Oceanic Nitrogen Reservoir Regulated by Plankton Diversity and Ocean Circulation

LETTER doi:10.1038/nature11357

Oceanic nitrogen reservoir regulated by plankton diversity and ocean circulation

Thomas Weber1 & Curtis Deutsch1

The average nitrogen-to-phosphorus ratio of marine phytoplankton We developed a simple ecosystem and biogeochemical model to (16N:1P) is closely matched to the nutrient content of mean ocean simulate the long-term coupling of N and P cycles in an observationally waters (14.3N:1P). This condition is thought to arise from biological constrained ocean general circulation model (GCM). The ecosystem 2 control over the ocean’s nitrogen budget, in which removal of comprises a general phytoplankton class (O) that assimilates NO3 32 bioavailable nitrogen by denitrifying bacteria ensures widespread and PO4 in the molar ratio RO, and a diazotrophic class (F) that 32 2 selection for diazotrophic phytoplankton that replenish this assimilates PO4 and releases NO3 from newly fixed N2. The 1–3 essential nutrient when it limits the growth of other species . maximum growth rate of N2-fixers (mF) is reduced relative to that of Here we show that in the context of a realistic ocean circulation other phytoplankton (mO), to reflect a constant energetic cost of model, and a uniform N:P ratio of plankton biomass, this feedback diazotrophy and a variable dependence on iron (Fe)16,17. Fe-limitation mechanism yields an oceanic nitrate deficit more than double its is patterned according to the distribution of atmospheric dust observed value. The critical missing phenomenon is diversity in the deposition, and its overall strength is varied between simulations metabolic N:P requirement of phytoplankton, which has recently (Fig. 1a, Supplementary Fig. 1). Denitrification is simulated in benthic been shown to exhibit large-scale patterns associated with grid boxes and anoxic regions of the water column (Supplementary species composition4. When we model these variations, such that Fig. 2), and its global rate is also varied within specified limits. The diazotrophs compete with high N:P communities in subtropical model’s P reservoir is conserved at the modern ocean value, but its N regions, the ocean nitrogen inventory rises and may even exceed inventory adjusts over millennial timescales until a steady-state the average N:P ratio of plankton. The latter condition, previously balance between N2-fixation and denitrification is achieved. See considered impossible, is prevented in the modern ocean by shallow Methods for model details. 5 circulations that communicate stoichiometric signals from remote Consistent with box models , we found that in a ‘Redfieldian’ ocean biomes dominated by diatoms with low N:P ratios. Large-scale where RO 5 16 everywhere, the steady-state value of SN/SP depends patterns of plankton diversity and the circulation pathways on two parameters: (1) the globally integrated rate of denitrification, connecting them are thus key factors determining the availability and (2) the competitive handicap faced by diazotrophs (mF/mO). The decrease in SN/SP at higher denitrification rates and stronger Fe of fixed nitrogen in the ocean. 32 2 The biologically mediated feedback between marine denitrification limitation (Fig. 1b) reflects the proportions of PO4 and NO3 1,2,5 required to support global N2 fixation rates that balance N removal and N2 fixation operates like a ‘nutrient thermostat’ that couples the 2 through denitrification. Because diazotrophs have slow growth rates, ocean’s fixed N reservoir (primarily in the form of nitrate, NO3 ) to its 32 2 2 they compete successfully for PO only when NO is low enough to less dynamic reservoir of P (primarily phosphate, PO 3 ). A central 4 3 4 hinder their competitors to a similar degree. The diazotrophic niche is element of this self-regulating mechanism is the partition of ecological 2 2 thus determined, through competitive dynamics, by the NO :PO 3 niches between diazotrophic (N -fixing) phytoplankton, which grow 3 4 2 ratio of ambient sea water, even though their growth is explicitly slowly but do not require an external supply of fixed N (refs 6, 7), and 2 independent of NO . When strong Fe limitation exacerbates their other plankton that grow quickly but are often N-limited owing to 3 8–10 competitive handicap, the deep ocean must accumulate a larger deficit persistent N removal in anoxic environments . The quantitative 2 of NO3 (lower SN/SP) to support the required N2-fixation rate. understanding of this mechanism rests on box models, in which Similarly, higher global rates of denitrification must be balanced by diazotrophs maintain the ocean’s ratio of major nutrient reservoirs enhanced N2 fixation that, for a given diazotrophic growth rate, can (SN/SP) close to, but slightly below, the N:P requirements of the 32 32 only be achieved with a greater excess of PO4 (lower SN/SP). plankton with which they compete for PO4 (refs 5, 11, 12). Throughout the observationally supported range of global Box model depictions make two major simplifications. First, denitrification rates (150–250 teragrams of N per year, Tg N yr21, diazotrophs are assumed to compete with plankton having a universal ref. 18), simulated SN/SP is considerably lower than its observed value Redfield ratio of 16N:1P, which sets a threshold for N2 fixation and of approximately 14.3. For the average plankton, this amounts to a constitutes the ‘set point’ of the nutrient thermostat, towards which 2 2 global deficit of NO3 relative to PO4 of 6–13 mM, which is 2–4 SN/SP is restored. In reality, the Redfield ratio is only an average value times larger than observed. Even when Fe limitation is eliminated, and plankton N:P varies systematically between marine species and unrealistically low denitrification rates (,100 Tg N yr21) are required 4,13–15 their preferred biomes . It has thus been hypothesized that the to reconcile the model with observations (Fig. 1b). Neither a shift in the regulation of SN/SP is biased towards the nutrient requirements of patterns of denitrification nor greater model complexity—adding those species cohabiting with diazotrophs in subtropical biomes3. dissolved organic matter, a complete iron cycle, or minor N budget Second, the circulation pathways that transport nutrients between terms—can eliminate this discrepancy (see Supplementary Notes). It surface regions with different species composition and N:P ratios are can only be resolved by expanding the ecological niche of diazotrophs. not represented in box models, but may have a central role in shaping This cannot be accomplished by increasing their growth rate, which the ecological niche of diazotrophs3. A realistic physical model is already reaches near parity with other plankton. It requires a process 32 2 required to identify the processes maintaining the N reservoir of an through which the availability of PO4 is enhanced relative to NO3 ocean with diverse plankton stoichiometry. in the subtropics. We investigated whether large-scale deviations from

1University of California Los Angeles, Los Angeles, California 90095, USA.

a South North c N fixation N2 fixation Pacific North Atlantic 2 1 Pacific 20 0.8 O / μ

F 0.6 15 Fe-limitation μF/μO μ

0.4 None 0.95 zonal mean Constant R R O Variable O Weak 0.91 O, 10 (R = 16) R R ( = 20) 0.2 O,ST O,ST Strong 0.56 0 0.01 0.1 1 10 100 60º S 40º S 20º S 0 20º N 40º N 60º N Dust deposition (g m–2 y–1) Latitude

b 300 d 300 11 3 1 10 12

) 250

–1 12 Estimated range –1 Estimated range 14

11 13 200 200 12 13 14 15 150 150 13

Denitrification (Tg N y 15

14 Denitrification (Tg N y 100 100 14 16 16 50 50 0.5 0.6 0.7 0.8 0.9 0.5 0.6 0.7 0.8 0.9 Strong Weak Strong Weak / / Fe-limitation μF μO Fe-limitation Fe-limitation μF μO Fe-limitation Figure 1 | Model scenarios and solutions. a, Growth rate of diazotrophs observational constraints on denitrification range; red line indicates observed relative to other phytoplankton, as a function of atmospheric dust deposition. SN/SP; grey dots are solutions of individual simulations. c, Variations in RO When Fe limitation is stronger, mF/mO is more variable between regions of high are incorporated using inferred community composition. Diazotrophs are and low deposition, and its mean value (mF=mO) is reduced. b, Predicted steady- predominantly confined to the green-shaded latitude bands. d,Asforb, except state SN/SP (black contours) for a Redfieldian ocean, across a range of for stoichiometrically diverse scenarios (RO,ST < 20). denitrification and Fe-limitation scenarios. Blue shading represents Redfield stoichiometry in nutrient uptake by non-fixing phytoplankton3 average export ratio of 16N:1P. As plankton stoichiometry becomes could provide such a mechanism, and resolve the gap between model more diverse, global SN/SP rises steadily (Fig. 2), reflecting the expan- predictions and measurements. sion of the diazotrophic niche caused by the high N:P requirements of Large-scale variations in RO have recently been shown to hold plankton cohabiting subtropical regions (increased RO,ST). However, throughout the Southern Ocean, where polar latitudes dominated by for every increase in RO,ST, the increase in SN/SP is only 40% as large, diatoms export low N:P organic matter, while Subantarctic latitudes much less than would be required for diazotrophs to keep SN/SPin are characterized by high N:P export ratios4. We added stoichiometric line with the needs of their local competitors. This implies that the diversity to the plankton in our model by extending this empirically ecological niche of diazotrophs is determined not only by local com- derived relationship between plankton biogeography and biomass N:P petition with high N:P plankton, but also by remote diatom-dominated (see Methods), while ensuring a global mean nutrient export ratio of communities with a lower N:P quota. These communities largely 16N:1P (Fig. 1c, Supplementary Figs 3 and 4). Because diatoms are occupy different ocean biomes, so their stoichiometric signatures must abundant in equatorial and high latitudes but scarce in the subtropics, be communicated over long distances by ocean circulation. the mean N:P of plankton that compete directly with diazotrophs— To illustrate the role of circulation in controlling SN/SP, we employ denoted RO,ST—is close to 20. This is consistent with the elemental a three-box model12 (Fig. 3a) in which the circulations that transport composition of the cyanobacteria that dominate oligotrophic waters19,20, the observed N:P of organic matter in the North Pacific Subtropical 21 Gyre , and theoretical predictions of high N:P allocation strategies 19 during resource competition22. 18 Constant R The introduction of stoichiometric diversity allows the model to O 17 R achieve the observed SN/SP values across a wide range of plausible Variable O Slope 1 denitrification and Fe-limitation scenarios (Fig. 1d). This increase is 16 15

driven by the high N:P requirements of oligotrophic phytoplankton, Σ N/ P which exacerbates N limitation in the subtropical gyres. A larger excess 14 32 of PO4 then remains to fuel N2 fixation, expanding the niche of 13 Slope 0 diazotrophs beyond that created by subsurface denitrification. This 12 allows a balanced N budget to be achieved at higher values of SN/SP. 16 17 18 19 20 21 R Under weak Fe limitation and low denitrification rates, SN/SP actually O,ST exceeds the average N:P of marine plankton (Fig. 1d)—a condition that Figure 2 | Response of SN/SP to the degree of stoichiometric diversity, cannot be attained in previous box-model depictions. The ocean’s ratio 21 RO,ST. Each simulation has 150 Tg N yr denitrification and no Fe limitation. of nutrient reservoirs thus depends not only on the mean N:P of its If the ‘set point’ of the nutrient thermostat were determined through local plankton, but also on their stoichiometric diversity. competition only, SN/SP would rise with RO,ST along a line of slope 1, yet a To investigate the sensitivity of SN/SP to different levels of much weaker response is observed in our model. Grey shading represents an stoichiometric diversity, we varied the N:P quota of diatom and non- estimate of error for the slope of this line, derived using different estimates of diatom endmember communities, while maintaining a constant diatom abundance as the basis for RO (see Methods and Supplementary Notes).

a 24 UW ST b F 0 R O O 22 Constant O Variable R 0.25 Ψ 20 O Slope 1 M M 18 0.5

R R Σ N/ P O,UW < 16 O,ST > 16 16 0.75 = Ψ /( + M ) Ψ 14 1 f Slope 0 12 16 17 18 19 20 21 22 23 24 25 R O,ST

Figure 3 | Role of ocean circulation illustrated in a three-box model. stoichiometry, which we vary by raising RO,ST and reducing RO,UW to maintain a, Structure of the model, with two surface regions (upwelling UW; subtropics a mean of 16. b, The response of SN/SP to stoichiometric diversity depends on ST) and two circulation pathways (vertical exchange M; overturning Y). fY, the fraction of subtropical source waters that first pass through UW. Diazotrophs are restricted to ST, and other plankton have diverse nutrients between the surface and deep ocean, and between diatom- currents and shallow overturning, creates a ‘biogeochemical tele- dominated upwelling regions (UW) and downwelling subtropical connection’ by which remote low N:P communities reduce the 32 gyres (ST), can be abstracted and manipulated (see Methods). When availability of PO4 to fuel N2 fixation. This counteracts the expanded surface nutrients are supplied only from vertical exchange (M) with niche of diazotrophs produced by local competition with high-N:P the deep ocean, the niche of diazotrophs is governed by their local subtropical communities, offsetting roughly half of the upwards competitors only, so every change in RO,ST produces an equal change pressure on SN/SP (Fig. 4b). Ocean circulation thus plays a critical in SN/SP, even though the global mean RO is anchored at 16 (Fig. 3b, role in the nutrient thermostat, reducing the bias of SN/SP towards fY 5 0). However, when the nutrients are predominantly supplied the stoichiometry of subtropical communities, and holding the through shallow overturning and near-surface lateral circulations nutrient content of sea water closer to the global average requirements 2 (Y), the signature of low N:P uptake—a higher residual NO3 : of phytoplankton. 32 PO4 ratio in surface waters—is transported directly from diatom- The ratio of oceanic N and P reservoirs is a simple but powerful dominated upwelling regions into the subtropics. This reduces the observational constraint on the dynamics of the marine N cycle. It 32 excess PO4 available to diazotrophs, and must be compensated by appears to be fundamentally incompatible with a universal Redfield 2 32 lower NO3 :PO4 in upwelling deep water to maintain a given N2- ratio of plankton biomass, lending global support to the large-scale fixation rate. The response of the ocean’s SN/SP to increases in RO,ST association between biogeography and plankton nutrient metabolism is thus damped as lateral circulations strengthen relative to vertical inferred from Southern Ocean nutrient data4. exchange. In the extreme case where all subtropical nutrients pass first The modern ocean’s SN/SP also places new bounds on global through surface communities with low N:P ratios (Fig. 3b, fY 5 1), the denitrification rates and the limitations to diazotroph growth. ocean N reservoir is entirely independent of spatial variations in RO, Denitrification rates at the upper end of the estimated range and SN/SP is regulated through the global-mean N:P of plankton, as (.250 Tg N yr21) are unable to yield the observed SN/SP ratio, even originally hypothesized by Redfield1. with a high degree of stoichiometric diversity (Fig. 1d). At the same In light of these box model results, the tendency of SN/SP in the time, the net effect of stoichiometric diversity among plankton taxa is GCM to track only about 40% of a change in RO,ST (Fig. 2) suggests that to expand the ecological niche of marine diazotrophs. In geochemical 23,24 about half the waters reaching subtropical sites of N2-fixation are first estimates of N2 fixation based on surface nutrients , this would influenced by low N:P plankton communities outside the subtropics. translate into higher diagnosed N2-fixation rates in the subtropics, The dependence of global SN/SP on the spatial pattern of plankton not lower rates24. Thus, stoichiometric diversity helps to close the N:P ratios can be computed by introducing taxon-dependent devia- long-standing gap between estimates of N sources and sinks25. tions from a constant-Redfield N:P (Fig. 1c, Supplementary Fig. 4) one From a mechanistic perspective, the expanded niche for diazotrophs grid cell at a time (see Methods). Regions can be divided between those yields a higher SN/SP for a given denitrification rate, but this is still whose stoichiometric deviations tend to raise SN/SP compared to the insufficient to achieve observed SN/SP when diazotrophs are strongly Redfieldian case (reddish positive values, Fig. 4a), and those that tend limited by airborne Fe (Fig. 1d). At most, the overall growth-rate to reduce it (bluish negative values, Fig. 4a). Large negative values are handicap of diazotrophs can approach 50%, and then only if found in the northern and equatorial Pacific and, to a lesser extent, in denitrification rates are at the lower end of the estimated range. If the polar regions of the Southern Ocean. The transport of nutrients the intrinsic cost of diazotrophy were greater than the conservative from these source regions into the subtropics, through surface Ekman value we use (mF/mO 5 0.95), or if diazotroph growth is also slowed by

a 4 b 3 2 60° N 2 ) –3 30° N 1 1 0° 0 R O < 16 0

R ΔΣ N/ Σ P 30° S –1 O > 16 Net ΔΣ N/ Σ P ( × 10 60° S –2 –1

–4 –2 0° 60° E 120° N 180° 120° W 60° W 0°

Figure 4 | Influence of individual surface regions on SN/SP. a, Values of b, Integral of DSN/SP over regions of high RO (.16) and low RO (,16), and DSN/SP represent the change in steady-state SN/SP (from the Redfield case) over the entire global domain (‘Net’). High- and low-RO regions exert opposite prompted by introducing the grid cell’s deviation of RO from 16 pressures on the ocean N reservoir, with a net increase in SN/SP over the (Supplementary Fig. 4), while holding RO 5 16 elsewhere (see Methods). Redfield case.

20 SEPTEMBER 2012 | VOL 489 | NATURE | 421 ©2012 Macmillan Publishers Limited. All rights reserved RESEARCH LETTER other factors not included here6, then the observed SN/SP would 8. Codispoti, L. A. in Productivity of the Ocean: Past and Present (eds Berger, W. H., Smetacek, V. S. & Wefer, G.) 377–394 (John Wiley and Sons, 1989). require even weaker Fe limitation (see Supplementary Notes). These 9. Ward, B. B. et al. Denitrification as the dominant nitrogen loss process in the findings imply a secondary role for atmospheric Fe deposition in Arabian Sea. Nature 461, 78–81 (2009). controlling rates of N2 fixation in the modern ocean. 10. Lam, P. & Kuypers, M. M. M. Microbial nitrogen cycling processes in oxygen minimum zones. Annu. Rev. Mar. Sci. 3, 317–345 (2011). Temporal changes in the mean N:P of plankton have been 11. Lenton, T. M. & Watson, A. J. Redfield revisited. 1. Regulation of nitrate, phosphate, hypothesized to drive long-term trends in ocean fertility and carbon and oxygen in the ocean. Glob. Biogeochem. Cycles 14, 225–248 (2000). storage, by shifting the set point of its nutrient thermostat26,27. Our 12. Lenton, T. M. & Klausmeier, C. A. Biotic stoichiometric controls on the deep ocean model shows that this set point is also controlled by the biogeography N: P ratio. Biogeosciences 4, 353–367 (2007). 13. Quigg, A. et al. The evolutionary inheritance of elemental stoichiometry in marine of distinct plankton taxa, and ocean circulation patterns that transport phytoplankton. Nature 425, 291–294 (2003). nutrients between biomes—two factors known to vary with climatic 14. Green, S. E. & Sambrotto, R. N. Plankton community structure and export of C, N, P conditions. Future stratification of the upper ocean and the expansion and Si in the Antarctic Circumpolar Current. Deep Sea Res. II 53, 620–643 (2006). of oligotrophic biomes expected under a warming climate28 could 15. Arrigo, K. R. et al. Phytoplankton community structure and the drawdown of nutrients and CO2 in the Southern Ocean. Science 283, 365–367 (1999). reshape the ecological niche of diazotrophs, and initiate a long-term 16. Berman-Frank, I., Cullen, J. T., Shaked, Y., Sherrell, R. M. & Falkowski, P. G. Iron perturbation in the ocean’s nutrient thermostat. availability, cellular iron quotas, and nitrogen fixation in Trichodesmium. Limnol. Oceanogr. 46, 1249–1260 (2001). 17. Kustka, A., Carpenter, E. J. & Sanudo-Wilhelmy, S. A. Iron and marine nitrogen METHODS SUMMARY fixation: progress and future directions. Res. Microbiol. 153, 255–262 (2002). We used an observationally constrained ocean GCM (ref. 29) with horizontal 18. DeVries, T., Deutsch, C., Primeau, F., Chang, B. & Devol, A. Global rates of water- resolution of 4u 3 4u and 24 vertical layers, and simulated tracer transport using column denitrification derived from nitrogen gas measurements. Nature Geosci. 5, the transport matrix method30. A simple ecosystem model was adopted from 547–550 (2012). 19. Heldal, M., Scanlan, D. J., Norland, S., Thingstad, F. & Mann, N. H. Elemental refs 5 and 12, but modified for a three-dimensional global domain. The model 2 32 composition of single cells of various strains of marine Prochlorococcus and includes four prognostic variables: NO3 ,PO4 , ‘general’ phytoplankton, and Synechococcus using X-ray microanalysis. Limnol. Oceanogr. 48, 1732–1743 diazotrophic phytoplankton. Plankton growth rates vary as a function of temper- (2003). ature, light, and nutrient concentrations, and the parameters governing these 20. Bertilsson, S., Berglund, O., Karl, D. M. & Chisholm, S. W. Elemental composition of relationships were tuned to optimize surface nutrient distributions (Supplemen- marine Prochlorococcus and Synechococcus: implications for the ecological tary Fig. 5). Diazotroph growth rates are scaled by atmospheric dust deposition to stoichiometry of the sea. Limnol. Oceanogr. 48, 1721–1731 (2003). 21. Karl, D. M. et al. Ecological nitrogen-to-phosphorus stoichiometry at station represent a heightened requirement for Fe in maintaining the nitrogenase enzyme. ALOHA. Deep Sea Res. II 48, 1529–1566 (2001). The strength of this scaling is varied to simulate differing degrees of Fe limitation. 22. Klausmeier, C. A., Litchman, E., Daufresne, T. & Levin, S. A. Optimal nitrogen-to- The spatial pattern of denitrification is governed by the degradation of organic phosphorus stoichiometry of phytoplankton. Nature 429, 171–174 (2004). matter in benthic grid cells and those where observed oxygen concentrations fall 23. Deutsch, C., Sarmiento, J. L., Sigman, D. M., Gruber, N. & Dunne, J. P. Spatial below 5 mM. The fluxes are then scaled in order to vary the global rate of N loss in a coupling of nitrogen inputs and losses in the ocean. Nature 445, 163–167 (2007). controlled manner between model scenarios. Each simulation was initialized with 24. Mills, M. M. & Arrigo, K. R. Magnitude of oceanic nitrogen fixation influenced by the nutrient uptake ratio of phytoplankton. Nature Geosci. 3, 412–416 (2010). observed nutrient distributions and integrated for at least 10,000 years, until the N 21 25. Codispoti, L. A. Biogeochemical cycles—is the ocean losing nitrate? Nature 376, budget was balanced to within 0.1 Tg N yr . Stoichiometric diversity among 724 (1995). phytoplankton was parameterized using observed distributions of silicic acid to 26. Falkowski, P. G. Rationalizing elemental ratios in unicellular algae. J. Phycol. 36, estimate the contribution of diatoms to nutrient export, and assuming different 3–6 (2000). biomass N:P ratios for diatoms and other non-diazotrophic taxa. For each simu- 27. Broecker, W. S. & Henderson, G. M. The sequence of events surrounding lation, the value of R was calculated as the mean N:P ratio of plankton Termination II and their implications for the cause of glacial-interglacial CO2 O,ST changes. Paleoceanography 13, 352–364 (1998). communities cohabiting surface grid boxes with diazotrophs at steady state. 28. Polovina, J. J., Howell, E. A. & Abecassis, M. Ocean’s least productive waters are expanding. Geophys. Res. Lett. 35, L03618 (2008). Full Methods and any associated references are available in the online version of 29. DeVries, T. & Primeau, F. Dynamically and observationally constrained estimates the paper. of water-mass distributions and ages in the global ocean. J. Phys. Oceanogr. 41, 2381–2401 (2011). Received 5 November 2011; accepted 27 June 2012. 30. Khatiwala, S. A computational framework for simulation of biogeochemical tracers in the ocean. Glob. Biogeochem. Cycles 21, doi:10.1029/2007GB002923 (2007). 1. Redfield, A. C. The biological control of chemical factors in the environment. Am. Sci. 46, 205–221 (1958). Supplementary Information is linked to the online version of the paper at 2. Redfield, A. C., Ketchum, B. H. & Richards, F. A. in The Sea Vol. 2 (ed. Hill, M. N.) www.nature.com/nature. 26–77 (Interscience, 1963). 3. Deutsch, C. & Weber, T. Nutrient ratios as a tracer and driver of ocean Acknowledgements We thank T. DeVries for providing the ocean circulation model. This work was funded by a NASA Earth Systems Science Fellowship (T.W.) and a grant biogeochemistry. Annu. Rev. Mar. Sci. 4, 113–141 (2012). 4. Weber, T. S. & Deutsch, C. Ocean nutrient ratios governed by plankton from the Gordon and Betty Moore Foundation (C.D.). biogeography. Nature 467, 550–554 (2010). Author Contributions T.W. developed the model and performed simulations and 5. Tyrrell, T. The relative influences of nitrogen and phosphorus on oceanic primary analyses. Both authors designed the study and wrote the paper. production. Nature 400, 525–531 (1999). 6. Karl, D. et al. Dinitrogen fixation in the world’s oceans. Biogeochemistry 57/58, Author Information Reprints and permissions information is available at 47–98 (2002). www.nature.com/reprints. The authors declare no competing financial interests. 7. Capone, D. G., Zehr, J. P., Paerl, H. W., Bergman, B. & Carpenter, E. J. Readers are welcome to comment on the online version of this article at Trichodesmium, a globally significant marine cyanobacterium. Science 276, www.nature.com/nature. Correspondence and requests for materials should be 1221–1229 (1997). addressed to T.W. ([email protected]).

METHODS Plankton N:P ratios. We assume that stoichiometric variability occurs primarily at the taxonomic level, and use the empirical relation derived by ref. 4: Circulation model. We use an observationally constrained ocean GCM which optimizes circulation to fit the linearized momentum equations, and observed ~ z { RO RO,diatwdiat RO,other(1 wdiat) ð7Þ distributions of temperature and salinity, and 14C (ref. 29). It has horizontal resolution of 4u 3 4u, and 24 vertical layers including two in the top 75 m. Here, wdiat is the fractional contribution of diatoms to nutrient export, and RO,diat Annual-mean flow fields are extracted as a matrix, A, facilitating tracer simula- and RO,other are the biomass N:P ratios of diatoms and other non-diazotropic tions using the transport matrix method30. phytoplankton respectively. Rather than simulate different taxonomic groups Ecosystem model. We adopt a simple ecosystem model (similar to refs 5 and 12) explicitly, we use a prior estimate of wdiat combined with equation (7) to apply a that includes four prognostic variables, representing the concentrations spatially varying pattern of RO to the single class of non-diazotrophic plankton. 2 32 Three different approaches are considered for estimating wdiat (Supplementary (in mM) of NO3 (N), PO4 (P), ‘general’ phytoplankton (O), and diazotrophic phytoplankton (F). Both phytoplankton types are simulated as organic phosphorous Fig. 3). Method 1 assumes that the relative abundance of diatoms scales with pools. The variables are governed by: the observed surface concentration of Si(OH)4 (ref. 35): ½Si dO ~ P N { w ~ ð8Þ mO min , O MO ð1Þ diat z dt PzKP NzKN ½Si KSi

dF P KSi is tuned to accommodate the observational constraint that diatoms contribute ~m F{MF ð2Þ 40–50% of global export production36. Methods 2 and 3 compute w from the dt F PzK diat P relative export fluxes of N and Si and an estimate of the Si:N ratios in diatom 37 38 dP biomass . Method 2 is based on observations only , diagnosing export fluxes ~AP { ðÞJO,UP z JF,UP z Qrem(M(O z F)) ð3Þ from the vertical gradients of Si and N between the thermocline and surface, dt whereas Method 3 diagnoses the fluxes in an ocean GCM36. We note that the dN two diagnostic methods are less appropriate in regions where N2 fixation con- ~AN { R J , z Q (M(R O z R F)) {D ð4Þ dt O O UP rem O F founds the diagnosis of N export. We used the simplest approach (Method 1), which agrees most closelywithsatellite-derived estimates ofdiatom biogeography39. The parameters of the ecosystem model are discussed below and their numeric Methods 2 and 3 are used to derive an estimate of uncertainty associated with the values listed in Supplementary Table 1. community-composition parameterization. Growth and mortality. The first terms on the right hand side of equations (1) and In our initial simulations (Fig. 1), R and R are held close to the values (2) represent plankton growth, as a function of environmental factors. The O,diat O,other diagnosed in ref. 4, with the added constraint that the mean N:P export ratio by maximum growth rate of O (m ) is given by: O non-fixing plankton is equal to the Redfield ratio of 16:1 (Supplementary Fig. 4). In ~ { 0 { { mO(T,I) mopt exp (k(T T ))(1 exp ( I=KI )) ð5Þ later simulations (Fig. 2), we vary RO,diat and RO,other to produce different degrees of stoichiometric diversity, but again ensure the global-mean constraint is satisfied. Here, m is the growth rate under optimal conditions, and k, T9,K,K and K opt I P N This allows us to identify changes in SN/SP that are caused only by changes in the control the sensitivity of growth to temperature (T), light (I) and nutrient con- spatial pattern of R , and not its global-mean value. The ratio of N fixation to P centrations. Sensitivity parameters are tuned to reproduce observed surface O 2 uptake by diazotrophs (R ) is assumed to be constant, but as in previous studies12, nutrient distributions31 in the model, ensuring a realistic pattern of biological F our results are not sensitive to the value of this parameter. nutrient drawdown (Supplementary Fig. 5). We account for the competitive Sensitivity testing. We rigorously tested the sensitivity of our results to handicap of diazotrophs by reducing their maximum growth rate (m ) relative F parameters and assumptions of the ecosystem model. See Supplementary Notes, to general plankton. It is scaled by a constant factor (d ), representing an intrinsic F Supplementary Figs 7–10 and Supplementary Table 2. energetic expenditure on nitrogenase activity, and by a Fe-limitation parameter to Three-box model. We use a three-box model of the ocean to assist our interpreta- represent the heightened requirement by diazotrophs for Fe: tion of the ocean GCM results. The geometry and nutrient fluxes of the model are ~ JFe shown in Supplementary Fig. 6. Its ecosystem and biogeochemistry components mF(T,I,Fe) mOdF ð6Þ JFezKFe are held as close to the ocean GCM version as possible for ease of comparison, with the following simplifications: JFe is the simulated distribution of atmospheric Fe deposition onto the surface 32 (1) Phytoplankton growth rates are set to the optimal value in ST, and reduced by a ocean , and the strength of the Fe-limitation is varied through KFe (Supplementary Fig. 1). In equations (1) and (2), M represents phytoplankton factor of 0.5 in the UW, so that residual nutrients remain in the surface as mortality, and includes a quadratic term that scales with total biomass observed. (2) The competitive handicap of diazotrophs is determined only by dF in ST (no Fe (M 5 m11m2B, where B 5 O 1 F, and m1 and m2 are rate constants) and can be thought of as representing grazing by zooplankton, which are not explicitly limitation), and mF is set to zero in UW. simulated. (3) Denitrification is distributed between the surface and deep ocean as in ref. 5. Nutrient cycling. Nutrients are transported by the circulation operator (matrix Regional control of SN/SP. For each surface grid cell (with x,y coordinates i,j), a A), and are assimilated into biological pools in the top 75 m through plankton simulation was conducted in which the local RO was set to its stoichiometrically diverse value, as computed from wdiat (Supplementary Fig. 4). In all other surface growth. In equations (3) and (4), JO,UP and JF,UP are the same as the first terms on regions, RO was held equal to the Redfield ratio. The difference between the steady- the right hand side of equations (1) and (2) respectively. RO is the biomass N:P state SN/SP in this simulation, and that in a uniform Redfieldian case, was then ratio of general phytoplankton, and RF is the amount of N fixed per unit P uptake computed: by diazotrophs. Following phytoplankton mortality, the recycling and reminera- ~ ~ ~ lization of organic matter is simulated using the operator Qrem. The majority is SN RO f (wdiat), x i, y j SN DSN=SPjx~i~ { ðÞðRO~16 9Þ recycled in the surface ocean, and restored to local inorganic pools. A small ~ y j SP RO~16, x=i, y=j SP fraction (we) is exported from the surface layers as organic particles, and remineralized over depth following a power-law relation33. This value is taken as a measure of the sensitivity of SN/SP to the uptake stoichi- N budget. Newly fixed N is assumed to derive from an abundant dissolved N2 pool ometry of plankton communities in the perturbed surface region. The efficiency of that is not simulated explicitly. In equation (4), D represents the sum of water- these computations was enhanced using a quasi-steady-state assumption for the 2 column and sediment denitrification, which are simulated as sinks of NO3 . ‘fast’ biological variables O and F, which reduces the model to a two-equation Water-column denitrification is proportional to the remineralization rate of system that can be solved directly for steady state using Newton’s method40. organic matter in grid cells with climatological oxygen concentrations below a Sensitivity testing demonstrated that the solutions derived from the quasi- critical threshold [O2]crit. Sediment denitrification is determined by the flux of steady-state-assumption approach and full four-equation model were almost organic matter to seafloor grid cells34. Because global denitrification rates are one indistinguishable. of the primary determinants of the steady-state N inventory, but are not well constrained observationally, D is scaled to maintain a specified global rate, and 31. Garcia, H. E., Locarni, R. A., Boyer, T. P. & Antonov, J. I. World Ocean Atlas 2005 Vol. 4 Nutrients (phosphate, nitrate, silicate) (US Government Printing Office, 2006). a constant partition among water column and sediments, thus controlling for 32. Mahowald, N. M. et al. Change in atmospheric mineral aerosols in response to these factors between simulations. Simulated distributions of N sources and sinks climate: last glacial period, preindustrial, modern, and doubled carbon dioxide are shown in Supplementary Fig. 2. climates. J. Geophys. Res. 111, doi:10.1029/2005JD006653 (2006).

33. Martin, J. H., Gordon, R. M., Fitzwater, S. & Broenkow, W. W. VERTEX: phytoplankton/ 37. Brzezinski, M. A. et al. A switch from Si(OH)4 to NO3-depletion in the glacial iron studies in the Gulf of Alaska. Deep-Sea Res. 36, 649–680 (1989). Southern Ocean. Geophys. Res. Lett. 29, 1564 (2002). 34. Middelburg, J. J., Soetaert, K., Herman, P. M. J. & Heip, C. H. R. Denitrification in 38. Sarmiento, J. L. & Gruber, N. Ocean Biogeochemical Dynamics (Princeton University marine sediments: a model study. Glob. Biogeochem. Cycles 10, 661–673 (1996). Press, 2006). 35. Egge, J. K. & Aksnes, D. L. Silicate as regulating nutrient in phytoplankton 39. Alvain, S., Moulin, C., Dandonneau, Y. & Loisel, H. Seasonal distribution and competition. Mar. Ecol. Prog. Ser. 83, 281–289 (1992). succession of dominant phytoplankton groups in the global ocean: a satellite view. 36. Jin, X., Gruber, N., Dunne, J. P., Sarmiento, J. L. & Armstrong, R. A. Diagnosing the Glob. Biogeochem. Cycles 22, doi:10.1029/2007GB003154 (2008). contribution of phytoplankton functional groups to the production and export of 40. Kwon, E. Y. & Primeau, F. Optimization and sensitivity study of a biogeochemistry particulate organic carbon, CaCO3, and opal from global nutrient and alkalinity ocean model using an implicit solver and in situ phosphate data. Glob. Biogeochem. distributions. Glob. Biogeochem. Cycles 20, doi:10.1029/2005GB002532 (2006). Cycles 20, doi:10.1029/2005GB002631 (2006).

1. Supplementary Notes

1.1 SENSITIVITY TESTING

We tested the sensitivity of our results to changes in model parameters that could affect ΣN/ΣP. We discuss the significance of each parameter and the test results below.

Distribution of denitrification The distribution of denitrification controls how rapidly denitrified waters are transported to sites of N2-fixation, so influences the strength of feedbacks between N sources and sinks. Anoxic regions of the water column occur in upwelling margins, which are an important source of nutrients to the surface of the subtropical Pacific and Indian Oceans. - - The NO3 :PO4 signature of water-column denitrification is thus efficiently transported to diazotrophic habitats, prompting a strong response from N2-fixation. Regions of deep- ocean sedimentary denitrification, however, are separated over much longer circulation pathways from diazotrophic habitats, and should elicit a weaker response from N2- fixation. We would therefore expect a higher steady-state N reservoir in an ocean where a greater fraction of total denitrification occurs in the water column (i.e. high FWC), rather than sediments. We tested the sensitivity to this parameter by varying FWC between 0 (sediment only) and 1 (all water column). Although our model exhibits the expected increase in ΣN/ΣP at high FWC, the response was relatively weak: ΣN/ΣP changes by ~0.5 over the whole range of simulations (Fig. S7). Isotopic constraints suggest that FWC is confined to the range 0.2-0.5, within which ΣN/ΣP varies by just 0.1. The inability of our Redfieldian (constant Ro) simulations to reproduce observed ΣN/ΣP values cannot therefore be attributed to the distribution of denitrification.

Additional N budget terms Our default simulations assume that N2-fixation and denitrification are the only sources and sinks of fixed N to and from the ocean. For a given rate of denitrification, additional sources of N would allow a balanced N budget to be achieved at lower N2-fixation rates, reducing the requirement for excess PO4 in the deep ocean. Additional sources of N should therefore raise steady-state ΣN/ΣP in our model. For a pre-industrial ocean, the most important of these sources was atmospheric deposition from lightning, which is thought to have supplied no more than 20TgN/yr to the ocean41. Incorporating an additional N source of this magnitude into our model, evenly distributed over the surface ocean, had a negligible effect on its N reservoir (Fig. S8). The simplifications made to the N budget in our default simulations are therefore not responsible for the low predicted ΣN/ΣP under Redfieldian conditions.

Diatom scheme The response of ΣN/ΣP to stoichiometric diversity in our model may be sensitive to the inferred distribution of diatoms upon which Ro is patterned. Using Method 1 (Si-based, see Methods) we found that for any increase in Ro,ST , ΣN/ΣP increased ~40% as much. We tested the robustness of this finding by repeating our experiments, but substituting

WWW.NATURE.COM/NATURE | 1 RESEARCH SUPPLEMENTARY INFORMATION

Methods 2 and 3 to estimate diatom distributions (see Methods). In each case, ΣN/ΣP raised as stoichiometric diversity became more pronounced, and reached observed values within the estimated range of global denitrification rates (Fig. S9). The response of ΣN/ΣP to Ro,ST was only slightly stronger using Method 2 (~50%), and slightly weaker using Method 3 (~35%), than in our default simulations. Our conclusions are thus robust against the precise distribution of diatoms assumed in simulations, as long as the broad, well-known, pattern of their biogeography is properly represented.

Dissolved organic matter dynamics Other non-Redfieldian processes may contribute to enhancing PO4 availability and raising ΣN/ΣP towards its observed value. For example, a faster re-supply of P than N from dissolved organic matter in the subtropics would expand the niche of diazotrophs relative to the Redfield case. We tested whether such dynamics might contribute to resolving the discrepancy of low ΣN/ΣP in the Redfield model by including explicit DON and DOP pools using two different formulations. The first assumes that the two pools behave similarly, with production rates proportional to their biological quotas and equal decay rates (“DOM Equal” scenario). The second accounts for potential differences in the dynamics of the pools. High quality measurements of dissolved organic matter are scarce, but the available data suggests a faster turnover (higher production and a faster decay rate) for DOP than for DON. Ref 23 derived the best-fit production and decay parameters to reproduce the observed distributions of DON and DOP in the Pacific. We use these parameters in our second formulation (“DOM Unequal” scenario). A comparison between steady-state ΣN/ΣP in our original simulations and those with explicit DOM cycles (Table S2) reveal no large changes under the parameter sets tested here, suggesting that these processes cannot reproduce the observed ΣN/ΣP.

Fe cycle In simulations discussed in the main text, the iron (Fe) limitation of diazotroph growth was parameterized as a simple function of Fe deposition onto the surface ocean (see Methods). We found that enhanced Fe limitation tended to exacerbate N loss in our model, enhancing the need for stoichiometric diversity of phytoplankton in order to reconcile observed and simulated N reservoirs. We tested whether the simplicity of the Fe parameterization may have biased this conclusion by repeating selected simulations in another version of our model, which includes a fully explicit and mechanistic model of the ocean Fe cycle42. Parameters are taken from ref 43, except that the Fe:P ratio of non- diazotrophic plankton (300µM/mol) is tuned for reproduction of surface nutrient distributions, and the Fe:P ratio of diazotrophs is varied (between 1-100 times that of non-diazotrophs) to represent different degrees of Fe limitation. We found that strengthening the Fe limitation of diazotrophs produced even larger N deficits in this version of the model (Fig. S10), supporting our conclusion that non-Redfieldian uptake ratios are necessary to explain the observed ΣN/ΣP.

Ocean circulation model We tested whether the discrepancy noted in our Redfieldian model – a low simulated N reservoir – was robust between ocean circulation models. We repeated a Redfieldian simulation (150TgN/yr denitrification, no Fe-limitation) in a finer resolution (2°x2°)

2 | WWW.NATURE.COM/NATURE SUPPLEMENTARY INFORMATION RESEARCH

version of the physical model described by ref. 29, in which flow fields are further constrained by observed CFC distributions. At steady state, this simulation yielded a ΣN/ΣP value of ~13.25, very similar to the value of 13.4 predicted by the coarse resolution model. This suggests that our results are relatively insensitive to the exact configuration of the physical model, so long as the large-scale nutrient transport pathways are well represented.

1.2 ADDITIONAL CONSIDERATIONS

Our initial findings of low ΣN/ΣP in a Redfieldian ocean, even when Fe-limitation of diazotrophs is relieved, should be considered conservative. In our model, diazotrophs are prescribed a very small intrinsic (non Fe-dependent) handicap to their growth rate, relative to other phytoplankton (µF/µo=0.95). Most studies assume a much larger “penalty” for N2-fixation, and impose greater sensitivity to other environmental conditions (such as temperature and light) on diazotrophs, as supported by observations. For example, during the IRONAGES project the diazotroph Trichodesmium was found to have average growth rates less than 10% that of its non-diazotrophic competitors (e.g. ref 44). Accounting for additional limiting factors in our model would further compound the discrepancy in ΣN/ΣP, requiring even lower rates of denitrification, weaker Fe-limitation, or greater variability in plankton N:P ratios to reproduce the observed value.

WWW.NATURE.COM/NATURE | 3 RESEARCH SUPPLEMENTARY INFORMATION

2. Supplementary Figures and Tables

Parameter Symbol Fixed Value Range Units Plankton growth rate under -1 optimal conditions µopt 1 day Temperature sensitivity parameter k 0.02 °C-1 Optimal growth temperature T' 30 °C -2 Light sensitivity parameter KI 10 Wm Intrinsic growth-rate handicap of diazotrophs δF 0.95 dimensionless Fe-sensitivity of diazotroph 2 growth rate KFe 0-0.5 g/m /yr Phosphate half-saturation constant KP 0.1 µM Nitrate half-saturation constant KN KP ×Ro µM -1 Linear mortality constant m1 0.1 day -1 -1 Quadratic mortality constant m2 10 µM day Si concentration at which diatoms become dominant KSi 8 µM

N:P ratio of diatoms Ro,di 5-16 mol/mol

N:P ratio of other non-fixers Ro,other 16-25 mol/mol

Si:N ratio of diatoms RSi:N 1-5 mol/mol

N:P ratio of diazotrophs Rf 50 mol/mol

Depth of euphotic zone z0 75 m

Export ratio φe 0.1 dimensionless Shape of Martin curve b 0.858 dimensionless critical oxygen concentration for water column denitrification [O2]crit 5 µM Fraction of total denitrification that occurs in 0.33 0 - 1 water column FWC (main text) (sensitivity testing) dimesionless

Table S1. Parameters of ecosystem and biogeochemistry model. See methods for governing equations.

Scenario No DOM DOM Equal DOM Unequal ΣN/ΣP 13.17 13.23 13.31

Table S2. Influence of DOM dynamics on ΣN/ΣP. Simulations have denitrification rates of 150TgN/yr and moderate Fe-limitation.

4 | WWW.NATURE.COM/NATURE SUPPLEMENTARY INFORMATION RESEARCH

Figure S1. Maps showing maximum growth rate of diazotrophs relative to other phytoplankton (µF/µO) for 2 two different Fe-limitation scenarios. a, Weak Fe-limitation: KFe = 0.05g/m /yr, mean µ’F/µʼO≈0.9. b, 2 Strong Fe-limitation: KFe = 0.25g/m /yr, mean µF/µO≈0.55.

Figure S2. Model distributions of N2-fixation (a,b) and denitrification (c,d). Fixation is shown for weak (a) and strong (b) Fe-limitation scenarios. Denitrification is shown separately for water-column (c) and sedimentary (d) fractions. Total fixation/denitrification is 150TgN/yr in each case.

WWW.NATURE.COM/NATURE | 5 RESEARCH SUPPLEMENTARY INFORMATION

Figure S3. Fraction of nutrient export attributed to diatoms (φdiat), estimated using three different methods: a, Method 1 scales the surface distribution of Si(OH)4; b, Method 2 uses the vertical gradients of N and Si between the surface and thermocline; c, Method 3 diagnoses φdiat from nutrient fluxes in an OGCM. Each of the three can be combined with biomass N:P ratios of diatom and non-diatom end-members to derive a spatial pattern of Ro.

Figure S4. Pattern of Ro applied to non-fixing phytoplankton in simulations used for Fig. 1 of main text. Derived using pattern of φdiat from Fig. S3a.

6 | WWW.NATURE.COM/NATURE SUPPLEMENTARY INFORMATION RESEARCH

Figure S5. Comparison between simulated and observed (World Ocean Atlas 2005) distributions of PO4 (a,b) and NO3 (c,d). Unlike PO4, the total inventory of NO3 deviates from observations in many of our simulations, leading to discrepancies between observed and simulated NO3 distributions. However, in simulations that reproduce a realistic N inventory, surface NO3 is close to observations (c,d), indicating that our ecosystem model produces realistic patterns of nutrient drawdown.

Figure S6. Schematic depicting geometry, circulation patterns, and nutrient fluxes of 3-box model. Adapted from ref 3.

WWW.NATURE.COM/NATURE | 7 RESEARCH SUPPLEMENTARY INFORMATION

Figure S7. Sensitivity of ocean N reservoir to the partition of denitrification between water column and sediments. The shaded region represents isotopic constraints on FWC – within this range, changes in the N reservoir are very small.

Figure S8. Sensitivity of ocean N reservoir to additional N sources to the ocean, representing N deposition by lightning. These sources are considered to account for less than 20TgN/yr input in the pre-industrial era, and have little influence on ΣN/ΣP in our model.

Figure S9. Sensitivity of N reservoir to method for inferring diatom distribution. Different methods yield slightly different response of ΣN/ΣP to Ro,ST, but do not change our conclusions. These results are used for the error estimate in Fig. 2 of the main text.

8 | WWW.NATURE.COM/NATURE SUPPLEMENTARY INFORMATION RESEARCH

Figure S10. Sensitivity of N reservoir to diazotroph Fe limitation in a model with full, mechanistic Fe cycle. The Fe limitation factor is the factor by which diazotroph Fe:P requirements are increased above those of general phytoplankton. Each simulation has 150TgN/yr denitrification.

SUPPLEMENTARY REFERENCES

Refs 1-30 refer to references in main text, and 31-40 refer to references in the Methods section.

41. Brandes, J.A., Devol, A.H., & Deutsch, C. New developments in the marine nitrogen cycle. Chemical Reviews 107, 577-589 (2007). 42. Parekh, P., Follows, M.J., & Boyle, E. Modeling the global ocean iron cycle. Global Biogeochemical Cycles 18 (2004). 43. Follows, M.J., Dutkiewicz, S., Grant, S., & Chisholm, S.W. Emergent biogeography of microbial communities in a model ocean. Science 315, 1843- 1846 (2007). 44. LaRoche, J. & Breitbarth, E. Importance of the diazotrophs as a source of new nitrogen in the ocean. Journal of Sea Research 53, 67-91 (2005).