Using batch cultures, the 15N/14N (hereafter δ15N) of diatom-bound organic matter was measured and compared to the δ15N of total diatom biomass during the progressive consumption of a nitrate pool in four polar diatom species (Fragilariopsis cylindrus, Fragilariopsis kerguelensis, Pseudo-nitzschia seriata, and Thalassiosira nordenskioeldii) and one temperate species (Thalassiosira aestivalis). In general, the δ15N of the dissolved nitrate in seawater was greater than that of the biomass, which was greater than that of the diatom-bound N. Rayleigh-type relationships were observed, allowing for estimation of the isotope effect (ɛ) for each species, with a range from 1.0‰ to 14.0‰ across all species. For all cultured strains, the δ15N values of the diatom-bound (δ15NDB) fraction was lower than those of the total diatom biomass (δ15Nbiomass). The isotopic offset between the biomass and diatom-bound N (δ15NDBoffset = δ15Nbiomass − δ15NDB) was relatively constant along the growth curve for each individual species but varied among species, with a range of 1.9‰–11.2‰. Weak relationships were determined when ɛ and the δ15NDBoffset were compared to cellular size and surface area:volume ratio. More significantly, with the exception of Pseudo-nitzschia seriata, a strong positive relationship was found between ɛ and δ15NDBoffset. While the culture data indicate a positive δ15NDBoffset across all studied diatom species, surface sediment data suggest a negative δ15NDBoffset for sedimentary assemblages. This indicates that either (1) the growth conditions of our cultures had some effect on δ15NDBoffset or (2) a low-δ15N component of the N that we measure as diatom frustule–bound is lost during early diagenesis. Given documented assemblage changes, our culture data for relevant species do not suggest that the higher δ15NDB observed in the Antarctic during ice ages can be explained by species related changes in the sedimentary bulk-to-diatom-bound isotopic difference. Future work on the diatom-bound material in cultured diatoms grown under in situ nutrient conditions, analysis of sediment trap and net tow material, and frustule dissolution experiments will more completely assess this paleoproxy.
 Primary productivity in the Southern Ocean is mainly limited by the availability of iron, and to a lesser degree light and silicic acid [Boyd et al., 1999; Dugdale et al., 1995; Martin et al., 1990a, 1990b; Nelson and Smith, 1991; Sedwick et al., 1997, 1999], leaving an excess of nitrate, an essential nutrient, in the surface ocean. Under these nitrate-replete conditions, the relative degree of uptake and assimilation of nitrate is recorded in the nitrogen isotopic composition of the resulting organic matter, where fractionation occurs during the reduction of nitrate to nitrite, by the enzyme nitrate reductase [Needoba et al., 2004; Shearer et al., 1992]. The 14N is preferentially incorporated into organic matter relative to 15N [Altabet et al., 1991; Cifuentes et al., 1989; Horrigan et al., 1990; Montoya et al., 1991; Wada and Hattori, 1978; Wada, 1980]. The result is an offset, whereby the 15N/14N (as δ15N = [(15N/14N)sample / (15N/14N)standard − 1] × 1000, where atmospheric N2 is the universal reference) of phytoplankton is lower than that of the substrate nitrate in seawater. During progressive consumption, the δ15N of both the nitrate substrate (δ15NO3−) and resulting organic matter (δ15Nbiomass) increases, as the concentration of nitrate decreases. Going back in time, relative changes in sedimentary δ15N are thought to reflect changes in the degree of nutrient consumption in the surface ocean [Altabet and Francois, 1994].
 Uncertainties associated with the sedimentary δ15N proxies make consistent interpretations difficult. Offsets in the δ15N of total sedimentary nitrogen (δ15Nbulk) are elevated by 3–5‰ relative to sinking particulate nitrogen (PN) captured in sediment traps [Altabet and Francois, 1994]. Differences of up to 3‰ can also occur in the δ15N of sinking PN (δ15NPN) between shallow and deep traps with δ15NPN ranging between −1‰ to 5‰ [Altabet and Francois, 2001; Lourey et al., 2003]. Summertime Southern Ocean δ15NPN values are low, between −0.1‰ and 1.7‰ during this period of peak sedimentation. However, wintertime δ15NPN is much higher during low productivity winter months when nutrient consumption is lowest, diverging from expectations based on isotope fractionation during nitrate assimilation. It was hypothesized that these differences are due to a reduction in the isotope effect during nitrate assimilation, or possibly more extensive isotopic alteration of sinking material during low-flux periods [Lourey et al., 2003]. These observations suggest that the assumption of a constant offset between biomass and bulk sediment δ15N is likely incorrect.
 The diatom-bound proxy has been applied in the Southern Ocean, where diatoms are the dominant photosynthetic organisms and are responsible for the majority of export production, producing an opal belt around Antarctica [Bathmann et al., 1997; Hart, 1942; Hasle, 1969; Lancelot et al., 2000]. Diatom-bound organic compounds are required for silica deposition during periods of growth, as cells reproduce and form a new frustule [Martin-Jezequel et al., 2000; Sumper and Kröger, 2004; Sumper et al., 2005; Sumper and Brunner, 2008]. Encased in silica, the organic fraction of diatoms' fossil remains makes a promising archive of surface ocean conditions. δ15NDB is a measure of the isotopic composition of only organic N that was bound within the silica frustule and thus avoids uncertainties related to alteration. In addition to removing the effects of alteration during sinking and burial, targeting the diatom-bound fraction eliminates any additional isotopic enrichment related to the incorporation of organic material from multiple trophic levels.
 However, the proxy presents uncertainties of its own. Large changes in δ15NDB within paleoceanographic records and differences between sites have been attributed to a spatially and temporally heterogeneous environment. An alternative explanation for these rapid, high-amplitude changes involves variation in the isotopic relationship between the total biomass of diatoms and the N that they incorporate into their frustules. Each diatom species makes their own suite of long chain polyamines (LCPA) or polypeptide silaffins, large N-rich molecules that serve as biomineralization compounds, forming a template for silica frustules [Kröger and Wetherbee, 2000; Kröger et al., 2002; Sumper et al., 2005; Sumper and Brunner, 2008]. Other nitrogen-containing compounds in the frustule include mycosporin-like amino acids [Ingalls et al., 2010], chitin [Brunner et al., 2010; Durkin et al., 2009], and probably others that have yet to be characterized. The specific biochemical and biophysical mechanisms of isotope fractionation during diatom frustule-bound N synthesis are unknown, as the specific processes and compounds involved in frustule formation are still being studied. Individual compounds may have distinct isotopic signatures. Since these compounds are found in different combinations and relative contributions, diatom species may vary in the δ15N relationship between their diatom-bound N and the rest of their biomass.
 To examine species-specific variation in the δ15NDB and δ15Nbiomass, we used large batch cultures to quantify the relationships between the growth media (δ15NO3−), total biomass N (δ15Nbiomass), and diatom-bound N (δ15NDB). Five diatom species encompassing a range of morphology, physiology, and taxonomy were selected. All but one are environmentally relevant species to Southern Ocean studies. Our results suggest that interspecies variations in offsets between the δ15Nbiomass and the δ15NDB are significant. Based on these observed species-related differences, we reevaluate changes in δ15NDB in a Southern Ocean sediment core.
2. Samples and Methods
2.1. Diatom Species and Culturing
 Taxonomic diversity was included in this investigation to determine if these differences produce any systematic variability in the isotopic composition of δ15NDB values. These diatom species ranged in reported size from approximately 3 to 75 μm, included fast and slow growing species with varying degrees of silicification, and included several chain formers. Experiments focused on five diatom species including Fragilariopsis kerguelensis, the dominant pennate diatom species in the Southern Ocean and largest in this study (∼25 μm) (supplied by P. Assmy, Alfred Wegener Institute for Polar and Marine Research, Bremerhaven, Germany), Fragilariopsis cylindrus (Van Heurck) Hasle (Provasoli-Guillard National Center for the Culture of Marine Phytoplankton (CCMP), strain 1102), a pennate ice edge indicator (∼6 μm), Thalassiosira nordenskioeldii Cleve (CCMP997), a centric cold water chain forming diatom (∼13 μm), and Pseudo-nitzschia seriata (Cleve) Peragallo (CCMP1309), a pennate domoic acid producing diatom and the smallest in this study (∼5 μm). The temperate centric Thalassiosira aestivalis (GSO100), isolated from the eastern North Pacific, was also used (∼15 μm).
 Semicontinuous batch cultures were maintained in 20 ml sterile filtered F/2 media [Guillard, 1975] by frequently transferring small volumes into fresh media during exponential growth, well before the stationary phase was reached. After multiple generations under constant conditions, the specific growth rate for each species was determined in serial, triplicate cultures. Fluorescence was monitored daily using a model 10 AU fluorometer (Turner Designs). Specific growth rates were determined using the relationship between in vivo chlorophyll-a fluorescence and the exponential increase in cell numbers [Brand et al., 1981]. The chlorophyll-a proxy for growth was verified by volumetric cell counts for each species using a Multisizer II (Coulter, Inc.) with a 140 μm aperture drawing 500 μL per sample. Cell counts were also conducted at 250x using an SZX16 microscope (Olympus) equipped with phase contrast optics.
 Glassware was acid washed and autoclaved. Whatman glass fiber filters (GFF's) were precombusted (450°C) in a muffle oven. Medium for experiments was prepared using 0.2 μm sterile-filtered Narragansett Bay seawater and 0.2 μm sterile filtered-nutrient stock solutions [Guillard and Ryther, 1962; Guillard, 1975]. To prevent precipitation of nutrients, F/2 medium was not autoclaved after filtering. Because specific nitrate reductase activity in light was observed to be five times higher than in the dark [Eppley et al., 1970], the cultures were maintained under continuous light at 100 μmol photons m−2s−1. Additionally, this light level has been shown to saturate the photochemical apparatus of Antarctic diatoms and would be found at this level or higher throughout the entire summer mixed layer [Knox, 2007; Sakshaug and Holm-Hansen, 1986]. Thalassiosira aestivalis was maintained at 18°C, while all other diatom cultures were kept at 2°C.
 Experiments began when each 20 L carboy, filled with 20 L of sterile-filtered F/2, was inoculated with ∼200 ml of an exponentially growing culture. To ensure near-complete nitrate consumption during the experiment, 150–250 μM NO3− was added, equaling only15–30% of the amount prescribed for F/2 media. Carboys were agitated to prevent settling of diatoms using a magnetic stir shaft with a 7.5 cm bladed propeller spinning at approximately 250 rpm. Tygon® tubing was threaded through the lid to approximately 1 cm above the propeller, fastened and sealed in place. This siphon tube was used to draw samples during the experiment and was otherwise sealed to prevent contamination. Carboy experiments were run in duplicate or triplicate for each species.
 Experiments were sampled daily over the course of two to five weeks, along the growth curve, until the stationary phase was reached. Daily measurements included fluorescence, nitrate concentration of the seawater, δ15NO3−, and the δ15Nbiomass. The δ15NDB and nitrogen content of diatom frustules was measured with less frequency due to the nitrogen requirement for each sample. Between four and ten discrete diatom-bound samples were gathered for each culture along the growth curve.
 Each fluorescence sample was preserved using 2% acid Lugols solution [Throndsen, 1978] for cell counts. Cell counts were made using a Multisizer II Coulter Counter on these preserved samples and the relationship of each culture's cell number versus fluorescence was then determined. Cell size was also measured using the Coulter counter and verified with 25 measurements per experiment on an inverted epifluorescent Nikon Diaphot 300 microscope, equipped with phase contrast at 400x. Cell volumes from the Coulter counter were not used as they assumed a spherical diatom. Instead, measurements of each axis of the diatom allowed for the calculation of cell surface area and volume. Water samples (∼100 ml) collected for seawater nitrate, phosphate, and silicate concentration and the δ15NO3− were gently (∼50 mm Hg vacuum) filtered through precombusted Whatman GFF's. Approximately 25 ml of filtrate was used to rinse the container and the remaining 75 ml aliquots were frozen for chemical analysis. The GFF was rinsed delicately with filtered seawater and frozen separately, to provide the δ15Nbiomass. For diatom-bound sampling, between 1 and 5 L of water (depending on amount of biomass present) was collected from the carboy, spun down in a centrifuge to collect the diatoms, and the remaining water was poured off. This provided a more rapid method for concentrating the sample and reduced biomass loss on filters. All samples were stored at −20°C for later analysis as a group.
2.2. Diatom-Bound Cleaning
 The high organic matter content of freshly harvested cells, relative to typical deep ocean sediments, required alteration of the existing methodology for chemically cleaning frustules [Brunelle et al., 2007; Robinson et al., 2004, 2005; Sigman et al., 1999a]. Frozen diatom-bound samples were thawed, rinsed with Milli-Q water and cleaned using (1) 35 ml of 2% by weight sodium lauryl sulfate at room temperature for ∼15 min, (2) a preliminary oxidation using ∼15 ml saturated potassium permanganate and ∼15 ml saturated oxalic acid, after 5 ml of 18 M sulfuric acid addition [Hasle and Fryxell, 1970], and (3) a final oxidation using 72% perchloric acid in a 100°C water bath for one hour. Samples were rinsed thoroughly with Milli-Q water between each consecutive treatment. An ultrasonic bath and vortex mixer were used to ensure complete reaction/rinse and prevent any clumping of sample material during each step. Diatom-bound samples were rinsed thoroughly with Milli-Q and then dried in a 60°C oven overnight prior to the oxidation of 2–20 mg of cleaned frustule by 1.5 ml of a 0.22 M potassium persulfate and 1.5 M sodium hydroxide solution in an autoclave at 121°C for 20 min [Bronk et al., 2000; Robinson et al., 2004].
2.3. Chemical Analysis
 Nitrate concentrations of both the growth media and the persulfate solutions containing oxidized diatom-bound N were measured by chemiluminescence after reduction to NO using a Teledyne Instruments (Model 200E) chemiluminescence NO/NOx analyzer and potassium nitrate reference standards [Braman and Hendrix, 1989]. Orthophosphate (USEPA method 365.5) [Grasshoff, 1976; Murphy and Riley, 1962] and silicate [Grasshoff, 1976; Parsons et al., 1984] concentrations were determined using a Lachat Instruments QuickChem 8000 flow injection analyzer.
 The δ15NO3− and δ15NDB were measured by gas chromatography–isotope ratio mass spectrometry (GC-IRMS) on a Thermo Delta V Advantage IRMS. Nitrate was converted to N2O for injection following the denitrifier method [Sigman et al., 2001]. The potassium nitrate reference material IAEA-N3 (4.7‰) was used to standardize δ15NDB and δ15NO3− results [Böhlke and Coplen, 1995]. The δ15Nbiomass was measured by combustion of the frozen, oven-dried (60°C) material. The glass fiber filters were packed into a tin cup for measurement by elemental analyzer coupled to the IRMS (EA-IRMS). Ammonium sulfate reference materials IAEA-N1 (0.4‰) and IAEA-N2 (20.3‰), and an in-house standard, amino caproic acid, were used to standardize EA-IRMS results [Böhlke and Coplen, 1995; Crosta and Shemesh, 2002; Révész et al., 1997]. Precision was determined by multiple measurements of standards within each run and replicate measurements of samples. All NO3− concentrations and 50% of the δ15NO3 samples were run in duplicate to within ± 1.1% for concentration and ± 0.3‰ for isotopic values. Approximately 50% of biomass samples were run in duplicate with δ15Nbiomass values within ± 0.4‰. Only 15% of the diatom-bound samples were run in duplicate, due to small sample sizes, with average δ15NDB precision of ± 0.8‰. Blank contributions and the associated errors in δ15NDB measurement were significant due to the blank associated with the alkaline persulfate reagent, typically 2–3 μM. Because of the small sample sizes, in particular early in the growth experiments, the blank accounted for up to 40% of the total N. Direct measurement of the persulfate reagent δ15N contribution is not consistently possible. The denitrifiers often die when the necessarily large volume of blank is added to the bacteria vial. However, estimates of the blank δ15N (δ15Nblank) were made by measuring the δ15N of several dilution series of amino acids with known isotopic compositions. Based on these experiments, a conservative range of 0 to −5‰ was assigned to the δ15Nblank. The actual δ15NDB was then calculated by mass balance using the measured value and this range. The error bars for each diatom-bound point are based on these error estimates, in addition to replicate measurements made when possible. Statistical significance was determined using a two-tailed, type-3 T test to provide p values for variables conservatively determined to be unpaired and independent with unequal variances.
3. Results and Discussion
3.1. Diatom Growth
 In replicate carboy experiments, exponential growth rates within each diatom species were comparable, with an average coefficient of variation of 14%, ranging from 1 to 27%. Specific growth rates between species ranged from 0.41 to 1.34 day−1, or 18–59 h doubling time (Table 1). This is slightly higher than growth rates seen in nature and may be due to 24 h light, the lack of predators, and the abundance of both macronutrients and micronutrients. Centric species grew significantly faster than pennate species (p = 0.008). Growth rate has been shown to affect silicification, with slower growths resulting in greater silicification, as more time is spent in the cell wall synthesis phase [Martin-Jezequel et al., 2000]. However, differences in growth rate do not influence the isotopic fractionation of N during NO3− assimilation [Needoba and Harrison, 2004].
Table 1. Growth Rates and Measured N Isotope Parameters for Pennate Diatoms and Centric Diatoms
Initial [NO3−] (μM)
Final [Si]/[PO4] (μM)
T(number) refers to transfer number, whereby Pseudo-nitzschia seriata T26 is the 26th transfer (batch) of the semicontinuous batch culture used for that specific experiment.
N content of diatom-bound samples were significantly lower, on average, than other experiments, indicating that alteration may have occurred. However, corresponding ɛ values (which do not depend upon cleaning) also differed, necessitating that both remain in the analysis.
Pooled standard deviations are reported.
[Si]/[PO4] were not measured for this experiment.
Thalassiosira aestivalis T34 initial [NO3−] and f are removed from mean calculations.
 The average initial concentration of nitrate was 212 ± 38 μM with an initial δ15NO3− of 8.9 ± 0.3‰ for all experiments. Phosphate and silicate were provided in excess of the expected demand to ensure that NO3− was the limiting nutrient (Table 1). Average initial concentrations were 33.5 ± 1.1 μM phosphate and 121.5 ± 3.4 μM silicate. In an attempt to account for their increased silica demands, carboys containing Fragilariopsis kerguelensis were spiked with 162.6 ± 3.4 μM silicate. On average, the fraction of initial nitrate remaining (f) at the end of experiments was 0.27 ± 0.16, indicating significant utilization of nitrate. The wide range in final silicate concentrations demonstrates the differences in silica requirement for each species (Table 1). Final silicate concentrations of 1.1 μM and 2.3 μM were observed in Fragilariopsis kerguelensis cultures, indicating that they may have become silica limited nearing the end of each experiment. All other cultures had significantly more silica present in their final sample (4.4 μM to 75.5 μM), reducing the likelihood of silica limitation during growth. At no point did the phosphate concentration fall below ∼15 μM in any experiment.
 The NO3− concentration decreased in each closed system carboy experiment during growth while the δ15NO3− increased in a logarithmic manner (Figure 1). The Rayleigh equation (1) defines this fractional distillation of the NO3− pool as a function of the isotope effect (ɛ), and the degree of nitrate utilization, or f:
The fraction of initial nitrate remaining at the time of measurement, f, is defined as f = [NO3−]measured/[NO3−]initial. Defined as the ratio of assimilation rates for each isotope [ɛ = (14k/15k − 1) × 1000, where 14k and 15k are the rate coefficients of the reaction for 14N- and 15N-containing reactant, respectively, ɛ reflects the preferential incorporation of 14N into biomass and the resulting isotopic offset between the organic matter and the nitrate pool at any instant. Therefore, as nitrate utilization increases (decreasing f), both the nitrate pool and the bulk biomass are progressively enriched in 15N (Figure 2a). The value of ɛ was estimated from the slope of the δ15NO3−f curves for each species and culture based on equation (1) (Figures 2b–2f).
 Most species exhibited relatively consistent ɛ over multiple experiments (Table 1). The two species with variable ɛ were Pseudo-nitzschia seriata and Fragilariopsis cylindrus. There was a wide range in estimated ɛ, with observed values between 1 and 14‰. The overall mean ɛ was 5.03 ± 3.42‰, which encompassed all species and experiments, except Pseudo-nitzschia seriata T35, within a 95% CI. The high degree of variability between all species was observed in other culturing experiments and has been attributed to a variety of factors including aeration, mechanical agitation, light levels, and nutrient concentrations [Montoya and McCarthy, 1995; Needoba and Harrison, 2004; Wada, 1980].
Fragilariopsis kerguelensis and Thalassiosira aestivalis experiment T34 did not follow the predicted relationship as closely as other species, with an r2 of 0.50 for both (Figure 1). Fragilariopsis kerguelensis showed very little change with f, such that the almost negligible slope produced the weak r2. Thalassiosira aestivalis T34 started with an exceedingly high concentration of NO3− (850.6 μM), which was not utilized as completely as other experiments. The low r2 here is attributed to fitting a curve to such a limited observed portion of f. However, even with a condensed δ15NO3− curve, with f ranging only from 1.0 to 0.78, the calculated ɛ still fell, within error, alongside the other ɛ values from replicate Thalassiosira aestivalis experiments.
 The δ15Nbiomass values were consistently lower than δ15NO3− values over the entire growth curve and increased logarithmically with decreasing f (Figure 2). Overall, measured δ15Nbiomass fit expectations based on the accumulated product equation of Mariotti et al. :
Using a δ15NO3t=0− of 8.9‰ and the calculated ɛ values from the NO3−/ln f curves, we estimated δ15Nbiomass for each specific species and carboy. The measured δ15Nbiomass values lie along the Rayleigh predictions (Figure 2), indicating that the cultures reflect a closed system, where the nitrate was assimilated into biomass with little or no recycling. While Rayleigh predictions were close for each species, several subtleties were present in specific cultures. Thalassiosira nordenskioeldii demonstrated negligible trends in δ15Nbiomass with observed increases of only 1‰ (Figure 2b). Predictions indicated that at least a 2‰ change should have been observed. Additionally, the δ15Nbiomass measurements for Thalassiosira nordenskioeldii were ∼1‰ lower than the predicted values. The biomass values for Pseudo-nitzschia seriata were approximately ∼1‰ higher than the predicted (Figure 2d). However, based upon the predicted versus measured differences in δ15Nbiomass, no cultures deviated significantly in either direction from the predicted result. There were no statistically significant differences between pennate and centric diatoms.
3.4. Diatom-Bound δ15N
 The δ15NDB values were consistently lower than δ15Nbiomass values over the entire growth curve. Overall, δ15NDB values tended to increase in much the same way as the δ15Nbiomass, with decreasing f (Figure 2). However, the δ15NDB measurements were significantly more variable than the substrate or the δ15Nbiomass measurements. The variability could be attributed to several factors including (1) small sample size, (2) a large persulfate blank (relative to the sample), and (3) excessive cleaning and alteration. An average of ∼3 mg/sample of cleaned opal was available for δ15NDB measurements, compared to 7–12 mg/sample typically measured from sediment cores. This was a larger problem for δ15NDB samples taken near the beginning of each experiment, before any significant growth had taken place and where up to 5 L of water was collected to obtain the required amount of diatoms for cleaning. As a result, error tended to decrease with decreasing f due to increased biomass recovery and a proportional reduction in blank contribution as the experiment progressed (Figure 2). In addition to the large blank contributions, cleaning fresh diatom material is difficult. Fresh diatom opal is more soluble and reagent sensitive than fossil diatom opal. Fractionation can occur due to “overcleaning” and possible frustule dissolution prior to the alkaline persulfate step. We developed several different cleaning protocols and tested them on cultured diatom material. The method used to clean the carboy experiments was chosen for its high degree of precision in both N content (±0.21 μmol N/g opal) and δ15NDB (±0.45‰). It is unlikely that this cleaning protocol fractionated samples, although we cannot completely discount it, given the range of size and morphology of the diatoms cultured. The N content in cleaned material for diatom-bound analysis of Pseudo-nitzschia seriata T35 and Fragilariopsis cylindrus T33 was significantly lower (p = 0.001) than the other experiments, potentially a result of overcleaning and alteration. However, the generally greater variability in these cultured δ15NDB measurements relative to our previous sedimentary δ15NDB is attributable to the small sample sizes and blank contamination. The nitrogen content of cleaned diatom-bound material is within the range found in other studies [Robinson et al., 2004, 2005; Sigman et al., 1999a].
 For each individual carboy experiment, the δ15NDB values appeared to be offset from the δ15Nbiomass by nearly constant amounts (δ15NDBoffset = δ15Nbiomass − δ15NDB) over the course of the growth curve. The δ15NDBoffset ranged from approximately 2–10‰, but was relatively consistent for each species with an overall average of 5.3‰ (Table 1). Much like the δ15NDB values, interspecies and replicate experiment variability in δ15NDBoffset and the associated measurement error decreased as f decreased and sample size increased. Given no clear trends, we assume that the ɛ for each specific species was constant, and a δ15NDB relationship could be described by
Six δ15NDB points were higher than their corresponding δ15Nbiomass measurement (negative values of δ15NDBoffset) and most of them occurred early in growth (i.e., high f). The average value of δ15NDBoffset did not change systematically with f. At high f, the range in δ15NDBoffset measurements was large, tapering in toward more consistent values as more NO3− was consumed. The initial scatter at high values of f is attributed to random analytical error associated with the aforementioned small sample sizes and blank. Pseudo-nitzschia seriata T35 was the single exception to the δ15NDBoffset relationship, whereby δ15NDB values were consistently higher than the δ15Nbiomass (Table 1). Pseudo-nitzschia seriata's negative δ15NDBoffset may be attributed to its small size. Alternatively, the low N content of cleaned material for Pseudo-nitzschia seriata T35 raises the possibility of alteration of the δ15NDB during the chemical cleaning.
 In an early study by Wada and Hattori , N isotope fractionation during diatom nitrate assimilation was observed to be inversely related to growth rate. However, subsequent work has not supported this relationship [Granger et al., 2004, 2008; Montoya and McCarthy, 1995; Needoba and Harrison, 2004; Needoba et al., 2004]. Consistent with the latter results, no trend was found to exist between growth rate and ɛ (Figure 3a). A slight positive trend was observed when Pseudo-nitzschia seriata was removed (r2 = 0.29), with higher growth rates corresponding with higher ɛ. However, this is counter to the expectation for weakly expressed fractionation under higher growth rates. The δ15NDBoffset had no relationship when compared to growth rate (not shown), indicating that factors other than growth drive fractionation during both uptake and frustule formation.
 In fact, the fractions associated with uptake and frustule formation appear to be related (Figure 2f), although how is not obvious. Nitrate is imported across the plasmalemma by active transport mechanisms [Granger et al., 2004, 2008; Tischner, 2000]. Active transport increases the internal concentration of NO3− relative to the external environment (e.g., C in work by Amoroso et al. ; N in work by Needoba et al. ), which then allows for nitrate efflux by passive mechanisms. ɛ is related to the ratio of efflux to influx. High relative efflux rates result in larger ɛ, and vice versa [Needoba et al., 2004]. Previous studies have demonstrated that C-isotope fractionation is inversely proportional to cell size [Burkhardt et al., 1999; Laws et al., 1995; Popp et al., 1998], the explanation of which was related to efflux/influx dynamic described above for isotope fractionation during nitrate assimilation. We found that ɛ had a negative linear correlation to cell size (Figure 3b), with larger cells having decreased fractionation. However, cells are not perfect spheres and the path length and exchange interface can be highly variable. The combination of shape and size may be approximated as the ratio of cellular surface area to volume (SA:V). In the case of elongated cells (high SA:V), there may be the potential for significant expression of isotopic fractionation due to the increased surface area for NO3− efflux relative to cell size [Popp et al., 1998]. All else held constant, a high SA:V ratio may lead to a lower ratio of nitrate reduction relative to efflux, thereby exchanging nitrate with the growth medium more readily, leading to less δ15N elevation of internal nitrate and more complete expression of the nitrate reduction isotope effect (i.e., higher ɛ) in the external medium. Observed increases in ɛ corresponded with increased SA:V, with a stronger relationship (r2 = 0.70) when pennate species were investigated alone (Figure 3c). Of course, it is unlikely that nitrate reduction scales simply as a function of cell volume, without any regulation of reduction (or uptake) rate by the organism.
 And while no systematic relationships observed when δ15NDBoffset were compared to size and SA:V (Figures 3d and 3e), a strong linear relationship between ɛ and δ15NDBoffset (δ15NDBoffset = 1.56ɛ + 0.05) was observed when Pseudo-nitzschia seriata was removed from the analysis (Figure 2f). This species had a high degree of error, as well as a wide range in reported δ15NDBoffset values. It is not known if small size and physiology plays a role in this variability, or if the alteration of small cells during cleaning resulted in analytical error. In any event, it is difficult to explain the observed correlation of the diatom-bound offset (the sum of frustule formation processes) to the ɛ of nitrate assimilation. After nitrate is acquired, fractionation occurs during the reduction of nitrate to nitrite, by the enzyme nitrate reductase [Mariotti et al., 1982; Needoba et al., 2004; Shearer et al., 1992]. This step is the first in a sequence that forms an internal pool of bioavailable nitrogen that is isotopically distinct and lighter than the external nitrate pool. Other enzymes piece together long-chain polyamines or biomineralization proteins, such as silaffins, which make up the diatom-bound organic matter. It is likely that each of these enzymes imposes a fractionation as well. It is possible that the allocation of organic N in the cell to either frustule-bound compounds such as silaffins and LCPAs or to N forms not related to the frustule could produce differences in the diatom-bound offset that are somehow related to size and thus to ɛ (Figure 3f). Larger diatoms, with a smaller SA:V, have proportionally more cytoplasm than a smaller diatom. However, the lack of relationship between δ15NDBoffset and size (or between δ15NDBoffset and SA:V) does not support this. Given how little we know about the isotopic fractionation associated with frustule formation, we cannot yet give a satisfying explanation for the observed variation in δ15NDBoffset, let alone its correlation with ɛ.
 While our single-species cultures in filtered seawater are useful for examining diatom-specific δ15NDBoffset, the conditions of the cultures are far from those found in nature. Southern Ocean diatom assemblages are heterogeneous mixtures of different diatom species that vary across time and space, closely connected to frontal systems, water masses, and sea ice distribution [Burckle et al., 1987; Eynaud et al., 1999; Fenner et al., 1976; Zielinski and Gersonde, 1997]. Additionally, present-day nitrate concentrations in the summertime Antarctic mixed layer average approximately 25 μM, much less than the nitrate concentrations in excess of 200 μM that apply to the starting conditions of our batch cultures. We observed significant variability in our cultured ɛ and δ15NDBoffset values and high variability in ɛ is typical of culture work [Needoba et al., 2003; Needoba and Harrison, 2004; Needoba et al., 2004; Wada and Hattori, 1978]. Far less variability is observed in ɛ from field studies [Sigman et al., 1999b; Wu et al., 1997]. In this vein, it is possible that δ15NDBoffset values in the ocean are also significantly less variable than our culture-based observations suggest. With this caveat in mind, we will discuss the potential implications of our results on paleoceanographic reconstructions.
3.5. Culture-Based Comparisons With Cores
 Most of the cultured species show changes in δ15NDB with increasing nutrient consumption, suggesting that a sedimentary δ15NDB record at least in part reflects the nutrient status of the surface ocean. However, there is an obvious discrepancy between our culture data and the δ15NDB measured in sediments. Much of the published sedimentary diatom-bound work indicates that, on average, the δ15N of bulk sediment (δ15Nbulk) is similar to, and often less than δ15NDB [Brunelle et al., 2007; Robinson et al., 2005; Robinson and Sigman, 2008; Sigman et al., 1999a]. δ15Nbulk in the open ocean has a δ15N higher than that of the sinking flux or, in turn, the phytoplankton biomass from which the sinking flux is ultimately derived. Thus, surface sediment δ15NDB and δ15Nbulk values suggest that sedimentary δ15NDB is higher than the δ15Nbiomass of the diatoms that produced the frustules in the sediment, opposite to our measurements of cultured diatoms.
 Higher δ15NDB relative to δ15Nbulk in sediments may be the result of postdepositional changes on either the δ15NDB or the δ15Nbulk. The rapid loss of a low δ15N fraction could cause the systematic enrichment of sedimentary diatom-bound material. There is no systematic reduction in the diatom-bound N content of cleaned opal with depth that would suggest loss of a portion of the frustule and fractionation of the bound N [Robinson et al., 2004; M. G. Horn and R. S. Robinson, Comparison of bulk and diatom-bound nitrogen isotopes in Southern Ocean downcore profiles, manuscript in preparation, 2011]. However, we cannot rule out the loss of such a low-δ15N fraction before the diatom opal reaches the seabed.
 Alternatively, the discrepancy may derive from physiological differences between cultures and open ocean diatoms. Possible differences include iron availability [Hutchins and Bruland, 1998; Leynaert et al., 2004; Takeda, 1998], light limitation, high nutrient contents, bottle effects, and more. More work will need to be done to test this.
 On top of this discrepancy in the sign of the δ15NDBoffset values, paleo-reconstructions of nutrient utilization must also come to grips with interspecies differences in δ15NDB. Sediments contain several different diatom species at any one time and the proportion of each species varies downcore. The relative contribution of each species' characteristic δ15NDBoffset, could significantly impact a record of δ15NDB. If we work from the measured values of δ15NDBoffset, we can make some preliminary estimates of how assemblage changes might impact nutrient status reconstructions.
 A mass balance, based upon diatom assemblage data and a range of δ15NDBoffset at core TN057-13PC4 provides us with a simplified estimate of assemblage-related (i.e., not utilization-related) changes between the LGM and the Holocene south of the APF (Figure 4) [Nielsen, 2004; Horn et al., 2011]. On the one hand, Fragilariopsis kerguelensis presently makes up the majority (60–90%) of polar Southern Ocean sediments in open ocean areas. During the LGM, its abundance dropped to >30% of the sediment [Crosta et al., 1998; Nielsen, 2004; Zielinski and Gersonde, 1997]. On the other hand, Fragilariopsis cylindrus makes up between 1 and 2% of sediments today and peaked at approximately 8% during the LGM [Nielsen, 2004; Zielinski and Gersonde, 1997]. Eucampia antarctica makes up a large portion of the imbalance between these two species with a peak abundance of 30% during the LGM and Holocene values around 3% [Nielsen, 2004]. Chaetoceros resting spore abundance also changes substantially, making 35% during the LGM but only 5–10% during the Holocene [Nielsen, 2004]. To simplify, we assume the remainder of the population, Eucampia antarctica, and Chaetoceros resting spores have a δ15NDBoffset of 6‰. This is both the average value of cultured species, excluding Fragilariopsis kerguelensis (Table 1), and the average centric value that is assumed for Eucampia antarctica, whose species-specific δ15NDBoffset is unknown. From these values, we calculate a community wide δ15NDBoffset value for the Holocene of ∼2.4‰ and the LGM of ∼4.9‰. This would correspond to a 2.5‰ increase in δ15NDB from LGM to Holocene based on assemblage changes, assuming no change in nutrient utilization. In contrast, decreases of 0–4‰ in δ15NDB from the LGM into the Holocene are observed at sites located across the Southern Ocean [Robinson et al., 2004, 2005; Robinson and Sigman, 2008]. This calculation does not support the assertion that the downcore δ15NDB changes were driven by assemblage alone. Of course, our understanding of δ15NDBoffset is incomplete, and the above calculation is simplistic. Adding to variations in δ15NDB, ɛ may also have changed due to variations in light limitation between the LGM and the Holocene [DiFiore et al., 2010]. If the observed ɛ-diatom-bound relationship holds in the field, this would impact the diatom-bound record through its effect on both δ15Nbiomass and the δ15NDBoffset.
 In all but one culture, it was observed that the δ15NO3− > δ15Nbiomass > δ15NDB. An average δ15NDBoffset (δ15NDBoffset = δ15Nbiomass − δ15NDB) of 5.3 ± 4.4‰ was observed, with an average isotope effect for nitrate assimilation (ɛ) of 5.03 ± 3.42‰. With few exceptions, specific values for δ15NDBoffset and ɛ are relatively consistent within species, but differ greatly between them. Differences between pennate and centric ɛ and δ15NDBoffset were statistically insignificant. ɛ and δ15NDBoffset were observed to be weakly negatively correlated with cell size (that is, decreased with increasing cell size). In addition, a weak positive correlation between ɛ and SA:V was observed for all diatoms; this relationship strengthened when pennate species are investigated alone. δ15NDBoffset is positively correlated with ɛ. However, open ocean testing is required, as ɛ's correlation and apparent control on δ15NDBoffset is not yet mechanistically understood. Our data indicate a positive δ15NDBoffset across all studied diatom species, whereas sedimentary data suggest a negative δ15NDBoffset between δ15Nbulk and δ15NDB [Brunelle et al., 2007; Robinson et al., 2004, 2005; Horn and Robinson, manuscript in preparation, 2011].
 Future work will need to investigate whether the growth conditions of our cultures had some effect on δ15NDBoffset, or if a low-δ15N component of the diatom-bound-associated N is lost during early diagenesis. Further culture studies using a chemostat maintained at environmentally relevant concentrations may also clarify the observed trends. Net tows and sediment trap data from the Southern Ocean may provide insight into the diagenetic processes that may affect δ15NDB from the time of formation to early sedimentation, and time series studies of fresh, organic-rich material buried in sediments for varying times may provide a first look at the processes in sediments that may affect δ15Nbulk and δ15NDB values. Additionally, gravitational SPLITT fractionation could be used to sort diatoms, potentially isolating single species or providing a more consistent assemblage for downcore analysis. Finally, interlaboratory comparisons are recommended, as there is no standardized method for isolating the organic matter in the diatom-bound fraction. Given assemblage changes, our culture data, for relevant species, do not suggest that the observed glacial/interglacial δ15N change in Antarctic sediment cores can be explained by assemblage change alone.
 E. Baker provided assistance with the flowing seawater facility and environmental chambers (supported by NSF EPSCoR award 00554548). J. Rines, A. Drzewianowski, J. Graff, and J. Krumholz helped with culturing. P. Assmy generously donated live cultures of Fragilariopsis kerguelensis. R. Campbell provided access to a Coulter counter. A. VanKeuren of the Nixon lab analyzed P and Si samples. This work was supported in part by U.S. NSF OPP-0453680, a URI Graduate School of Oceanography alumni grant, and a URI graduate assistant united fellowship.