparral (n=10), desert under-shrub (n=8) and desert interstitial (n=8) sites to isolate soils from leaching and plant uptake effects. At least 100 g surface soil (0–10 cm depth) was homogenized—using the methods described above—and brought to field capacity with high purity 14.6 M-ohm/cm water in half-gallon, resealable polyethylene bags. Prior to sealing the bags, subsamples were collected in 60 ml Falcon tubes and immediately placed on ice for analyses of initial nutrient, isotope and genetic conditions. The bags were then pressed to expel air, sealed to promote an anaerobic environment and buried in place. Field-incubated samples were collected after 7 days and analyzed for chemistry, isotopes and nir abundance.

Soil moisture

Pre-weighed tins were filled with approximately 15 g field-moist, fine-fraction soil and weighed. Soil-filled tins were placed in a drying oven set at 55 °C. Dry weights were recorded after 48 h or when soil was completely dry. Moisture was calculated as water weight of the total sample and was used to transform field-moist weights in future analyses into soil dry weights.

KCl extractable soil nitrate

Soil inorganic N was extracted in-field with a 2 M potassium chloride (KCl) solution within 4 h of sample collection. The method was adapted from Keeney and Nelson to the field-setting (Keeney and Nelson, 1982). HDPE Nalgene containers were triple-rinsed daily in high-purity water for three days prior to usage. The 2 M KCl solution was prepared with high-grade KCl and high-purity water in a clean Nalgene container sealed with a twist-on cap and parafilm to prevent leakage during travel. A 50 ml volume of 2 M KCl and approximately 10 g field-moist soil were both measured into 60 ml-capacity sterile polypropylene specimen containers, which were then tightly sealed with twist-on caps. Samples were shaken vigorously for 5 min, left to settle upright for 5 min, shaken again for 2 min, and then left to settle upright for 2 h. The supernatant was filtered through funnels lined with ashless Whatman No.1 paper filters, which had been pre-washed with the 2 M KCl solution. The filtrate was collected in clean 60 ml-capacity HDPE Nalgene containers, which were immediately stored in a freezer or kept on constant ice in a cooler for later quantitative analyses in our UC Davis lab.

Colorimetric methods were used to quantify KCl extractable soil NO3−. A Griess reagent was added to samples in semi-micro cuvettes, with a 1:1 ratio in most cases, adjusted for greater sensitivity with lower concentrations (Doane and Horwáth, 2003). Vanadium(III) chloride reagent produced a red color indicating NO3−, which was measured after 8 h at 540 nm in a spectrophotometer (Miranda et al., 2001; Doane and Horwáth, 2003). NO3− was calculated based on a standard curve created with stock potassium nitrate solution. A minimum 1 μM NO3− in the KCl extract was required for isotope analyses.

15N/14N and 18O/16O analysis

Natural abundance of 15N/14N and 18O/16O in soil NO3− extracts were analyzed via a ThermoFinnigan PreCon-GasBench interfacing a Delta V Plus isotope-ratio mass spectrometer (Rock and Ellert, 2007) at UC Davis’ Stable Isotope Facility. Samples were stored at −20 °C and thawed and mixed thoroughly before analysis. The denitrifier method was used to convert NO3− (and any trace NO2−) to N2O, which was purged from vials, passed through an Ascarite CO2 scrubber, and concentrated in two subsequent liquid N2 cryo-traps (Sigman et al., 2001; Casciotti et al., 2002). A He stream (25 ml min−1) carried the gas sample through this trace gas concentration system and then through an Agilent GS-Q capillary column (1.0 ml min−1, 30 m × 0.32 mm, 40 °C).

International NO3− isotope standards USGS 32, USGS 34, and USGS 35 (National Institute of Standards and Technology, Gaithersburg, MD, USA) were used to calibrate the instrument, in addition to batch standards to account for drift. International reference standards were used to calculate isotope composition –N2 in air for δ15N and Vienna Standard Mean Oceanic Water (VSMOW) for δ18O.

NO3− isotope composition was corrected for the small KCl blank as follows:

where X=N or O isotopes of nitrate, and f=fraction nitrate of total nitrate.

Soil DNA extractions

DNA was extracted from ~0.25 g fine-earth fraction (<2 mm) soil—corrected for moisture—into 100 μl aliquots using Mo-Bio PowerSoil DNA Isolation Kit (Mo Bio Laboratories, Carlsbad, CA, USA) following manufacturer instructions. To verify a successful extraction and to quantify double-stranded DNA (dsDNA), the Qubit fluorometer (Thermo Fisher Scientific, Carlsbad, CA, USA) assay was used with Quant-iT dsDNA Broad Range reagent and buffer. DNA extracts yielding detectable dsDNA were then stored at −20 °C according to the Mo-Bio PowerSoil DNA Isolation Kit.

qPCR to determine functional gene abundances

Real-time qPCRs were used to detect and quantify specific functional genes of interest: nitrite reductase genes nirK and nirS (NO2− → NO, the denitrification step with most isotopic fractionation), and bacterial 16S rRNA (the gene encoding for a highly conserved region in prokaryotic organisms, estimating total bacterial population). The genes of interest were amplified with specific target primers and thermal conditions and observed using fluorescent dye. The primers used for nirS were nirSCd3aFm and nirSR3cdm (Throback et al., 2004). The nirK primers were nirK876 and nirK1040 (Henry et al., 2004). The 16S rRNA primers were 341F and 534R (Lopez-Gutierrez et al., 2004). See Supplementary Materials for full primer sequence descriptions and thermal conditions.

Standards were pre-made with serial dilutions of target DNA cloned into a plasmid (Scow Laboratory, UC Davis, Davis, CA, USA). Each total reaction had 5 μl template DNA (various dilutions for optimal detection), 12.5 μl SYBR GreenER qPCR SuperMix (Invitrogen, stored at 4 °C), 0.5 μM forward and reverse primers, and nanopure autoclaved deionized water for a 25 μl total volume. All standards, environmental samples and controls were run in triplicate in Optical 96-Well Fast Thermal Cycling Plates via Applied Biosystems 7300 Real-Time PCR System.

Gene copies/g dry soil=[(mean quantity target gene detected)/5 μl sample] × (dilution factor × 100 μl original aliquot)/g dry soil. Mean quantity target gene detected=10^((n−b)/m), where n=the number of cycles required for the fluorescent signal to exceed the background level, and m and b are the slope and y-intercept of the standard curve. A slope of −3.5±0.4 is acceptable with R2>0.99.

Statistics and analysis

Means and standard errors of the natural logarithm of KCl extractable soil nitrate (NO3−), δ15N of NO3−, δ18O of NO3−, and nirS abundance were regressed for sites for which > three parameter sets of values were detected. Using the natural logarithm of NO3− allowed us to perform linear regressions. Data distributions were normal or near-normal for all parameters based on normal quantile plots. Least-squares regression analyses determined the linear fit of all scatterplots, and we reported R-squared and P-values of regression slopes. P-values less than 0.05 were considered significant; at a 95% confidence level we reject the null hypothesis that there is no effect among variables.

We were unable to assemble full data sets of all target parameters for each site, owing to detection limits on NO3− concentrations (KCl sample extracts with <1 μM NO3− prevented samples from being analyzed for stable N and O isotopes). We focused results on sample sites wherein at least three samples were collected with detectable ensembles of NO3− concentrations, and abundance of isotopes and nirS. In the case of oak-woodland sites, criteria of triplicate nirS abundance data from at least three samples were not met; however, we were able to measure oak woodland N and O isotopes. Hence we include these data in analysis of isotopes but not the fully coupled analysis of isotopes, gene abundances and NO3− concentrations. Significant outliers detected via leverage and Cook’s distance analyses were removed.

Stepwise regressions for δ15N- NO3− and nirS abundance were performed for incubation samples with the following data at a 95% confidence level: average field soil moisture for top 10 cm, average field %C for top 10 cm soil, mean air temperature, soil moisture in the incubation, nirS abundance, nirK abundance, 16S rRNA abundance, nirS/16S rRNA, nirK/16S rRNA, ln[NO3−], O and N isotopes from NO3−. These stepwise regressions were repeated for incubation samples from 7-day incubations which displayed decreased NO3− concentrations and had δ15N>0, representing samples with denitrification as the dominant NO3− loss process that had occurred. Results from these analyses steered the focus of the figures and discussion toward select relationships.

Results

Soil nitrate and isotope composition patterns

Soil NO3− concentrations differed substantially across sites, generally tracking changes in MAP and soil C concentrations among terrestrial biomes (Table 1,Figure 2). NO3− concentrations were lowest in the mesic redwood site, and increased into the drier ecosystems, with a peak concentration observed for the desert sites, particularly in samples collected beneath creosote vegetation (Figure 2). The δ15N of KCl extractable NO3− was negatively related to its concentration; highest mean δ15N-NO3− was observed for redwood sites (control 11.67±1.09‰, clearcut 10.79±0.53‰), intermediate values were apparent in the chaparral (1.56±1.37‰), and lowest values were observed for the desert sites (under shrub 0.85±0.32‰, interstitial −0.88±2.53‰) (Figure 2). Highest mean δ18O-NO3− values were observed in redwood control (11.03±2.19‰) and chaparral (11.61±1.21‰), intermediate in redwood clearcut (4.22±0.71‰), and lowest in desert sites (under shrub 0.47±0.69‰, interstitial −1.28±0.76‰) (Figure 2). δ15N-NO3− positively related with soil C (R2=0.22, P<<0.05).

Figure 2
figure 2

Surface soil NO3−, δ15N-NO3− and nirS across sites. Includes sites with full study data set (N and O isotopes, extractable NO3−, nirS). n=9 (redwood control, 13 for NO3−), 3 (redwood clearcut), 11 (chaparral, 14 for NO3−), 15 (desert under creosote), and 14 (desert interstitial). Bars are standard error.

Full size image

Linear regressions indicated several statistically significant relationships among soil NO3− isotopes and concentrations. Field surface δ15N-NO3− and δ18O-NO3− were positively correlated, with a linear slope for all surface soil isotope values at 0.55 (R2=0.25, P=0.0002) (Figure 3). If excluding sites without sufficient corresponding nirS data (that is, no oak woodland), the slope lowered to 0.47 (R2=0.17, P=0.005). A significant, negative relationship was observed between soil NO3− concentrations and δ15N-NO3− across all sites examined (R2=0.19, P=0.003) (Figure 5).

Figure 3
figure 3

δ18O-NO3− vs δ15N-NO3−. Bold symbols are mean values per site. Bars are standard error. Linear regression for field surface soil datapoints is y=0.55x +1.71 (R2=0.25, P=0.00026, n=49). Outliers identified via leverage analyses were removed. If excluding sites without sufficient corresponding nirS data (that is, no oak woodland), y=0.47x+1.86 (R2=0.17, P=0.005, n=44).

Full size image

A positive relationship between δ15N-NO3− and δ18O-NO3− was observed in the in-situ 7-day surface soil (10 cm) incubations at the redwood and chaparral sites (Figure 6). In these ecosystems, decreased NO3− concentrations were strongly related to an increase in δ15N-NO3− (R2=0.50, P=0.01, n=11). In contrast, there was no clear linear relationship for N and O isotopes from desert incubations, despite declines in NO3− concentrations during the incubation experiments (R2=0.00, n=15; Figure 6).

nirS gene abundance patterns

Mean nirS abundance averaged ~107 copies g−1 dry soil across sites. Abundances were highest in the redwood site (7.43 × 107±1.22 × 107 copies g−1 in redwood control, 6.04 × 107±1.32 × 107 copies g−1 in redwood clearcut) and lowest in the desert sites (2.64 × 106±1.22 x105 copies g−1 under creosote bush, 9.66 × 106±2.65 × 106 copies g−1 in the interstitial spaces; Figure 2). nirS abundance positively correlated with soil C (R2=0.51, P<<0.05).

Relationships between isotope composition and nirS

nirS and δ15N-NO3− were significantly positively related across all sites (R2=0.35, P=0.00002) (Figures 2 and 4), and both parameters exhibited significant negative relationships with NO3− concentrations (nirS R2=0.20, P=0.002; δ15N-NO3− R2=0.19, P=0.003; Figures 2 and 5). Although the distribution of field nirS data appears bi-modal (Figure 2), the Breusch-Pagan Test revealed homoscedasticity of the error terms (P>0.05) for a linear regression model of ln[NO3−] vs nirS (Figure 4). nirS abundances and δ15N-NO3− were highest in the undisturbed redwood site, which corresponded with the lowest soil extractable NO3− concentrations (Figure 2). The desert soils under the creosote vegetation exhibited the highest soil extractable NO3− concentrations, lowest nirS abundance and δ15N-NO3− near 0‰ (Figure 2).

Figure 4
figure 4

nirS vs δ15N-NO3−. Bold symbols are mean values per site. Bars are standard error. Linear regression for all datapoints is y=3.27E^6x +1.20E^7 (R2=0.35, P=0.00002).

Full size image
Figure 5
figure 5

nirS (top) and δ15N-NO3− (bottom) vs ln[NO3−]. Soil extractable nitrate concentration prior to logarithmic transformation was in units of μM NO3− g−1 dry soil. Bold symbols are mean values per site. Bars are standard error. Linear regressions for all datapoints: nirS y=−9.55E+06x + 3.28E+07 (R2=0.20, P=0.002); δ15N-NO3− y=−1.66x + 5.24 (R2=0.19, P=0.003).

Full size image

Positive denitrifier gene-isotope relationships were likewise observed in the 7-day incubation experiments conducted in the redwood and chaparral sites. The decrease in NO3− was associated with both an increase in δ15N-NO3−>0 (n=11) and increase in nirS (R2=0.15, P=0.24) in these ecosystems. Furthermore, nirS/16S rRNA exhibited a significant positive relationship with δ15N-NO3− (R2=0.70, P=0.001; Figure 7) in the short-term incubation experiments. From stepwise multiple regressions of desert incubations with NO3− loss and positive N isotopes, δ15N-NO3− responded negatively to ln[NO3−], positively to nirS abundance, negatively to nirS/16S rRNA and slightly negatively to δ18O-NO3− (multiple R=0.86, standard error s.e.=7.4, n=15, significance F<0.05). For stepwise regressions of corresponding redwood and chaparral incubations, δ15N-NO3− responded negatively to ln[NO3−], not significantly with nirS abundance, and positively to δ18O-NO3− (multiple R=0.95, s.e.=2.01, n=11, significance F<<0.05).

Other environmental and microbial variables included at the start (temperature, carbon, moisture, nirK, nirK/16S) did not have a significant relationship with δ15N-NO3− by the end of the stepwise regressions. These results were from stepwise regressions set at 95% confidence. Differences in linear response to δ18O-NO3− alone are shown in Figure 6. Denitrifier linear responses to δ15N-NO3− are shown in Figure 7.

Figure 6
figure 6

δ18O-NO3− vs δ15N-NO3− incubations. Includes initial and final subsamples from 7-day incubations for which [NO3−] decreased and δ15N-NO3−>0. Linear regression for redwood and chaparral incubations is y=0.58x +8.03 (R2=0.50, P=0.01, n=11). Linear regression for desert incubations is y=0.017x +2.05 (R2=0.00, n=15).

Full size image
Figure 7
figure 7

Denitrifiers vs δ15N-NO3− in redwood and chaparral incubations. Data include any initial and final subsamples from redwood and chaparral 7-day in-situ incubations for which [NO3−] decreased and δ15N-NO3−>0 (n=11). Linear regression for nirS/16S rRNA is y=0.0075x + 0.0053 (R2=0.70, P=0.001). Regression for nirS is y=2.0 × 106x +3 × 107 (R2=0.15, P=0.24).

Full size image

Discussion

Our findings support the first hypothesis—that variations in terrestrial δ15N are controlled primarily by biogenic gaseous emissions of N (Houlton and Bai, 2009), principally isotopic discrimination via denitrifying organisms’ enzyme-systems (Mariotti et al., 1981). This result agrees with the elevated δ15N of N inputs compared to that of N leaching losses in our sites (Mnich and Houlton, 2015), and furthermore is consistent with global scale imbalance between input δ15N and the residual N of the terrestrial biosphere (Houlton and Bai, 2009; Vitousek et al., 2013). In addition, the negative relationships between ecosystem NO3− availability and δ15N/nirS support the second hypothesis—that microbial denitrification can consume NO3− to low levels in terrestrial ecosystems, even in temperate environments where N limitation is widespread (LeBauer and Treseder, 2008). The incubation portion of our study further identifies the capacity for denitrifiers to rapidly consume NO3− to low levels within a given site. Overall, these findings are at odds with general formulations of denitrification in contemporary ecosystem models (for example, CENTURY, Davidson and Seitzinger, 2006), which simulate an increase in this process with increasing soil NO3− supplies. Rather, the evidence points to direct control of denitrifiers over soil NO3− availability patterns, which, over time, help to explain the persistence of N limitation observed for grassland to forest ecosystems (Vitousek and Howarth, 1991, LeBauer and Treseder, 2008).

One possible explanation for the denitrifier-driven effect on soil NO3− is that plants in these sites rely on NH4+ and/or direct uptake of amino acids to meet their N demands, thereby allowing denitrifiers to consume NO3− to relatively low levels. This interpretation is consistent with elevated δ15N and δ18O where NO3− is low vs high; plant reliance on NO3− would not impart such significant isotope effects. Moreover, denitrifiers may directly out-compete plants and other microbes for NO3− in moist and C-rich environments, perhaps taking advantage of strong redox potential gradients across soil aggregates in the forest and savanna sites (Sexstone et al., 1985; Ebrahimi and Or, 2016). Indeed, the significant positive correlations between nirS (R2=0.51, P<<0.05) and δ15N-NO3− (R2=0.22, P<<0.05) vs soil C concentrations across our terrestrial sites point to the likelihood of this mechanism. It is worthwhile to note that aggregate pore sizes can influence anoxic zones and greenhouse gas diffusion/emissions, even in unsaturated and variable field conditions (Ebrahimi and Or, 2016), although we did not study aggregates closely in these sites.

Temperature is another environmental factor to consider, as our sites exhibited a gradient in mean air temperatures (Table 1). Temperature increases have been shown theoretically and empirically to influence the kinetics of denitrification-driven isotopic fractionation (Maggi and Riley, 2015). The affinity for biological NO3− fractionation increases with temperature, and Maggi and Riley, 2015 accurately modeled this nonlinear phenomenon at a range of temperatures (20-35 °C), even with heterogeneous soils. While air temperature was not as closely related to isotopic data as other variables were in our study, it is not outside the realm of possibility that soil temperatures, which were not measured, influenced the patterns observed.

Furthermore, other N transformation processes could play a role in influencing available NO3− pools and potentially δ15N, including: nitrification, heterotrophic immobilization and potentially dissimilatory NO3− reduction to NH4+ (DNRA). Nitrification is a fractionating process that acts to lower the δ15N of NO3− vs that of NH4+ substrates, whereas both heterotrophic microbial assimilation and DNRA would be expected to impart an isotope effect on NO3− that resembles that of denitrification.

However, additional evidence for principal denitrifier controls over NO3− concentrations is apparent in dual-isotope relationships in our study (Figure 3). Specifically, a positive relationship between δ18O and δ15N with a slope near 0.6 is diagnostic of terrestrial denitrification, because denitrifying bacteria preferentially convert both light isotopes in NO3− to gaseous products (Lehmann et al., 2003). Heterotrophic NO3− immobilization also fractionates against O and N isotopes in NO3−; though at a much higher slope, typically between 1.0 and 2.0 (Granger et al., 2010). Our observed 0.55 slope for field sites (R2=0.25, P=0.00026) is therefore consistent with the imprint of terrestrial denitrifiers on soil NO3− consumption (Figure 3), as seen in previous work in tropical and temperate forest soils (Houlton et al., 2006; Fang et al., 2015). We cannot specifically address why the slopes differ across studies, but past work on marine bacteria cultures indicates that chemolithotrophs fractionate a slope closer to 0.5, whereas heterotrophic denitrifiers appear to fractionate O and N isotopes at a higher slope (Frey et al., 2014). Differences between microbial effects on O and N isotope relationships generally reflect differences in kinetic rate transfers of isotopes from substrates to products and O exchange reactions with water and air. Indeed, variations in the isotopic composition of O source for NO3− could explain the δ18O-NO3− differences observed across our ecosystem sites (for example, redwood control vs clearcut in Figure 2).

Moreover, the spatial patterns observed across sites are similarly observed in our short-term incubation experiments (that is, 7 days), thus pointing to the importance of microbial denitrifiers in driving soil NO3− availability patterns in both space and time. Aerobic nitrification and even atmospheric deposition may have slightly diminishing effects on N and O isotopes of NO3− in an open system; and so the relationship between N and O isotopes of NO3− in anaerobic incubations is of particular interest. Indeed, δ15N and δ18O of NO3− in the soil incubation experiments reveal a slope that matches expectations for marine and freshwater denitrifiers (Lehmann et al., 2003; Granger et al., 2008) and observations for other terrestrial ecosystems (Houlton et al., 2006; Fang et al., 2015), similar to the spatial relationships observed for all samples across sites (that is, ~ 0.6) (Figure 6). Not only did we observe a positive relationship between nirS and δ15N-NO3− mirroring the cross-system results; but there was a significant positive correlation between nirS/16S rRNA and δ15N-NO3− in our soil incubation experiments as well (R2=0.70, P=0.001) (Figure 7). This demonstrates that the microbial community composition was rapidly enriched in organisms with denitrifying genes, which explains the increase in δ15N and δ18O that accompanied declining NO3− concentrations. Such community streamlining has been observed in aquatic environments (Jayakumar et al., 2009), but, to our knowledge, these are the first observations for terrestrial soil-systems. In sum, these findings support our conclusion that denitrifiers played a major role in determining NO3− pool sizes across sites, with isotopic and molecular evidence revealing complementary patterns across scales, from induced short-term consumption events to the determination of cross-site patterns among biomes.

Importantly, the inverse relationships between δ15N-NO3− and soil NO3− concentration is observed across sites (Figures 2 and 5); however, the calculated isotope effects were much lower than those observed in pure cultures of denitrifying bacteria. The isotope effect of denitrification ranges from 0‰ to ~−33‰ (Högberg, 1997), with an average of approximately −20±1.0‰ observed for laboratory studies (Delwiche and Steyn, 1970; Granger et al., 2008; Houlton and Bai, 2009). The magnitude of the isotope effect can be lower in the field, particularly in environments where non-homogenous interactions allow for locally complete NO3− consumption, as in a ‘closed system’ Rayleigh model (Brandes and Devol, 2002; Houlton et al., 2006). In this case, NO3− can be completely consumed in micro-environments leaving little or no residual NO3− to express denitrification’s isotope effect (Houlton et al., 2006). Furthermore, non-fractionating sinks such as plant uptake and microbial assimilation can consume NO3− without altering 15N/14N. In our sites the relationship between δ15N and ln[NO3−] suggests an integrated isotope effect—a net isotope effect reflecting all NO3− sources and sinks—that varies from -2.12 in the redwood site to 1.92 in the desert interstitial site, with an overall effect across sites of −1.66.

One exception to the inverse relationship between δ15N-NO3− and soil NO3− is evident in the desert soil, particularly beneath vegetation canopies. The soil devoid of vegetation had notably lower [NO3−] than those collected under L.tridentata ‘islands of fertility’; and yet the interstitial sites had the same, if not lower, δ15N-NO3− as the vegetated ones (Figure 2). The weakly positive relationships in interstitial spaces suggest that nitrification, ammonia volatilization or other abiotic processes could be responsible for gaseous N production in such arid environments (McCalley and Sparks, 2009). It is possible that in arid and wind-prone desert ecosystems, uptake of NO3− by heterotrophic microbes and plants could mask any denitrification-driven enrichment of heavy N and O isotopes. Any substantial non-fractionating sinks for NO3− could greatly reduce the isotope effect of denitrifiers. We cannot rule out the role of episodically driven effects on denitrification in the desert, where precipitation occurs over very short, high-intensity events.

The field portion of our study provides useful, integrated snapshots of O and N isotopes of NO3− in a variety of field settings—taking into account the collective effects of ecosystem N cycling. The incubation experiments help to isolate the field-based evidence for denitrifier-driven controls over soil NO3−, albeit under optimal conditions of low O2 availability. Future work in such dynamic systems could benefit from long-term, daily field studies coupled with field-sensors of soil moisture, pH, temperature and redox.

Implications for linking microbial gene abundance to ecosystem functioning

Our results have implications for understanding linkages among microbial functional gene abundances and ecosystem pattern and process. As modern genetic techniques have tremendously increased, questions over the utility of microbial gene abundance measures have become paramount, representing a key area of active research in the microbial ecology and ecosystem biogeochemistry. Our finding of a positive correlation between gene abundance and isotope composition provides a new, integrative tool for connecting DNA to variations in ecosystem NO3− pools and natural isotope composition; it reveals a link between genes encoding for a given enzyme (for example, nirS) and its role in a key ecosystem process (for example, denitrification) across biomes. We suspect that such a ‘gene abundance to isotope’ relationship is likely best observed in wet seasons or generally moist sites, owing to denitrifiers’ ability to thrive in high moisture and anaerobic soil environments where C is abundant (Dawson and Murphy, 1972; Burgin et al., 2010; Szukics et al., 2010; Groffman 2012).

Previous studies on relationships between nirS/K and terrestrial denitrification rates have been mixed, which could be explained by mismatches in the scale of measurements or artifacts of disturbance (Wallenstein et al., 2006). Structure and abundance of microbial communities bearing nirS appear less sensitive to long-term fertilizations than those with nirK (Chen et al., 2010), and actual transcription of nirS is less sensitive to changes in pH than nirK (Liu et al., 2014). Thus, our focus on nirS likely minimized the chances of localized factors from obscuring relationships between community structure, gene abundance and actual gene expression.

Terrestrial denitrification has remained a poorly understood aspect of N cycling research (Davidson and Seitzinger, 2006). Our results demonstrate a significant link between nirS and natural isotope abundance, over scales ranging from short-term soil incubation experiments to the factors structuring longer-term soil NO3− pools across forest to desert ecosystems. Kinetic isotopic discrimination is the result of denitrifying organisms’ enzymatic preferences for light isotopes in NO3−. Thus, while gene abundance data alone do not represent the diversity or activity of soil microbial communities, the relationships between nirS and heavy isotope enrichment point to tight connections between denitrifier gene abundances and microbial enzyme activity across our sites. Where N availability is highest in the desert biome, the imprint of denitrification appears weakest; where N availability was lowest, coupled molecular and isotopic data point to substantial NO3− consumption by denitrifying bacteria. These findings suggest that models of denitrification should be re-formulated to include a more direct influence of denitrifiers in determining terrestrial N availability.