The stable isotopic compositions of bulk, clay-bound, organic, and compound-specific nitrogen were determined for mid to late Holocene Black Sea sediments from a set of box and gravity cores. The data demonstrate that cyanobacterial N2fixation provided ∼55% of phytoplankton-derived N preserved in the top 1–2 cm of the sediments. Prior to widespread agricultural and industrial development in the catchment, N2fixation was more prominent, providing 70–80% of phytoplankton N. Organic and clay-bound nitrogen fractions record different down-coreδ15N trends that reflect phytoplankton and detrital sources, respectively, and in samples with low organic matter content, the clay-bound fraction comprises up to 38% of bulk nitrogen. Compared with bulk samples, pyropheophytina (Pphe a), which is a chlorophyll a (Chl a) degradation product, provides a more accurate record of changing phytoplankton δ15N values during the Holocene. An examination of the δ15NPphe a values in light of published and new estimates of the isotopic difference between biomass and Chl a suggests that most of the preserved Pphe a was derived from eukaryotic algae, not cyanobacteria. We infer from these data that cyanobacterial biomass is rapidly recycled in the photic zone, with 15N-depleted NH4+ released during heterotrophy and assimilated by other phytoplankton. A conceptual model for N2 fixation in the Black Sea is presented, drawing upon water column nutrient and hydrographic data as well as regional climate variability to explain the proposed temporal variability in N2 fixation.
 In redox-stratified marine basins like the Black Sea, widespread denitrification and/or anaerobic ammonium oxidation (anammox) near the redox boundary (chemocline) lead to combined-N (NO3−, NO2−, and NH4+) deficits that are expected to stimulate bacterial N2 fixation [Montoya and Voss, 2006]. Estimates of N2 fixation vary widely in Black Sea N cycle models, with some ignoring its contribution [Grégoire and Soetaert, 2010; Grégoire et al., 2004], and others holding that it contributes up to 80% of combined-N for primary productivity [McCarthy et al., 2007; Romaniello and Derry, 2010]. N:P nutrient ratios near and below 5:1 suggest that Black Sea subsurface waters may support significant N2 fixation when upwelled (Figure 1) [Fuchsman et al., 2008; Konovalov et al., 2008]. In contrast, nutrient-rich deep waters in the oceans generally conserve the Redfield ratio of 16:1, reflecting remineralization of phytoplankton biomass [Redfield, 1958; Tyrrell, 1999]. In oceanic oxygen-minimum zones, the loss of combined N yields PO43− excess (N:P ratios below 16:1) that stimulates N2 fixation [Deutsch et al., 2007]. N2-fixing cyanobacteria are a critical component of the marine N cycle [Capone et al., 1997; Zehr et al., 2001], maintaining levels of combined N close to the Redfield ratio. Even though algal blooms may take up nutrients with a lower N:P ratio (e.g., 9:1) [Mills and Arrigo, 2010], we expect the low nutrient N:P ratios in the Black Sea to favor N2 fixation. PO43−excess has likely persisted during the mid to late-Holocene in the Black Sea [Arthur and Dean, 1998], so we propose that N2 fixation has been essential for maintaining phytoplankton productivity throughout that time.
 Density- and redox-stratification in the Black Sea produce four distinct layers with different physical and chemical characteristics (Figure 1). The surface mixed layer (SML) from 0 to 40 m is fully oxygenated and contains only trace nutrient concentrations, the result of phytoplankton growth and biomass export to deeper depths during decomposition and remineralization [Codispoti et al., 1991]. The cold intermediate layer (CIL) between ∼40–80 m is characterized by a strong density gradient that forms the lower boundary of seasonal surface water mixing, allowing remineralized nutrients to accumulate at its base. The chemocline is located within the suboxic layer (SOL), a water mass that spans depths between ∼80–100 m in the deep basin [Murray et al., 1991]. Strong chemical gradients in the SOL enhance denitrification and anammox, fundamentally affecting nitrogen cycling in the basin by converting combined N to gaseous N2 and N2O [Jensen et al., 2008; Kuypers et al., 2003; Murray et al., 1995; Ward and Kilpatrick, 1991]. The combined-N deficit is exacerbated in the anoxic waters, where PO43− release from mineral and organic phases [Van Cappellen and Ingall, 1994] outpaces NH4+ release from organic matter (OM) decomposition, leading to PO43− accumulation and low N:P ratios in the anoxic layer (AOL), below ∼100 m [Codispoti et al., 1991; Fuchsman et al., 2008].
 The Black Sea is a brackish marine basin with limited connection to the global ocean through the Bosporus Strait. Several large rivers, notably the Danube, Dnieper, and Don in Eastern Europe flow into the sea leading to a positive water balance and net outflow through the Bosporus [Murray et al., 1991]. Deep-water anoxia developed nearly synchronously about 7.8 ka (kiloannum before present) at water depths below ∼200 m [Arthur et al., 1994; Jones and Gagnon, 1994]. At some time prior to 7.8 ka, probably near 10 ka when the Mediterranean Sea level increased to the Bosporus sill height, Mediterranean waters began flowing into the Black Sea and mixing into the deep water mass. The ensuing density stratification and estuarine circulation pattern trapped nutrients in the deep water, which, through surface-deep mixing, stimulated primary productivity [Arthur and Dean, 1998]. Ultimately, redox-stratification is maintained by relatively high productivity in the surface waters that supports aerobic and anaerobic respiration in the deeper waters [Meyer and Kump, 2008].
 The Holocene Black Sea has long been considered a possible analog for ancient marine basins during oceanic anoxic events (OAEs) [Degens and Stoffers, 1976; Glenn and Arthur, 1985]. Similar to black shales deposited during OAEs, the Black Sea has laminated OM-rich sediments, organic biomarkers, and other chemical indicators for anoxic deep waters [Arthur and Sageman, 1994; Lyons et al., 2009; Sinninghe Damsté et al., 1993; Wilkin and Arthur, 2001]. Due to combined-N losses in redox-stratified oceans during OAEs, N2-fixing cyanobacteria were likely an important component of the marine ecosystem [Kuypers et al., 2004]. Organic biomarker studies of black shales deposited during OAEs reveal high 2-methylhopane indices (2-MeHI) that indicate cyanobacteria may have been abundant in the oceans [Cao et al., 2009; Kuypers et al., 2004; Summons et al., 1999]. Many black shales also have δ15N values near and below 0‰, consistent with N2 fixation [Cao et al., 2009; Junium and Arthur, 2007; Kuypers et al., 2004; Rau et al., 1987]. This interpretation is based on the isotope effect associated with nitrogenase, the key enzyme in N2 fixation, which results in cyanobacterial biomass that is 15N-depleted by about 2‰ relative to N2 [Carpenter et al., 1997; Hoering and Ford, 1960]. Thus, sinking particulate OM δ15N values are low in waters dominated by N2-fixing cyanobacteria, and where bottom waters are anoxic, the lowδ15N signal is preserved in sediments [Sachs and Repeta, 1999]. The combined biomarker and δ15N evidence points toward cyanobacterial N2-fixation being an important process in ancient deep marine anoxic basins. In contrast, most normal marine sedimentaryδ15N values are greater than 4‰, and in some settings like coastal Chile where denitrification is especially prominent, as high as 15‰ [Altabet and Francois, 1994; De Pol-Holz et al., 2009]. NO3− is 15N-enriched by denitrification, anammox, and Rayleigh fractionation associated with uptake, a signal that is acquired by phytoplankton and exported to the underlying sediments.
 Sedimentary N is prone to diagenetic processes that can alter its isotopic composition. Very high Corganic:Ntotalratios in OM-rich marine sediments attest to preferential loss of N relative to C, but isotopic fractionation due to sedimentary ammonification of labile N cannot account for very lowδ15N values [Junium and Arthur, 2007]. Compound-specificδ15N values can be used to determine if δ15Ntot values are overprinted by diagenetic effects [Higgins et al., 2010; Junium, 2010; Kashiyama et al., 2008; Ohkouchi et al., 2006]. These N-isotope studies reportδ15N values of individual chlorins, geoporphyrins, and maleimides that are derived from chloropigments, most commonly chlorophyll a (Chl a) [Chikaraishi et al., 2008; Kashiyama et al., 2007]. Chl a is degraded to many different compounds in Holocene sediments [Keely, 2006], including pyropheophytin a (Pphe a), which previously has been found in abundance in Black Sea surface sediments [Sachs and Repeta, 2000]. δ15N values of chlorins and geoporphyrins, when interpreted in light of modern water column and culture observations, confirm that although δ15Ntotvalues in OM-rich sediments generally preserve a primary phytoplankton signal [Chicarelli et al., 1993; Higgins et al., 2012; Sachs and Repeta, 1999], bulk records across isotope excursions may be attenuated relative to molecular values [Junium, 2010].
 In this study, we examine Holocene Black Sea N isotope signals that potentially reflect changes in phytoplankton nutrient sources. We focus especially on cyanobacterial N2fixation during the mid to late Holocene, when the sea became redox-stratified. We report N stable isotope data on bulk and refractory fractions to help constrain primary and detrital N signatures. Past phytoplanktonδ15N values, as captured by Pphe a, are variable, spanning a 4‰ range during the past 7.8 ka. These data suggest that variability in N cycling was linked to the N:P ratio of upwelled nutrients, the strength of the surface-deep density gradient, and regional climate variability.
 Sediment cores were collected during leg 1 of the R/V Knorr 1988 Black Sea expedition [Murray, 1991]. We analyzed samples from 7 giant gravity cores (GGC) spanning water depths from 205 to 2190 m (Figure 2). GGCs retrieved up to 4 m of sediment, and all GGCs used in this study recovered the Unit II/III boundary, which has been dated to ∼7.8 ka [Arthur and Sageman, 1994; Jones and Gagnon, 1994]. Three cores also retrieved ∼1 m of Unit III sediments, but the bases of these cores are undated. The GGCs were supplemented with 4 box cores (BC) from depths between 184 m and 2164 m. These short cores (∼0.5 m) retrieved the “fluff layer” at the sediment-water interface and the upper layers of Unit I, an interval that is poorly recovered in the longer gravity cores. BC 55 recovered all of Unit I, including the Transition Sapropel, and the upper 10 cm of Unit II. The other three BCs only retrieved Unit I sediments above the Transition Sapropel.
2.2. N and C Stable Isotopes
 Stable N and C isotope analyses were performed in the Penn State Biogeochemical Stable Isotope Lab using a coupled elemental analyzer-isotope ratio mass spectrometer (EA-IRMS) comprised of a Costech ECS 4010 EA and a ThermoFinnigan Deltaplus XP mass spectrometer linked via a ThermoFinnigan Conflo III interface. Stable isotope compositions are presented in standard δnotation. Our analysis of in-house and international standards yields precision better than 0.2‰.δ15N values are presented relative to air and δ13C values relative to Vienna Pee Dee Belemnite.
 Samples for bulk stable isotope analysis were freeze-dried, powdered, and acidified with buffered acetic acid (pH 4) to remove calcium carbonate. After repeated rinsing, the samples were freeze-dried and powdered for bulk organic carbon (Corg) and total nitrogen (Ntot) stable isotope analysis. For a subset of 20 samples, differences between δ15N values for decarbonated and whole samples were less than 0.2‰. Selected samples from GGC 71 were oxidized with 10% hydrogen peroxide at 40°C to remove reactive organic matter, yielding residual refractory carbon (Cref) and nitrogen (Nref) [Junium and Arthur, 2007]. These fractions are analogous to Cresidual and NKOBr [Freudenthal et al., 2001] and Nfix [de Lange, 1992]. Exchangeable NH4+ (Nex) was extracted from GGC 71 samples by suspending 200 mg of non-acidified powder in 10 ml of 2N KCl and shaking for one hour to liberate loosely bound NH4+. The KCl solutions were passed through 0.45-μm membrane filters and collected for NH4+ analysis. To calculate organic N (Norg) content we assume inorganic N is composed solely of Nref and Nex and thus Norg = Ntot – Nref – Nex. In practice, Nex is less than 0.01% of Ntot, so we calculate δ15Norg using equation (1):
2.3. N and C Compositions
 Total and refractory fractions were analyzed for weight percent C and N content by EA, concurrently with isotope ratio determinations. N2 and CO2 peak areas (Isodat v1.42) were converted to weight percent composition using response factors generated from standards of known composition (sucrose and caffeine). All weight percent concentrations are presented for decarbonated samples, eliminating variability caused by high calcite or aragonite concentrations in some samples. Nex concentrations were determined spectrophotometrically using reagents for the Nitrogen, Ammonia Salicylate Method (HACH Method 8155; HACH Company, Loveland, Colorado) which was adapted from a method developed by Reardon et al. . The resulting absorbances were converted to NH4+ concentrations using a response factor (R2 = 0.99) determined for NH4Cl standard solutions.
2.4. Pigment Isotope Measurements
 Cores for pigment extraction were continuously refrigerated at +4°C (GGCs), or frozen at −20°C (BCs) at the Woods Hole Oceanographic Institution since collection in 1988. Thermal pigment degradation and alteration of the pigment δ15N signal during storage is unlikely, as the frozen and refrigerated cores were maintained colder than in situ sediments under 8.9°C Black Sea bottom water [Murray et al., 1991]. The refrigerated cores were split previously, and orange Fe-oxides were observed, so we scraped off the exposed orange/tan surface to reveal dark laminations prior to taking samples. Sedimentary pigment distributions for BC 55, GGC 59, and GGC 69 were analyzed using published solvent extraction and reversed-phase high performance liquid chromatography (RP-HPLC) methods on an Agilent 1200 series HPLC system [Airs et al., 2001]. Separations were achieved on two consecutive Waters Spherisorb ODS2 columns (3 μm, 150 mm × 4.6 mm I.D.) protected by a pre-column (10 mm × 5 mm I.D.) and a Phenomenex Security Guard C18 guard column. The mobile phase consisted of a gradient of 0.5 M ammonium acetate, methanol, acetonitrile, and ethyl acetate [Airs et al., 2001]. Multistage mass spectrometry (MSn) was performed on an Agilent 6310 LC/MS in positive ion mode using an APCI source and the following tuning settings: 4000 nA corona current, 60 psi nebulizer pressure, 5 l min−1 drying gas flow, 350°C drying temperature, and 400°C vaporizer temperature. Compounds were identified by comparison with published elution times, mass spectra, and MSn fragmentation patterns [Airs et al., 2001].
 Pphe a peak collection intervals for stable N isotope analysis were determined based on light absorbance at 665 nm, and entire peaks were collected to account for isotopic heterogeneity across single peaks [Bidigare et al., 1991]. The collected Pphe a fractions were dried under N2, dissolved in hexane:acetone (92:8), and injected into an isocratic flow of the same composition onto a 5-μm, 250 mm × 4.6 mm Agilent SIL silica gel column [Sachs and Repeta, 2000]. Pigments purified using this two-dimensional HPLC approach have been shown to be suitably pure for isotope analysis [Ohkouchi et al., 2005; Sachs and Repeta, 2000]. The purified Pphe a samples were dissolved in acetone (40 μl), transferred to solvent-rinsed (methanol, dichloromethane, and hexane) smooth-sided tin capsules (Costech 2.9 × 6 mm), and dried at 40°C before nano-EA analysis [Polissar et al., 2009]. Eluent blanks collected on HPLC and analyzed for N isotopes were indistinguishable from blank tin capsules. We confirmed sample purity based on molar C:N ratios and by comparing the quantity of C and N analyzed by nano-EA with sample amounts calculated based on visible light absorbance at 665 nm.
3.1. Organic Carbon
 Gravity cores from all depths record (1) an abrupt increase in weight percent Corg at the Unit II/III boundary, (2) maximum Corg content in Unit IIb1, and (3) a gradual decrease in Corg content through Units IIa and I (Figure 3 and Figure S1 in the auxiliary material). These trends have been reported previously [e.g., Arthur and Dean, 1998; Calvert and Fontugne, 1987], but we provide a basin-wide record of decarbonated Corg data that depicts the balance between organic matter and terrigenous detrital material without the influence of variable inputs of calcite and aragonite. We also present a comprehensive record of δ13Corg values for Holocene Black Sea sediments (Figures 3and S2) that confirms basin-wide variability in carbon cycling [Arthur and Dean, 1998]. The sedimentary unit designations are well established, and are based on color changes that result from variable inorganic and organic carbon content [Arthur et al., 1994; Hay et al., 1991; Ross and Degens, 1974]. Sediments at the top of Unit III generally have higher δ13Corg values (−26 to −24‰) in the deep basin, and lower values (−29 to −25‰) in shallow water cores. Above the Unit II/III transition, all of the cores record nearly identical δ13Corgtrends, suggesting that most organic matter was derived from basin-wide plankton growth rather than local blooms and terrigenous organic matter.
3.2. Sedimentary Nitrogen
δ15Ntot values follow consistent temporal trends in all cores in this study (Figures 3 and S3). Unit III δ15Ntot values are between 3 and 6‰, and decrease to near 1‰ in Unit IIb2, coincident with the abrupt increase in Corg content. In Unit IIb1, δ15Ntot values increase to ∼4‰ before dropping stepwise to minimum values between 0 and 1‰ in Unit IIa. Unit I δ15N values are similarly low except for a ∼1‰ increase in the Transition Sapropel and a 2–3‰ increase in the surface sediments. Corg:Ntot ratios range between 10:1 and 20:1, with maximum values near the base of Unit IIb and a trend toward lower values in Unit IIa (Figure S4). These values fall between the Redfield ratio of 6.7:1 and the C:N ratios of ancient black shales that are commonly between 30:1 and 50:1 [Junium and Arthur, 2007].
 Analysis of GGC 71 sediments reveals that Nex is a minor component of inorganic N and negligible in its effect on δ15Ntot values, a conclusion reached in other similar studies [de Lange, 1992; Freudenthal et al., 2001]. We measured only trace Nexconcentrations between 0.7 ppm and 1.2 ppm in GGC 71 samples, without a consistent trend down-core. Operationally defined as all N resistant to oxidation by 10% H2O2, Nref accounts for up to 38% of Ntotin samples from GGC 71. We measured little down-core variability in Nref, in contrast with Ntot and Norg which are both significantly higher in Unit IIb, but there is an increase the Nref:Al with depth (Figure 4), suggesting an increase in the mineral phase N content. The stable isotopic composition of Nrefis relatively constant down-core, with an averageδ15N value of 3.1 ± 0.4‰ in Units I and II, compared with lower and more variable values for δ15Norg (2.3 ± 0.9‰).
3.3. Pigment Nitrogen
 Chl a is the most abundant tetrapyrrole pigment in Black Sea surface waters and its degradation products are abundant in the sediments. We detected 14 compounds that derive from Chl a, including Pphe a, pheophytin a, hydroxypheophytin a, and 11 different pyropheophorbide a steryl esters [King and Repeta, 1994]. Pphe a is the most abundant Chl a derivative in all Unit I and II samples, so we chose this compound for stable isotope analysis. We did not detect Pphe a in Unit III sediments, only trace concentrations of more degraded chlorins that we did not attempt to analyze for N isotopes. Pphe a δ15N values range between −4.8‰ and −0.6‰ (Figure 5); the 4.2‰ range of values is similar to the range of δ15Ntot values, and both measurements follow similar stratigraphic trends.
4.1. Nitrogen in OM-Lean Sediments
4.1.1. Inorganic N and δ15Ntot Values
 Stratigraphic trends in δ15Ntot values can reflect variability in primary and detrital N sources [Freudenthal et al., 2001]. While inorganic detrital N may be ignored in OM-rich marine sediments, it can comprise a significant proportion of Ntotin OM-lean samples [de Lange, 1992]. Detrital clay deposition is focused along the Black Sea margins [Ross and Degens, 1974], diluting OM concentrations nearshore [Arthur and Dean, 1998]. NH4+ can replace K+ irreversibly in interlayer positions of illite [Young and Aldag, 1982], the most abundant clay mineral in Black Sea sediments [Müller and Stoffers, 1974], suggesting that it may be important to consider inorganic N phases when examining N isotopes of marginal sediments. GGC 71 δ15Nref values are consistently between 3 and 4‰ in Units I and IIa, but values as low as 1.9‰ occur in Unit IIb2 samples that also have low total and organic δ15N values (Figure 4). Low Cref:Nref and high Nref:Al ratios in Unit IIb suggest that additional N was incorporated into the refractory fraction of these older OM-rich sediments, possibly as NH4+ derived from OM degradation. Incorporation of NH4+ into nonexchangeable clay interlayer positions is consistent with our finding that Nexdoes not increase down-core. While it is possible that lowδ15Nref values in Unit IIb2 are the result of a change in detrital clay source, the presence of an in situ source for relatively 15N-depleted NH4+suggests post-depositional substitution is more likely. In a Cretaceous black shale,δ15Ntot and δ15Nref values are indistinguishable [Junium and Arthur, 2007], further evidence of pore water NH4+incorporation into the mineral phase in OM-rich sediments. The apparent substitution could be a function of burial duration or pore water NH4+ concentration, which increases with depth in Black Sea sediments [Manheim and Chan, 1974].
 The effects of Nref on δ15Ntot values are expected to be most pronounced in samples with low OM content and δ15Norg values significantly higher or lower than δ15Nref, for example in GGC 71 samples that have Ntot contents less than 0.2% by weight. Calculated δ15Norg values illustrate the influence of Nref on measured δ15Ntot data for GGC 71 (Figure 4). The difference between δ15Norg and δ15Ntotvalues is small in OM-rich samples but as large as 1‰ in OM-lean samples from Unit IIa.δ15Ntotvalues remain useful in this study, however, as most samples are OM-rich, and in relatively OM-poor samples (which still have >2% Corg) the δ15Ntot values record the significant stratigraphic trends; though they do not capture the full magnitude of the δ15Norg excursions. In other marine or lacustrine systems that typically have less than 2% Corg, detrital N should be considered when interpreting δ15Ntot values.
4.1.2. Inorganic N and Corg:Ntot
 Inorganic N inputs affect Corg:Ntot ratios as well as δ15Ntot values [Calvert, 2004]. The average inorganic N content in GGC 71 is reflected in the positive intercept of the linear regression of Ntot versus Corg assuming relatively stable inputs of detrital N and variable Norg (Figure 6). As a result, stratigraphic variability in Corg:Ntot can be a function of changes in the relative amounts of organic and inorganic N rather than reflective of changes in OM sources or diagenetic effects [Calvert, 2004]. Corg and Ntot of GGC 71 samples are strongly correlated, with a linear regression intercept of 0.057% N, representing the predicted average amount of inorganic N in this core (Figure 6). The average measured Nref content of 0.063 ± 0.010% is similar to the intercept value, so Nref does indeed reflect the inorganic N fraction in these samples. Corg:Norg values are relatively constant in Units I and II, with an average value of 17.4 ± 1.5 (mole:mole) and no apparent stratigraphic trend (Figure 4). Corg:Ntot, on the other hand, trends from maximum values near 17 in Unit IIb2 to a minimum of 11 in Unit I of GGC 71. Thus, the upward, decreasing trend in Corg:Ntot in this core may be explained by the decreasing Norg content superimposed over nearly constant inorganic N concentrations.
4.2. Nitrogen Fixation
 There has been significant variability in N cycling during the deposition of OM-rich sediments in the Black Sea, as evidenced by the range ofδ15Ntot values between +4.5‰ and −0.1‰ (Figure 3). This is in contrast with, for example, Pleistocene Mediterranean sapropels, which are consistently 15N-depleted reflecting cyanobacterial N2 fixation [Meyers and Bernasconi, 2005]. In the Mediterranean Sea, high sedimentary δ15N values between 3‰ and 6‰ are only found in OM-poor layers deposited under a fully oxygenated water column with15N-enriched NO3− as the primary N source [Higgins et al., 2010]. Between 7.8 and 7.4 ka, at the onset of Black Sea sapropel deposition, average sedimentary δ15N values dropped from 3.3‰ to 1.0‰ (Figure 3). Water column anoxia had been established in the whole basin below ∼100 m, supporting widespread water column anammox and denitrification that led to combined-N deficits and N2 fixation. Blumenberg et al.  interpreted the presence of bacteriohopanepentol, in conjunction with generally low δ15Ntot values, as evidence of abundant cyanobacteria throughout the deposition of Units I and II. Our model agrees with their analysis, though we add that the intervals with higher δ15N values (7.3–5.0 ka, 1.9–1.5 ka, and ∼0.4 to 0.0 ka) record decreases in the relative inputs from cyanobacterial N2 fixation.
 N2 fixation has been detected in the Black Sea [McCarthy et al., 2007] and may be carried out by unidentified populations of small unicellular cyanobacteria or by other non-photosynthetic bacteria or archaea [Zehr et al., 2001, 2006]. This is in contrast to other settings like the eutrophic Baltic Sea and oligotrophic ocean gyres where prominent surface blooms of colonial diazotrophic cyanobacteria develop [Castenholz and Garcia-Pichel, 2000; Mohlin and Wulff, 2009]. Synechococcus are relatively abundant in the pelagic waters of the modern Black Sea [Uysal, 2006], but this genus does not include diazotrophs. Blumenberg et al.  suggest that Nostoc or Gloeocapsa may produce bacteriohopanepentol, and other cyanobacteria, notably Lyngbya, Oscillatoria, and Microcystis, have been identified in the littoral Black Sea [Aysel et al., 2004; Stoica and Herndl, 2007]. These genera include planktonic diazotrophs that may have proliferated at times during the Holocene.
 Phytoplankton and sinking particulate δ15N values range between 1 and 6‰ in the Black Sea [Coban-Yildiz et al., 2006; Fry et al., 1991; Fuchsman et al., 2008], so surface sediment δ15Ntot values (2.9 ± 0.5‰) are consistent with a phytoplankton origin, probably integrating seasonal, geographical, and interannual variability. Deep nutrients have δ15N values near 8‰ (NO3−) in the CIL and 6–8‰ (NH4+) at the top of the AOL, 15N-enriched by denitrification and anammox near the chemocline [Fuchsman et al., 2008; Velinsky et al., 1991], and riverine NO3− inputs most likely reflect the high δ15N values (>8‰) of Eastern European rivers [Johannsen et al., 2008]. As combined-N is assimilated completely in the SML, the relatively low phytoplanktonδ15N values suggest the importance of N2 fixation. Based on isotope balance (Figure 7), we estimate that ∼55% of phytoplankton N has an N2 fixation origin, assuming δ15Nplankton and δ15Nnitrate values of 3‰ and 8‰, respectively. However, direct measurements of uptake rates indicate that N2 fixation may comprise just 4.6% of phytoplankton productivity in the Black Sea, with recycled NH4+ and upwelled NO3− accounting for most of the phytoplankton N uptake [McCarthy et al., 2007]. In agreement, we propose that eukaryotic algae acquire the isotopic N2 fixation signal via NH4+ recycling in the SML, and the signal is transferred to the sediments via sinking particulate organic matter.
4.3. Compound-Specific N Isotopes
δ15Ntot values generally approximate the isotopic composition of sedimentary OM, but they also include inorganic N signatures. Variable accumulation of terrigenous OM indicated by δ13Corg values and hydrogen indices during sapropel deposition [Arthur and Dean, 1998; Hay et al., 1991] could also affect δ15Ntot trends. Freeman et al. , through an analysis of the carbon isotopic compositions of hydrocarbon biomarkers, demonstrated that low δ13Corg values in the Black Sea may also be derived from phytoplankton or bacteria growing on high concentrations of 13C-depleted CO2(aq). Thus, the bulk N isotopic record may reflect marine phytoplankton balanced by bacterial and/or terrestrial organic matter inputs.
 The isotopic difference (Δ15Ncell-Chla = δ15Ncell – δ15NChl a) between whole algae and Chl a is relatively constant for cultured diatoms, haptophytes, and green and brown algae. Sachs et al.  reported that δ15NChl a values are 5.08 ± 0.87‰ lower than δ15N values of the whole algal cells, a relationship that also holds for natural phytoplankton assemblages and OM-rich sediments [Higgins et al., 2010; Junium, 2010; Ohkouchi et al., 2006; Sachs and Repeta, 1999]. Published Δ15Ncell-Chla values for cyanobacteria are variable, ranging between +5.7‰ and −15.1‰ [Beaumont et al., 2000; Higgins et al., 2011; Katase and Wada, 1990; Sachs et al., 1999]. Δ15Ncell-Chla values are essentially independent of nutrient source (NO3−, NH4+, or N2) and fall into two distinct ranges, −9.8 ± 1.8‰ for freshwater cyanobacteria and −0.9 ± 1.3‰ for marine strains [Higgins et al., 2011]. We measured δ15N values of biomass and Chl a produced by cyanobacterial cultures obtained from Carolina Biological Supply Company (Lyngbya sp., Tolypothrix distorta, and Gloeocapsasp.), grown on B-HEPES medium with NO3− as the N substrate. These cyanobacteria yielded Δ15Ncell-Chla values between −8.7‰ and −10.4‰ (Table 1), consistent with other freshwater strains [Higgins et al., 2011]. We also determined that Δ15Ntot-Chla = −10.1‰ for a natural marine intertidal Lyngbyamat sample collected from Assawoman Island on the Atlantic coast of Virginia, USA. Thus, at least one strain of mat-forming, filamentous, marine/hypersaline cyanobacteria fractionates N isotopes similarly to freshwater cyanobacteria, suggesting that there is more heterogeneity in N isotopic fractionation among marine cyanobacteria than previously observed. The Δ15Ntot-Chla values demonstrate that cyanobacteria and eukaryotic algae partition N isotopes differently during chlorophyll synthesis. Following standard models for intracellular isotopic fractionation [Hayes, 2001], variations in the biosynthetic pathway of tetrapyrrole compounds, possibly with respect to the partitioning of protoporhyrin IX to Mg-protoporphyin IX (chlorophyll pathway) and heme b (heme pathway, including phycobilins in cyanobacteria), may explain the variability in Δ15Ncell-Chla, though we have not explored this scenario in depth.
Table 1. The δ15Nbiomass and δ15NChl a Values of Cyanobacteria
 Down-core trends inδ15NPphe a values generally parallel δ15Ntotvalues throughout the OM-rich Black Sea sediments (Figure 5). These sediments yield an average Δ15Ntot-Pphea value of 5.1 ± 0.9‰ for 13 samples from Units I and II, the same value reported for cultured algal eukaryotes [Sachs et al., 1999]. The one significant outlier is from the Transition Sapropel, an interval that we discuss further in section 4.5. The low Δ15Ntot-Pphea value of 2.8‰ may indicate a shift to a dominant eukaryotic algae with an irregular Δ15Ntot-Chla, a new non-algal source of15N-depleted OM, or increased cyanobacterial productivity or preservation of cyanobacterial pigment derivatives. Thus,δ15Ntot values in this study appear to be valid indicators of average phytoplankton δ15N values at the time of deposition, and degradation products of cyanobacterial Chl a are not preserved extensively in the sediments, as has been also suggested for black shales from OAE2 [Higgins et al., 2011, 2012]. This conclusion assumes that N2 fixation in the Black Sea is by cyanobacteria and that strains in the Black Sea partition N isotopes similarly to the few species in the current data set, as Δ15Ntot-Chla has not been determined for cyanobacteria in the Black Sea. Biomass δ15N values for N2-fixing cyanobacteria are near −1‰ so, using average Δ15Ncell-Chla values of −10‰ or −1‰, we predict that N2-fixing cyanobacteria would produce Chla with a δ15N value near +9‰ or 0‰. In contrast, δ15NPphe a values in Units IIb2 and IIa, intervals we associate with increased N2 fixation, range between −3‰ and −5‰. This disparity provides further evidence that cyanobacterial pigments were recycled in the SML, though some biomass reached the sediments as evidenced by cyanobacterial biomarkers [Blumenberg et al., 2009; Sinninghe Damsté et al., 1993].
 Nitrogen-containing molecules like Chlamay be preferentially scavenged in the N-limiting surface waters, partially accounting for the elevated C:N ratio in the sediments (Figure 8). One biogeochemical model of N cycling in the Black Sea predicts that more than 95% of sinking particulate organic N is recycled in the upper 100 m of the water column [Grégoire and Beckers, 2004]. Konovalov et al.  proposed a similar scenario for recycling cyanobacterial biomass in the SOL. Their model draws on a different source for N2-fixing biomass, one proposed to be deeper in the water column; nonetheless they explain that the biomass is rapidly recycled to release15N-depleted NH4+. As most phytoplankton biomass is recycled in the upper water column, a short-duration cyanobacterial bloom could influence theδ15N composition of the whole phytoplankton community, even if most of the Chl a degradation products in the sediments were produced by eukaryotic algae.
4.4. Surface Mixed Layer N:P Ratios
 The average Unit II Corg:Ntot ratio of 18:1 is higher than the Redfield ratio (Figure 8). The C:N ratio of Black Sea phytoplankton is similar to the Redfield ratio [McCarthy et al., 2007], so deviation from this value in the sediments suggests preferential N loss via ammonification of sinking and sedimentary OM. PO43− is released from sinking organic matter and MnO2 and FeOOH mineral phases at the top of the AOL [Murray et al., 1995]. PO43− liberation from sinking particulate matter reduces its burial efficiency, reflected in sedimentary Corg:Pexcess ratios that average close to 1000:1, much higher than the Redfield ratio (Figure 8) [Arthur and Dean, 1998]. The C:N and C:P data can be combined to generate a Unit II sedimentary Norg:Pexcessratio of 57:1, which indicates that organic N was buried ∼3.5 times more efficiently than P throughout the mid to late Holocene, when the Black Sea was continuously redox-stratified. It follows that combined-N deficits in the water column have persisted and N2-fixing organisms have probably been an important component of the ecosystem throughout the past 7.8 ka.
 Although average phytoplankton N:P uptake ratios are close to 16:1 in the oceans, this ratio can vary significantly depending on phytoplankton phylogeny and growth conditions [Geider and La Roche, 2002]. Algal N:P uptake ratios can be significantly lower than the Redfield ratio, particularly under nutrient replete conditions with PO43− available in excess of NO3−. Cyanobacterial N:P ratios have not been studied extensively, and the N requirements of phycobilisomes cause cyanobacterial N:P ratios to be relatively high, especially while fixing N2 in oligotrophic settings [Geider and La Roche, 2002]. Mills and Arrigo  employed an oceanic N cycle model with N:P uptake ratios of 9:1 for eukaryotic algal blooms, 25:1 for unicellular cyanobacteria, and 50:1 for N2-fixing cyanobacteria. In our calculations we adopt the general approach ofTyrrell , that PO43− limits total phytoplankton productivity and that eukaryotic algae consume PO43−until the supply of combined-N is depleted. The remaining PO43− is assimilated by N2-fixing cyanobacteria. We assume that the PO43− influx to the SML comes from the CIL, the maximum mixing depth of surface waters [Gregg and Yakushev, 2005; Murray et al., 1991]. This nutrient pool has an N:P ratio near 5:1 [Fuchsman et al., 2008], so we expect upwelling CIL waters to stimulate significant cyanobacterial N2 fixation in the SML.
 The proportions of upwelled PO43− taken up by eukaryotic algae and cyanobacteria are represented by the following equations:
where Palg is the fraction of PO43− assimilated by eukaryotic algae and Pfix by N2-fixing cyanobacteria. If we assume a constant N:P uptake ratio of 16:1, 30% of the upwelled CIL PO43− would be taken up by eukaryotic algae and 70% by N2-fixing cyanobacteria. Usingδnitrate = 8‰, we calculate a δplankton value of 1.6‰ (Figure 7), which is about 1.3‰ lower than core top values. This approach implies that (N:P)CIL would need to be higher than the maximum observation of 5:1 [Fuchsman et al., 2008]. Equations (2) and (3) can only be used to estimate the relative contributions of combined N and N2 fixation if we assume a constant N:P uptake ratio. Blooms of diatoms, haptophytes, and dinoflagellates have been observed in late winter and spring in the Black Sea, so it is possible that these rapidly growing algae have relatively low uptake ratios.
Equations (4) and (5) were derived to calculate the fractions of phytoplankton N produced by algae and N2-fixing cyanobacteria with different N:P uptake ratios.
 These N fraction equations can be substituted into the isotope balance relationship (Figure 7), which can be reorganized to relate the N:P uptake ratio of N2-fixing cyanobacteria (N:P)fix to the N:P ratios of upwelled water, Pfix (equation (3)), and the δ15N values of NO3−, plankton, and N2-fixing biomass (equation (6)):
 Using observed δ15N parameters (δnitrate = 8, δfix = −1, and δplankton = 3) and assuming a eukaryotic algal N:P uptake ratio of 9:1, we calculate that (N:P)CIL would need to be 7.3:1 for an N2-fixing cyanobacterial N:P uptake ratio of 50:1 (Figure 9a). The 50:1 ratio was determined for the filamentous marine cyanobacterium Trichodesmium [Geider and La Roche, 2002]; for unicellular N2-fixing cyanobacteria an N:P uptake ratio of 25:1 may be more applicable, resulting in a calculated (N:P)CIL ratio of 6.1:1. This value is closer to observations, but there is no evidence for CIL N:P ratios higher than 5:1. δfix or δnitrate values of +0.7‰ or +12‰, respectively, would be required to force the model to an (N:P)CIL ratio of 5:1. In the past, increased denitrification or anammox may have caused greater 15N enrichment of nitrate toward the +12‰ value, but the resulting decrease in (N:P)CIL would diminish its effect on δplankton by stimulating increased N2 fixation, requiring even higher δnitrate values. Alternatively, the higher N:P requirement of this model may indicate that combined N from riverine sources also contributes directly to phytoplankton productivity in the offshore SML. Recent calculations of anthropogenic riverine nutrient fluxes to the Black Sea propose an N:P ratio of 40:1 between 1978 and 1988, the decade before the surface sediments were collected [Ludwig et al., 2009]. Only 5% of the combined-N flux to the offshore SML would need to come from riverine sources to increase the N:P influx ratio to 6.6:1. Thus, in the modern sea it appears that a component of offshore phytoplankton productivity is supported by riverine nutrients, and during the Holocene this component was diminished.
4.5. Model for Sedimentary N Isotope Distributions
 Most of the mid to late Holocene was characterized by δ15Nplankton values near 1‰, indicating that N2 fixation was more prevalent than in the modern sea, probably due to lower anthropogenic nutrient inputs from rivers and weaker density stratification allowing deeper mixing, resulting in N:P influx ratios between 3.5:1 and 5.5:1 (Figure 9b). This scenario provides a strong lever to induce N2 fixation, as the deeper waters near the base of the SOL contain PO43−in excess of combined-N concentrations (Figure 1) [Codispoti et al., 1991; Fuchsman et al., 2008]. We refer to growth on nutrients derived from deeper mixing as “mode-1,” and it is the dominant mode of the mid to late Holocene. We present an additional simulation (Figure 9c) that uses a lower δnitrate value of 6‰ to account for a potential decrease due to remineralization of additional cyanobacterial biomass or weakened redox stratification and decreased anammox and denitrification. The lower δnitrate value would increase the predicted N:P influx ratio to between 4.5:1 and 6.2:1. Using δnitrate values between 6‰ and 8‰, we predict that ffixis between 0.7 and 0.8 for mode-1, approximately 30–40% higher than in the modern sea (Figure 7). The depth of surface-deep mixing in the modern Black Sea is limited by the CIL, and the strong density gradient does not permit the upward mixing of SOL waters into the SML [Oguz, 2002]. Thus, in the “mode-2” modern sea, seasonal mixing only draws on the nutrient pool in the CIL (Figure 1), a scenario that also applies to the Transition Sapropel and Unit IIb1 and results in biomass with higher δ15N values (Table 2).
Table 2. Summary of Properties Related to Mode-1 and Mode-2
Calculated from δ15NPphe a, assuming Δ15Nbiomass-Pphea = 5‰.
 The density structure of the Black Sea allows nutrients to accumulate below the CIL by limiting the depth of seasonal mixing and the thickness of the SML. Dense CIL waters are generated in the winter and may have two sources, cold surface water (near 0°C) from the northwest shelf and cold surface water (6–7°C) in the gyres [Oguz, 2002; Oguz and Besiktepe, 1999]. Vertical density stratification in the CIL is very strong, as indicated by the steep salinity gradient and Brunt-Vaisala frequency calculations [Murray et al., 1991]. Wintertime observations from 2003 showed that a cold weather pattern cooled SML waters in the gyres to the point that they mixed deeply into the CIL [Gregg and Yakushev, 2005]. This mixing would have delivered nutrient-rich waters from the CIL to the SML. Thus, upwelling of nutrient-rich CIL waters and OM production by mode-2 may depend in part on strong seasonality. The addition of cold water to the CIL also maintains a strong density gradient, which ultimately inhibits the extension of deeper mixing below the chemocline and may increase anammox and denitrification.
 Mode-2 most likely dominated during Unit IIb1 deposition from 6.9 to 5.1 ka, coincident with the 6 ka “climate optimum” when wet and warm conditions dominated in Northern Europe [Prentice et al., 1996]. High Caspian Sea levels, pollen records from Romania and the Crimean Peninsula, and the interruption of scytonemin deposition in the Black Sea illustrate that a warm and wet climate regime was in place in the Black Sea region at that time (Figure 3) [Chepalyga, 1985; Cordova and Lehman, 2005; Feurdean et al., 2008; Fulton et al., 2012]. High influx of fresh river water to the SML maintained the salinity gradient and ultimately strengthened stratification, likely inhibiting upwelling of SOL waters. Isorenieratane and bacteriochlorophyll e concentrations are high in Unit IIb1 [Fulton et al., 2012; Sinninghe Damsté et al., 1993; Wakeham et al., 1995] indicating that the chemocline was likely shallower than 100 m during Unit IIb1 deposition, further evidence for stable density stratification.
 Unit IIb2(7.8–6.9 ka) contains evidence for deep mixing (mode-1) that explains its lowδ15N values. High hydrogen indices and low δ13Corg values [Arthur and Dean, 1998] suggest that most of the preserved biomass is of marine origin and that plankton were assimilating DIC with especially low δ13C values and/or high concentrations of CO2(aq) [Freeman et al., 1990; Freeman et al., 1994]. In the Black Sea, these characteristics are found in waters near and below the chemocline [Fry et al., 1991]. During Unit IIb2 deposition, SML δ13CDIC values may have been especially low and CO2(aq) concentrations may have been high because of anomalously deep mixing. Seekreide deposition at this time resulted from the precipitation of aragonite crystals in the water column [Degens and Stoffers, 1980]. Seekreide deposits point toward breakdown of stratification and mixing of surface and deep waters. Thus, density stratification was relatively unstable during the deposition of Unit IIb2, and the transition to Unit IIb1 resulted from increasingly stable density stratification that interrupted seekreide formation. As PO43− concentrations were still relatively low at that time [Arthur and Dean, 1998], intense mixing of surface and deep waters helped stimulate productivity and sapropel formation.
 Unit IIa, deposited from 5.1 to 2.1 ka, accumulated during a long interval of mode-1 productivity. Unit IIa lacks isorenieratane and bacteriochlorophylle, providing evidence for a deeper chemocline during that time [Fulton et al., 2012; Sinninghe Damsté et al., 1993], and additional evidence from GGC 48 sedimentation patterns, pyrite framboid size distributions, and biomarkers also point toward a deeper chemocline [Huang et al., 2000; Wilkin and Arthur, 2001]. Alkenone hydrogen isotope values and dinoflagellate distributions suggest the Black Sea may have been characterized by relatively high surface water salinity at times during Unit IIa deposition [van der Meer et al., 2008]. Low Caspian sea level suggests that the regional climate was dry [Chepalyga, 1985], and abundant desert-soil-crust scytonemin in Black Sea sediments along with Crimean and Romanian pollen records provide further evidence for relatively dry conditions in the Black Sea catchment [Cordova and Lehman, 2005; Feurdean et al., 2008; Fulton et al., 2012]. The Subboreal Phase (5–2.5 ka) records a shift toward lower Northern Hemisphere summer insolation and thus cooler temperatures [Wanner et al., 2008]. Taken together, these data demonstrate that though bottom waters remained anoxic, the surface-deep salinity gradient was weakened by decreased freshwater influx, allowing deeper mixing of oxygenated surface waters, eroding the chemocline, and upwelling nutrient-rich, low N:P waters from the SOL/AOL.
 The Subatlantic Phase (2.5 ka-present) in Europe is characterized by a trend toward warm and wet conditions that approach Atlantic Phase (7.5–5 ka) climate conditions. Unit I sediments deposited from 2.1 to 0.0 ka reflect this transition, with a mix of mode-1 and mode-2 conditions. Mode-1 appears to dominate most of Unit I, with the exception of the most recent ∼0.4 ka and the Transition Sapropel deposited from 1.9 to 1.5 ka. The Transition Sapropel was deposited under warm and wet conditions (Figure 3) that resulted in increased freshwater and terrigenous OM influx [Hay et al., 1991]. Carbon and oxygen isotope data from a stalagmite in Romania, the Crimean pollen record, and the termination of scytonemin depositon in the Black sea provide evidence for this warm and wet “optimum” in the Black Sea region [Onac et al., 2002; Cordova and Lehman, 2005; Fulton et al., 2012]. There is not a clear climate signal that explains the increasing δ15N values at the top of Unit I, and they most likely reflect the increased influence of anthropogenic activity on Black Sea nutrient dynamics. Land-use changes are well documented in the watershed, as agricultural and industrial practices have led to the eutrophication of rivers, including the Danube [Ludwig et al., 2009]. An increasing N influx of over the past several hundred years could account for a decrease in N2 fixation and increase in phytoplankton δ15N values.
 The above model accounts for changing biomass δ15N values and predicts that phytoplankton biomass produced during mode-1 (e.g., Unit IIa) deposition would have relatively lowδ15N values. Conversely, the 15N-enriched values in the Transition Sapropel and IIb1suggest increased mode-2 productivity. Though N2fixation is supported by mode-2 conditions, it is not expected to be as prevalent because the upwelled waters have higher N:P ratio. The modern Black Sea has many of the characteristics of mode-2, with relatively high core topδ15Ntot values, relatively low levels of N2-fixation in the water column, a strong salinity gradient, and a shallow chemocline. Modern conditions, however, are strongly influenced by anthropogenic eutrophication and therefore are not directly applicable to nutrient cycling in the Holocene Black Sea.
 Nitrogen in deep-basin Black Sea sediments derives primarily from phytoplankton biomass. Marginal sediments, on the other hand, receive a greater influx of clay minerals than deep basin sediments, and total N from GGC 71 contains up to 38% inorganic N. Accounting for inorganic N results in a difference of up to ∼1‰ betweenδ15Ntot and δ15Norg, though δ15Ntotvalues record the same general trends in all cores, regardless of clay content. Compound-specificδ15N values of Pphe aconfirm that the down-core trends result from temporal variability in theδ15N values of phytoplankton and δ15NPphe a and δ15Ntotvalues follow similar trends, with a 5.1‰ offset characteristic of algae. This sedimentary analysis demonstrates that nitrogen cycling in the Black Sea is controlled by basin-wide processes which produce the same isotopic signatures at all locations.
 Regional climate was the ultimate driver of variations in sedimentary δ15N values in the Black Sea. The cool and dry Subboreal climate regime weakened the salinity gradient in the Black Sea, allowing for deeper mixing of the surface waters and upwelling of PO43−-rich deep waters. These conditions persisted to a lesser degree in the Subatlatic Phase, and diazotrophic cyanobacterial growth was favored by combined-N deficits. Such was the case during most of the past 5 ka when Units IIa and I were deposited and average sedimentaryδ15N values are near 1‰. A warm and wet climate, on the other hand, strengthens the salinity gradient, inhibiting deep mixing and limiting N2-fixation. Such were the conditions during the 6 ka “climate optimum” when Unit IIb1 was deposited with δ15N values up to 4.5‰, suggesting that N2 fixation was less prevalent. Thus, seasonal mixing depth determines nutrient dynamics in the surface waters, with variability in the productivity of N2-fixing cyanobacteria a primary control on phytoplanktonδ15N values.
 The authors thank Jim Broda and the Woods Hole Oceanographic Institution for access to core samples. We acknowledge Pratigya Polissar and Dennis Walizer for technical assistance during the analysis of stable isotopes, Geneviève Elsworth for sample processing, and Christopher Junium for many insightful discussions. We also acknowledge the careful reviews of Meytal Higgins and an anonymous reviewer that greatly improved this manuscript. The National Science Foundation funded this research with a grant (NSF EAR-0525464) to M.A. and K.F., and graduate support for J.F. was granted via the Penn State Biogeochemical Research Initiative for Education (NSF DGE-9972759).