New Production Regulates Export Stoichiometry in the Ocean

The proportion in which carbon and growth-limiting nutrients are exported from the oceans’ productive surface layer to the deep sea is a crucial parameter in models of the biological carbon pump. Based on >400 vertical flux observations of particulate organic carbon (POC) and nitrogen (PON) from the European Arctic Ocean we show the common assumption of constant C:N stoichiometry not to be met. Exported POC:PON ratios exceeded the classical Redfield atomic ratio of 6.625 in the entire region, with the largest deviation in the deep Central Arctic Ocean. In this part the mean exported POC:PON ratio of 9.7 (a:a) implies c. 40% higher carbon export compared to Redfield-based estimates. When spatially integrated, the potential POC export in the European Arctic was 10–30% higher than suggested by calculations based on constant POC:PON ratios. We further demonstrate that the exported POC:PON ratio varies regionally in relation to nitrate-based new production over geographical scales that range from the Arctic to the subtropics, being highest in the least productive oligotrophic Central Arctic Ocean and subtropical gyres. Accounting for variations in export stoichiometry among systems of different productivity will improve the ability of models to resolve regional patterns in carbon export and, hence, the oceans’ contribution to the global carbon cycle will be predicted more accurately.


Introduction
The concept of new vs. regenerated primary production sensu Dugdale and Goering (1967) [1] has profoundly shaped the understanding of biological production and carbon sequestration in the ocean. The revelation that new production based on allochthonous nutrients supplied to the surface layer approximates export of organic matter to the deep ocean [2] has been widely adopted as a law of nature in ecosystem models used to quantify the contribution of the oceans' biological carbon pump to the global carbon cycle. While export from a steady-state system must be balanced by an equivalently large input of new nutrients, recycling of carbon and growth-limiting nutrients by the pelagic food web profoundly impacts the potential of a system to export carbon [3,4]. Because organic nitrogen is remineralised faster than carbon in large parts of the ocean [5], the common assumption of constant elemental stoichiometry (e.g. Redfield) for converting nitrogen export to carbon typically underestimates the strength of the biological carbon pump and therefore carbon export (e.g. [6]). Assuming that C:N is constant over spatially heterogeneous environments also constrains the ability of models to resolve regional patterns in carbon export, which are crucial for understanding ecosystem functioning. Model experiments with a depth-dependent particulate organic C:N ratio significantly increased carbon sequestration compared to a constant ratio model [7]. Furthermore, carbon overconsumption in response to increased atmospheric CO 2 is suggested to increase the spread of suboxic regions in the ocean due to respiration of excess carbon exported from the surface layer [8]. These findings suggest that our ability to predict how perturbations to the ocean environment affect the marine ecosystem in part depend on our understanding of the stoichiometric coupling between carbon and nutrients. At present, knowledge of the regulation of exported C:N is limited which poses a constraint to further development of ecosystems models. Because nitrogen recycling within the euphotic zone varies regionally with primary production [2], export stoichiometry and, hence, the efficiency of the biological carbon pump should reflect nutrient availability or trophic state (oligo-, meso-, eutrophy) of pelagic ecosystems.
We use published and unpublished data on vertical flux of particulate organic carbon (POC) and nitrogen (PON) from the European Arctic Ocean to test the hypothesis of constant export stoichiometry. The same methodology was adopted in all studies in this region, which allows comparisons of the data among regions in order to determine spatial patterns. The Arctic Ocean was also not included in a previous synthesis of export stoichiometry [9]. Furthermore, we assess the anticipated relationship between nitrate-based new production as a measure of trophic state and export stoichiometry over large geographical scale based on published data from other primarily nitrogen-limited parts of the World Ocean.

Data Collection Analytical Methods
Vertical export in the European Arctic Ocean was determined by means of short-term deployed surface-tethered drifting sediment traps at a total of 64 stations or occasions (Fig. 1). Between 1 and 14 sediment traps were used to collect sinking particulate matter in the 20-200 m depth range. A typical array consisted of traps situated at 20, 30, 40, 50, 60, 90, 120 and 200 m depth, and was attached to a surface-tethered float or a drifting ice floe and deployed for circa 24 h without preservatives. The sediment traps consisted of duplicate parallel cylinders (height:diameter ratio 6.25) mounted in a gimballed frame equipped with a vane. The cylinders maintain vertical orientation perpendicular to the current direction at the moderate current velocities prevailing under all deployments [10]. The trapping efficiency for particulate organic carbon is 90-110%, determined by comparison with the 234 Thorium method, and the traps therefore provide reliable estimates of particle fluxes [10].The observations were a priori grouped into seven sub-regions based on the dominant water mass, ice conditions, and bottom depth ( Table 1). The observations are temporally constrained to the productive season (March-August), but the temporal resolution and distribution of observations over the season vary among regions.
Particulate organic carbon (POC) and nitrogen (PON) were determined in GF/F filter samples of the sediment trap content using an elemental CHN analyser. The filters were dried and exposed to HCl fumes in order to remove inorganic carbon prior to analyses [11]. POC and PON values were corrected for background contamination determined in blank filters. POC and PON fluxes in units of mass were converted to molar unit using the atomic weights of C and N.
Exported POC:PON and new production from other parts of the World Ocean were obtained from the literature. Only observations from the upper 500 m of the water column were included in order to allow comparisons with the data from the Arctic Ocean (Table 2). New production in the European Arctic Ocean was simulated [6] and published values from other regions were mostly based on nitrate inventories ( Table 2). Data on new production were in most cases available in units of carbon based on conversion from nitrate uptake to carbon using either observed DIC:NO 3 drawdown ratios or Redfield C:N. Where new production was available only in units of nitrogen (W Sargasso Sea), it was converted to carbon using the Redfield C:N ratio.

Statistical Analyses
Exported POC:PON data were not normally distributed wherefore regional variations and deviations from Redfield were assessed by a nonparametric Kruskal-Wallis test with the Multiple Comparisons post hoc test in StatisticaH and a Wilcoxon signed rank test in R (www.r-project.org), respectively. Linearity of the POC:PON ratio was assessed by the type two regression Standard Major Axis of log-transformed POC on log-transformed PON flux using the package lmodel2 in R because the variation was uncontrolled in both directions, and to overcome the bias associated with ordinary least square regression when testing the slope against a given value (in this case 1) [12]. In log-log space, a slope equal to one implies that C and N occur in the same proportions across the whole range of C and N export values, and the SMA regression was used to test this hypothesis, following Sterner et al. (2008) [12]. Slopes were considered to be significantly different from one, indicating deviation from constancy, when the 95% confidence interval did not include the value one. The relationship between mean POC:PON and new production was assessed by a linear regression following assessments of normal distributions (Shapiro-Wilk tests, P = 0.69 and P = 0.39, respectively). The Svalbard shelf break area was excluded from the regression due to strong advection of organic matter reflective of a more southerly Atlantic production regime into this area by the West Spitsbergen Current [13,14], material that contributes to vertical export in addition to autochthonous production.

Results and Discussion
Regional Patterns in the European Arctic Ocean and was higher in the Central Arctic Ocean than elsewhere (P,0.01, Multiple Comparisons post hoc test) except over the NE Greenland shelf (P = 0.062). Variations were significant over short spatial scales relating to geophysical boundaries shown by the higher POC:PON ratio in the seasonally ice-covered northern Barents Sea than in the southern part dominated by warmer (.0u) Atlantic Water (P = 0.002). This pattern corresponds to the higher abundance of diatoms and flagellates (Phaeocystis sp.) in the phytoplankton communities of the Arctic and Atlantic sections, respectively [11]. Diatoms are exported to greater extent than Phaeocystis sp. [15] and represent a carbon-rich fraction of the exported organic matter likely owing to their high lipid content. Seasonal biases in the data, e.g. towards pre-bloom conditions in late winter in the Fram Strait and over the NE Greenland shelf and towards summer in the Central Arctic Ocean, could further increase the differences among regions.
The mean exported POC:PON was above Redfield (6.625, a:a) in all seven regions (Wilcoxon signed rank test, P,0.0001 in all cases). Conversion of nitrogen-based export estimates to POC using the Redfield ratio therefore underestimates carbon export in the European Arctic Ocean, consistent with most other oceanic regions [9]. The error is highest in the Central Arctic Ocean where the mean exported POC:PON of 9.7 implies that POC export is c. 40% (i.e. 9.7/6.6) higher than Redfield-based estimates. In order to estimate the impact of variable stoichiometry on annual POC export in the European Arctic Ocean, potential POC export was calculated by multiplying spatially integrated annual new production by the POC:PON ratio of the exported material. The potential POC export is 208 Tg C yr 21 for the entire area when using the observed POC:PON ratios specific to each region (''Observed Variable C:N'', Table 3). This is almost 30% higher than potential POC export when applying the Redfield C:N ratio (165 Tg C yr 21 ), and c. 10% higher than estimates based on a constant C:N ratio of 7.6 as used in the model described in Wassmann et al. 2006 [6]. Regional variations in the stoichiometry of exported organic matter therefore significantly impacts modelled POC export, and accounting for this variation is an improvement from constant stoichiometry models. However, these estimates would benefit further from more data from the Central Arctic Ocean which has relatively few observations and which also represents the largest area.  Table 4). This suggests that   Table 2. Mean exported POC:PON ratio (mean6SD (SE)) of particulate organic matter sampled by means of sediment traps, and mean annual new production (NP) in regions included in Fig. 4, with reference to source. nitrogen recycling is weaker during high-flux events corresponding to spring-bloom conditions prior to nitrogen depletion compared to low-flux events when detrital material likely comprises a larger fraction of the exported material. Among individual regions, the NW Norwegian Shelf revealed a depletion of POC relative to PON (slope = 0.832, Fig. 3H) whereas an equally strong enrichment was observed in the Central Arctic Ocean (slope = 1.22, Fig. 3B). In three regions the exported POC:PON ratio was constant (i.e. SMA slope not significantly different from one). Factors acting at regional scale therefore contribute to shaping export stoichiometry. In addition to variations in the relative contribution of dominant phytoplankton taxa (diatoms and flagellates, see above) among regions, organic matter released from sea ice represent another carbon-rich fraction of vertically exported material in ice-covered areas [16][17][18] that could partially explain the enrichemnt in POC over PON in the Central Arctic Ocean and northern Barents Sea. Colonisation of sinking particles by bacteria and heterotrophic flagellates, which was observed in a temperate upwelling system, would have an opposite effect by increasing the nitrogen content of initially nitrogen-deprived organic matter [39]. The same process could occur in the Arctic Ocean as well, but this remains speculative until more detailed data become available. In combination with dissolution and degradation of particulate organic matter, these factors contribute to shaping elemental fluxes from the water column and, therefore, to the net carbon export efficiency in terms of POC:PON stoichiometry.

Global Patterns and Regulation of Export Stoichiometry
From a modelling point of view where regional variations in carbon export integrated over time is of primary interest, a unifying concept for regulation of export stoichiometry and, hence, carbon export efficiency, is desirable. We tested the hypotheses that nitrate-based new production is a driver of export stoichiometry across systems characterised by different trophic state (Table 2). A linear regression of the mean exported POC:PON ratio in each region on new production revealed a significant negative relationship (R 2 = 0.66, P = 0.0004, Fig. 4), with exported POC:PON decreasing by 0.28 per mole carbon produced based on nitrate uptake. Standard errors of the slope and intercept were 0.0585 and 0.2811, respectively. Albeit the variation in exported POC:PON is high (range of ca 1) for any new production rate, the regression suggests stronger nitrogen recycling in nutrient-poor oligotrophic environments, resulting in higher carbon export per unit limiting nutrient as compared to more nutrient-rich mesotrophic environments. More efficient nitrogen recycling under oligotrophic regions is governed by the pelagic food web structure, which is characterized by an ''inverted biomass pyramid'' (i.e. heterotrophic to autorophic biomass ratio .1), resulting in consumer-control of nutrient pools and flows [19,20]. Under eutrophic conditions where autotrophs dominate, nitrogen recycling is less efficient which again favours export of the accumulated organic matter [20]. However, even in the most productive regions the C:N ratio of exported particulate organic  Table 3. Spatially integrated potential export of particulate organic carbon in Tg C yr 21 in the European Arctic Ocean (excluding the NW Norwegian shelf) calculated based on different C:N ratios. matter exceeds the Redfield ratio ( Table 2). The observed relationship is consistent with the overall increasing proportion of PON over POC with increasing vertical flux (Fig. 3A), since the highest fluxes typically are recorded during phytoplankton blooms when new production is highest. Regions where primary production is controlled by other factors than nitrate deviate from this pattern. This is exemplified in Fig. 4 by a High Nutrient Low Chlorophyll area of the subarctic North Pacific where nitrate never is depleted in the surface layer [21], and the W Sargasso Sea where phosphorus excerts stronger control of primary production than nitrogen [22]. Hence, exceptions to what appears to be a general dependence of exported POC:PON on new production are explained by different controls of primary production in these oceanographic regions, and dominance of other phytoplankton functional groups [23]. Since nitrate limits primary production in most parts of the world's oceans [24], nitrate-based new production emerges as a driver of export stoichiometry across climatically-defined domains (arctic, temperate, tropics).

Conclusions
This study shows that the C:N ratio of exported particulate organic matter is consistently above Redfield, implying that biogeochemical models based on Redfield stoichiometry underestimate POC export in the ocean. Potential POC export in the European Arctic Ocean is 10-30% higher when accounting for the variation in exported POC:PON compared to estimates based on constant stoichiometry. This study further shows that the exported POC:PON ratio varies among nitrogen-limited systems depending on new production, reflecting the system's trophic state. Hence, the error induced by constant stoichiometry is largest in nutrient-poor oligotrophic systems, which represent the major part of the world's oceans. Scaling export stoichiometry to trophic state improves the representation of the biological carbon pump in models, and more realistic POC export estimates are thus achieved. This is essential in order to correctly predict pathways of primary production in marine ecosystems, particularly in response to changes in inorganic nutrient supply and productivity.  Table 4. doi:10.1371/journal.pone.0054027.g003   Table 2 for abbreviations for individual sites. Regions shown for comparison but which are not included in the regression (in grey) are Svalbard, NE subarctic Pacific (OSP) and W Sargasso Sea (BATS). doi:10.1371/journal.pone.0054027.g004