Environmental forcing of nitrogen fixation in the eastern tropical and sub-tropical North Atlantic Ocean.

During the winter of 2006 we measured nifH gene abundances, dinitrogen (N(2)) fixation rates and carbon fixation rates in the eastern tropical and sub-tropical North Atlantic Ocean. The dominant diazotrophic phylotypes were filamentous cyanobacteria, which may include Trichodesmium and Katagnymene, with up to 10(6) L(-1)nifH gene copies, unicellular group A cyanobacteria with up to 10(5) L(-1)nifH gene copies and gamma A proteobacteria with up to 10(4) L(-1)nifH gene copies. N(2) fixation rates were low and ranged between 0.032-1.28 nmol N L(-1) d(-1) with a mean of 0.30 ± 0.29 nmol N L(-1) d(-1) (1σ, n = 65). CO(2)-fixation rates, representing primary production, appeared to be nitrogen limited as suggested by low dissolved inorganic nitrogen to phosphate ratios (DIN:DIP) of about 2 ± 3.2 in surface waters. Nevertheless, N(2) fixation rates contributed only 0.55 ± 0.87% (range 0.03-5.24%) of the N required for primary production. Boosted regression trees analysis (BRT) showed that the distribution of the gamma A proteobacteria and filamentous cyanobacteria nifH genes was mainly predicted by the distribution of Prochlorococcus, Synechococcus, picoeukaryotes and heterotrophic bacteria. In addition, BRT indicated that multiple a-biotic environmental variables including nutrients DIN, dissolved organic nitrogen (DON) and DIP, trace metals like dissolved aluminum (DAl), as a proxy of dust inputs, dissolved iron (DFe) and Fe-binding ligands as well as oxygen and temperature influenced N(2) fixation rates and the distribution of the dominant diazotrophic phylotypes. Our results suggest that lower predicted oxygen concentrations and higher temperatures due to climate warming may increase N(2) fixation rates. However, the balance between a decreased supply of DIP and DFe from deep waters as a result of more pronounced stratification and an enhanced supply of these nutrients with a predicted increase in deposition of Saharan dust may ultimately determine the consequences of climate warming for N(2) fixation in the North Atlantic.

Introduction

Nitrogen is a key nutrient, limin class="Chemical">ting primary production throughout much of the world's upper oceans [1]. In tropical and sub-tropical oligotrophic oceanic environments, biological fixation of dinitrogen (N2) provides an important source of new nitrogen for primary production and carbon export [2]. Nevertheless, our knowledge of the diversity, abundance and distribution of diazotrophs (N2 fixing micro-organisms) is limited and the factors that control N2 fixation in the marine environment are still poorly understood [3].

Until recently, the majority of N2 fixation studies in the Atlantic Ocean have focused solely on n class="Species">Trichodesmium [4]. However, with the increasing application of molecular genetic analyses, more information is becoming available on the diversity, abundance and distribution of diazotrophs [5], [6]. It is now clear that a broad suite of diazotrophs inhabit the oceans including diatom endosymbionts, Crocosphaera and other uncultured small unicellular diazotrophs, e.g. group A and C cyanobacteria, gamma A proteobacteria, and cluster III nifH phylotypes [5], [7], [8], [9]. Furthermore, large scale patterns in the distribution of diazotrophs are recognized [10]. In the western part of the North Atlantic Ocean Trichodesmium biomass and N2 fixation rates are reported to be high [4], [10], whereas in the eastern part of the North Atlantic N2-fixing unicellular cyanobacteria are reported to be responsible for a significant part of the N2 fixation [3], [11], [12].

Several envn class="Chemical">ironmental factors have been reported to control N2 fixation by Trichodesmium. Laboratory experiments and field observations suggest that N2 fixation by Trichodesmium is limited to water temperatures between 20°C–32°C [13]. However, co-variation of low oxygen, low nutrients and high light, due to a more strongly stratified water column, could underlie the influence of temperature on N2 fixation [14]. N2 fixation in natural populations dominated by Trichodesmium can be controlled by phosphorus or iron availability [15], [16], [17] or both [18]. Additionally, enhanced concentrations of the nutrients nitrate (NO3 -), ammonium and organic nitrogen sources including urea and glutamate may inhibit N2 fixation of Trichodesmium [19], [20], [21].

By contrast the environmentn class="Chemical">al factors that control the abundance and activity of unicellular diazotrophs are unknown. Langlois et al. (2008) reported that most diazotrophic phylotypes in the North Atlantic Ocean are almost completely restricted to regions with NO3 - concentrations <0.5 µM and are limited to warmer seawater temperatures (20–30°C). The majority of the diazotrophic community in general and the uncultured group A unicellular cyanobacteria in particular, have been observed in regions with seawater temperatures of about 22°C. The uncultured γ-proteobacterium A group (gamma A) and filamentous group (Trichodesmium spp.) have been observed in conditions with mean temperatures of ca. 24 and 25°C, respectively [3]. However, N2 fixation by marine unicellular diazotrophs at seawater temperatures as low as 15–19°C has been reported [22], [23].

The tropical and sub-tropicn class="Chemical">al North Atlantic Ocean, in the vicinity of the Cape Verde islands, is a region receiving enhanced Saharan dust inputs [24] and some of the highest N2 fixation rates and diazotroph abundances have been reported here [3], [4], [10], [11], [25], [26]. To investigate which environmental variables determine the distribution of the dominant diazotrophs and N2 fixation rates in this area we measured nifH gene abundance, N2 fixation and CO2 fixation rates, and an extensive set of environmental variables including nanomolar concentrations of phosphate (PO4 3-) and NO3 - + NO2 - (hereafter termed NO3 -), dissolved Fe (DFe) and dissolved Al (DAl) as a proxy of dust inputs, and a diverse set of biological variables. We used boosted regression trees analysis (BRT) to investigate the relative contribution of the environmental variables in explaining the distribution of diazotrophs and the rates of N2 fixation.

Materials and Methods

Sampling

This study was conducted during a cruise in the vicinity of the Cape Verde islands (26 January to 26 February 2006) on board the research vessel FS Posein class="Chemical">don (cruise P332) (Fig. 1). Surface seawater was pumped into a trace metal clean laboratory container using a Teflon diaphragm pump (Almatec A-15, Germany) connected by an acid-washed braided PVC tubing to a towed fish positioned at ca. 3 m depth alongside the ship. Unfiltered seawater was collected for N2 fixation measurements in an acid-cleaned 25 L low density polyethylene carboy (Nalgene). Simultaneously, material was collected by vacuum filtration (0.2 bar) of 1.5–2 L onto 0.22 µm Durapore (Millipore) filters for analysis of nifH genes. Within 10 minutes after sampling, these filters were stored at −80°C until extraction in the laboratory. Samples for analysis of DFe and DAl, Fe-binding ligands, and nanomolar PO4 3- and NO3 - were collected from the towed fish and filtered in-line using a filter capsule (Sartorius, Sartobran 300) with a 0.2 µm filtration cut-off. Samples for total dissolved phosphate (TDP), total dissolved nitrogen (TDN) and ammonium (NH4 +) were sampled from the surface water bottles of the CTD rosette frame. Oxygen concentrations were measured using a Seabird sensor on the CTD frame. Photosynthetic variables were measured using active fluorescence measurements performed by fast repetition rate fluorometry (FRRF) using a Chelsea Instruments FASTtracka FRRF (Chelsea Technologies Group, UK) mounted on the CTD frame. Samples for Chlorophyll a (Chl a, 0.7 µm GF/F filtered and stored at −80°C), dissolved silicate (SiO4 4-) and flow cytometry were taken from the non-toxic underway surface seawater supply (sampling depth ca. 4 m). Underway temperature and salinity were determined using a thermosalinograph (Meerestechnik Elektronic, Germany).

Figure 1

The cruise track of the P332 cruise in January-February 2006.

N2 and C fixation rates

Rates of N2 fixation (n class="Chemical">15N uptake) and CO2 fixation (13C uptake) were measured using the stable isotopes of N (15N2) and C (H13CO3 -) [27]. All collection and incubation steps were carried out using trace-metal clean techniques. N2- and CO2 fixation rates were determined in an unfiltered, non-concentrated homogeneous surface seawater sample collected in a 25 L carboy, and subsequently incubated in quadruplicate in gas tight 4 L polycarbonate bottles (Nalgene) equipped with Teflon lined butyl septum caps. Any air bubbles were removed from the bottles following the filling process. Trace quantities of 15N2 (99 atom %, Cambridge Isotope Laboratories) were added (4 mL, 9% of ambient N2) using a gas-tight syringe (Chromatographie-Zubhör, Germany). The H13CO3 - solution (99 atom %, Brand company) was added to a final concentration of 25 µM or 1.2% of ambient TCO2 as H13CO3 - [16]. Incubations were performed on deck and the temperature was controled with circulation of surface seawater. At night, incubators were covered with a black plastic sheet to protect the incubations against the influence of the ship's lights and during daytime the light was attenuated to 20% surface irradiance using blue filters (Lagoon blue #172, Lee Filters, UK). Underway seawater samples were typically collected in the morning and evening but on occasions in the afternoon. Samples were typically incubated for a minimum of 24 h and a maximum of 31.8 h.

All incubations were terminated by on-board filtration under gentle vacuum through a pre-combusted GF/F filter (Whatman, UK). Filters were frozen (−20°C) and stored for further ann class="Chemical">alysis. At the home laboratory the filters were dried at 60°C, acid-fumed with concentrated HCl to remove carbonates, and then stored over desiccant until analysis. Each filter was packed into a tin capsule and pelletized for elemental and isotopic analysis.

Carbon and n class="Chemical">nitrogen total mass and isotope ratios were measured at the Stable Isotope Laboratory at Stanford University, Stanford, CA (USA) using a continuous-flow mass spectrometer (Finnegan Delta Plus XL, Germany), coupled to an elemental analyzer (Costech ECS 4010, USA). Acetanilide was used as a mass calibration standard. Isotope values were calibrated using international carbon (IAEA-CH-6) and nitrogen (IAEA-NO-3) reference materials with assigned δ 13C and δ 15N values of −10.4‰ and +4.7‰. Blank corrections were made following [28]. Analytical precision (1 standard deviation) was±0.2‰ for δ15N and±0.1‰ for δ13C. Carbon and nitrogen isotope ratios are reported in ‰ relative to Vienna–PeeDee belemnite for δ13C and atmospheric N2 for δ15N. The N2- and CO2 fixation rates were calculated by isotope mass balance [27]. The solubility of N2 in seawater was calculated according to Hamme and Emerson [29]. Removal of outliers (1.5 times below or above the first and third quartile, respectively) resulted in 65 N2 fixation rates and combined with missing values in 54 carbon fixation rates at 18 stations.

Reported averages are accompanied by “n”, where “n” is either the number of individual incubation bottles (bottles) in which the variable was measured for each replicate bottle or the number of experiments (experiments) where the same vn class="Chemical">alue for an environmental variable was given to all replicates of one experiment.

Nutrients, trace metals and environmental data

Values for n class="Chemical">DFe and the organic complexation of Fe were taken from Rijkenberg et al. [30]. Dissolved Al (<0.2 µm) was determined using the fluorometric lumogallion method with a spectrofluorometer (model Aminco, American Instruments Co.) [31]. NO3 - and PO4 3- were measured at nanomolar concentrations using a system comprised of a conventional segmented-flow autoanalyser connected to two 2-metre liquid waveguide capillary cells (WPI Inc, USA) and using miniaturized spectrophotometric systems (Ocean Optics Inc., USA) [32]. NO3 - and PO4 3- were determined colorimetrically using the sulphanilamide-NEDD and molybdenum blue methods, respectively, achieving detection limits of 1.5 nmol L−1 NO3 - and 0.8 nmol L−1 PO4 3- [32]. Micromolar concentrations of silicate were measured on a Scalar Sanplus autoanalyser [33]. Ammonium was measured at nanomolar concentrations based on the reaction of NH4 + with orthophtaldialdehyde in the presence of sulphite [34]. Reagents were added immediately after collection of the samples and fluorometric analysis (excitation at 370 nm and emission at 420 nm) was conducted on a Turner Design fluorometer (TD700) after a 24 hours incubation.

Seawater samples for measurement of TDN were filtered using combusted (450 °C, 4–6 h) glass-fibre filters (Whatman, GF/F). The filtrate was transn class="Chemical">ferred to a combusted (450 °C, 4–6 h) glass ampoule and stabilised by acidification to pH 2 using hydrochloric acid and subsequently flame-sealed. The ampoules were stored at 4°C until analysis. TDN was measured using high-temperature combustion on a Shimadzu TOC 5000A total carbon analyser (Shimadzu Corp, Japan) coupled with a Sievers NCD 255 nitrogen chemiluminescence detector (Sievers Instruments, Inc, US) [35]. DON concentrations were calculated by subtracting the ammonium and NO3 - + NO2 - concentrations from the TDN concentrations.

Seawater samples for totn class="Chemical">al dissolved phosphorus (TDP) were filtered using glass-fibre filters (Fisherband® MF 300; nominal pore size 0.7 µm) and subsequently irradiated using a UV lamp to oxidise organic phosphorus compounds to PO4 3- which was subsequently measured by colorimetry using a Skalar Sanplus autoanalyser according to Kirkwood [33]. Dissolved organic phosphorus (DOP) was calculated by subtracting phosphate concentration from the measured TDP.

Chl a was determined on duplicate 500 mL seawater samples filtered through 25 mm diameter glass-fiber filters (Fisherbrand MF 300). Filters were frozen at −80°C until onboard ann class="Chemical">alysis by fluorometry. Samples were extracted in 7 mL of 90% acetone for 24 h at 4°C; Chl a concentrations (>0.7 µm) in the extracts were measured using a TD-700 Turner Designs fluorometer following calibration with fresh Chl a standard (Sigma, UK).

Seawater samples of 1.6 mL were fixed with a finn class="Chemical">al concentration of 1% paraformaldehyde for 24 hours at 4°C and subsequently frozen at −80°C until processed by flow cytometry (FACSCalibur, Becton Dickinson, BD Biosciences, Oxford, UK). Two groups of cyanobacteria, Prochlorococcus spp. and Synechococcus spp, and a broad group of picoeukaryotes were identified by their characteristic autofluorescence [36]. The heterotrophic prokaryotes (Bacteria and Archaea) were counted after staining of the whole microbial population with the nucleic acid stain SYBR Green I and subsequent subtraction of the autofluorescent cells [36].

qPCR

The nifH abundances of filamentous (Trichodesmium spp.), unicellular Group A, and Gamma A n class="Chemical">diazotrophs were determined using TaqMan primers and probe sets as described in Langloiset al. (2008). qPCR reactions contained 1x TaqMan master mix (Applied Biosystems), 40 ng µl−1 bovine albumin (BSA), 5 pmol µL−1 each forward and reverse primers, 100 nmol L−1 TaqMan probe, and 5 µl DNA (average of 4 ng DNA). Plasmid standards, described in Langlois et al. (2008), were run in duplicate, as were template controls (NTC). Samples were run in triplicate. All reactions were run on an ABI Prism 7000 instrument (Applied Biosystems) using the default cycling program with 45 cycles. Raw data were analyzed using the ABI 7000 system SDS software (version 1.2.3) with RQ study application. Primer amplification efficiencies were 97% for filamentous and Group A and 95% for Gamma A; calculated using the formula E = 10−1/slope-1 [37]. No amplification was observed in the NTCs, thus setting the potential detection limit to 1 copy L−1. When the elution and filtration volumes were accounted for the actual detection limit was 50 copies L−1.

Environmental variables

Environmentn class="Chemical">al variables were either collected at the time of sampling or taken from the nearest stations or underway sample points. The environmental variables available for analysis included: i) oceanographic variables such as oxygen, salinity, the water-mass as identified by a temperature salinity plot, and the mixed layer depth [25], ii) as factor the dust event on 3&4 Feb 2006, iii) NO3 -, PO4 3-, SiO4 4-, NH4 +, DON and DOP, iv) biological variables including the photosynthetic efficiency, Fv/Fm, and the PSII cross-section (σPSII), flow cytometer counts of heterotrophic prokaryotes, picoeukaryotes, Prochlorococcus, Synechococcus and the nifH gene abundance of gamma A proteobacteria, unicellular group A and filamentous cyanobacteria, and the N2 fixation rates, and v) trace metals DAl and DFe, free and total concentrations of Fe-binding ligands and their conditional stability constant log K'. Incubation time and the time of 15N2-tracer addition were also considered in the model to check whether these influenced the N2 fixation rate response variable.

Boosted regression trees analysis (BRT)

BRT was used to identify which environmentn class="Chemical">al variables could describe (1) log nifH gene abundances of gamma A proteobacteria, (2) filamentous and group A cyanobacteria, (3) total nifH gene abundance (sum of group A proteobacteria, gamma A and filamentous cyanobacteria), and (4) N2 fixation rates [38], [39]. As the data was continuous and normally distributed, the model was fit using a Gaussian error distribution and link function [38]. BRT is a relatively new statistical technique and is based on a combination of regression trees and boosting. Boosting increases the emphasis on poorly modelled observations and iteratively fits regression trees to the data. BRT has advantages over standard techniques in that it can fit complex non-linear relationships (Elith et al. 2008) and can deal with missing values. Following the model simplification procedure of Elith et al. (2008), we identified the variables that provided the best model performance. Model performance was assessed using a Pearson's correlation and deviance measures, i.e. goodness of fit (Elith et al. 2008). All evaluation statistics were calculated with 6-fold cross- validation and the BRT models were run 30 times to ensure stable estimates of model evaluation (Elith et al. 2008). Replicates of experimental treatments were placed into a single fold so that the models could be evaluated on independent treatment data. All BRT models were fitted in R (v2.6.0, www.Rproject.org) using the ‘gbm’ library [40].

Results and Discussion

Study area

Our study area was situated between 12–30°N and 20–30°W, with most of the N2- and n class="Chemical">CO2 fixation experiments performed in proximity of the Cape Verde islands. The surface waters to the north and south of the Cape Verde islands consisted of the westward flowing North Equatorial Current (NEC) and North Equatorial Counter Current (NECC), respectively [41]. The surface seawater temperature varied between 18°C in the vicinity of the Canary Islands and up to 24.8°C at ∼12°N, south of the Cape Verde islands (Fig. 2A). The salinity of the surface seawater varied between 35.62 and 37.36 southeast and northwest of the Cape Verde islands, respectively (Fig. 2B).

Figure 2

The sea surface (A) temperature (°C) and (B) salinity during the P332 cruise.

Nitrogen fixation rates and primary productivity

Winter N2 fixation rates ranged between 0.032–1.28 nmol N L−1 d−1 with an overn class="Chemical">all mean of 0.30±0.29 nmol N L−1 d−1 (n = 65 bottles) (Fig. 3). N2 fixation rates were lowest north and east of the Cape Verde islands (0.21±0.15 nmol N L−1 d−1, range 0.032–0.70 nmol N L−1 d−1, n = 55 bottles) and highest to the south and west of the Cape Verde islands (0.91±0.31 nmol N L−1 d−1, range 0.35–1.28 nmol N L−1 d−1, n = 10 bottles) (Fig. 3 and 4A). Using acetylene reduction assays, Staal et al. [12] also observed low winter time N2 fixation rates of 0.154±0.091 nmol N L−1 d−1 in light controlled (200 µm m−2 s−1) incubations southeast of the Cape Verde islands. Autumn N2 fixation rates between 2.4–151.2 nmol N L−1 d−1 (non-size fractionated and non-trace metal clean sampled) [26] have been reported for the same study region. Unicellular N2 fixation rates of 3.48 nmol N L−1 d−1 have been reported in other regions of the tropical and sub-tropical North Atlantic Ocean [42]. Whereas Voss et al. [11] reported occasional high rates of N2 fixation of up to 75 nmol N L−1 d−1 in surface waters (non-size fractionated and non-trace metal clean sampled during a high dust deposition event in autumn) along 10°N in the eastern North Atlantic, though most values were below 6 nmol N L−1 d−1. The higher N2 fixation rates as measured by Voss et al. [11] coincided with enhanced N2 fixation rates and diazotroph abundance along 10°N as reported by Moore et al. [16] and Staal et al. [12]. N2 fixation rates in the eastern Atlantic Ocean along ca. 10°N may be higher as a result of warmer seawater temperatures, lower oxygen concentrations and a higher supply of Fe and phosphate due to Saharan dust inputs [16]. However, differences in season and location may complicate direct comparison of N2 fixation rates.

Figure 3

Bubble plot of the average N2 fixation rates (nmol L−1 d−1) based on 3 to 4 replicates.

Figure 4

The environmental parameters measured during the P332 cruise in the Cape Verde region.

The (A) N2 fixation rate, (B) CO2 fixation rate, (C) total nifH gene abundance, (D) filamentous nifH gene abundance, (E) group A nifH gene abundance, (F) gamma A nifH gene abundance, (G) NO3 - + NO2 -, (H) NH4 +, (I) DON, (J) PO4 3-, (K) DOP, (L) DAl, (M) Prochlorococcus spp. abundance, (N) Synechococcus spp. abundance, (O) picoeukaryote abundance, (P) heterotrophic prokaryote abundance, (Q) Chl a, (R) σPSII, (S) Fv/Fm, (T) DFe, (U) total Fe-binding ligands, (V) O2, (W) temperature as a function of latitude.

Primary productivity was measured in the same bottles as N2 fixation and ranged between 0.07–1.59 µmol C L−1 d−1 with a mean of 0.55±0.34 µmol C L−1 d−1 (n = 54 bottles) (Fig. 4B). Little latitudinn class="Chemical">al variation in rates of primary production was observed as seen for N2 fixation rates. Converting N2 fixation to C-uptake rates via Redfield equivalents showed that the contribution of N2 fixation to primary production was between 0.03–5.24%, with an average of 0.55±0.87% (n = 54 bottles). In our study, the average contribution of N2 fixation to primary production was substantially lower than the values of 5.8–12.2% reported by Voss et al. [11]. Metabolic control of N2 fixation by other nitrogen sources during our study was unlikely. Surface water NO3 - concentrations ranged between 1.5–128 nmol L−1 (Fig. 4G; n = 13 experiments) and were well below the concentrations (>0.5 µmol L−1) reported to inhibit N2 fixation [20], [43]. The ammonium concentrations ranged between 2.3–208 nmol L−1 (n = 15 experiments) and DON between 4.7–7.8 µmol L−1 (n = 15 experiments) (Fig. 4H–I). Fixed nitrogen sources such as urea or ammonium can inhibit N2 fixation in Trichodesmium [19], although in the laboratory Trichodesmium has been reported to regulate ammonium metabolism and N2 fixation within its circadian rhythm [44]. Calculated from primary production and compensated for the nitrogen provided by N2 fixation, the turnover time of the DIN pool (DIN = NO3 - + NO2 - + NH4 +) was less than 1 day in 19 out of 33 incubations. This implies that the DIN pool including N2-fixed nitrogen did not provide sufficient nitrogen to sustain the observed primary production. Access to DON or rapid recycling of nitrogen were likely required for provision of additional nitrogen sources [45].

The (A) N2 fixation rate, (B) n class="Chemical">CO2 fixation rate, (C) total nifH gene abundance, (D) filamentous nifH gene abundance, (E) group A nifH gene abundance, (F) gamma A nifH gene abundance, (G) NO3 - + NO2 -, (H) NH4 +, (I) DON, (J) PO4 3-, (K) DOP, (L) DAl, (M) Prochlorococcus spp. abundance, (N) Synechococcus spp. abundance, (O) picoeukaryote abundance, (P) heterotrophic prokaryote abundance, (Q) Chl a, (R) σPSII, (S) Fv/Fm, (T) DFe, (U) total Fe-binding ligands, (V) O2, (W) temperature as a function of latitude.

The PO4 3- concentrations of the surface n class="Chemical">water used in our N2 fixation incubations ranged between 10–92 nmol L−1, with a mean of 44±27 nmol L−1 (Fig. 4J; n = 13 experiments). Surface dissolved inorganic phosphate concentrations below 1 nmol L−1 were reported in the western North Atlantic Ocean, with DIN:DIP ratios between 20–30, suggesting that PO4 3- may control primary production and N2 fixation rates [46]. However, very low inorganic DIN:DIP ratios with a mean of 2±3.2 (n = 94, range 0–15) in the vicinity of the Cape Verde islands suggest that nitrogen, depending on the availability of DON and nitrogen recycling rates, was the primary limiting nutrient [16], [18]. Indeed, in our N2 fixation incubations the turnover time of the dissolved inorganic phosphate pool (DIP) based on primary production varied between 1.2–36 days (n = 40 bottles). Primary production and N2 fixation in our incubations were unlikely to be controlled by the availability of PO4 3-. Based on the conversion of primary production to PO4 3- demands via a C:P Redfield ratio of 106, PO4 3- concentrations were neither limiting primary production anywhere else in the vicinity of the Cape Verde islands (mean 39±24 nmol L−1 PO4 3- (n = 96), range 10–100 nmol L−1).

The DFe concentrations in the surface waters were on average 0.22±0.05 nmol L−1 (n = 18 experiments) and ranged between 0.11 and 0.33 nmol L−1. The n class="Chemical">Fe requirement for primary production, using a conversion of 60.5 µmol Fe per mol of fixed C as based on average phytoplankton nutrient requirements [47], varied between 4.0 10−3–0.1 nmol L−1 DFe (n = 54 bottles). Thus the DFe concentrations were 2.3–73 times higher than the calculated Fe requirements. Conversion of N2 fixation rates using Redfield stoichiometry into C-uptake according to Voss et al. [11], assuming 28 µmol Fe per mol of fixed C for phototrophic diazotrophs as based on Trichodesmium [48], yield Fe requirements between 6.0 10−6–2.4 10−4 nmol L−1 Fe (n = 65 bottles) which are ca. 104 times lower than the DFe concentrations. Iron:C ratios as high as 450±242 µmol Fe per mol C have been reported for Trichodesmium in Australian waters [15] resulting in Fe requirements between 1.0 10−4–3.8 10−3 nmol Fe (n = 65 bottles) in our incubations, suggesting that even using high Fe:C ratios there still is approximately a 600 fold surplus of DFe. These calculations indicate that diazotrophic activity itself will not induce Fe limitation. However, the large uncertainties in Fe:C ratios which vary with region, growth conditions, phytoplankton population [49], and uncertainties in the biological availability of Fe [50] make it difficult to assess (micro)nutrient control of diazotrophy. In fact, along 10°N in the eastern tropical North Atlantic, Voss et al. [11] determined that Fe concentrations were higher than the estimated requirement for diazotrophy, while at the same time Mills et al. (2004) showed that Fe (in combination with P) stimulated N2 fixation rates in this region. Over 99% of the DFe fraction in the vicinity of the Cape Verde islands was complexed to organic Fe-binding ligands [30]. It is still unknown what fraction of this organically complexed Fe pool is directly or indirectly available for uptake by phytoplankton and diazotrophs [50], [51]. Clearly, calculated quotas and stoichiometric ratios may not apply as well as we assume and more detail regarding the availability to diazotrophs of the organically complexed Fe is required.

Environmental variables predicting nifH gene abundance and N2 fixation rates

BRT analysis was used to investigate which envn class="Chemical">ironmental variables explain the distribution of group A, gamma A, filamentous and total nifH gene abundances (the sum of group A, gamma A and filamentous nifH gene abundances) as well as N2 fixation rates. The group A nifH gene abundance was poorly described by the BRT analysis and is therefore not further discussed. However, BRT performed well for the gamma A and filamentous phyloptypes as well as the total nifH gene abundance and N2 fixation rates with respective cross validated Pearson correlations of 0.94, 0.53, 0.98 and 0.47 (Table 1) and robust model performance for a relatively small data set as shown by small 95% confidence intervals in the mean partial dependence plots based on bootstrapping of 30 BRT runs (Fig. 5– 8).

Table 1

Mean predictive performance of the BRT models and contributions of explanatory variables to the prediction of the N2 fixation rates, the total concentration of nifH genes, the concentration of nifH genes of filamentous cyanobacteria and Gamma A, a gamma-proteobacterium.

dependent vasriable	model characteristics		explanatory variables	Contribution%
total nifH genes	learning rate	0.05	Prochlorococcus	41.3±1.2
	tree complexity	3	Synechococcus	24.5±1.4
	trees fitted	2173±200	picoeukaryotes	19.8±1.2
	total dev.	5.334	DAl	9.4±1.1
	mean cv res. dev.	2.76±0.12	total Fe-binding ligands	3.1±0.7
	cv corr.	0.98±0.01	DFe	1.8±0.3
	D̂2	0.48
Gamma A nifH genes	learning rate	0.01	Synechococcus	23.8±0.7
	tree complexity	3	Prochlorococcus	22.8±0.6
	trees fitted	3697±547	DON	17.3±0.8
	total dev.	8.221	picoeukaryotes	15.5±0.8
	mean cv res. dev.	3.59±0.08	Heterotr. bacteria	14.6±1.0
	cv corr.	0.94±0.01	Group A nif genes	6.2±0.8
	D̂2	0.56
Filamentous nifH genes	learning rate	0.01	CO2 fixation rate	29.0±3.4
	tree complexity	3	DAl	17.6±2.3
	trees fitted	4185±354	Prochlorococcus	15.0±0.9
	total dev.	15.446	picoeukaryotes	13.8±1.1
	mean cv res. dev.	7.08±0.2	temp.	8.1±0.6
	cv corr.	0.53±0.01	Fv/Fm	7.4±0.6
	D̂2	0.54	NH4 +	4.9±0.6
			Gamma A nif genes	4.3±0.6
N2 fixation rates	learning rate	0.05	CO2 fixation rate	21.0±2.4
	tree complexity	3	NO3 - + NO2 -	13.4±1.5
	trees fitted	1820±632	PSII cross-section	11.8±1.0
	total dev.	0.081	phosphate	10.5±0.6
	mean cv res. dev.	0.053±0.003	DON	9.7±0.7
	cv corr.	0.47±0.02	oxygen	9.0±0.6
	D̂2	0.36	DFe	7.7±0.6
			picoeukaryotes	6.7±0.5
			Chl a	5.9±0.6
			temp.	4.3±0.6

The learning rate, tree complexity and number of trees fitted (trees fitted), as well as the total deviance for a saturated model (total dev.), are given for each model. Model performance measures, estimated using 30 model runs and 6-fold cross validation, included mean residual deviance and its standard error, the mean proportion of the total deviance explained (D 2) and the mean Pearson correlation (cv corr.) and its standard error.

Figure 5

Mean partial dependence plots for the environmental variables describing the total nifH gene abundance.

The partial dependence plots with 95% confidence intervals (light grey, based on bootstrapping of 30 BRT runs and indicating robustness of model performance) for the 6 environmental variables best at explaining the variation in total nifH gene abundance (the sum of Group A, gamma A and filamentous nifH genes). The 6 environmental variables are A) Prochlorococcus, B) Synechococcus, C) picoeukaryotes, D) DAl, E) total Fe ligands and F) DFe. The y-axis is centered to have zero mean over the log distribution of the fitted total nifH gene abundance. A common scale is used for the x-axis. Rug plots at inside top of plots show the distribution of sites across that variable, in deciles.

Figure 6

Mean partial dependence plots for the environmental variables describing the gamma A nifH gene abundance.

Mean partial dependence plots with 95% confidence intervals (light grey, based on bootstrapping of 30 BRT runs and indicating robustness of model performance) for the 6 environmental variables best at explaining the variation in gamma A nifH gene abundance. The 6 environmental variables are A) Prochlorococcus, B) Synechococcus, C) heterotrophic prokaryotes, D) picoeukaryotes, E) DON and F) group A nifH genes. The y-axis is centered to have zero mean over the log distribution of the fitted gamma A nifH gene abundance. A common scale is used for the x-axis. Rug plots at inside top of plots show the distribution of sites across that variable, in deciles.

Figure 7

Mean partial dependence plots for the environmental variables describing the filamentous nifH gene abundance.

Mean partial dependence plots with 95% confidence intervals (light grey, based on bootstrapping of 30 BRT runs and indicating robustness of model performance) for the 8 environmental variables which explain the variation in filamentous nifH gene abundance best. The 8 environmental variables are A) C-fix rate, B) DAl, C) Prochloroccocus, D) picoeukaryotes, E) Fv/Fm, F) NH4 +, G) temperature and H) gamma A nifH genes. The y-axis is centered to have zero mean over the log distribution of the fitted filamentous nifH gene abundance. A common scale is used for the x-axis. Rug plots at inside top of plots show the distribution of sites across that variable, in deciles.

Figure 8

Mean partial dependence plots for the environmental variables describing the N2 fixation rates.

Mean partial dependence plots with 95% confidence intervals (light grey, based on bootstrapping of 30 BRT runs and indicating robustness of model performance) for the 10 environmental variables which explain the variation in N2 fixation rates best. The environmental variables are A) C-fix rate, B) phosphate, C) nitrate + nitrite, D) sigma-PSII, E) DON, F) oxygen, G) DFe, H) picoeukaryotes, I) Chl a, and J) temperature. The y-axis is centered to have zero mean over the distribution of fitted N2 fixation rates. A common scale is used for the x-axis. Rug plots at inside top of plots show the distribution of sites across that variable, in deciles.

The partial dependence plots with 95% confidence intervn class="Chemical">als (light grey, based on bootstrapping of 30 BRT runs and indicating robustness of model performance) for the 6 environmental variables best at explaining the variation in total nifH gene abundance (the sum of Group A, gamma A and filamentous nifH genes). The 6 environmental variables are A) Prochlorococcus, B) Synechococcus, C) picoeukaryotes, D) DAl, E) total Fe ligands and F) DFe. The y-axis is centered to have zero mean over the log distribution of the fitted total nifH gene abundance. A common scale is used for the x-axis. Rug plots at inside top of plots show the distribution of sites across that variable, in deciles.

Mean partial dependence plots with 95% confidence intervn class="Chemical">als (light grey, based on bootstrapping of 30 BRT runs and indicating robustness of model performance) for the 6 environmental variables best at explaining the variation in gamma A nifH gene abundance. The 6 environmental variables are A) Prochlorococcus, B) Synechococcus, C) heterotrophic prokaryotes, D) picoeukaryotes, E) DON and F) group A nifH genes. The y-axis is centered to have zero mean over the log distribution of the fitted gamma A nifH gene abundance. A common scale is used for the x-axis. Rug plots at inside top of plots show the distribution of sites across that variable, in deciles.

Mean partial dependence plots with 95% confidence intervn class="Chemical">als (light grey, based on bootstrapping of 30 BRT runs and indicating robustness of model performance) for the 8 environmental variables which explain the variation in filamentous nifH gene abundance best. The 8 environmental variables are A) C-fix rate, B) DAl, C) Prochloroccocus, D) picoeukaryotes, E) Fv/Fm, F) NH4 +, G) temperature and H) gamma A nifH genes. The y-axis is centered to have zero mean over the log distribution of the fitted filamentous nifH gene abundance. A common scale is used for the x-axis. Rug plots at inside top of plots show the distribution of sites across that variable, in deciles.

Mean partial dependence plots with 95% confidence intervn class="Chemical">als (light grey, based on bootstrapping of 30 BRT runs and indicating robustness of model performance) for the 10 environmental variables which explain the variation in N2 fixation rates best. The environmental variables are A) C-fix rate, B) phosphate, C) nitrate + nitrite, D) sigma-PSII, E) DON, F) oxygen, G) DFe, H) picoeukaryotes, I) Chl a, and J) temperature. The y-axis is centered to have zero mean over the distribution of fitted N2 fixation rates. A common scale is used for the x-axis. Rug plots at inside top of plots show the distribution of sites across that variable, in deciles.

The learning rate, tree complexity and number of trees fitted (trees fitted), as well as the total deviance for a saturated model (totn class="Chemical">al dev.), are given for each model. Model performance measures, estimated using 30 model runs and 6-fold cross validation, included mean residual deviance and its standard error, the mean proportion of the total deviance explained (D 2) and the mean Pearson correlation (cv corr.) and its standard error.

The abundances of Prochlorococcus (Fig. 4m), Synechococcus (Fig. 4n) and picoeukaryotes (Fig. 4O) contributed strongly to the explanation of the distribution of the total nifH gene and gamma A nifH gene abundances (Fig. 5 and 6, Table 1). Prochlorococcus and picoeukaryotes were n class="Chemical">also important in explaining filamentous nifH gene abundance (Fig. 7). In all cases nifH genes were more abundant with higher abundance of Prochlorococcus and lower abundance of Synechococcus and picoeukaryotes. High nifH gene abundance coinciding with the non-N2 fixing Prochlorococcus abundance suggests co-occupation of the same environment. Prochlorococcus spp. have a very wide oceanic distribution and, like the majority of the oceanic N2-fixing organisms, thrive at enhanced temperatures with maximum abundances at 26–29°C [52]. Prochlorococcus as well as N2-fixing organisms are most abundant in, but not restricted to, oligotrophic waters while Synechococcus may have high abundances in nutrient-rich tropical environments [3], [52], [53].

The abundance of group A nifH genes (Fig. 4E) contributed 6% to the prediction of gamma A nifH gene abundance (Fig. 6), and gamma A nifH gene abundance contributed 4% to the prediction of filamentous nifH gene abundance (Fig. 7). As expected the positive relationship in both cases indicates that the different phyloptypes live side by side in the marine envn class="Chemical">ironment. However, nifH gene distributions of the group A, gamma A and filamentous diazotrophic phylotypes did not explain the N2 fixation rates (Fig. 8) indicating that the presence of nifH genes does not necessarily indicate an active metabolic pathway leading to N2 fixation. We further observed that heterotrophic prokaryote cell abundances (Fig. 4P) contributed 15% to the prediction of gamma A nifH gene abundance (Fig. 6).

Dissolved Al (Fig. 4L), a proxy for dust input, and consequently a proxy for the input of n class="Chemical">aeolian Fe and PO4 3- contributed 9% to the explanation of total nifH gene abundance and 18% to filamentous nifH gene abundance with higher nifH gene abundances at higher DAl concentrations (Fig. 5 and 7) [30], [54]. Our budget calculations (see above) suggest that DFe and PO4 3- did not limit primary production or N2 fixation. However, as DAl represents the cumulative effects of dust inputs rather than signifying individual short-term dust events, the general distribution of N2 fixing organisms may be determined by general patterns of dust inputs as also shown by Langlois et al. (2008). Contributions of DFe and total Fe-binding ligand concentrations (Fig. 4U), important to keep aeolian Fe in solution [30], were relatively small in explaining total nifH gene abundance and did not explain gamma A and filamentous nifH gene abundance. The lower DFe concentrations found at higher total nifH gene abundance may be the result of biological uptake. Photosynthesis as well as N2 fixation requires Fe, with N2 fixation requiring about 5–10 times more Fe than NO3 - utilization [55]. BRT showed that both higher N2 fixation rates and higher filamentous nifH gene abundance coincided with a higher CO2 fixation rate altogether resulting in relative high demands for Fe with consequently lower DFe concentrations at higher total nifH gene abundances (Fig. 7 and 8).

The CO2 fixation rate was the main contributor in explaining filamentous nifH gene abundance (29%) as well as n class="Chemical">N2 fixation rates (21%) (Fig. 7 and 8). Considering that the N2 fixation rate contributed up to 5% of the N required for new production, the high contribution of the CO2 fixation rates in predicting the distribution of filamentous nifH gene abundance and N2 fixation rates may be a consequence of fixed N release into the marine environment. Alternatively, environmental circumstances that promote high CO2 fixation rates also promote the presence of filamentous diazotrophs resulting in higher N2 fixation rates in our incubations. Also temperature contributed to the filamentous nifH gene abundance as well as N2 fixation rates, with higher abundance and rates in waters with temperatures above 22–23°C (∼8 and 4% contribution respectively). The similarity in the model response as shown by the partial dependence plots of CO2 fixation rate and temperature for the filamentous nifH gene abundance and N2 fixation rates suggests that the filamentous phylotypes may have been responsible for the main part of the N2 fixation rates. The low contribution of temperature to explaining N2 fixation is perhaps surprising as it has been shown that N2 fixation by Trichodesmium occurs at temperatures between 20°C and 34°C, with an optimum of 27°C [13], [14]. The surface water temperatures encountered during the cruise varied between 18 and 25°C, hence below the reported optimum values. The BRT model results suggest a step-up in N2 fixation between 22 and 23°C coinciding with the optimum temperature for the presence of group A and filamentous cyanobacteria (Langlois et al. 2008). Lower oxygen solubility in combination with faster metabolic processes such as respiration, facilitating oxygen scavenging, may underlie this high temperature preference [3], [14], [56], [57]. This idea concurs with our finding that N2 fixation rates varied negatively with oxygen concentrations.

N2 fixation was higher at lower dissolved n class="Chemical">oxygen concentrations (9% contribution) (Fig. 4V and 8). This is likely due to the oxygen sensitive nature of the nitrogenase enzyme, resulting in a decrease in nitrogenase activity with increasing oxygen concentrations [14], [58], [59]. While Trichodesmium and most unicellular microorganisms grow and fix N2 under fully aerobic conditions, the BRT analysis predicts that lower environmental concentrations of oxygen may be advantageous for N2 fixation.

The photophysiological variables as Fv/Fm (7%, Fig. 4S) and σPSII (12%, Fig. 4R) both contributed to the prediction of filamentous nifH gene abundance and n class="Chemical">N2 fixation, respectively. Fv/Fm, the photochemical efficiency, and σPSII, the photosystem II effective absorption cross section, both give information about effects of physiological stress on the photosystem and changes in community structure [60]. The magnitude in variability of Fv/Fm and σPSII due to changes in phytoplankton community structure often exceeds that induced by nutrient limitation [60]. As discussed above, the most likely limiting nutrient in the vicinity of the Cape Verdes was nitrogen. Nitrogen limitation may cause physiological stress on the photosystem in the overall community and strengthen the competitiveness of N2-fixing organisms resulting in the observed negative relationships between the photophysiological variables and filamentous nifH gene abundance and N2 fixation rates (Fig. 7 and 8). Alternatively, the change of filamentous nifH gene abundance and N2 fixation with these photophysiological variables may represent a change in community structure. Chl a concentrations, contributing 6% to predicting N2 fixation rates (Fig. 4Q), were low at high N2 fixation rates and may also have been the result of changes in community structure [61] (Fig. 8).

The distribution of diazotrophs as well as the n class="Chemical">N2 fixation pathway is regulated by the presence of nitrogen containing compounds [3], [43]. DON (Fig. 4I) contributed 17% to the prediction of gamma A nifH gene abundance, NH4 + (Fig. 4H) contributed 5% to the prediction of filamentous nifH gene abundance, while NO3 - (Fig. 4G), and DON contributed 13% and 10%, respectively, to the prediction of N2 fixation rates (Fig. 6– 8).

In the presence of micro-molar concentrations of NO3 -, a biochemicn class="Chemical">al down regulation of the N2 fixation pathway occurs [19], [20], [43]. However, at relatively low concentrations (300–400 nmol L−1 NO3 -), N2 fixation in Trichodesmium spp. has been reported to recover [20]. Overall, the NO3 - pool in our experiments was <130 nmol L−1 (and total DIN <350 nmol L−1), below NO3 - concentrations at which N2 fixation rates would be inhibited. Although an increase in N2 fixation rates with increasing NO3 - up to 30 nmol L−1, as suggested by the BRT analysis (Fig. 8), may appear surprising, we know that NO3 - concentrations were low in our experiments and may have been quickly exhausted. It is known that Trichodesmium prefers to utilize NH4 + and NO3 - over N2 fixation [20], [21]. As supported by the BRT analysis that predicts higher filamentous nifH gene abundance at higher NH4 + concentrations, more diazotrophs such as Trichodesmium may have been present in the seawater from the incubations with slightly higher NH4 + and NO3 - concentrations. As a consequence, with diminishing concentrations of NH4 + and NO3 - the higher abundance of diazotrophs may have resulted in the fixation of more N2. The positive relationship between filamentous nifH gene abundance and NH4 + may also be explained by the release, and subsequent regeneration, of amino acids and DON by Trichodesmium during N2 fixation [62], [63].

DON contributed 17% to the prediction of gamma A nifH gene abundance and 10% to the prediction of n class="Chemical">N2 fixation rates. DON can be produced as well as utilized by diazotrophs [43], [62]. Gamma A nifH gene abundance was higher at lower DON concentrations consistent with the hypothesis that N depleted environments promote the growth of N2-fixing organisms. The positive relationship between N2 fixation rate and DON could be explained by the exudation of fixed N in the form of, for example, amino acids [62]. Both explanations are plausible considering that enhanced gamma A nifH gene abundance does not mean that N2 fixation is active.

PO4 3- (Fig 4J) contributed 11% to explaining n class="Chemical">N2 fixation rates. PO4 3- forms an important nutrient that potentially limits N2 fixation [18]. Fitted values predicted that N2 fixation increased at ca. 40 nmol L−1 PO4 3- (Fig. 8). This suggests that N2 fixation may be released from PO4 3- limitation above a PO4 3- concentration of ca. 40 nmol L−1. Concentrations this low are common in the central and Western North Atlantic [16], [46], [64]. With the exception of the most northern sites our study area generally had PO4 3- concentrations exceeding this limit (Fig. 4J).

It is clear that no single factor controls the distribution of diazotrophs and n class="Chemical">N2 fixation but that these are determined by a combination of variables. This is especially true for areas and seasons where no nutrients, trace metals or other environmental parameters are directly limiting or inhibiting diazotrophy. Not all environmental variables considered in this study may have been causal for N2 fixation and the distribution of diazotrophs. However, from experimental and field studies it is known that a-biotic environmental variables such as NO3 - + NO2 -, DON, Fe, PO4 3-, temperature and oxygen directly affect N2 fixation in the North Atlantic Ocean [16], [17], [18], [21], [56], [57]. In this study, BRT analysis showed that these a-biotic environmental variables all contributed to determining N2 fixation rates and that some of them also affected the distribution of gamma A, filamentous and total nifH gene abundances in the eastern tropical and sub-tropical North Atlantic Ocean.

Potential consequences of climate warming on N2 fixation

Identification of the a-biotic environmentn class="Chemical">al variables that have the potential to limit or inhibit N2 fixation allows us to assess how a change in these environmental variables due to climate warming may affect future N2 fixation rates. The warming of the earth system will result in enhanced sea surface temperature (SST), with predicted enhanced deposition of Saharan dust to the surface oceans [65]. N2 fixation rates in the eastern tropical and sub-tropical North Atlantic Ocean are likely to be affected by enhanced SST, reduced oxygen concentrations, reduced nutrient supply and an increase in light availability due to enhanced water column stratification. According to our results N2 fixation rates will increase at lower oxygen concentrations and higher temperatures and are thus consistent with previous work. In addition, N2 fixation may be promoted by an increase in light availability, an enhanced CO2 concentration and changes in N/P stoichiometry induced by shifts in the phytoplankton community [66], [67], [68].

However, in the eastern tropical and subtropicn class="Chemical">al North Atlantic an increase in N2 fixation may be hampered by PO4 3- limitation and, although not shown in this study, potentially by Fe limitation, should the availability of either element decrease. Due to future enhanced stratification, we may experience a decrease in upward mixing into the surface waters of deeper waters with excess phosphate (excPO4 3- = PO4 3- - (NO3 - + NO2 -)/16; excPO4 3- = 14 e 0.02 x Apparent Oxygen Utilization, R2 = 0.71, p<0.001, n = 25) and high concentrations of Fe, derived from the oxygen minimum zone situated south and west of the Cape Verde. Increased atmospheric Fe and PO4 3- inputs from the Sahara, may potentially off set the reduced deep water supply; the atmospheric inputs are however more sporadic. As a consequence, the balance between a reduced supply of nutrients due to stratification and an enhanced nutrient input with increasing deposition of Saharan dust may ultimately determine the consequences of climate change for N2 fixation in the eastern tropical and sub-tropical North Atlantic.