Icehouse – Greenhouse Variations in Marine Denitrification

Long-term secular variation in the isotopic composition of seawater fixed nitrogen (N) is poorly known. Here, we document variation in the N-isotopic composition of marine sediments ( δNsed) since 660 Ma (million years ago) in order to understand major changes in the marine N cycle through time and their relationship to first-order climate variation. During the Phanerozoic, greenhouse climate modes were characterized by lowδNsed (∼ −2 to +2 ‰) and icehouse climate modes by high δNsed (∼ +4 to +8 ‰). Shifts toward higherδNsed occurred rapidly during the early stages of icehouse modes, prior to the development of major continental glaciation, suggesting a potentially important role for the marine N cycle in long-term climate change. Reservoir box modeling of the marine N cycle demonstrates that secular variation inδNsed was likely due to changes in the dominant locus of denitrification, with a shift in favor of sedimentary denitrification during greenhouse modes owing to higher eustatic (global sea-level) elevations and greater on-shelf burial of organic matter, and a shift in favor of water-column denitrification during icehouse modes owing to lower eustatic elevations, enhanced organic carbon sinking fluxes, and expanded oceanic oxygen-minimum zones. The results of this study provide new insights into operation of the marine N cycle, its relationship to the global carbon cycle, and its potential role in modulating climate change at multimillion-year timescales.


Introduction
Nitrogen (N) plays a key role in marine productivity and organic carbon fluxes and is thus a potentially major influence on the global climate system (Gruber and Galloway, 2008). Variation in marine sediment N-isotopic compositions during the Quaternary (2.6 Ma to the present) has been linked to changes in organic carbon burial and oceanic denitrification rates during Pleistocene glacial-interglacial cycles (François et al., 1992;Altabet et al., 1995;Ganeshram et al., 1995;Haug et al., 1998;Naqvi et al., 1998;Broecker and Henderson, 1998;Suthhof et al., 2001;Liu et al., 2005Liu et al., , 2008. At this timescale (i.e., ∼ 10 5 yr), the marine N cycle is thought to act mainly as a positive climate feedback, but negative feedbacks involving the influence of both N fixation and denitrification on oceanic fixed-N inventories have been proposed as well (Deutsch et al., 2004). Although pre-Quaternary δ 15 N sed variation has been reported, including highly 15 N-depleted (−4 to 0 ‰) Jurassic-Cretaceous units (Rau et al., 1987;Jenkyns et al., 2001;Junium and Arthur, 2007) and highly 15 N-enriched (+6 to +14 ‰) Carboniferous units (Algeo et al., 2008), the Phanerozoic record of marine N-isotopic variation and its relationship to long-term (i.e., multimillion-year) climate change have not been systemically investigated to date (Algeo and Meyers, 2009). Additional study of the marine N cycle is needed to better understand its relationship to organic carbon burial and long-term climate change and to more accurately parameterize N fluxes in general circulation models. In this study, we document variation in δ 15 N sed from 660 Ma to the present, demonstrating a strong relationship to first-order climate cycles, with   Table 1). The mean long-term trend is given by a LOWESS curve (red line) and uncertainty envelope ( ± 1σ ; green field). The LOWESS curve, which varies over a ∼ 10 ‰ range, accounts for 74 % of total variance in the δ 15 N sed data set. At top, epochs of moderate (light blue) and heavy (dark blue) continental glaciation are from Montañez et al. (2011); the ages of all marine sedimentary units and climate events have been adjusted to the timescale of Gradstein et al. (2012). Tr = transitional interval.
lower δ 15 N during greenhouse intervals and higher δ 15 N during icehouse intervals. This pattern suggests that long-term variation in the marine N cycle is controlled by first-order tectonic cycles, and that it is linked to (is a possibly a driver of) long-term climate change.

Methods
This study is based on the N-isotope distributions of 153 marine units ranging in age from the Neoproterozoic (660 Ma) to the early Quaternary (∼ 2 Ma) (Fig. 1). Among these units are 35 that were analyzed specifically for this study (see isotopic methods, Appendix A), 33 that were taken from our own earlier research publications, and 85 that were taken from other published reports. For each study unit, we determined the median (50th percentile), standard deviation range (16th and 84th percentiles), and full range (minimum and maximum values) of its δ 15 N distribution (Table 1). We also report organic δ 13 C distributions as well as means for %TOC (total organic carbon), %N, and molar C org : N ratios, where available (Table 1). The ages of all units were adjusted to the 2012 geologic timescale (Gradstein et al., 2012). A LOWESS (LOcally WEighted Scatterplot Smoothing) curve was calculated for the entire data set per the methods of Appendix B.

Results
Our δ 15 N sed data set exhibits a mean plus/minus one standard deviation of +2.0 ± 3.1 ‰ with a range of −5.2 to +10.4 ‰ (  (Fig. 1). The modeled LOWESS curve for the Phanerozoic exhibits a minimum of −2.8 ‰ in the Cambrian and a maximum of +8.0 ‰ in the Mississippian. The uncertainty attached to this mean trend varies from ±0.9 to 2.9 ‰ through the Phanerozoic but is mostly < ±2 ‰ (based on plus/minus one standard deviation). The most abrupt changes in δ 15 N sed are associated with a ∼ 6 ‰ rise during the mid-Mississippian and a ∼ 5 ‰ rise during the Late Cretaceous. The Phanerozoic δ 15 N sed curve shows a strong relationship to first-order climate cycles, with low values during the greenhouse climate modes of the mid-Paleozoic and mid-Mesozoic and high values during the icehouse climate modes of the Late Paleozoic and Cenozoic (Fig. 1). The δ 15 N sed data set exhibits pronounced secular variation (i.e., a range of > 10 ‰) and strong secular coherence (i.e., 74 % of total variance is accounted for by the LOWESS curve). The secular coherence of the data set is significant in view of the relatively short residence time of nitrate in seawater (∼ 3 kyr) (Tyrrell, 1999;Brandes and Devol, 2002), which theoretically offers potential for strong δ 15 N NO − 3 variation at intermediate (10 3 -10 6 yr) timescales (Deutsch et al., 2004). Indeed, sub-Recent marine sediments exhibit a ∼ 14 ‰ range of δ 15 N variation (Tesdal et al., 2012), reflecting local water mass effects linked to (1) strong N fixation, which can lower δ 15 N NO − 3 by several per mille, as in the Cariaco Basin and Baltic Sea, and (2) strong water-column denitrification, which can raise δ 15 N NO − 3 by > 10 ‰, as in upwelling systems in the Arabian Sea and the eastern tropical Pacific (Brandes and Devol, 2002;. However, the cumulative δ 15 N distribution for sub-Recent sediments yields a mode of 5-6 ‰ with a standard deviation of ±2.5 ‰ (Tesdal et al., 2012), which conforms well to the isotopic composition of modern seawater nitrate (+4.8 ± 0.2 ‰) (Sigman et al., 2000); note that the mean value of 6.7 ‰ reported by Tesdal et al. (2012) is skewed toward the high side by an overrepresentation of upwelling-zone sediments. Thus, the δ 15 N sed values of paleomarine units (Table 1) can be viewed as a random sample of a population of sediment δ 15 N sed values of a given age, the average of which is close to the δ 15 N NO − 3 of contemporaneous seawater. Although we cannot discount the possibility that some of our units are nonrepresentative of seawater δ 15 N NO − 3 of a given age, the broadly coherent pattern of secular variation recorded by our data set is not consistent with it being primarily a record of random local water mass effects (see Sect. 4.4).

The marine nitrogen cycle
Long-term secular variation in δ 15 N sed and, by extension, in the δ 15 N of seawater fixed nitrogen can be interpreted in terms of dominant processes of the marine N cycle, the main features of which are now well understood. Most bioavailable N is fixed by diazotrophic cyanobacteria with a fractionation of −1 to −3 ‰ relative to the atmospheric N 2 source (δ 15 N air ∼ 0 ‰) (Brandes and Devol, 2002;. Apart from assimilatory uptake, the major sinks for seawater fixed N are denitrification within the sediment or in the water column and the anammox process. Denitrification involves the bacterial use of nitrate as an oxidant in the respiration of organic matter with a maximum fractionation of ∼ −27 ‰ (but commonly with an effective fractionation of ∼ −20 ± 3 ‰), resulting in a strongly 15 N-enriched residual seawater nitrate pool. Denitrification in suboxic marine sediments typically yields much lower net fractionation (∼ −1 to −3 ‰ ) owing to near-quantitative utilization of porewater nitrate Lehmann et al., 2004). The anammox reaction, in which ammonium and nitrate (or nitrite) are converted to N 2 , may eliminate more fixed N than denitrification in some marine environments (Kuypers et al., 2005), although the isotopic fractionation associated with this process is not well known (Galbraith et al., 2008).
The N-isotopic composition of marine sediment depends on the δ 15 N of seawater fixed N, fractionation during assimilatory uptake, and subsequent alteration during decay in the water column and sediment . Both ammonium and nitrate can be used as N sources in primary production, with fractionations of −10 (± 5) ‰ and −3( ± 2) ‰, respectively (Hoch et al., 1994;Waser et al., 1998). Nitrate is by far the more important source of N for eukaryotic marine algae, but ammonium is utilized by some modern microbial communities (Higgins et al., 2012) and may have been the main N substrate for eukaryotic algae during some oceanic anoxic events (OAEs; Altabet, 2001;Higgins et al., 2012). Assimilatory uptake enriches the residual fixed N pool in 15 N and can result in shifts in the δ 15 N NO3 -of local water masses (Hoch et al., 1994), but quantitative utilization of fixed N by marine autotrophs at annual timescales normally limits fractionation due to this process (Sigman et al., 2000;Somes et al., 2010). These processes determine the N-isotopic composition of primary marine organic matter, before modification by diagenesis.

Influence of diagenesis on sediment δ 15 N
Diagenesis has the potential to alter the N-isotopic composition of organic matter. First, selective degradation of amino acids can produce shifts of a few per mille in δ 15 N sed (Prahl et al., 1997;Gaye-Haake et al., 2005). Second, aerobic bacterial decomposition of organic matter results in deamination, i.e., the release of isotopically light NH + 4 to sediment porewaters (Macko and Estep, 1984;Macko et al., 1987;Holmes et al., 1999), which results in 15 N enrichment of the organic residue by a few per mille (Altabet, 1988;Libes and Deuser, 1988;François et al., 1992;Saino, 1992;Lourey et al., 2003). Subsequent nitrification can enrich the porewater NH + 4 pool in 15 N by 4-5 ‰, potentially leading to changes in bulk-sediment δ 15 N if NH + 4 diffuses back to the water column (Brandes and Devol, 1997;Prokopenko et al., 2006). However, if NH + 4 generated within the sediment is captured by clay minerals, then the bulk sediment may show little or no change in δ 15 N relative to the organic sinking flux (Higgins et al., 2012). Because decay processes can have variable effects on δ 15 N sed , net fractionation can be either positive or negative relative to the unaltered source material. Surface sediments tend to be enriched in 15 N by 1-5 ‰ relative to particulate organic nitrogen in the water column, possibly because the latter has undergone less extensive deamination (Brandes and Devol, 1997;Gaye-Haake et al., 2005;Prokopenko et al., 2006;Higgins et al., 2010). Differences between the δ 15 N of the sinking and sediment fractions show water-depth dependence, reflecting greater oxic degradation of organic matter settling to the deep-ocean floor, although this effect is relatively small . In contrast, rapid burial of organic matter in continental shelf and shelf-margin settings can yield sediment δ 15 N values that are little modified from those of the organic export flux (Altabet and François, 1994;Altabet, 2001;Robinson et al., 2012).
Studies of N subfractions have been undertaken with the goal of recovering a N-isotopic signature that is comparatively free of diagenetic effects. Chlorin N (Higgins et al., 2010) is 15 N-depleted relative to bulk-sediment N due to a ∼ 5 ‰ fractionation during photosynthesis . Some studies have claimed large (up to 5 ‰ ) shifts in bulk-sediment δ 15 N as a consequence of diagenesis . However, the studies of N subfractions cited above exhibit a systematic offset of 3-5 ‰ between bulk-sediment and compound-specific δ 15 N values that is consistent with the effects of photosynthetic fractionation overprinted by, at most, small (< 2 ‰) diagenetic effects. Following early diagenesis, deeper burial rarely causes more than minor changes in N-isotopic compositions, as shown by (1) δ 15 N variation of only a few per mille over a wide range of metamorphic grades (Imbus et al., 1992;Busigny et al., 2003;Jia and Kerrich, 2004), and (2) δ 15 N values for metamorphosed units that are virtually indistinguishable from those of coeval unmetamorphosed units (e.g., compare the Eocene-Jurassic Franciscan Complex with age-equivalent units; Table 1). Ancient marine sediments are thus considered to be fairly robust recorders of the ambient isotopic composition of seawater fixed N (Altabet and François, 1994;Altabet et al., 1995;Higgins et al., 2010;Robinson et al., 2012).

Influence of organic matter source on sediment δ 15 N
All of the data used in this study represent bulk-sediment Nisotopic compositions, thus including both organic and inorganic nitrogen. The amount of mineral N present in most marine sediments is so small that it typically has little influence on bulk-sediment δ 15 N (Holloway et al., 1998;Holloway and Dahlgren, 1999). In contrast, clay-adsorbed N (principally ammonium) can be quantitatively important, with concentrations of ∼ 0.1-0.2 % in some marine units (e.g., Fig. 3 in Meyers, 1997;Fig. 3 in Lücke and Brauer, 2004;and Fig. S1 in Algeo et al., 2008). However, clay-adsorbed N is mostly derived from sedimentary organic matter, and the organic-toclay transfer of nitrogen is often at a late diagenetic stage, thus limiting translocation of N within the sediment column (Macko et al., 1986). These considerations suggest that the presence of a small inorganic N fraction in the study units is unlikely to affect our results.
In compiling the δ 15 N sed data set used in the present study, our principal concern was that admixture of large amounts of terrestrially sourced organic N might bias the marine δ 15 N record. A number of different procedures can be used to  screen samples for the presence of terrestrial organic matter, including petrographic analysis to identify maceral types (Hutton, 1987), biomarker analysis of steroids, polysaccharides, and hopane and tricyclic ratios (Huang and Meischein, 1979;Frimmel et al., 2004;Peters et al., 2004;Grice et al., 2005;Sephton et al., 2005;Wang and Visscher, 2007;Xie et al., 2007;Algeo et al., 2012), and hydrogen and oxygen indices (HI-OI) (Espitalié et al., 1977(Espitalié et al., , 1985Peters, 1986). Such proxies are generally reliable in distinguishing organic matter sources, subject to some caveats (Meyers et al., 2009a). These types of proxies were available only for a subset of the present study units (Table 1), but, where available, they generally confirmed the dominance of marine over terrestrial organic matter. Studies of modern continental shelf sediments show a rapid decline in the proportion of terrestrial organic matter away from coastlines (Hedges et al., 1997;Hartnett et al., 1998). The study units of Proterozoic to Jurassic age were mostly epicontinental and, hence, deposited close to land areas (Fig. 2), although there was little terrestrial vegetation for export to marine systems prior to the Devonian (Kenrick and Crane, 1997   and distal continent-margin sites that were at a significant remove from land areas (Fig. 2) and, hence, unlikely to have accumulated large amounts of terrestrial organic matter. Sediment C org : N ratios potentially also provide insights regarding organic matter sources (Meyers, 1994(Meyers, , 1997. Terrestrial organic matter is characterized by high C org : N ratios (∼ 20-200) owing to an abundance of N-poor cellulose in land plants (Ertel and Hedges, 1985). In contrast, fresh marine organic matter exhibits low C org : N ratios (∼ 4-10) owing to a lack of cellulose and an abundance of N-rich proteins in planktic algae (Müller, 1977). Diagenesis can result in either lower C org : N ratios through preferential preservation of organic N as clay-adsorbed ammonium, or higher C org : N ratios through preferential loss of proteinaceous components (Meyers, 1994). Covariation between δ 15 N, δ 13 C org , and C org : N ratios can reveal mixing relationships in estuarine (Thornton and McManus, 1994;Ogrinc et al., 2005) and marine sediments (Müller, 1977;Meyers et al., 2009b). In our Phanerozoic data set, δ 13 C org and δ 15 N exhibit no relationship (r 2 = 0.01; Fig. 3), but δ 15 N exhibits moderate negative covariation with C org : N (r 2 = 0.21; p(α) < 0.001; Fig. 4). The source of the latter relationship is uncertain. Although conceivably representing a marine-terrestrial mixing trend, this interpretation is unlikely given that the majority of units with low δ 15 N and high C org : N values come from openmarine settings of Cretaceous-Recent age that presumably contain little terrestrial organic matter. The linkage of higher C org : N ratios (to ∼ 40) with lower δ 15 N values is particu-   Fig. 4. δ 15 N sed versus C org : N ratio. Average composition of modern marine plankton shown by red star, and approximate compositional range of terrestrial (i.e., soil-derived) organic matter by green rectangle. Note that the pattern of negative covariation between δ 15 N sed and C org : N is not clearly associated with a terrestrial endmember and does not provide evidence of pervasive mixing of marine and terrestrial organic matter in our study units. larly characteristic of organic-rich sediments deposited under anoxic conditions (e.g., Junium and Arthur, 2007). This pattern has been attributed to enhanced cyanobacterial N fixation under N-poor conditions in restricted anoxic marine basins (Junium and Arthur, 2007) but potentially might be due to enhanced assimilatory recycling of 15 N-depleted ammonium in such settings (Higgins et al., 2012).
The relatively N-poor nature of terrestrial organic matter means that, even if present in modest quantities, it is unlikely to have had much influence on bulk sediment δ 15 N. For example, in a 50 : 50 mixture of marine and terrestrial organic matter, ∼ 80-95 % of total N will be of marine origin because of the lower C org : N ratios of marine organic matter (∼ 4-10) relative to terrestrial organic matter (∼ 20-200) (Meyers, 1994(Meyers, , 1997. Where mixing proportions have been quantified, the terrestrial organic fraction is more commonly in the range of 10-20 % (e.g., Jaminski et al., 1998;Algeo et al., 2008), in which case > 95 % of total N is marinederived. Although we cannot conclusively demonstrate that our Phanerozoic marine δ 15 N trend (Fig. 1) is uninfluenced by terrestrial contamination, we infer that such influences were probably minimal, and that the observed pattern of secular variation in δ 15 N sed broadly reflects the isotopic composition of contemporaneous seawater fixed N.

Influence of depositional setting on sediment δ 15 N
One important issue is whether our δ 15 N sed data set records variation in a global parameter (i.e., seawater nitrate δ 15 N) or represents mainly local water mass effects in which sediment δ 15 N varied as a function of depositional setting. Owing to unevenness in the distribution of depositional settings in our data set through time, we cannot answer this question definitively, but the following analysis provides some insight as to the relative importance of local versus global controls on sediment δ 15 N.
We classified the 153 study units into five categories of depositional setting: (1) oceanic, i.e., unrestricted deep marine; (2) oceanic-mediterranean, i.e., restricted deep marine; (3) upwelling, i.e., open continental margin/slope with a known upwelling system; (4) shelf, i.e., open continental margin without upwelling; and (5) epeiric sea, i.e., a cratonic-interior shelf or basin. When viewed as a function of time (Fig. 5), it is apparent that there is a major change in depositional settings in the mid-Mesozoic: all pre-Cretaceous units are from either shelf or epeiric-sea settings, whereas Cretaceous to Recent units are mostly oceanic or oceanic-mediterranean with a small number from other settings. The reason for this shift is that nearly all pre-Cretaceous units were collected in outcrop and represent cratonic deposits, whereas the majority of Cretaceous and younger units were collected during deepsea (DSDP, ODP, or IODP) cruises and represent deep-ocean deposits. Thus, the ultimate control on the age distribution of depositional settings in our data set is the age distribution of present-day oceanic crust. Several significant observations can be gleaned from the age distribution of depositional settings (Fig. 5). First, relatively young (i.e., Neogene) oceanic units exhibit an average δ 15 N sed (+4.2 ± 0.8 ‰) that overlaps with and is only marginally depleted relative to the N-isotopic composition of present-day seawater nitrate (+4.8-5.0 ‰) (Sigman et al., 2000). This observation is consistent with the inference that δ 15 N sed is a relatively robust recorder of seawater δ 15 N NO − 3 (Altabet and François, 1994;Altabet et al., 1995;Higgins et al., 2010;Robinson et al., 2012). Second, the range of δ 15 N sed variation shown by Neogene units as a function of depositional setting is limited: on average, upwelling units (+5.5 ± 2.1 ‰) are just 1.3 ‰ enriched and oceanicmediterranean units (+1.2 ± 2.2 ‰) just 3.0 ‰ depleted in 15 N relative to oceanic units (Fig. 3). While these differences are statistically significant (at p(α) < 0.01), they are much smaller than the > 10 ‰ range of δ 15 N sed variation observed through the Phanerozoic (Fig. 1). Third, secular variation in δ 15 N sed is coherent across the mid-Mesozoic "junction" at which pre-Cretaceous epeiric/shelf units yield to Cretaceous and younger oceanic/oceanic-mediterranean units (Fig. 5). This observation is significant because it suggests that different kinds of depositional settings are recording a common signal that shows up in both cratonic interiors and the deep ocean. While local influences are likely to have modified the  N-isotopic composition of some study units, the foregoing observations are not consistent with the hypothesis that our Phanerozoic δ 15 N sed record is dominated by such influences. We infer that there is a dominant underlying secular signal present in the δ 15 N sed data set that is independent of setting type and that reflects a global control, i.e., seawater δ 15 N NO − 3 . In our data set, most of the Phanerozoic is characterized by a relatively low density of data (averaging one data point per 4-5 million years). However, a few narrow (≤ 1-Myrlong) time slices are represented by multiple data points, providing a basis for assessing spatial variance at certain times in the past. One such interval is the Cenomanian-Turonian boundary (Fig. 6). During this interval, the range of variation in unit-mean δ 15 N sed values is just 1.7 ‰ (i.e., −2.9 to −1.2 ‰) for sites ranging from high northern to high southern paleolatitudes. While the majority of these units represent oceanic-mediterranean settings in the young North Atlantic and South Atlantic basins, similar δ 15 N sed values are nonetheless observed in epeiric (−2.9 ‰ in England) and upwelling settings (−1.6 ‰ in Morocco) (Jenkyns et al., 2007) as well as outside the Atlantic region (−2.6 ‰ on the Kerguelen Plateau) (Meyers et al., 2009a). Thus, these data imply a relatively uniform N-isotopic composition for global seawater nitrate at the Cenomanian-Turonian boundary. Further, almost all of these regions exhibit a +4 to +5 ‰ shift in δ 15 N sed for units of latest Cretaceous to early Paleogene age ( Fig. 1; Table 1), consistent with our hypothesis of a global shift in seawater δ 15 N NO −   Table 1.
Another time slice with multiple data points is the Permian-Triassic boundary (PTB; Fig. 7). The range of δ 15 N sed variation observed at the PTB is somewhat greater than for the Cenomanian-Turonian boundary, but the geographic distribution of units is wider and their setting types are more diverse as well. Late Permian units predating the PTB crisis exhibit a δ 15 N sed range of 4.6 ‰ (i.e., +0.3 ‰ to +4.9 ‰) but show spatially coherent variation: low values characterize the central Panthalassic Ocean (+0.3 ‰), intermediate values the Tethyan region (mostly +2.0 to +3.6 ‰), and high values the northwestern Pangean margin (+3.5 to +4.9 ‰). This pattern is likely to reflect regional variation in the intensity of water-column denitrification, which was higher in the high-productivity oceanic cul-de-sac formed by the Tethys Ocean (Mii et al., 2001;Grossman et al., 2008) and in the northwest Pangean upwelling system (Beauchamp and Baud, 2002;Schoepfer et al., 2012Schoepfer et al., , 2013. Despite major changes in seawater temperature and dissolved oxygen levels in conjunction with the PTB crisis (Romano et al., 2012;Sun et al., 2012;Song et al., 2013), marine units show remarkably little change in δ 15 N across the PTB: Lower Triassic unit means range from −0.4 ‰ to +5.3 ‰, and the magnitude of the PTB shift at individual locales varies from −2.8 ‰ to +0.4 ‰ with an average of −0.9 ‰. Negative δ 15 N sed shifts at the PTB have been attributed to enhanced N fixation rates (Luo et al., 2011). However, these shifts are consistent with our hypothesis of lowered seawater δ 15 N NO3values as a consequence of enhanced rates of sedimentary (relative to water-column) denitrification during greenhouse climate intervals such as that of the Early Triassic (Romano et al., 2012;Sun et al., 2012).

Marine nitrogen cycle modeling
We employed a reservoir box model to investigate possible controls on long-term secular variation in seawater δ 15 N NO − 3 (see Appendix C for model details). Seawater δ 15 N NO − 3 can be approximated from a steady-state isotope mass balance that assumes N fixation (f FIX ) as the primary source and sedimentary (f DS ) and water-column (f DW ) denitrification as the two largest sinks for seawater fixed N (Brandes and Devol, 2002;Deutsch et al., 2004;. We assumed that the marine N cycle is in a homeostatic steadystate condition at geologic timescales (DeVries et al., 2013), and thus that losses of fixed N to denitrification are balanced by new N fixation (i.e., f DW + f DS = f FIX ), which is consistent with the strong spatial coupling of these processes in the modern ocean (Galbraith et al., 2004;Deutsch et al., 2007;Knapp et al., 2008). Our baseline scenario utilized fluxes and fractionation factors based on the modern marine N cycle -i.e., f FIX = 220 Tg a −1 , f DS = 160 Tg a −1 , f DW = 60 Tg a −1 , ε FIX = −2 ‰, ε DS = −2 ‰, and ε DW = −20 ‰ -where ε represents the fractionations associated with N source and sink fluxes (f ), and the photosynthetic fractionation factor (ε P ) linked to nitrate utilization is 0 ‰. This scenario yields an equilibrium seawater δ 15 N NO − 3 of +4.9 ‰ that matches the composition of fixed N in the present-day deep ocean (Sigman et al., 2000).
The most important influence on global seawater δ 15 N NO − 3 variation in our model is the fraction of denitrification that occurs in the water column (F DW , calculated as f DW / (f DW + f DS ); Fig. 8). In our baseline scenario (ε DW = −20 ‰, ε P = 0 ‰), the modern seawater δ 15 N NO − 3 of ∼ +4.9 ‰ corresponds to F DW of 0.27 (point 1, Fig. 8b), which is close to recent estimates of 0.29 (DeVries et al., 2012) and 0.36 (Eugster and Gruber, 2012). However, the same δ 15 N NO − 3 composition can be achieved with other model parameterizations. Laboratory culture studies indicate that ε DW might be as low as −10 ‰ in some marine systems (Kritee et al., 2012). Reducing ε DW to −15 ‰ and −10 ‰ yields F DW of 0.37 and 0.62 (points 2 and 3, Fig. 8b); the former is still within the range of F DW estimates for modern marine systems (cf. Eugster and Gruber, 2012) although the latter is not. Our baseline scenario assumes no net fractionation linked to photosynthetic assimilation of seawater nitrate (ε P = 0 ‰) (Brandes and Devol, 2002;Granger et al., 2010), but uptake of ammonium is accompanied by a significant negative fractionation (Hoch et al., 1994;Waser et al., 1998), and a recent study lends support to the hypothesis that recycled ammonium was a major source of fixed N for eukaryotic algae during some OAEs (Higgins et al., 2012). We modeled the effects of variable photosynthetic fractionation with ε P values of −4 ‰ and −8 ‰, which yield δ 15 N NO − 3 equal to +4.9 ‰ when F DW is 0.48 and 0.72, respectively (points 4 and 5, Fig. 8b). These F DW values are improbably large for the modern (icehouse) marine N cycle, but nonzero values of   (Fig. 1). Site-specific changes in δ 15 N sed across the PTB range from −2.8 ‰ to +0.4 ‰ with an average of −0.9 ‰, which is consistent with our hypothesis of intensified sedimentary denitrification during greenhouse climate intervals such as the Early Triassic. Data sources in Table 1 with additional data from S. Schoepfer and T. Algeo (unpubl. data). ε P may have been important during greenhouse intervals (see below).
The variations in seawater δ 15 N NO − 3 between icehouse and greenhouse climate modes observed in our long-term δ 15 N sed record ( Fig. 1) are an indication of major secular changes in the marine N cycle. In our baseline scenario, the peak icehouse δ 15 N NO − 3 of ∼ +8 ‰ yields F DW of ∼ 0.45 (point 6, Fig. 8b), indicating an increase in water-column denitrification relative to the modern ocean. Although the same δ 15 N NO − 3 can be achieved with ε DW of −15 ‰ and −10 ‰, the resulting F DW values (0.62 and 0.98; points 7 and 8, Fig. 8b) are improbably large. Lack of evidence for ammonium recycling during icehouse modes makes nonzero ε P values unlikely, which in any case would yield equally improbable values of F DW . The minimum greenhouse δ 15 N NO − 3 of ∼ −3 ‰ cannot be achieved in our baseline scenario even when F DW is reduced to 0 (point 9, Fig. 8b). However, evidence for strong ammonium recycling in greenhouse oceans (Higgins et al., 2012) indicates that ε P may have been nonzero at those times. Decreasing ε P to −4 ‰ and −8 ‰ yields F DW of 0.10 and 0.33 (points 10 and 11, Fig. 8b), the former representing a large decrease in F DW relative to the modern ocean. Since ε P of −8 ‰ represents an absolute minimum (i.e., recycling of nearly all seawater N as ammonium), ε P = −4 ‰ is a more reasonable estimate for anoxic marine systems with mixed utilization of recycled ammonium and cyanobacterially fixed N (Higgins et al., 2012). Note that, at low F DW , variation in ε DW has little effect on δ 15 N NO − 3 . In summary, the most likely scenario to account for long-term secular shifts in δ 15 N NO − 3 ( Fig. 1) within existing N-budget and N-isotopic constraints is for (1) F DW to vary between ∼ 0.2 and 0.5 (permissive of ε DW values between −15 and −20 ‰ ) with ε P = 0 ‰ during icehouse climate modes, and (2) F DW to decrease to ∼ 0.1-0.2 with a shift in ε P to ca.

Controls on long-term variation in the marine nitrogen cycle
Although enhanced water-column denitrification has been inferred during the warm climate intervals that produced OAEs (Rau et al., 1987;Jenkyns et al., 2001;Junium and Arthur, 2007), strong 15 N depletion of contemporaneous sediments is inconsistent with globally elevated water-column denitrification rates. Our inference of reduced water-column denitrification during greenhouse climate modes (Fig. 9a) contradicts the existing paradigm linking OAEs to high watercolumn denitrification rates (Rau et al., 1987;Jenkyns et al., 2001;Junium and Arthur, 2007). A reconciliation of these views is possible if rates were high regionally in semirestricted marine basins such as the proto-South Atlantic but reduced on a globally integrated basis. Our results are also at odds with the observation that modern upwelling zones exhibit peak δ 15 N sed values in conjunction with deglaciations rather than glacial maxima (François et al., 1992;Altabet et al., 1995;Ganeshram et al., 1995). While the latter relationship is valid at intermediate timescales, our results indicate that F DW is higher on a time-averaged basis (i.e., integrating glacial-interglacial variation) for icehouse modes than for greenhouse modes. This inference is supported by a study of Plio-Pleistocene sediments in the eastern tropical Pacific, in which δ 15 N sed rose by ∼ 2 ‰ following a cooling event at 2.1 Ma (Liu et al., 2008), consistent with an increase in time-averaged water-column denitrification rates. We infer that transient, albeit repeated, shifts in favor of watercolumn denitrification (i.e., higher F DW ) during the interglacial stages of icehouse climate intervals have resulted in a sustained (i.e., multimillion-year) shift toward higher seawater δ 15 N NO − 3 that has been captured by the long-term δ 15 N sed record (Fig. 1). Such a long-term climate-related shift in seawater δ 15 N NO − 3 can occur if the positive shift associated with each glacial epoch is larger than the negative shift associated with each interglacial epoch, resulting in progressively more 15 N-enriched compositions for icehouse intervals relative to greenhouse intervals.
Several mechanisms might potentially link variations in seawater δ 15 N NO − 3 to long-term climate cycles. Sea-level elevation is known to influence the locus of denitrification in marine systems (Deutsch et al., 2004). High sea-level as a function of the fraction of watercolumn denitrification (F DW ). The dashed diagonal lines represent variable fractionation during water-column denitrification (ε DW ), and the solid diagonal lines variable fractionation during photosynthetic uptake of seawater fixed N (ε P ). Colored fields show the isotopic range of marine δ 15 N sed during greenhouse (green) and icehouse (blue) climate modes as well as modern seawater δ 15 N NO − 3 (gray) (Sigman et al., 2000). Arrows at bottom show F DW of 0.27 (this study), 0.29 (DeVries et al., 2012), and 0.36 (Eugster and Gruber, 2012). The red curve represents our "most likely scenario" of concurrent changes in F DW and ε P as a function of greenhouse-icehouse climate shifts. See text for discussion of numbered points. elevations during greenhouse climate modes favor sedimentary denitrification owing to greater burial of organic matter on continental shelves (Fig. 9a), whereas low sealevel elevations during icehouse climate modes favor watercolumn denitrification through elevated organic carbon sinking fluxes to the thermocline and expansion of oceanic oxygen-minimum zones (Fig. 9b). A first-order sea-level control on the marine N cycle is consistent with existing records of Phanerozoic eustasy and continental flooding (Fig. 10). Eustasy shows a strong relationship to firstorder climate modes, with long-term rises or highstands during greenhouse intervals and long-term falls or lowstands during icehouse intervals. δ 15 N sed exhibits a distinct pattern of negative covariation with eustatic elevation for the Phanerozoic as a whole (r 2 = 0.18; Fig. 11). This relationship is even stronger for Cretaceous-Recent units alone (r 2 = 0.37), probably because both the δ 15 N sed and eustatic records are more securely defined for this interval than for the pre-Cretaceous. δ 15 N sed also exhibits negative covariation with long-term continental flooding records (Fig. 10), although the relationship is not as strong as for eustasy due to several factors: (1) greater vintage of the flooding records, (2) provinciality of the Sloss (1963) record (which represents only North America), and (3) low resolution of the Ronov (1984) record (which provides only 2-3 area estimates for most geologic periods).
A second potential mechanism linking long-term variation in the marine N cycle to first-order Phanerozoic climate cycles may be through tectonic controls. In this scenario, changes in oceanic gateways and circulation patterns can alter the locus of denitrification through changes in upwelling intensity or thermocline ventilation. In our long-term δ 15 N sed record, the mid-Early Mississippian and mid-Late Cretaceous feature as intervals of potentially rapid changes in seawater δ 15 N NO − Icehouse climate mode Greenhouse climate mode  circulation patterns. The Early Mississippian was a time of closure of an equatorial seaway in the Rheic Ocean region, which probably led to a change from circum-equatorial to meridional ocean circulation (Saltzman, 2003). The Late Cretaceous coincided with widening of the central and south Atlantic basins and a translocation of deepwater formation into the North Atlantic region (MacLeod and Huber, 1996; Barrera et al., 1997;Frank and Arthur, 1999). These examples show that the marine N cycle is intimately linked to first-order tectonic and climatic cycles, although further investigation will be needed to determine the exact nature of these connections.
The hypothesis that the marine N cycle has been a driver of long-term climate change is speculative but cannot be dismissed entirely. The critical issue is the nature of links between plate tectonics and global climate. Past work has focused largely on the role of the carbon cycle, i.e., changes   (2005) and Haq and Schutter (2008) as given in Snedden and Liu (2010).
(c) Continental flooding data from Sloss (1963) and Ronov (1984) as given in Miller et al. (2005). Phanerozoic climate modes at top are from Fig. 1; all ages have been adjusted to the Gradstein et al. (2012) timescale. Tr = transitional interval.
in atmospheric pCO 2 linked to mantle degassing, rates of uplift and continental weathering, and changes in marine organic carbon burial rates as a function of oceanic circulation and seawater redox conditions (Mackenzie and Pigott, 1981;Raymo and Ruddiman, 1991;Falkowski et al., 2000;Zachos et al., 2001;Berner, 2006a). The marine N cycle is intimately connected to burial of marine organic carbon Galloway et al., 2004), but whether it is a passive responder to changes in carbon fluxes (as generally assumed) or an active control on such changes is uncertain. One mechanism by which the N cycle might be a driver is through switches between equatorial and polar sites of deepwater formation (Barrera et al., 1997;Frank and Arthur, 1999), with attendant effects on sites of deepwater nutrient upwelling. Even if the marine N cycle is a passive responder to carbon-cycle forcings, it may play an important role as an amplifier of climate change. For example, enhanced N 2 O production in low-oxygen regions of the ocean during extended intervals of climatic warming might serve as a positive climate feedback (Naqvi et al., 1998;Bakker et al., 2013) that promotes a bimodality of long-term climate conditions (i.e., greenhouse versus icehouse modes; Fig. 1).

Conclusions
The present analysis of δ 15 N variation in 153 marine sedimentary units ranging in age from the Neoproterozoic to the Quaternary is the first to assess long-term variation in the marine N cycle and controls thereon. Variation in δ 15 N sed , which serves as a proxy for seawater nitrate δ 15 N, exhibits strong secular coherence since 660 Ma, with 74 % of total variance accounted for by a LOWESS trend. This pattern is surprising because the short residence time of fixed N in modern seawater (≤ 3 kyr) suggests that short-term variation in the marine N cycle has the potential to dominate the sedimentary N-isotope record and produce no coherent long-term patterns. Average δ 15 N sed ranges from lower values (∼ −2 to +2 ‰ ) during greenhouse climate modes of the mid-Paleozoic and mid-Mesozoic to higher δ 15 N (∼ +4 to +8 ‰) during icehouse climate modes of the late Paleozoic and Cenozoic. This pattern suggests that long-term variation in the marine N cycle is controlled by first-order tectonic cycles, and that it is linked to -and possibly a driver of -longterm climate change. We tentatively link long-term variation in the marine nitrogen cycle to global sea-level changes and shifts in the dominant locus of denitrification, with sedimentary denitrification and water-column denitrification dominant during greenhouse highstands and icehouse lowstands, respectively, a relationship confirmed by reservoir box modeling. These results also challenge the widely held idea that oceanic anoxic events (OAEs) were associated with elevated rates of water-column denitrification. Rather, the present study shows that globally integrated water-column denitrification rates must have been lower during greenhouse intervals (when OAEs developed) relative to icehouse intervals.    Fig. C2. Comparison of output of fully integrated and simplified N cycle models. At t ≤ 0, the model is at equilibrium and represents our "baseline" scenario (f FIX = 220 Tg a −1 , f DW = 60 Tg a −1 , f DS = 160 Tg a −1 , ε FIX = −2 ‰, ε DS = −2 ‰, ε DW = −20 ‰, ε P = 0 ‰, and F DW = 0.27), which yields a seawater δ 15 N NO − 3 value of +4.9 ‰ (i.e., equivalent to modern seawater nitrate; Sigman et al., 2000). At t = 0, changes in F DW to values ranging from 0 to 0.5 result in evolution of δ 15 N NO − 3 at a rate reflecting the response time of the system (which is closely related to the ∼ 3 kyr residence time of nitrate in seawater; Tyrrell, 1999). Differences in δ 15 N NO − 3 between the fully integrated model (blue curves) and simplified model (red curves) are generally < 0.4 ‰, indicating that the output of the simplified model is robust. We parameterized the model based on the modern marine N budget. The total mass of seawater nitrate is ∼ 8.0 × 10 5 Tg N (Brandes and Devol, 2002). Whether the present-day marine N cycle is in balance is a matter of debate (Codispoti, 1995;Brandes and Devol, 2002). Recent studies have documented strong spatial coupling of cyanobacterial N fixation and water-column denitrification in the modern ocean (Galbraith et al., 2004;Deutsch et al., 2007;Knapp et al., 2008), implying that short-term losses of fixed N are locally compensated. At longer timescales, losses of fixed N to denitrification must be balanced by new N fixation in order to maintain a N : P ratio in global seawater close to that of marine phytoplankton (16 : 1) (Tyrrell, 1999). Estimates of the fluxes associated with cyanobacterial N fixation (f FIX ), sedimentary denitrification (f DS ), and watercolumn denitrification (f DW ) vary widely in older literature, although recent analyses of large data sets are beginning to converge on a consensus range of values (DeVries et al., 2012(DeVries et al., , 2013Eugster and Gruber, 2012;Groβkopf et al., 2012). Estimates of f FIX include 120-140 Tg a −1 (Gruber and Sarmiento, 1997; Galloway et al., 2004;Gruber and Galloway, 2008), 131-134 Tg a −1 (Eugster and Gruber, 2012), and 177 ± 8 Tg a −1 (Groβkopf et al., 2012). Total oceanic denitrification has been estimated at 145-185 Tg a −1 (Gruber and Sarmiento, 1997;Galloway et al., 2004), 120-240 Tg a −1 (DeVries et al., 2013), 230 ± 60 Tg a −1 (DeVries et al., 2012), 230-285 Tg a −1 (Middelburg et al., 1996), 240 Tg a −1 (Gruber and Galloway, 2008), and > 400 Tg a −1 (Codispoti, 2007). The relative importance of sedimentary versus water-column denitrification was not well known in the past (Gruber and Sarmiento, 1997), but recent marine N budgets have provided independent estimates of each flux. Estimates for f DW include 52 ± 13 Tg a −1 (Eugster and Gruber, 2012) and 66 ± 6 Tg a −1 (DeVries et al., 2012), while estimates for f DS range from 93 ± 25 Tg a −1 (Eugster and Gruber, 2012) to 164 ± 60 Tg a −1 (DeVries et al., 2012) and > 300 Tg a −1 (Codispoti, 2007).

Supplementary material related to this article is
The δ 15 N of seawater nitrate in the deep ocean (i.e., the largest reservoir of fixed N) is +4.8 to +5.0 ‰ (Sigman et al., 2000). The δ 15 N of present-day atmospheric N 2 is 0 ‰ (Mariotti, 1984), a value inferred to have been nearly invariant through time (Berner, 2006b). The fractionation associated with cyanobacterial N fixation of atmospheric N 2 (ε FIX ) is estimated to be −1 to −3 ‰ (Macko et al., 1987;Carpenter et al., 1997). Fractionation during water-column denitrification (ε DW ) has a maximum value of ∼ −27( ± 3) ‰ (Gruber and Sarmiento, 1997;Barford et al., 1999;Voss et al., 2001;Murray et al., 2005), although the effective fractionation may be closer to −20 ‰ (Brandes and Devol, 2002). Recent culture studies have suggested that this fractionation might even be as low as −10 to −15 ‰ (Kritee et al., 2012), an idea that we explore in our modeling simulations. We did not parameterize the anammox reaction  separately owing to significant uncertainties concerning the scale of this process and any associated fractionation. While this reaction is a major sink for seawater fixed N, possibly larger than water-column denitrification in some oceanic regions (Mulder et al., 1995;Kuypers et al., 2005), it is thought that field-based estimates of fractionation due to water-column denitrification have incorporated any effects related to anammox (Thamdrup et al., 2006). Denitrification in suboxic marine sediments (ε DS ) typically yields a small net fractionation (∼ −1 to −3 ‰) owing to near-quantitative utilization of porewater nitrate (Lehmann et al., 2004;Galbraith et al., 2008). However, this fractionation can range from ∼ 0 ‰ in organic-rich, reactive sediments to as high as −5 to −7 ‰ in organic-lean, unreactive sediments (Lehmann et al., 2007). An estimate of −0.8 ‰ for the global mean fractionation due to sedimentary denitrification (Kuypers et al., 2005) does not take into account effects associated with the upward diffusive flux of 15 N-enriched ammonium in reactive sediments (Higgins et al., 2012). The fractionation associated with assimilation of seawater fixed N by eukaryotic marine algal (ε P ) can be as large as −5 to −8 ‰ for nitrate (Lehmann et al., 2007) but is more typically −1 to −3 ‰ (Macko et al., 1987;Carpenter et al., 1997). We assumed a net fractionation of 0 ‰ based on complete photosynthetic utilization of seawater nitrate at longer timescales. The fractionation associated with ammonium uptake by marine algae is −10( ± 5) ‰ (Brandes