Nitrogen inputs and losses in response to chronic CO 2 exposure in a subtropical oak woodland

Rising atmospheric CO 2 concentrations may alter the nitrogen (N) content of ecosystems by changing N inputs and N losses, but responses vary in field experiments, possibly because multiple mechanisms are at play. We measured N fixation and N losses in a subtropical oak woodland exposed to 11 years of elevated atmospheric CO 2 concentrations. We also explored the role of herbivory, carbon limitation, and competition for light or nutrients in shaping the response of N fixation to elevated CO2. Elevated CO2 did not significantly alter gaseous N losses, but lower recovery and deeper distribution in the soil of a long-term15N tracer indicated that elevated CO2 increased leaching losses. Elevated CO 2 had no effect on nonsymbiotic N fixation, and had a transient effect on symbiotic N fixation by the dominant legume. Elevated CO2 tended to reduce soil and plant concentrations of iron, molybdenum, phosphorus, and vanadium, nutrients essential for N fixation. Competition for nutrients and herbivory likely contributed to the declining response of N fixation to elevated CO2. These results indicate that positive responses of N fixation to elevated CO2 may be transient and that chronic exposure to elevated CO 2 can increase N leaching. Models that assume increased fixation or reduced N losses with elevated CO2 may overestimate future N accumulation in the biosphere.


Introduction
Nitrogen (N) is the element most frequently limiting to plant growth (LeBauer and Treseder, 2008). Nitrogen inputs and losses from terrestrial ecosystems determine ecosystem N pool size, and in turn influence the potential for carbon (C) uptake when plant growth is N limited. Carbon uptake and storage are thus sensitive to the balance of N inputs and losses (Pepper et al., 2007;Gerber et al., 2010;Esser et al., 2011). Here, we synthesize the effects of 11 years of chronic exposure to increased CO 2 concentrations on N inputs and losses from a subtropical oak woodland.
Nitrogen fixation is the major biological pathway through which the biosphere accumulates N. Nitrogen fixation has a high demand for reducing power to break the triple covalent bond shared by the two atoms in the N 2 molecule (Benemann and Valentine, 1972). Symbioses between bacteria capable of N fixation and photosynthetic organisms have evolved Published by Copernicus Publications on behalf of the European Geosciences Union. 3324 B. A. Hungate et al.: Nitrogen inputs and losses in response to chronic CO 2 exposure multiple times, likely an adaptive pairing because of the high energetic cost of N fixation and N being frequently limiting to plant growth (Sprent, 1985). Increased photosynthesis with rising concentrations of atmospheric CO 2 has been postulated to shunt more labile C from the plant to bacterial root symbionts, increasing rates of N fixation. In some cases, elevated CO 2 has been found to disproportionately increase the growth of N-fixing plants (Norby and Sigal, 1989;Arnone and Gordon, 1990;Hartwig et al., 2000;Soussana and Hartwig, 1996;Zanetti et al., 1996;Hebeisen et al., 1997;Feng et al., 2004) and increase N fixation, though this potential is not always realized under field conditions (Schäppi and Korner, 1997;Arnone, 1999;Hoosbeek et al., 2011).
Some symbiotic (i.e., free-living) heterotrophic bacteria can also fix N. Elevated CO 2 can increase the amount of C that plants produce belowground through root growth, turnover, and exudation (Drake et al., 2011;Hagedorn et al., 2013;Lagomarsino et al., 2013) and thereby alleviate C limitation of nonsymbiotic N fixation. This may explain why elevated atmospheric CO 2 has been found to stimulate nonsymbiotic N fixation by soil bacteria in a salt marsh (Dakora and Drake, 2000) and a rice paddy soil (Hoque et al., 2001). However, in a temperate pine forest and desert soil, elevated CO 2 had no effect on nonsymbiotic N 2 fixation (Hofmockel and Schlesinger, 2007;Billings et al., 2003). Thus, effects of elevated CO 2 on nonsymbiotic N 2 fixation are equivocal.
N-fixing organisms require high concentrations of iron (Fe), phosphate (P), and molybdenum (Mo), or in some instances vanadium (V) (Williams, 2002). (Smith, 1992). Responses of N-fixation to elevated CO 2 can be limited by availability of these nutrients (Niklaus et al., 1998;Jin et al., 2012). The response of N fixation to elevated CO 2 across multiple studies was only significant when non-N nutrients were added as fertilizer; without nutrient amendments, the effect of CO 2 on N fixation was negligible and not significant (van Groenigen et al., 2006). Elevated CO 2 often increases plant growth and element accumulation (Luo et al., 2006), including in the subtropical woodland studied here . Therefore, increased element uptake by non-fixing plants could restrict nutrient availability for N-fixing organisms, potentially limiting their response to elevated CO 2 .
Because of the high energy requirements of N fixation, shading by the canopy can limit the growth of N-fixing plants (Gutschick, 1987;Vitousek et al., 2002). Therefore, if elevated CO 2 promotes growth of the dominant species, enhancing its leaf area, the growth of N-fixing plants could be suppressed. Herbivory could also influence the response of N-fixing plants to elevated CO 2 . Herbivores often prefer the tissue of N-fixing plants to that of other plants because Nfixing plants have a higher protein content than non-fixing plants (Ritchie and Tilman, 1995;Hulme, 1994Hulme, , 1996. For this reason, factors promoting the growth of N-fixing plants could in turn stimulate selective herbivory. With the exception of episodic losses during disturbance, N losses from terrestrial ecosystems occur primarily as gaseous products of nitrification and denitrification (NO,N 2 O,and N 2 ) and through leaching of NO − 3 and organic N. Elevated CO 2 can alter N losses if input of labile C to the rhizosphere enhances denitrification rates (Smart et al., 1997;Robinson and Conroy, 1999;Baggs et al., 2003a, b), or by altering soil water content because of reduced evapotranspiration (Hungate et al., 1997a;Arnone and Bohlen, 1998;Robinson and Conroy, 1999). Elevated CO 2 could also reduce ammonium availability to nitrifiers, suppressing nitrification and N losses through gaseous fluxes (Hungate et al., 1997b) and NO − 3 leaching (Torbert et al., 1998). Across studies conducted to date, elevated CO 2 has been found to increase N 2 O efflux from terrestrial ecosystems (van Groenigen et al., 2011); effects on N leaching have not been synthesized.
During the first 6 years of the experimental treatment of the subtropical woodland studied here, elevated CO 2 increased N 2 fixation during the first year of treatment, but the response subsequently disappeared . Here, we extend this record to include symbiotic N 2 fixation during the full 11 years of the CO 2 experiment, and we also assess responses of nonsymbiotic N 2 fixation to elevated CO 2 . We also investigate possible mechanisms shaping the responses to CO 2 , testing the hypotheses that selective herbivory, light competition, and changes in nutrient availability modulate the response of N 2 fixation to elevated CO 2 . We also report new data on rates of N gas losses and tracer 15 N recovery in deep soil to assess effects of CO 2 on N leaching.

Site description
This work was conducted at the Smithsonian Environmental Research Center's elevated CO 2 experiment at Kennedy Space Center, Cape Canaveral, Florida, USA (28 • 38 N, 80 • 42 W). The experiment consisted of 16 open-top chambers, each 2.5 m high with an octagonal surface area of 9.42 m 2 . Eight chambers were maintained at ambient atmospheric CO 2 concentrations and eight chambers were maintained at approximately 350 µL L −1 above ambient atmospheric CO 2 concentration. The soils at the site were acidic Spodosols (Arenic Haplohumods and Spodic Quartzipsamments). The vegetation was Florida coastal scrub oak palmetto (Dijkstra et al., 2002;Johnson et al., 2003;Seiler et al., 2009), dominated by three oaks (Quercus myrtifolia, Q. geminata, and Q. chapmanii) and several less abundant species, including saw palmetto (Seranoa repens), shiny blueberry (Vaccinium myrsinites), rusty Lyonia (Lyonia ferruginea), and tarflower (Befaria racemosa). A native vine, Elliott's milkpea (Galactia elliottii) constituted only 1 % of aboveground productivity  but is important for its ability to fix nitrogen.

Symbiotic N 2 fixation by Galactia elliottii
We estimated G. elliottii annual aboveground production as the annual flux of G. elliottii mass collected in litter traps (Stiling et al., 1999(Stiling et al., , 2002(Stiling et al., , 2009). Litter of G. elliottii was sorted and measured separately beginning in 1999; in 1997 and 1998, G. elliottii litter fall was measured together with other "non-oak" species. As described previously , for 1997 and 1998, we estimated that G. elliottii litter constituted 68 % of the non-oak litter. This estimate is the average of the proportion of G. elliottii litter mass in in the total non-oak litter fraction as measured from 1999 to 2002 (63 %) and the estimated proportion of G. elliottii litter mass in 1997 and 1998 based on a mixing model using N concentration in G. elliottii litter (high in nitrogen percentage) versus other species (lower in nitrogen percentage); this mixing model yielded an estimate of 73 %.
We measured N 2 fixation rates using isotope dilution. A nitrogen-15 tracer was added on 19 June 1998 (0.18 g N m −2 (NH 4 ) 2 SO 4 99.9 atom % 15 N). To the extent that G. elliottii fixes N 2 via symbiotic bacterial fixation from the atmosphere, G. elliottii leaves will be lower in δ 15 N than oak leaves, whose N is derived from the soil, directly increased by the δ 15 N value of the added tracer. The proportion of N in G. elliottii that was derived from N 2 fixation (p f ) was calculated using the standard model for 15 N dilution (Shearer and Kohl 1986), where the 15 N signature of unlabeled G. elliottii is the atmospheric end member: where the subscript O refers to the dominant oak (Q. myrtifolia); G to the N-fixer, G. elliottii, after adding the 15 N tracer; and Go to G. elliottii before adding the 15 N tracer. This calculation relaxes the common assumption that the δ 15 N value of N obtained by fixation is equal to the δ 15 N of atmospheric N 2 ; it therefore accounts for biological fractionation during N 2 fixation (Yoneyama et al., 1986) and provides a more precise estimate of the end member of the mixing model. The δ 15 N value of G. elliottii prior to label application was −2.2 ‰. A second assumption of this method is that the reference plant (Q. myrtifolia) obtains its N from the soil rather than from N fixation. Departures from this assumption will cause the mixing model to underestimate the proportion of N derived from fixation, unlikely to be a serious error at our site because the mixing model indicated that G. elliottii obtained nearly all its N from fixation. The isotope dilution method also requires that the δ 15 N tracer is sufficiently strong for the δ 15 N value of the reference plant to remain distinct from the atmospheric source. This was the case throughout the experiment: the average δ 15 N in oak leaves was 131.0 ± 5.3 ‰. By the final harvest, this value declined to 84.3 ± 4.2 ‰, still clearly distinct from the N 2 -fixer value (−2.2 ‰), providing sufficient resolution in the mixing model such that the standard deviation for the proportion of N derived from fixation (p f ) was 0.012. The isotope dilution method also assumes that G. elliottii, if it takes up N from the soil, accesses approximately the same soil N available to the reference plant. Evidence suggests that this assumption is reasonable, as both fine roots (67 ± 13 %) and G. elliottii nodules (74 ± 9 %) were concentrated in the top 30 cm of soil. We measured N concentration and δ 15 N in G. elliottii foliage and senesced litter on an elemental analyzer inline with an isotope-ratio mass spectrometer. Total N 2 fixation was calculated as the product of G. elliottii biomass production, N concentration, and the proportion of N derived from fixation. To calculate p f using Eq. 1, we used δ 15 N values from senesced leaves because they were gathered in litter traps and therefore captured an integrative sample of G. elliottii tissue at the plot scale (Stiling et al., 2009). For N concentration, we used the percentage of N in green leaves to avoid underestimating N 2 fixation due to retranslocation during senescence.
We measured nodule biomass by handpicking soil from cores taken in July 2007 (0-100 cm) (Hungate et al., 2013b;Day et al., 2013). Nodules were washed free of sand, oven dried, and weighed. For comparison, we also report here data obtained during the first year of the experiment , when we measured nodule mass and number in buried columns of C-horizon sand placed in the top 15 cm of soil. The earlier assay using ingrowth cores captures new nodule growth, whereas the cores at the final harvest measure the standing crop of nodules. Though not directly comparable, both assess responses of N 2 -fixing nodules to the elevated-CO 2 treatment and thus are presented together here.

Herbivory
From 2000 to 2006, litter of the oaks and of G. elliottii was scored for damage by insect herbivores and sorted, counted, and weighed separately by species and insect damage category as described previously (Stiling et al., 1999(Stiling et al., , 2009). The rate of litterfall of herbivore-damaged leaves, L (g m −2 year −1 ), was determined for each species as the total mass of damaged litter divided by the area of the litter traps in each chamber. Live green leaves of all species were also sampled during the experiment in order to determine the average mass of an undamaged green leaf for each species. Seven plants were randomly chosen for each of the three oak species in September 2004. Fifty leaves were sampled haphazardly over the entire canopy of each plant. For G. elliottii, total aboveground biomass was sampled in 1/6th of each chamber in September 1999, separated into leaves and stems, and the leaves were counted. For all four species, leaves were dried in a ventilated oven at 70 • C for 72 h and weighed, and the average leaf mass was calculated. We used these measurements to estimate the amount of tissue consumed by insect herbivores. For each species, we multiplied the number of damaged leaves collected in the litter traps (# leaves m −2 year −1 ) by the average mass of a green leaf of that species (g leaf −1 ) to determine the production of all the leaves of which insects consumed at least a part, P, (g m −2 year −1 ). Note that this differs from total leaf production in that P only includes leaves that were damaged by insect herbivores. We then calculated herbivore consumption (C) as the difference of P − L. Our use of green leaf mass in calculating C is justified because herbivore damage often causes early leaf abscission (Faeth et al., 1981;Williams and Whitham, 1986;Stiling and Simberloff, 1989. Finally, we calculated total leaf production as the total number of leaves in litterfall multiplied by average green mass and then calculated the proportion of total leaf production that was consumed by herbivores, for each species, as Cdivided by total leaf production.

Soil micronutrient analyses
In July 2005, October 2006, and July 2007, soil was collected from the A (0-10 cm) and E (10-30 cm) horizons. Samples were collected from five locations within each chamber. In 2007 we also analyzed samples from the E2 horizon (30-60 cm). The cores from each plot were composited, yielding one A and E sample for each plot for 2005 and 2006, and one A, E, and E2 sample in 2007. We sampled foliage of G. elliotii, collecting fully exposed and expanded leaves from five plants per plot in May 2003 and in July 2007. Leaves were oven dried at 60 • C. Total ecosystem element mass data for 2007 were reported previously (Duval et al., 2014); here, they are shown as separate components (plant and soil), together with data from 2003, 2005, and 2006.
Soils were air dried and passed through a 2 mm sieve. Soilavailable Mo, Fe, and V concentrations were determined after ammonium oxalate extraction (Liu et al., 1996), and soilavailable P was determined by extraction with NaOH (Carter, 1993). Extracts were filtered, diluted 10 times, suspended in 10 mL of 0.32 M trace metal grade HNO 3 , and analyzed by Inductively coupled plasma mass spectrometry (ICP-MS) as described below.
For 2007, plant samples were dried at 60 • C and ∼ 750 mg was ashed at 600 • C. Five hundred milligrams of G. elliottii leaves was run through a MARS microwave digester using EPA protocol 3052x, consisting of a 30 min cycle at 200 • C with 9 mL of HNO 3 , 3 mL of HCl, and 2 mL of HF. All reagents were trace-metal-grade concentrated acids. The resulting solutions were dried on a hot plate and resuspended in 10 mL of 0.32 M trace-metal-grade HNO 3 prior to ICP-MS analysis.
Element concentrations were determined on a Thermo X Series quadrupole ICP-MS at the Keck Isotope Biogeochemistry Laboratory, Arizona State University, Tempe, Arizona, and a Thermo X2 Series quadrupole ICP-MS at the Isotope Geochemistry Laboratory, Northern Arizona University, Flagstaff, Arizona. In these analyses, we used the standard references cody shale (SCo-1) for soils and peach (NIST 1547) and apple leaves (NIST 1515) for plants. In the ICP-MS analysis, we had > 90 % recovery of all elements measured.

Acetylene reduction for nonsymbiotic N 2 fixation
We measured acetylene reduction to ethylene in soil incubations to estimate N 2 fixation by free-living soil heterotrophs. Soil samples were collected in January and March 2006. At each date, five cores (4 cm diameter x 15 cm deep) were collected from each chamber, shipped overnight to Florida International University (Miami, Florida), composited for each plot, and sieved through a 2 mm sieve. Soils were composited by treatment and 3 g subsamples were filled into 20 mL glass vials (n = 4-5). In January 2006, the experimental design included two levels of glucose addition (0 and 324 µg C g −1 soil) and two levels of soil moisture (3 % and 10 % volumetric). In March 2006, the experimental design included three levels of glucose addition (0, 6.7, and 34.4 µg C g −1 soil) and two levels of phosphorus addition (0 and 6.5 µg P g −1 soil). For both experiments, treatments were crossed in a fully factorial design with n = 4 for January and n = 5 for March. Acetylene (1 mL) was added to each vial, and the vials were sealed and incubated for 6 h at room temperature. Headspace samples were analyzed via gas chromatography for ethylene production using an HP5890 gas chromatograph equipped with a flame ionization detector.

Nitrous oxide and nitrogen oxide gas fluxes
We measured soil production of nitrous oxide (N 2 O) and nitric oxide (NO x ) using static chambers (Hutchinson and Livingston, 1993) during 2005. For N 2 O, headspace samples were collected in syringes and analyzed by gas chromatography. For NO x , the static chamber was plumbed to a chemiluminescent detector and the computer that logged real-time [NO x ]. Chambers (1.8 L) were constructed from a 10.2 cm diameter PVC pipe closed with a PVC cap. The bottom 3 cm of each chamber was tapered to allow the chamber to slide smoothly into PVC rings of similar diameter installed in each plot in 2004. Once the chamber was in place, headspace air (15 mL) was sampled through a rubber septum (fixed to the top of each chamber) using a 20 mL nylon syringe equipped with a nylon stopcock and a 23-gauge needle. Three subsequent headspace samples were taken at 15 min intervals. Syringes were maintained under pressure using a rubber band until analysis, within 12 h of sample collection. Samples were analyzed on a gas chromatograph system (Shimadzu) with Haysep-Q60/80-and Porapack-Q60/80packed columns and equipped with an electron capture device to determine N 2 O concentrations. Field fluxes were calculated using linear regression of concentrations over time.
The flux rates were expressed as µg N 2 O-N m −2 d −1 .

Tracer 15 N distribution and recovery in deep soils
During the final harvest of the experiment in 2007, each plot was cored to the water table or to a depth of 3 m (whichever was deeper). Core depth averaged 260.6 cm (SEM 13.5), not significantly different between the CO 2 treatments (P = 0.90; t test). We modeled the depth distribution of tracer 15 N using the function where d is depth, Y is the cumulative proportion of tracer 15 N recovered up to depth d, and β is a fitted parameter (Gale and Grigal, 1987). We used Microsoft Excel's Solver function to find the values of β, minimizing the sum of squares of the errors in Y between measured and modeled values for each plot. We used a t test to determine if elevated CO 2 altered β and used nonlinear regression to explore relationships between the vertical distribution of recovered 15 N and total 15 N recovery at the ecosystem scale (mg 15 N m −2 ).

Statistical analyses
We used analysis of variance (ANOVA) to test for effects of elevated CO 2 , using repeated measures to test for effects of time and interactions between CO 2 and time, and split-plot ANOVA to test for differences among soil depths. When necessary, data were log-transformed (N 2 fixation, soil element concentrations) to meet assumptions of ANOVA. For nodule biomass in 1996 and 1997, we used a Kruskal-Wallis test because data did not meet assumptions of ANOVA and values of zero precluded log-transformation. We used an α threshold of 0.10. We omitted one data point from the analysis of foliar Mo of G. elliottii in 2007 based on a Grubb's outlier test; omitting the data point did not influence the significance of any statistical comparison. We used simple and multiple linear regressions to explore relationships between N 2 fixation and possible drivers. We also used ANOVA for the acetylene reduction assay to test effects of CO 2 , glucose, and water (in January 2006) or phosphorus (March 2006) on rates of N 2 fixation by free-living soil microorganisms. We note that compositing soils from plots by CO 2 treatment compromised the independence of replicates within CO 2 treatments. The absence of any significant effects of CO 2 on nonsymbiotic N 2 fixation (see below) provides some protection from the dangers of false inference caused by pseudoreplication (Hurlbert, 1984). We assessed effects of N 2 fixation as a function of time since disturbance by calculating time since disturbance as the number of years that had elapsed between the date of the measurement and the most recent disturbance, whether by fire at the beginning of the experiment or by hurricane in September 2004 (Hungate et al., 2013a). We expressed the effect of elevated CO 2 on N 2 fixation as the absolute difference between elevated and ambient CO 2 plots.

N 2 fixation by G. elliottii
N 2 fixation by Galactia elliottii varied over time (P < 0.001), with high rates in 1998 and low rates in 2000 and 2005 (Fig. 1). Elevated CO 2 increased N 2 fixation by G. elliottii during the first six months of experimental treatment ( Fig. 1a and as reported in Hungate et al., 1999). However, by the third year of exposure to elevated, CO 2 , N 2 fixation was not different between the two treatments (CO 2 × time interaction, P = 0.070). Subsequently (years 5-7), elevated CO 2 suppressed G. elliottii N 2 fixation. Rates equalized again by year 9 (2005). Accumulated over the 11-year exposure period, G. elliottii fixed 3.92 ± 0.50 g N m −2 year −1 in the ambient CO 2 treatment compared to 3.51 ± 0.71 g N m −2 year −1 in the elevated-CO 2 treatment, a nonsignificant difference (repeated measures ANOVA, P = 0.376). The effect of elevated CO 2 dissipated with time since disturbance (Fig. 1b). G. elliottii derived nearly all of its foliar N from fixation, increasing from 92 % in 1998 to 100 % from 2005 to 2007 (Table 1). Given the small range of variation in reliance on atmospheric N 2 , temporal changes in N 2 fixation (Fig. 1) were driven by effects of time and treatment on the productivity of G. elliottii. Nevertheless, the proportion of N derived from the atmosphere by G. elliottii was sensitive to temporal variation and to the CO 2 treatment. Across all years, elevated CO 2 reduced the reliance of G. elliottii on N from the atmosphere, though the effect was small, from an average of 96.9 % for the ambient CO 2 treatment to 94.7 % for the elevated treatment. Foliar nitrogen percentage of G. elliotti was lower in the elevated-CO 2 treatment (Table 1), an effect that varied over time. During the first year of the experiment, elevated CO 2 stimulated nodule biomass (effect of CO 2 in 1996, P = 0.062), but elevated CO 2 had no effect on nodule mass at final harvest (Fig. 2, Table 2), a pattern similar to that found for N 2 fixation. N 2 fixation by G. elliottii (g N m −2 year −1 ) was positively correlated with G. elliottii foliar N concentration, negatively correlated with the total mass of the oak leaves, negatively correlated with herbivory, and unrelated to leaf area index (Table 3). Therefore, as the dominant plants grew larger, N 2 fixation declined, suggesting competition. The absence of any relationship with leaf area index argues against competition for light. N 2 fixation also declined as foliar nitrogen percentage declined, which can indicate limitation by non-N nutrients ). Finally, high proportional consumption of G. elliottii tissue was associated with lower rates of N 2 fixation, suggesting some control of N 2 fixation by herbivory (Table 3).  Table 3. Multiple regression of rates of N 2 fixation (g N m −2 year −1 ) as a function of foliar N concentration in G. elliottii (g g −1 × 100 %), herbivory (% of leaf production consumed), leaf area index (m 2 m −2 ), and total oak biomass (kg m −2 ). The overall regression is significant (F 4,103 = 8.543; P < 0.001; adjusted r 2 = 0.22).

Soil-extractable micronutrient concentrations
Concentrations of soil-extractable Fe, Mo, and V declined from 2005 to 2007 (Fig. 3, Table 4). Elevated CO 2 significantly reduced soil-extractable concentrations of Mo and V, and tended to reduce Fe. Some assays of soil P availability collected during the experiment indicated that elevated CO 2 reduced soil P availability (ion exchangeable P in 1997, extractable P in 2001), though this effect was not apparent at the final harvest (Fig. 4) However, the rate of N 2 fixation was not related to soil-extractable nutrient concentration (data not shown, P > 0.10 for all regressions).

Herbivory
Herbivores in this subtropical woodland consumed a higher percentage of G. elliottii leaf production (24.0 ± 1.8 %) than of oak leaf production (14.1 ± 0.6 %, P < 0.001, Fig. 6). The percentage consumed of G. elliottii was not affected by the

Nonsymbiotic N 2 fixation
Elevated CO 2 had no effect on acetylene reduction in soil laboratory incubations (Tables 5 and 6). Acetylene reduction was also unresponsive to phosphorus addition or soil water content. Acetylene reduction increased with glucose addition, consistent with carbon limitation of heterotrophic nonsymbiotic N 2 fixation in these soils.

Nitrous oxide and nitric oxide fluxes
Across all measurements conducted during the CO 2 enrichment experiment, N 2 O efflux averaged 0.346 mg N 2 O-N m −2 d −1 for the ambient CO 2 treatment and 0.377 mg N 2 O-N m −2 d −1 for the elevated-CO 2 treatment, a nonsignificant difference of 0.035 mg N 2 O-N m −2 d −1 (5 and 9 % confidence limits, −0.070 and 0.132). Scaled to the entire 11-year period and assuming that N 2 O constitutes 10 % of the total N losses in denitrification (with 90 % as N 2 production), our best estimate is that elevated CO 2 increased  compared to the controls (5 % and 95 % confidence limits, 0.01 to 0.47). If these rates are typical for N 2 O and NO x losses over the experiment, the elevated-CO 2 -treated plots lost 1.2 g N m −2 more N in gaseous fluxes compared to the ambient CO 2 treatment, but this difference was not significant (−3.3 to 5.3; 5 and 95 % confidence limits).

15 N tracer recovery and distribution as an indicator of N leaching
Nitrogen movement from the O and A horizons into deeper soil (> 15 cm) was measured directly during the first 3 years of the experiment using resin lysimeters , with a tendency for CO 2 to reduce vertical N movement. Yet whole-system recovery and distribution of the 15 N tracer were consistent with increased 15 N losses via leaching in the elevated-CO 2 treatment. Elevated CO 2 reduced total 15 N recovery , and of 15 N that remained in mineral soil, more was found deeper in the profile in the elevated-CO 2 treatment as indicated by a higher β value (β = 0.81 ± 0.01 for ambient CO 2 and 0.89 ± 0.03 for elevated CO 2 , P = 0.03). These findings were associated such that reduced 15 N recovery occurred in plots where the recovered 15 N was distributed deeper in the soil profile (Fig. 7). Thus, elevated CO 2 reduced 15 N recovery and promoted 15 N movement down the soil profile, consistent with increased N leaching losses.

Effects of CO 2 on processes regulating ecosystem N accumulation
Findings reported here explain the absence of any stimulation of N accumulation in response to 11 years of exposure of this subtropical oak woodland to elevated carbon dioxide concentration (Hungate et al., 2013b). Elevated atmospheric CO 2 either elicited no response in processes favoring N accumulation (nonsymbiotic N 2 fixation) or caused only a transient and quantitatively negligible response (symbiotic N 2 fixation). Reduced gaseous N losses from the elevated-CO 2 -treated plots were observed for NO x , but NO x was a minor component of the ecosystem N budget such that the changes observed were insufficient to promote N accumulation. Moreover, the reduced 15 N recovery and pattern of greater 15 N recovery deep in the soil indicates that elevated CO 2 enhanced leaching losses of N. Our results indicate that processes promoting N loss were more responsive to elevated CO 2 than were processes promoting N accumulation.

Effects of CO 2 on N 2 fixation
N 2 fixation by the leguminous vine, G. elliottii, was only temporarily responsive to the CO 2 treatment (Fig. 1). Growth and N 2 fixation in N 2 -fixing plants can respond positively to elevated CO 2 (Cernusak et al., 2011), but in many cases these responses are absent, e.g., N-fixers in temperate grasslands (Garten et al., 2008;Zhang et al., 2011), Alnus (Temperton et al., 2003;Millett et al., 2012), and ocean cyanobacteria (Czerny et al., 2009;Law et al., 2012). Chronic CO 2 exposure was found to reduce cyanobacterial abundance in desert crusts (Steven, 2012). Our finding that nonsymbiotic 3332 B. A. Hungate et al.: Nitrogen inputs and losses in response to chronic CO 2 exposure N 2 fixation in soil was also insensitive to the CO 2 treatment is consistent with observations in other forest ecosystems (Hofmockel and Schlesinger, 2007). In general, responses appear to be more muted under field conditions in long-term experiments than in short-term (van Groenigen et al., 2006;Leuzinger et al., 2011), so our finding that elevated CO 2 did not increase N 2 fixation is not unusual. Nitrogen availability, light limitation, herbivory, and non-N nutrient limitation are plausible explanations for the absence of a significant response of N 2 fixation after the first year of the experiment.

Nitrogen availability
Increased N availability is well known to reduce N 2 fixation (Streeter, 1988), as uptake of N from the soil is energetically favorable to fixation. In the scrub oak ecosystem studied here, elevated CO 2 increased plant N uptake from soil, in part by enhanced turnover of soil organic matter (Carney et al., 2007;Langley et al., 2009;Hungate et al., 2013b). The slight reduction in the reliance of G. elliottii on atmospheric N in response to elevated CO 2 (Fig. 1, Table 1) is consistent with a CO 2 -stimulation of soil N availability. N 2 fixation can decline with shading during canopy development, yet in this system leaf area index was not a significant predictor of N 2 fixation (Table 3), possibly because the N 2 -fixers are vines, capable of climbing rapidly to the top of the short canopy to alleviate light limitation (Kurina and Vitousek, 1999). On the other hand, the negative relationship between N 2 fixation and total oak biomass (Table 3) suggests competition for other resources between the oaks and the N 2 -fixing vine, possibly competition for nutrients.

Non-N nutrients
When supplied with sufficient non-N nutrients, N 2 -fixing plants will increase nodule biomass and rates of N 2 fixation (van Groenigen et al., 2006;Rogers et al., 2009), while at the same time maintaining foliar N concentration in response to elevated CO 2 . Under nutrient-limiting conditions, foliar N declines, nodule growth diminishes, and rates of N 2 fixation are depressed , a pattern consistent with the positive correlation between foliar N concentration and N 2 fixation shown here (Table 3). The responses of N 2 fixation and growth of N-fixing plants to elevated CO 2 may depend on availability of non-N nutrients. In our system, there was some evidence that the availability of soil P declined, particularly during the first 6 years of the experiment (Fig. 4, and see Johnson et al., , 2003, a response also observed in a rice-wheat rotation, see Ma et al., 2007). However, reduced P availability is not a universal response to increased CO 2 concentration (Dijkstra et al., 2012;Khan et al., 2008Khan et al., , 2010. Other elements critical for N 2 fixation such as Mo, Fe, and V either declined significantly in soil and in foliage or showed a tendency to decline (Figs. 3,5), suggesting their role in modulating the response of N 2 fixation to elevated CO 2 . Natali et al., 2009 found that total soil metal content increased in the surface soils of a loblolly pine and sweetgum plantations, though total metal concentration may be a poor indicator of extractable metal concentration and metal availability to plants. Foliar concentrations of most metal elements were substantially lower at this scrub oak site compared to the loblolly pine and sweetgum plantations (Natali et al., 2009) (Duval et al., 2014). Sandy texture and low pH are two soil properties typically associated with low metal availability because sandy soils have low ion exchange capacity and because metal availability is very sensitive to pH (Vlek and Lindsay, 1977;Sposito, 1984;Goldberg et al., 1996;Kabata-Pendias, 2001). Thus, compared to other ecosystems, this subtropical woodland may be more likely to exhibit micronutrient limitation of ecosystem processes, such as N 2 fixation.
Several experiments have demonstrated the importance of P availability for the responses of N 2 fixation to elevated CO 2 : the stimulation of growth and N 2 fixation by elevated CO 2 was higher with P addition for clover (Edwards et al., 2006), Azolla (Cheng et al., 2010), soybean (Lam et al., 2012), chickpea, and field pea (Jin et al., 2012). The CO 2stimulation of N-fixer growth and N 2 fixation in several grasslands experiments was higher with P addition compared to controls without supplemental P (Niklaus et al., 1998;Grunzweig and Korner, 2003). In general, with supplemental nutrients added, elevated CO 2 often increases N 2 fixation (Lee et al., 2003;Otera et al., 2011), but in experiments without exogenous nutrient addition, N 2 fixation is generally not responsive to elevated CO 2 ( van Groenigen et al., 2006).

Herbivory
Herbivory has been postulated to reduce N 2 fixation in ecosystems because of preferential feeding on the more nutritious tissue of N 2 -fixers compared to other plants (Vitousek and Howarth, 1991). This hypothesis is consistent with our finding that herbivores in this subtropical woodland consumed a larger proportion of G. elliottii leaf production than leaf production of oaks (Fig. 6). Herbivory could also limit the response of N 2 fixation to environmental change, if grazing on the tissue of N 2 -fixers increases. Increasing herbivore pressure on the dominant N 2 -fixing vine in our experiment was associated with reduced rates of N 2 fixation (Table 3), which supports this idea. Elevated CO 2 decreased the proportion of leaves with herbivore damage in this scrub oak woodland (Stiling and Cornelissen, 2007), as has been found in other systems (Lindroth, 2010;Robinson et al., 2012). Insect herbivores often respond to reduced leaf nitrogen concentration by consuming more leaf tissue . In the scrub oak woodland studied here, this response resulted in no effect of elevated CO 2 on the proportion of N-fixer leaf production consumed by herbivory (Fig. 6).

C limitation of nonsymbiotic N 2 fixation
Our finding that acetylene reduction increased with glucose addition suggests C limitation of heterotrophic N 2 fixation in soil (Tables 5 and 6), as has been found previously (Billings et al., 2003). Neither P addition nor altered soil water content influenced rates of acetylene reduction, indicating that these factors were not limiting to soil N 2 fixation.

Temporal dynamics
Evidence for a single, dominating mechanism underlying the response of N 2 fixation to elevated CO 2 was weak in our experiment, possibly because the process is limited by multiple factors whose influences shift over time. The experimental site was struck by a hurricane in 2004 (Li et al., 2007a, b), a disturbance that preceded the high soil nutrient concentrations observed in 2005, which subsequently declined in 2006 and 2007 (Fig. 3). This disturbance also preceded a large positive response of oak production to elevated CO 2 found both above-and belowground (Day et al., 2013;Hungate et al., 2013a), as well as the disappearance of the CO 2 suppression of G. elliottii growth and N 2 fixation (Fig. 1). Consistent with this, the positive correlation between foliar Mo concentration and N 2 fixation rates found in the two years before the hurricane  was no longer apparent by the final harvest (regression between foliar [Mo] and N 2 fixation for 2007, r 2 = 0.138; P = 0.638). Yet, rates of N 2 fixation were low after the hurricane disturbance in both treatments (Fig. 1), and there was no association between soilextractable micronutrient availability and rates of N 2 fixation during the period 2005-2007; therefore, the availability of micronutrients, if they played a role, were not the only factor limiting N 2 fixation at this time.
The temporal variation in the response of N 2 fixation, in particular the finding that initially strong positive responses dissipate with time (Leuzinger et al., 2011), may be a general feature of global change experiments. While elevated CO 2 can stimulate N 2 fixation in some species and under some conditions, responses in many field studies are far more muted (van Groenigen et al., 2006) consistent with findings reported here. Thus, increased N 2 fixation is not a certain outcome of rising atmospheric CO 2 concentrations.

N Losses
N fixation is the major biological process mediating N inputs to terrestrial ecosystems, but N losses through gaseous and leaching pathways exert an equally important influence on total N pool size. Our finding that elevated CO 2 had no significant effect on N 2 O production in this subtropical woodland is consistent with several past studies finding no effect of CO 2 on N 2 O production (Mosier et al., 2002;Phillips et al., 2001), though increased N 2 O production has been detected in others (Hagedorn et al., 2000;Ineson et al., 1998;Kammann et al., 2008;Lam et al., 2010;Smith et al., 2010). On average, elevated CO 2 tends to increase soil N 2 O emissions by around 20 % ( van Groenigen et al., 2011), a larger stimulation than the nonsignificant increase of 8 % we observed in this subtropical woodland. The trend for elevated CO 2 to reduce NO x losses from this subtropical woodland indicates that NO x losses were likely not the major pathway of increased N loss from this system in response to elevated CO 2 . By contrast, the concurrence of reduced tracer 15 N recovery (Hungate et al., 2013b) and deeper distribution of tracer 15 N throughout the soil profile (Fig. 7) supports the notion that elevated CO 2 stimulated N leaching in this experiment. Increased leaching with elevated CO 2 has been observed (Korner and Arnone, 1992) and may be caused by a combination of increased plant water-use efficiency resulting in greater downward water flux through the soil profile (Jackson et al., 1998), along with increased turnover of soil organic matter in response to rising CO 2 (Drake et al., 2011;Hungate et al., 2013b). Some experiments have documented reduced N leaching with elevated CO 2 , so our finding of increased leaching is likely not universal. Furthermore, during the first three years of this experiment, we found no effect of elevated CO 2 on vertical movement of N from the surface to deeper soil layers . The use of the 15 N tracer to integrate the cumulative effects of elevated CO 2 on nitrogen losses in this experiment may be both more temporally integrative and sensitive to changes in the ecosystem-scale distribution and retention of N in response to elevated CO 2 .

Summary
Eleven years of chronic exposure to increased CO 2 concentrations elicited disequilibrium in the N cycle, with increased rates of internal N transformations, no change in N inputs, and increased N losses. Elevated CO 2 accelerated the rate of soil N mineralization (Langley et al., 2009;McKinley et al., 2009), which likely contributed to increased N uptake by plants . Nitrogen losses also increased, with increased turnover of N through plant tissues, as evidenced by the increased 15 N dilution in plants  along with no change in net plant N capital . Elevated CO 2 also appeared to enhance N losses at the scale of the soil profile: the pattern of lower 15 N recovery in plots exhibiting greater downward movement of 15 N in the soil profile suggests increased leaching. Thus, processes that make nutrients available to plants can also promote nutrient losses. Finally, we found no evidence that elevated CO 2 enhanced N inputs via N 2 fixation. Together, these results describe an ecosystem in which more rapid cycling of N with elevated CO 2 is unlikely to be sustained. These empirical findings contrast with model projections in which elevated CO 2 enhances N 2 fixation and reduces leaching (Thornley and Cannell, 2000). Given the strong influence of N cycling and N accumulation on the C cycle, the changes 3334 B. A. Hungate et al.: Nitrogen inputs and losses in response to chronic CO 2 exposure in N cycling reported here, if general, would tend to dampen the biosphere's capacity to assimilate and store C in the face of rising atmospheric CO 2 concentrations (Thornton et al., 2007;Churkina et al., 2009;Arneth et al., 2010;Zaehle and Dalmonech, 2011).