Introduction

Biological N2 fixation is the main source of fixed nitrogen (N) in the oceans and a key process in regulating primary productivity and thereby carbon fluxes (Gruber, 2005). N2 fixation is carried out by a limited, but diverse, group of bacteria and Methanotrophic Archaea, known as diazotrophs (Zehr et al., 2003b). The Baltic Sea is among the world’s largest brackish seas and regularly experiences extensive blooms of diazotrophic heterocystous cyanobacteria (Stal et al., 2003) providing N-input almost as large as the riverine load (Larsson et al., 2001). Cyanobacterial N2 fixation is also significant in open oceans (Goebel et al., 2007), but molecular techniques targeting the nifH gene of the nitrogenase enzyme have revealed that heterotrophic diazotrophs are also present and nifH transcribed in diverse marine and estuarine environments (reviewed in Riemann et al., 2010). However, their significance to N2 fixation and the factors controlling their activity and distribution patterns are not yet understood. A better understanding of their ecology is therefore essential for resolving the role of this functional group in oceanic and coastal nutrient cycling.

Nitrogenase activity is thought to be tightly regulated by ammonium (NH4+) availability (Kanemoto and Ludden, 1984; Klugkist and Haaker, 1984) and is inactivated and destroyed by oxygen (O2; Dixon and Kahn, 2004). Consequently, heterotrophic diazotrophs likely require suboxic conditions for active N2 fixation. In the oxygenated water column, such loci may be associated with live or dead particulate matter (reviewed in Riemann et al., 2010). Alternatively, these organisms may exploit low O2 biomes, such as pelagic transition zones between oxic and anoxic layers and oceanic oxygen minimum zones (OMZs). Such areas, representing zones with large N-losses through anammox and heterotrophic denitrification (Lam and Kuypers, 2011), can be possible sites for heterotrophic N2 fixation (Zehr et al., 2006; Riemann et al., 2010). Concordantly, N2 fixation rates and diverse putative heterotrophic diazotrophs were recently reported from hypoxic waters in the eastern tropical South Pacific (Fernandez et al., 2011) and the Southern Californian Bight (Hamersley et al., 2011). Similarly, nifH genes and transcripts related to heterotrophs were reported from the Arabian Sea OMZ (Jayakumar et al., 2012). Interestingly, a recent study documented the co-occurrence of N2 fixation and denitrification in a meromictic lake (Halm et al., 2009). Combined, these findings suggest that two central N transformation processes, thus far thought to be segregated, can co-occur in OMZs, challenging the conventional view of N cycling in the oceans and unveiling the intriguing possibility of NH4+-rich suboxic and anoxic waters as environments for heterotrophic N2 fixation. However, knowledge on N2 fixation in suboxic waters is limited (Ulloa et al., 2012) and molecular data and N2 fixation rates in truly anoxic marine systems is lacking.

The Baltic Sea proper is characterized by a permanent halocline, which prevents vertical mixing. Organic matter mineralization causes O2 deficiency (<2 ml l−1) in deep waters, covering an area of 12 000–70 000 km2 (Conley et al., 2002). At and below the chemocline, hydrogen sulfide (H2S) and NH4+ accumulates and the potential for denitrification and anammox has been demonstrated (Hietanen et al., 2012). The aim of this study was to explore the chemocline and anoxic waters as potential loci for heterotrophic N2 fixation. The 454-pyrosequencing provided high resolution data on the distribution of nifH phylotypes and through quantitative polymerase chain reaction (QPCR) and reverse transcription QPCR (RT-QPCR), we determined the abundance and expression of seven prominent nifH gene clusters. The analysis of five sampling profiles, four from the Gotland Deep and one from the Bornholm Basin, from three sampling years, revealed a large diversity of heterotrophic diazotrophs at and below the chemocline. We also report transcription of nifH and N2 fixation rates in anoxic NH4+-rich water and thereby extend the currently known regime of marine N2 fixation.

Materials and methods

Environmental sampling

Samples were collected at the Gotland Deep (240 m total depth, 57°20.09’N, 20°03.09’E) in the summers of 2007 (9 July, morning), 2008 (15 August, noon) and 2011 (14 and 18 July, morning). Samples were collected between 5 and 233 m using 5 l Free-Flow bottles (Hydrobios) attached to a PVC-coated stainless steel rosette equipped with sensors for O2 (Sea-Bird SBE 43), conductivity, temperature and depth (Seabird Inc., Bellevue, WA, USA), with high depth resolution around the chemocline (129–132 m in 2007, 121–127 m in 2008 and 110–121 m in 2011, Figure 1). In 2011 (13 July), samples were collected at 10 m intervals from the Bornholm Basin (89 m total depth, 55°15.05’N, 15°59.06’E). For each sample, 1–3.8 l were filtered onto a 10 μm polycarbonate filter (47 mm, GE Water & Process Technologies, Trevose, PA, USA) and 0.6–2 l of the filtrate was filtered onto a 0.2 μm (47 mm) Supor filter (PALL Corporation, Port Washington, NY, USA). After filtration, RNA filters were soaked in RNAlater (Ambion, Carlsbad, CA, USA). Filters were placed in 0.5 ml RNAlater or 1 ml Tris-EDTA (10 mM Tris-HCl, 1 mM EDTA, pH 8.0) for RNA and DNA filters, respectively, and the cryotubes were frozen in liquid N or at −20 °C (2011). A summary of nucleic acid samples and analytical methods is provided in Table 1. Samples for bacterial enumeration were fixed with glutaraldehyde (1% final concentration) and counted using flow cytometry (Troussellier et al., 1999). Chemical profiles of H2S, NH4+ (not analyzed in 2007), phosphate (PO43−), nitrite (NO3−) and nitrate (NO2−) were determined for the Gotland Deep and PO43− and NO2− for the Bornholm Basin (Grasshoff et al., 1983).

Figure 1
figure 1

Environmental data from the sampled depth profiles at the Gotland Deep (a) and the Bornholm Basin (b) in the Baltic Sea proper. Depths of sampling for nutrients and nucleic acids are indicated by gray horizontal lines. Concentrations of inorganic nutrients are in μmol l−1, temperature in °C and O2 in ml l−1.

Full size image
Table 1 Summary of samples obtained and analyzed by 454-pyrosequencing and/or QPCR
Full size table

N 2 fixation rates

On the 14 July 2011, rates of N2 fixation were calculated using 15N2-incorporation (Montoya et al., 1996) in samples from the oxygenated surface and anoxic waters at the Gotland Deep. Water was transferred directly from the Free-Flow bottle into triplicate 1.2 l acid-washed polycarbonate bottles pre-flushed with argon. The bottles were filled from the bottom up using silicone tubing. 1.0 ml of 15N2 tracer gas (98%, Campro Scientific, Berlin, Germany) was added through a gas-tight septum. The bottles were inverted 50 times and incubated for 24 h in an on-deck container with running surface seawater covered with fabric reducing the light to 13% of the surface light intensity (5 m samples) or at 8 °C in the dark (200 m samples). Post-incubation size-fractionation was performed to quantify free-living N2 fixation: 600 ml of water was filtered through a 10 μm polycarbonate filter (GE Water & Process Technologies) and thereafter through a pre-combusted (450 °C for 8 h) GF/F filter (0.7 μm, Whatman, GE Healthcare, Little Chalfont, UK). The remaining 600 ml were filtered through a GF/F filter. The background δ15N levels, two time-zero controls, and un-amended controls were included for each depth. Filters were stored at −20 °C, dried at 60 °C for 24 h, and analyzed on an isotope ratio mass spectrometer (PDZ Europa, Northwich, UK) at the Laboratory of Applied Physical Chemistry, Gent, Belgium. In these analyses, particulate N concentrations were >33 μg N l−1 and did not limit detection. No significant differences were observed between background δ15N levels and the time zero and un-amended controls.

Nucleic acids extractions and complementary DNA (cDNA) synthesis

Community DNA was extracted using an enzyme/phenol-chloroform protocol (Riemann et al., 2008). Community RNA was extracted using the RNeasy mini kit (Qiagen, Hilden, Germany; Boström et al., 2007b). For each sampling year and filter type, blank filters were included as negative controls. DNA and RNA concentrations were measured using NanoDrop 2000 (Thermo Scientific, Waltham, MA, USA). First-strand cDNA synthesis was done using the nifH3 primer (Zehr and McReynolds, 1989) and the TaqMan reverse transcription kit (Applied Biosystems, Foster City, CA, USA) including a negative control with only RT-PCR grade water (Ambion) and a second set of reactions without reverse transcriptase. The DNA and cDNA was stored at −20 °C.

PCR amplification and 454-pyrosequencing

A total of 16 DNA and 4 cDNA samples were amplified with nifH3 and nifH4 primers (Zehr and Turner, 2001) followed by 30 cycles with custom bar-coded nifH1 and nifH2 primers (Supplementary Table S1) using Pure Taq Ready-To-Go PCR Beads (GE Healthcare). As the cDNA samples from 41, 132 and 233 m yielded faint PCR products they were first amplified with nifH1 and nifH2 primers (Zehr and McReynolds, 1989) followed by 10 PCR cycles with barcoded primers (Farnelid et al., 2011). Non-transcribed RNA samples were included as control for incomplete DNA digestion. The PCR products were gel purified (Gel extraction kit, Qiagen) and quantified (NanoDrop 2000). For PCR samples from negative extraction and RT-PCR grade water controls, although no bands were visible, a slice of the expected product size (402 bp) was excised, purified, and cloned (TOPO TA, Invitrogen, Carlsbad, CA, USA). No clones were obtained. The PCR products were mixed in equal amounts and sequenced using a GS FLX pyrosequencing system (Roche, Basel, Switzerland). Sequences are deposited in the National Center for Biotechnology Information Sequence Read Archive under accession number SRA058502.

Sequence quality controls and analysis

The data set was filtered to only include sequences with correct tag and primer sequences. These were then trimmed, sequences were clipped at 219 bp, translated into amino acids (73 aa), and sequences having in-frame stop codon(s) were removed. Unique reads were aligned as previously described (Farnelid et al., 2011). Reads were screened to check for homology to all reported nifH contaminant sequences, but no identical reads were found. To study the phylogeny of the dominant clusters, the reads were clustered at 96% similarity using CD-hit (Li and Godzik, 2006) and clusters containing <100 sequences were removed. The remaining 222 clusters and related sequences in the database were aligned in MEGA (Tamura et al., 2007) and a phylogenetic tree was created (p-distance). To allow for cross-sample comparisons, a random re-sampling to identical sequencing depth (5000 sequences) was done using an in-house developed Perl script. The SChao1 richness estimator (Chao, 1984) and the Shannon diversity index (Shannon, 1948) were calculated and rarefaction curves were generated using Analytical Rarefaction 1.3 (http://www.uga.edu/strata/software/index.html). Bray–Curtis distances between samples were calculated from the abundances of the nifH clusters using the R package Vegan (http://www.r-project.org/). Principal coordinates analysis and hierarchical clustering based on these pairwise distances were performed with the R packages Made4 and Cluster, respectively.

Quantitative PCR and RT-QPCR

Eight of the identified nifH clusters were targeted by QPCR. For seven of the clusters, primers and probes were designed using the software Primer Express 3.0 (Applied Biosystems). The primer/probe set for the ALHOU cluster was designed to also amplify the B24DA cluster (Supplementary Table S2) and is referred to as ALHOU hereafter. For the Nodularia-like nifH cluster (CM8AH), the forward primer and probe developed by Boström et al. (2007b) and a reverse primer within the 219 bp 454-read was used (Supplementary Table S2). Primers and probes had mismatches with all non-target sequences (Supplementary Table S2), and the specificity was verified through BLAST searches against the NCBI database. A QPCR standard for the EVHVF target was obtained from an available clone (EU916338, Farnelid et al., 2009). For the other targets, the 219 bp long sequences were synthesized and cloned into pUC19 vectors (Epoch Life Science, Missouri City, TX, USA). Plasmids were extracted (EZNA Plasmid Mini Kit, Omega bio-tek, Norcross, GA, USA), linearized using HindIII (10 U μl−1, Roche), purified (EZNA gel extraction kit, Omega bio-tek) and quantified (NanoDrop 2000). The specificities of the primer and probe sets were evaluated for each target with 105 copies of the non-target standards. Unspecific amplification was very low (<34 copies of 105 detected). To examine potential PCR inhibition in the samples, 5, 2 or 0.5 ng of mixed DNA from each profile and size fraction was run in QPCR with 104 copies of standard. There was no inhibition in any of the samples at 2 ng per reaction.

The QPCR and RT-QPCR reactions were run on a StepOnePlus Real-Time PCR System (Applied Biosystems). Each reaction (25 μl) consisted of TaqMan Universal PCR Master Mix (Applied Biosystems), 900 nM of each primer (HPLC purified, Sigma-Aldrich, St Louis, MO, USA), 250 mM probe (Applied Biosystems), RT-PCR grade water and 1–2 μl standard or sample (DNA 2 ng, cDNA 1.2 ng). The thermal cycling conditions were 2 min at 50 °C, 10 min at 95 °C followed by 40 cycles of 15 s at 95 °C and 1 min at 60 °C. All DNA samples, standard dilutions (106–101 copies), and no-template controls (RT-PCR grade water) were run in triplicate and cDNA samples were run in duplicate. No-template controls were undetectable. The efficiency of the primer/probe sets was 80–103% (Supplementary Table S3). In some samples, at least one of the replicates did not produce an amplification signal above the lowest point on the standard curve (10 nifH copies); in these, targets were considered detectable but not quantifiable. To calculate the percentage of heterotrophic nifH phylotypes, the abundances of the quantified nifH clusters were summed and related to the bacterial abundance in the sample, assuming one nifH copy per genome and one genome per bacterium (Langlois et al., 2008).

Results and Discussion

Environmental data

The Gotland Deep station is characterized by a stable halocline (60–80 m) dividing the water column into an upper oxygenated layer and an underlying suboxic and anoxic layer. In 2007, the chemocline was at 127–130 m (Figure 1a). In 2008, the halocline was deeper (90–100 m), possibly due to a recent mixing event, and the chemocline was at 121–127 m. In 2011, the halocline and the chemocline were located around 65–75 m and 120–125 m, respectively (Figure 1a). At 200 m, NH4+ concentrations were high (19.7 and 19.4 μM; Figure 1a). Coinciding with the disappearance of O2, peaks in NO3− concentration were observed in 2008 (2.57 μM, 100 m) and in 2011 (7.25 μM, 110 m and 5.82 μM, 109 m). At the Bornholm Basin, O2 was below detection at 75 m (Figure 1b). For all sampling profiles, the bacterial abundance was highest at the surface (2.6 × 106−4.2 × 106 ml−1) and lowest at the halocline (1.3 × 105−3.8 × 105 ml−1; Supplementary Figure S1). No peaks were observed directly at the chemocline, as previously reported (Höfle and Brettar, 1995). Instead the abundance increased with depth or remained stable below the chemocline.

Diversity and composition of nifH DNA and cDNA libraries

Pyrosequencing of nifH genes PCR amplified from the Gotland Deep in 2007 and 2008 (Table 1) produced a total of 311 857 sequences encoding 28 812 unique proteins (Supplementary Figure S2). The number of high-quality sequences varied from 5814 to 40 670 per sample. Rarefaction analysis at 96% similarity clustering indicated that the 5000 re-sampled sequences were a good representation of the total diversity (Lundin et al., 2012; Supplementary Figure S3). The cDNA samples featured the least number of clusters (Supplementary Figure S4D) suggesting that only a limited number of nifH gene clusters were actively transcribed. Similar differences between DNA and RNA libraries were previously reported (for example, Man-Aharonovich et al., 2007). We speculate that the discrepancy is caused by temporal changes in environmental conditions with resulting transient niches for activity of specific N2-fixers. However, it should also be noted that the current sampling and extraction protocol may lead to biases in gene expression (Feike et al., 2012) and thereby underestimation of the occurrence of transcripts.

The nifH gene and transcript libraries showed that the diazotrophic community consisted mainly of alpha-, beta- and gammaproteobacteria belonging to the canonical nifH cluster I and diverse anaerobes within nifH cluster III (Chien and Zinder, 1996), including, for example, Desulfovibrio spp. and archaeal methanogens (Figure 2). Previous studies have reported diverse non-cyanobacterial nifH phylotypes within these clusters in both surface (Man-Aharonovich et al., 2007; Farnelid et al., 2011; Mulholland et al., 2012) and deep waters (Hamersley et al., 2011; Jayakumar et al., 2012). Among the nifH sequences within cluster III in this study (Figure 2), none were identical to previously reported clones from the surface of the Baltic Sea proper (Farnelid et al., 2009). Interestingly, cluster III phylotypes were not enriched in the >10 μm size fraction relative to the <10 μm size fraction (Supplementary Figure S2B). Hence, in contrast with previous suggestions (for example, Braun et al., 1999; Man-Aharonovich et al., 2007) particles did not appear to provide an advantageous environment for these bacteria.

Figure 2
figure 2

Neighbor-joining phylogenetic tree of 96% aa similarity clustered nifH sequences with affiliating canonical nifH clusters (Chien and Zinder, 1996) indicated by colored branches. The highest relative abundance was calculated by normalizing the number of sequences in each cluster and is shown as being in the oxic (red), at the chemocline (purple) or in anoxic waters (black), derived from the O2 concentrations in Figure 1, for each type of sample; 2007 cDNA <10 μm (A), 2007 DNA >10 μm (B), 2008 DNA <10 μm (C) and 2007 DNA <10 μm (D). White indicates that no sequences within the cluster were detected. Accession numbers for reference sequences are indicated in squared brackets and clusters targeted by QPCR in this study are marked in bold. Bootstrap values (1000 replicates) >50% are indicated with size proportional gray circles. For clarity, only clusters with >100 sequences are included in the tree. The EVHVF, CM8AH and Pseudanabaena-like clusters were collapsed, respectively, and the numbers of subclusters are indicated in parentheses. The full branching of the subclusters is shown in Supplementary Figure S2A.

Full size image

The occurrence of alternative nitrogenases (cluster II) in the nifH libraries was surprisingly low (208 sequences) considering their prevalence among clones (Farnelid et al., 2009) or isolates from the Baltic Sea (Boström et al., 2007a). Although libraries consisted mostly of non-cyanobacterial sequences, nifH reads affiliated with Nodularia, known to dominate cyanobacterial blooms (Stal et al., 2003; Boström et al., 2007b), were common in surface samples in 2007. Pseudanabaena-like nifH gene sequences were highly represented in oxic samples both sampling years (Figure 2 and Supplementary Figure S2). For the non-cyanobacterial clusters, the highest relative abundance generally occurred at or below the chemocline and interestingly, among the ones detected in cDNA, all feature the highest relative abundance at the chemocline (Figure 2). Thus, the chemocline provides environmental conditions that promote growth of diverse assemblages of putative heterotrophic diazotrophs that actively express nitrogenase.

The principal coordinates analysis showed that the composition of putative diazotrophs in the oxic samples was distinct from those at the chemocline and anoxic waters. In contrast, the latter two groups were intermingled (Figure 3 and Supplementary Figure S5). The oxic samples displayed large variation in composition, partly related to the size fraction analyzed (with the exception of 2007, 41 m, <10 μm; Figure 3). Generally, the nifH diversity increased with depth and the gradual decrease in O2 (Supplementary Figures S3 and S4). The Chao richness estimators and Shannon diversity indexes, respectively, were similar in 2007 and 2008, decreasing abruptly around the chemocline in <10 μm samples (Supplementary Figures S4A, B). Our observations of increased nifH diversity with depth suggest that suboxic water contains numerous ecological niches for putative heterotrophic diazotrophs. In addition, most of the identified clusters were found during both years (2007 and 2008; Figure 2), suggesting that they are a stable component of the Baltic Sea bacterial community.

Figure 3
figure 3

Principal coordinates analysis (PCoA) plot based on the Bray–Curtis distances calculated between samples. The samples are indicated as being in the oxic (squares), at the chemocline (circles) or in anoxic waters (stars), derived from the O2 concentrations in Figure 1. Symbols for samples from 2007 and 2008 are black and gray, respectively. Open symbols show samples from the >10 μm size fraction.

Full size image

Distribution and abundances of nifH phylotypes and transcripts

One drawback of the nested PCR approach is the high susceptibility to contamination through handling or presence of nifH genes in PCR reagents (Zehr et al., 2003a; Goto et al., 2005). To circumvent this potential risk, nifH phylotypes may be quantified using QPCR (Hewson et al., 2007). We specifically designed primers and probes for the EVHVF cluster (100% aa similarity to Burkholderia vietnamiensis; AM110708; Figure 2) known to affiliate with reported contaminants (Farnelid et al., 2009). The distribution of EVHVF was patchy among samples and sampling years with generally low abundance (Figure 4 and Supplementary Table S3) suggesting that its presence was not due to contamination of extraction or PCR reagents. Nevertheless, the high relative abundance of EVHVF in three of the cDNA libraries (Supplementary Figure S2A) does not mirror the abundance suggested by QPCR. Similarly, the AT2LU cluster (99% aa similarity to AY223933; Zehr et al., 2003b) could not be quantified in any of the samples although it was present in the 129 m cDNA library (2731 sequences; Figure 2 and Supplementary Figure S2B). Hence, in concordance with recent findings (Turk et al., 2011), some nifH phylotypes may be preferentially amplified. In addition, the coverage of degenerate nifH primers differs between phylogenetic groups (Gaby and Buckley, 2012). Therefore, the relative proportions of clusters in sequence libraries should be interpreted with caution.

Figure 4
figure 4

Abundances of specific nifH gene clusters in the free-living (<10 μm) fractions (a) and the particle-associated (>10 μm) fractions (b) determined by QPCR at the Gotland Deep. Cluster AT2LU is not shown as it was never quantifiable. Detection of transcripts is indicated with a star and when quantified, the number of transcripts l−1 is indicated next to the horizontal bars. Sample depth (m) is indicated in each panel. Based on data shown in Figure 1, the profiles are divided into the oxic, the chemocline or the anoxic depth strata. Note the different scales on the X axis for a and b.

Full size image

Of the clusters quantified, only nifH from the cyanobacterium Nodularia sp. (CM8AH) was present in surface samples. Abundances were up to 1.9 × 106 copies l−1 and transcripts up to 2.1 × 105 copies l−1 (Figure 4 and Supplementary Table S3), which is comparable to a previous report from the Baltic Sea proper (Boström et al., 2007b). N2 fixation by Nodularia sp. is likely limited to illuminated waters and, therefore, nifH gene copies from these cyanobacteria at and below the chemocline (Figure 4 and Supplementary Table S3) is probably derived from settling senescing cells.

Two gammaproteobacteria-like nifH clusters, EQF91 and ALHOU, were targeted by QPCR (Figure 2 and Supplementary Table S2). EQF91 was most closely related to clones from hypoxic water in the Southern Californian Bight (97% aa similarity; HQ660860; Hamersley et al., 2011) suggesting that these phylotypes may be characteristic for suboxic regions. Accordingly, EQF91 was never detected above 100 m at the Gotland Deep or above 68 m at the Bornholm Basin, but was consistently present at and below the chemocline, with abundances up to 5.2 × 105 copies l−1 (Figures 4 and 5 and Supplementary Table S3). On the 18 July 2011 at 119 m, EQF91 transcripts were detected but below the limit of quantification (Figure 4). The ALHOU cluster was related to Pseudomonas stutzeri (96% aa similarity) and a prevalent cluster from the South Pacific OMZ (93% aa similarity; HM801245; Fernandez et al., 2011). The abundance of ALHOU reached up to 3.0 × 106 copies l−1. The cluster was consistently present in suboxic/anoxic waters in all sampling profiles but transcription was only detected at 88 m in the Bornholm Basin (Figures 4 and 5 and Supplementary Table S3). However, as transcripts within the ALHOU and B24DA clusters were present in both the 129 and 132 m cDNA libraries (Supplementary Figure S2), constraints on the amount of template cDNA or disturbances during sampling (see Feike et al., 2012) may explain why transcripts were not more frequently detected or quantified.

Figure 5
figure 5

Abundances of specific nifH gene clusters in the free-living (<10 μm) fractions (a) and the particle-associated (>10 μm) fractions (b) determined by QPCR at the Bornholm Basin. Cluster AT2LU is not shown as it was never quantifiable. Detection of transcripts is indicated with a star and when quantified, the number of transcripts l−1 is indicated next to the horizontal bars. Sample depth (m) is indicated to the left. Note the different scales on the X axis for a and b.

Full size image

Three clusters within nifH cluster III were targeted by QPCR (Figure 2). The D0CY3 cluster was 99% similar (aa) to sequences from North American coastal waters (Mulholland et al., 2012) while the ECI27 cluster was 100% similar (aa) to nifH clones (JN638704) from an unpublished study of the redoxcline of the Black Sea. Very high abundances of the D0CY3 (up to 3.3 × 106 copies l−1) and ECI27 (up to 2.2 × 107 copies l−1) clusters were observed at and below the chemocline (Figure 4 and Supplementary Table S3). These abundances are among the highest reported (but see Halm et al., 2012; Mulholland et al., 2012) and magnitudes higher than previous reports on cluster III phylotypes in the Pacific and Atlantic oceans (Church et al., 2005; Langlois et al., 2008). The activity of anaerobic N2-fixers is largely unknown but interestingly, transcripts of a cluster III phylotype were recently quantified to 1.9 × 103 copies l−1 at 200 m in the South Pacific Gyre (Halm et al., 2012). In our study, transcripts of D0CY3 (3.2 × 104 and 6.5 × 103 copies l−1) at the two consecutive samplings for 200 m Gotland Deep samples (Figure 4 and Supplementary Table S3) and the co-occurrence of genes and transcripts also at the Bornholm Basin (88 m, Figure 5) further highlights the potential significance of cluster III phylotypes as N2-fixers in suboxic/anoxic waters. In summary, the QPCR results show that abundances of several heterotrophic phylotypes increased dramatically at and below the chemocline. nifH transcripts of both Proteobacteria and cluster III were also detected indicating that bacteria within these groups are active in N2 fixation (Figures 4 and 5 and Supplementary Table S3).

N fixation

Rates of N2 fixation were measured in surface (5 m) and deep waters (200 m) of the Gotland Deep on 14 July 2011 using the traditional 15N2-incorporation method; an approach yielding conservative estimates of N2 fixation (Mohr et al., 2010). The mean rate of N2 fixation in surface water was 7.6±1.76 nmol l−1 day−1 of which 2.0±0.74 nmol l−1 day−1 was in the <10 μm size fraction. As some Nodularia cells may occur solitary or in short series (Lehtimäki et al., 2000) and Pseudanabaena may pass a 10 μm filter (Stal et al., 2003; Farnelid et al., 2009), N2 fixation in the <10 μm size fraction is not evidence for heterotrophic N2 fixation. In the anoxic sample, the mean rate was 0.44±0.26 nmol l−1 day−1 of which 0.24±0.26 nmol l−1 day−1 was in the <10 μm size fraction. These rates are comparable to those of hypoxic waters in the Southern Californian Bight (average 0.07 nmol l−1 day−1; Hamersley et al., 2011) and the Peruvian OMZ (average 1.27 nmol l−1 day−1; Fernandez et al., 2011). In the same water sample, transcription of the D0CY3 cluster was 3.2 × 104 copies l−1 (Figure 4a) strongly suggesting that these anaerobic bacteria were largely responsible for the N2 fixation at 200 m.

In anoxic water, N2 fixation took place at a NH4+ concentration of 19.7 μM (Figure 1). Similarly, N2 fixation has been measured off Peru and in a meromictic lake at measurable NH4+ levels (Halm et al., 2009; Fernandez et al., 2011). In the laboratory, NH4+ downregulates or shuts down nitrogenase expression (for example, Kanemoto and Ludden, 1984; Klugkist and Haaker, 1984). However, the high NH4+ concentration at 200 m did not reduce the abundance or inhibit transcription of ECI27 and D0CY3 (Figure 4a). Similarly, N2 fixation and/or transcripts (notably within nifH cluster III) were reported from NH4+-rich microbial mats (Steppe and Paerl, 2002; Omoregie et al., 2004), representing a natural microscale analog of hypoxic waters columns (Zehr et al., 2006). Sustained N2 fixation following NH4+ amendment has also been observed in a salt marsh (Hanson, 1977) and a gammaproteobacterial culture (Boström et al., 2007a). Thus, despite the high energy demand of N2 fixation compared with NH4+ uptake, availability of NH4+ does not always preclude N2 fixation (reviewed in Knapp, 2012). An intriguing explanation could be that some nitrogenases, because of their early origin before the oxygenation of the atmosphere when high ambient levels of NH4+ prevailed (Raymond et al., 2004), may not be tightly regulated by NH4+ (Zehr et al., 2006). Nonetheless, the maintenance of N2 fixation rates and transcripts in NH4+-replete waters highlights the incomplete understanding of the biology of these organisms, and underlines the need for surveys of diazotrophic activities in marine environments hitherto believed to be void of diazotrophy.

Implications of heterotrophic N 2 fixation in suboxic waters

In this study, we demonstrate that suboxic and anoxic waters of the Baltic Sea harbor diverse and active heterotrophic N2-fixing communities. The chemocline of the Gotland Deep features high microbial activity (Höfle and Brettar, 1995), dark CO2 fixation (Jost et al., 2008) and loss of inorganic N (Hietanen et al., 2012). The growth of bacteria involved in N-loss processes is often carbon limited (Lam and Kuypers, 2011). It is thus perplexing to find microorganisms investing in the expensive process of N2 fixation under conditions that are N-replete and most probably carbon limited.

N2 fixation in the anoxic waters was low compared with surface waters (Larsson et al., 2001). However, unlike cyanobacteria, heterotrophic diazotrophs are not limited by light or water temperature. Hence, they could potentially fix N2 in a large pelagic realm. Several of the nifH phylotypes that we quantified were abundant in suboxic/anoxic waters during all sampling years and in both basins, ∼340 km apart (Figures 4 and 5). Accordingly, if consistently present, they may occupy a volume of up to 3000 km3 in the Baltic Sea (<2 ml O2 l−1; Hansson et al., 2011), and account for a fixation of up to 11 Gg N year−1, equal to up to 6% of the N fixed yearly in surface waters of the Baltic Sea proper (Larsson et al., 2001), if assuming that our rate measured at 200 m is representative. Although this rough and speculative calculation merely illustrates the large potential of heterotrophic diazotrophs, determining their exact contribution to N input is difficult because of their generally low concentration, the very low transcript copy numbers, and accordingly, rather low rates of N2 fixation. Still, heterotrophic N2-fixers accounted for up to 2.1% of the total bacteria at the chemocline (Supplementary Figure S1). In comparison, biogeochemically significant anammox bacteria ranged from 2.2% to 3.1% of the bacterial community in the Peruvian OMZ (Hamersley et al., 2007). Hence, future N2 fixation measurements with extensive spatio-temporal resolution including also the deeper suboxic and anoxic zones of the Baltic Sea, could potentially reveal integrated rates of high N budgetary significance.

Heterotrophic diazotrophs are almost ubiquitous in marine surface waters (Riemann et al., 2010; Farnelid et al., 2011) and occur in OMZs (Fernandez et al., 2011; Hamersley et al., 2011; Jayakumar et al., 2012) around the globe. Combined with the present extension of their niches to NH4+-rich sulfidic-anoxic waters of one of the largest brackish waters on earth, the prevailing perception of vertical and geographical boundaries for marine N2 fixation has been dramatically challenged; thus, we foresee a future broadening of the marine N2 fixation regime with potentially large consequences for N biogeochemistry estimates.