Introduction

Dinitrogen (N2) fixing prokaryotes (diazotrophs) supply bioavailable nitrogen to planktonic communities, fueling primary production and contributing to carbon export in the ocean [1]. Nitrogen inputs by diazotrophs may become even more important in the future ocean, as global warming enhances water column stratification constraining nitrogen availability for primary producers [2]. Early studies suggested that diazotrophs were only present in low latitude warm oligotrophic waters of the (sub)tropical ocean. However, over the past decade it has become clear that diazotrophs are also found in cold and nutrient-rich environments such as estuaries, shelf seas, polar regions and the ocean’s dark pelagic realm [3]. Devoid of light, N2 fixation in the dark ocean has been attributed to heterotrophic non-cyanobacterial diazotrophs presumably relying on reduced organic compounds for energy and carbon supply [4,5,6]. However, cyanobacterial diazotrophs have been repeatedly observed in mesopelagic to bathypelagic depths (200–4000 m; Table S1; [7]).

The filamentous cyanobacterium Trichodesmium thrives in tropical and subtropical ocean’s photic zone where it can introduce 60–80 Tg N y−1, representing roughly half of the reactive nitrogen input to the global ocean [8]. Trichodesmium has gas vesicles which confer cells with buoyancy, restricting their vertical distribution in the water column [9, 10]. Accordingly, Trichodesmium biomass is thought to be fully remineralized within the photic zone [11]. Recently, however, molecular and imaging studies have documented intact Trichodesmium cells and expression of nitrogenase genes down to 4000 m depth across the world’s oceans (Table S1). The mechanisms through which Trichodesmium sinks into the dark ocean may include gravitational sinking, downwelling events [12], mineral ballasting [13], or sudden autocatalytic cell death in response to nutrient limitation [14]. Sinking velocities can be fast enough for surface photoautotrophic cells to reach the dark ocean while remaining viable [15,16,17,18], but whether sinking Trichodesmium remains metabolically active and carries out N2 fixation in the dark ocean is not known.

If diazotrophically active Trichodesmium fixes N2 in the dark ocean it would not only affect the global marine nitrogen inventory but conceivably also stimulate microbial processing of particulate organic matter in the (sub)tropical regions where it thrives, affecting vertical carbon export and remineralization in the dark ocean [19]. In the present study we combine 15N2 incubation on sediment traps, sinking simulation laboratory experiments and cell metabolism modeling to document that Trichodesmium can fix N2 while sinking far below the photic zone, constituting a hitherto unaccounted reactive nitrogen and organic matter source in the dark ocean.

Materials and methods

Sinking particle sampling in the South Pacific Ocean

Sinking particles were collected in the western tropical South Pacific during the GEOTRACES GPpr14 expedition (https://doi.org/10.17600/18000884) onboard the R/V L’Atalante from November 1st to December 5th 2019. Surface tethered mooring lines (~1000 m long) were deployed at two stations: S05M (21.157°S, 175.153°W, 5 days) and S10M (19.423°S, 175.133°W, 4 days). The mooring lines featured three sediment traps placed at 170, 270, and 1000 m. These depth levels were chosen to match the base of the photic layer, the usual depth to calculate flux attenuation (100 m deeper than the photic layer) and the base of the mesopelagic layer, respectively. Each trap consisting of four particle interceptor tubes mounted on an articulated cross-frame. Out of the four tubes deployed per depth, two were used for this study. The first tube was filled with 6 l of 0.2 µm filtered seawater followed by 2.5 l of a 50 g l−1 saline brine labeled with 15N2 gas. The brine was prepared in a 4.5 l polycarbonate bottle fitted with a septum screwcap and labeled with high-purity 15N2 gas (Euroiso-top) injected through the septum and the bubble thoroughly mixed with the brine for several hours using a magnetic stirring plate. The brine had an enrichment ~5.5 15N atom % as determined by membrane inlet mass spectrometry [20]. The second tube was filled with 6 l RNAlater solution [21] and 2.5 l of non-labeled brine.

Immediately upon recovery of the traps onboard, the upper layer of the tubes was carefully removed with a peristaltic pump until the density gradient of the brine was reached. The seawater layer overlying the brine prevented potential intrusions of surface seawater when recovering the trap line onboard. We confirmed that no surface contamination occurred since the particulate organic carbon fluxes measured in brine-filled traps agreed with those measured in parallel traps kept closed during recovery onboard (‘RESPIRE’ traps [22]; M. Bressac, personal communication). The brine of the 15N2-labeled traps was transferred to magnetic stirrer plates to ensure homogeneous aliquot sampling. Three aliquots of 50 ml were filtered onto precombusted 25 mm GF/F filters (GE Healthcare, Little Chalfont, UK) for bulk elemental analysis coupled isotope ratio mass spectrometry (EA-IRMS), and another three were filtered onto 1 µm polycarbonate filters (Nucleopore, Whatman, Maidstone, UK) and fixed with 1.6% microscopy grade paraformaldehyde for single-cell nanoscale secondary ion mass spectrometry (nanoSIMS) analyses (see below). Samples integrating biomass between 2000 and 200 m (bottlenet [16]) were used to measure natural 15N atom % enrichment of bulk biomass and Trichodesmium cells. In parallel, the whole brine volume of the tube filled with RNAlater was immediately filtered onto 0.2 µm polysulfone filters (Supor, Pall Gelman, Port Washington, NY, USA). The filters were transferred to bead beater tubes containing a mix of 0.1 mm and 0.5 mm silica beads, flash-frozen in liquid nitrogen and stored at −80 °C until RNA extractions (see below).

Bulk and Trichodesmium-specific N2 fixation rates

The 15N/14N ratio of dissolved nitrogen was measured by membrane inlet mass spectrometry and the bulk 15N/14N ratio and particulate nitrogen concentration of particles with an Integra CN EA-IRMS (SerCon Ltd, Chesire, UK) as described elsewhere [23]. The 15N atom % enrichment of Trichodesmium filaments was analyzed on a nanoSIMS 50 L (CAMECA, Gennevilliers, France) at the Leibniz Institute for Baltic Sea Research (IOW, Germany). Sample filters were mounted with conductive tape on 10 × 5 mm aluminum stubs (Ted Pella Inc., Redding, CA) and gold-coated to a thickness of ca. 30 nm (Cressington auto sputter coater). A 1 pA 16 keV Cesium (Cs+) primary beam was scanned on a 512 × 512 pixel raster with a raster area of 15 × 15 µm, and a counting time of 250 µs per pixel. Samples were pre-sputtered with 600 pA Cs+ current for 2 min in a raster of 30 × 30 µm to remove the gold and surface contaminants and reach the steady state of ion formation. Negative secondary ions 12C, 13C, 12C14N, 12C15N and 31P were detected with electron multiplier detectors, and secondary electrons were simultaneously imaged. Sixty serial quantitative secondary ion mass planes were generated, drift corrected and accumulated to the final image. Mass resolving power was >8000 to resolve isobaric interferences. Data was processed using the Look@nanoSIMS software [24]. Isotope ratio images were generated by dividing the 13C ion count by the 12C ion count, and the 12C15N ion count by the 12C14N ion count pixel by pixel. Individual Trichodesmium filaments (trichomes) were identified in nanoSIMS secondary electron 12C, 12C14N images. These images were used to define regions of interest (ROIs). For each ROI, the 13C/12C and 15N/14N ratios were calculated based on the ion counts averaged over the ROIs.

Bulk N2 fixation rates were calculated following the equations of Montoya et al. [25]. Trichodesmium-specific volumetric N2 fixation rates were calculated using the following equation:

$$Volumetric\;N_2fixation\;rate = \frac{{(A_{Tricho} - A_{TrichoNat})}}{{(A_{N_2} - A_{TrichoNat})}} \times \frac{{PN_{Tricho}}}{t}$$

where ATricho is the 15N atom% enrichment of individual Trichodesmium cells incubated with 15N2, ATrichoNat is the natural 15N atom% enrichment of Trichodesmium as analyzed by nanoSIMS (see above), AN2 is the 15N atom% enrichment of dissolved N2 measured by membrane inlet mass spectrometry (see above), PNTricho is the nitrogen biomass of Trichodesmium and t is the incubation time. PNTricho was calculated by converting Trichodesmium nifH gene copies l−1 (as provided by Bonnet et al. [26]) to carbon considering the average of the mmol C: nifH ratio provided in Meiler et al. [27]. Carbon was converted to nitrogen considering a C:N ratio of 6:1 [28]. Trichodesmium abundance was calculated by converting Trichodesmium nifH gene copies l−1 into cells l−1 considering a ratio of 12 and 103 nifH gene copies per Trichodesmium cell, obtained empirically for stations S05M and S10M, respectively, by comparing quantitative PCR and microscopy Trichodesmium counts [26]. Finally, Trichodesmium cell-specific N2 fixation rates (fmol N cell−1 d−1) are calculated by dividing volumetric rates by Trichodesmium abundance (cells l−1).

RNA extractions, nifH gene sequencing and bioinformatics

RNA was extracted using the RNeasy mini kit (Qiagen) including a 1 h on-column DNase digestion. PCRs on extracted RNA controlled for complete DNA digestion. Reverse transcription (RT) were performed with TaqMan Reverse Transcription (Applied Biosystems) using reverse primer nifH3 (5’-ATR TTR TTN GCN GCR TA-3’) and 5 μl of RNA extract. Triplicate nested PCR reactions were conducted using degenerate nifH primers nifH1 (5’-TGYGAYCCNAARGCNGA-3’write here), nifH2 (5’-ADNGCCATCATYTCNCC-3’), nifH3 and nifH4 (5’-TTYTAYGGNAARGGNGG-3’) [29]. The PCR mix was composed of 5 μl of 5X MyTaq red PCR buffer (Bioline), 1.25 μl of 25 mM MgCl2, 0.5 μl of 20 μM forward and reverse primers, 0.25 μl Platinum Taq and 5 μl of cDNA (1 μl from the first PCR reaction was used as template in the second reaction). The reaction volume was adjusted to 25 μl with PCR grade water. Triplicate PCR products were pooled and purified using the Geneclean Turbo kit (MP Biomedicals). Samples were sequenced using the Illumina MiSeq platform with 2 × 300 bp paired-end reads. Demultiplexed paired-end sequences were dereplicated, denoised, assembled and chimeras discarded using the DADA2 pipeline [30]. This generated 6 929 ASVs (146,000 ± 18,000 reads per sample). Sequences have been deposited in the Sequence Read Archive under accession number PRJNA742179. The ASVs were translated to amino acid sequences using FrameBot [31] and filtering for homologous genes was done following the NifMAP pipeline [32]. This reduced the number of ASVs to 6 503 accounting for 842 515 reads (96% of all reads). Taxonomic ranks were assigned according to the nifH gene reference database collated and maintained by the Zehr research group (v. June 2017; https://www.jzehrlab.com/nifh).

Hydrostatic pressure experiments with Trichodesmium cultures

The effect of increasing hydrostatic pressure and decreasing temperature on Trichodesmium was tested using a sinking particle simulator [33]. Exponentially growing cultures of Trichodesmium erythraeum IMS101 grown at 27 °C under a 12 h:12 light:dark cycle on YBCII medium [34] were transferred to four autoclaved 500 ml high pressure titanium bottles (HPBs) placed inside incubators (Memmert IPP 750 + , Schwabach, Germany) programmed to decrease temperature at the desired pace. For 360 h, the pressure within the HPBs was increased from 0 to 3 MPa using a piloted pressure generator based on a motorized syringe controlled by a computer. Pressure was logged continuously by means of a Metrolog (Metro-Mesures, Mennecy, France) and controlled by the software with a precision of 0.2%. Concomitantly, temperature of the HPBs was decreased from 27 to 14 °C. These conditions simulate those experienced by cells when sinking from the surface to ~300 m depth with a conservative sinking speed of 20 m d−1 (at the lower end of empirically measured Trichodesmium sinking velocities; Table S2). Triplicate analytical culture aliquots were sampled from two HPBs at 192 and from another two HPBs at 360 h, corresponding to simulated depths of 160 and 300 m, respectively. N2 fixation incubations at these time points were done under the corresponding hydrostatic pressure conditions by 100 ml culture aliquots to pressurized titanium flasks containing 30% volume of culture medium previously enriched with 15N2 gas (prepared as explained above) and incubated for 24 h. Particulate and dissolved samples were analyzed by EA-IRMS and MIMS as described above. Cultures under a 12 h: 12 h dark:dark cycle and temperature decrease pace identical to that of pressurized cultures were used as a control.

Sinking Trichodesmium cell metabolism model

We hypothesized that Trichodesmium fixes N2 while sinking into the dark ocean at the expense of carbon accumulated from photosynthesis before starting to sink. Sinking Trichodesmium can thus only fix N2 until reaching the depth where its carbon storage is depleted. To compute how much carbon storage Trichodesmium needs to acquire before sinking to be able to fix N2 at 1000 m depth, we adapted a previously published coarse-grained cell metabolism model [35]. The model simulates photosynthesis, N2 fixation, and respiration for the entire trichome, resolving diffusion boundary layers for oxygen [36] transport and distinguishing between photosynthetic and non-photosynthetic cells (Fig. S1). Photosynthetic cells fix carbon with harvested light energy, accumulate carbon, and synthesize new biomass, whereas non-photosynthetic cells use carbon stored by other cells to fix N2 (Fig. S1).

To adapt the model to Trichodesmium colonies sinking into the mesopelagic layer we considered the variability in temperature and oxygen observed from the ocean surface to 1000 m in our study (Fig. S2) as well as the inhibiting effect of increasing nitrate concentrations on N2 fixation. Walden’s rule [37] was used to simulate the variation of oxygen diffusivity, as in previous studies [38], and the Arrhenius equation was used to simulate the temperature dependencies of metabolisms [35]. The adapted model was applied to Trichodesmium sinking from the bottom of the mixed layer (~40 m; Fig. S2) over a wide range of calculated velocities (ranging from 12 to ~600 m d−1; Supplementary Methods) considering a ratio of initial carbon storage level relative to non-storage biomass (RSto) ranging between 0 and 2. This value of RSto range is conservative, since it varies between 1 and 10 according to carbohydrate and lipid to protein ratios in various phytoplankton species [39]. To test the effect of nitrate inhibition on Trichodesmium’s N2 fixation in the mesopelagic, we considered a decrease in the diazotrophically active cells by 70 and 50% in each trichome. All model equations are detailed in Inomura et al. [35] and the adapted code is fully available in Zenodo (https://zenodo.org/record/5153594; https://doi.org/10.5281/zenodo.5153594).

Statistical analyses

15N atom % enrichment values were checked for normality using a Shapiro-Wilk test and significant differences between samples tested with Wilcoxon test, using R software package dplyr in RStudio Version 1.2.5033.

Results and discussion

Trichodesmium fixes N2 and expresses nifH in the mesopelagic ocean

The isotopic enrichment of single Trichodesmium filaments in the traps ranged between 0.428 ± 0.002 and 0.463 ± 0.033 15N atom % (Fig. 1). The 15N atom % enrichment of all Trichodesmium filaments analyzed was significantly higher than that of filaments not incubated with 15N2 collected over the same depth range (0.364 ± 0.006 15N atom %; Wilcoxon test p = 0.004 and p = 0.014 for S05M and S10M, respectively; Fig. 1). The derived cell-specific N2 fixation rates ranged between 36 and 214 fmol N cell−1 d−1 (Table S3), in the same range of previous Trichodesmium cell-specific N2 fixation measurements in surface waters of the South Pacific [40, 41]. While the salinity in the traps (~50 ppt) was higher than that of ambient waters (~34 ppt), previous studies have shown that increased salinity reduces but does not impair Trichodesmium growth up to 42 ppt [42]. Comparing in Trichodesmium cultures grown on 34 and 50 ppt showed a decrease in N2 fixation rates by 63 ± 15% (Supplementary Methods). Hence, the high salinity of the 15N2-labeled brine may have reduced but did not completely inhibit N2 fixation in Trichodesmium, implying that in situ rates could be higher than measured in our traps. Overall, these cell-specific rates indicate that Trichodesmium sinking into the mesopelagic zone can fix N2 at rates comparable to the surface.

Fig. 1: Trichodesmium-specific 15N atom % enrichment.
figure 1

A, B Boxplots showing the range, average and outliers of 15N/14N ratios measured in natural (non 15N2-labeled) and 15N2-labeled samples from 170 m, 270 m and 1 000 m depth at stations S05M and S10M, respectively. The number of trichomes scanned per depth is shown over each box (n values). All Trichodesmium filaments analyzed were significantly enriched in 15N. Examples of nanoSIMS 15N/14N ratio images of Trichodesmium filaments sampled at station S05M showing the 15N/14N isotopic ratio enrichment of Trichodesmium filaments according to the color bar. Filaments shown in panel (C) are those not incubated with 15N2 (natural). The filaments shown in panels (DF) are those incubated with 15N2 and collected from sediment traps deployed at 170, 270, and 1000 m, respectively. The same pattern is repeated for station S10M in panels (GJ) The scale bar in nanoSIMS 15N/14N ratio images is 5 µm.

Amplicon sequencing revealed that nifH gene transcripts annotated as Trichodesmium were present in all sediment trap samples, accounting for 26–84% (average 56%) of the sequence reads (Fig. 2). This indicates that Trichodesmium constituted a substantial fraction of the diazotroph community that actively transcribed the nifH gene in all samples (Fig. 2). However, comparing the 15N2 atom % enrichment of bulk sediment trap material and individual filaments, Trichodesmium accounted for 1 to 70% of the N2 fixation activity measured in the traps (Table S3). Hence, other organisms contributed to N2 fixation within the traps, which could be driven by either surface diazotrophs attached to particles or by true mesopelagic N2 fixation driven by diazotrophs residing in deep waters. The contribution of Trichodesmium to mesopelagic N2 fixation in the subtropical region studied here is notorious but likely restricted to the (sub)tropical regions where this cyanobacterium abounds. The predicted warming and expansion of oligotrophic subtropical gyres towards higher latitudes may expand the effect of sinking Trichodesmium to higher latitudes in the future [43].

Fig. 2: Diazotroph community expression profiles.
figure 2

Diazotroph community nifH gene expression profiles. Relative abundance of nifH cDNA reads originating from sediment traps at 170 m, 270 m, and 1000 m at stations S05M (A) and S10M (B). Relative abundances were calculated based on 146,000 ± 18,000 reads per sample.

Crocosphaera, the only other cyanobacterial group present in the cDNA sequence data, comprised 10% of the cDNA sequence reads of station S05M at 270 m, but less than 0.1% in the other samples. Sequences annotated as the genus Chlorobium (Bacteroidota) comprised 60% of cDNA sequences recovered from 1000 m depth at station S05M and constituted a substantial fraction of the expression profiles of station S10M at 170 m and 270 m, comprising 16% and 20% of cDNA sequences, respectively, but their contribution was less than 1% in the rest of the samples. This indicates that, when specific conditions are met, Chlorobium can contribute substantially to the nifH transcript pool. A noteworthy contribution to the expression profiles was Alphaproteobacteria classified as the genus Yangia (Rhodobacteraceae [44]), which comprised approximately 10% of cDNA sequences of S05M but were virtually undetected at S10M. Both Bacteroidota and Rhodobacteraceae have been identified as epibionts of Trichodesmium [45] and were likely attached to the colonies as they sunk from the surface ocean into the traps. The other most prevalent group was Deltaproteobacteria, with genera such as Desulfatibacillum, Desulfobacter, Desulfolobus, and Desulfovibrio together comprising 12.5% of cDNA reads on average. This group has not been previously identified as an epibiont of Trichodesmium but is commonly observed in sinking particles intercepted with sediment traps in the dark ocean [21].

Sinking Trichodesmium simulation experiments

We confirmed the ability of Trichodesmium to fix N2 under mesopelagic conditions using a particle sinking simulator (see Methods; [43]). Trichodesmium cultures submitted to increasing hydrostatic pressure for 192 and 360 h in the dark (equivalent to 150 and 300 m simulated depth) had 15N atom % enrichment values of 0.375 ± 0.012 and 0.370 ± 0.002 atom %, respectively (Fig. 3A), indicating low but measurable N2 fixation. Control Trichodesmium cultures kept at ambient lab pressure conditions had 15N atom % enrichments of 0.390 ± 0.012 and 0.393 ± 0.008 after 192 and 360 h of incubation, respectively (Fig. 3B). This difference in pressurized (Fig. 3A) and non-pressurized (Fig. 3B) N2 fixation rates is in agreement with previous metabolic rate slowdown in epipelagic prokaryotes submitted to mesopelagic pressure levels [46] and indicates that increasing pressure reduces but does not completely incapacitate N2 fixation in Trichodesmium. We however note that the while the sinking simulator allows changing pressure and temperature along the simulated sinking process, it does not allow changing nutrient concentrations. The increasing concentrations of nitrate with depth in the mesopelagic may inhibit N2 fixation. Thus, the rates measured in these sinking simulations represent a non-inhibited upper limit of diazotrophic activity.

Fig. 3: Trichodesmium culture sinking simulation experiments.
figure 3

A 15N atom % enrichment values of triplicate Trichodesmium cultures submitted to a 12 h:12 h dark:dark light cycle under increasing hydrostatic pressure (0–3 MPa) and decreasing temperature (27–14 °C) on a sinking particle simulator for 192 and 360 h (simulating 160 and 300 m depth, respectively). B 15N atom % enrichment values of triplicate Trichodesmium cultures submitted to a 12 h:12 h dark:dark at ambient laboratory pressure and decreasing temperature (27 to 14 °C) for 192 and 360 h.

How does Trichodesmium obtain energy to fix N2 at depth?

Trichodesmium uses photosynthetically fixed carbon to fuel the energetically expensive process of N2 fixation [47]. Hence, the dark conditions of the mesopelagic ocean should be expected to halt photosynthesis and N2 fixation in Trichodesmium, eventually leading to its death. We hypothesized that Trichodesmium can fix N2 in the dark ocean using stored intracellular carbon acquired by photosynthesis before starting to sink. To test how much carbon storage Trichodesmium would need to fix N2 until reaching 1000 m of depth (corresponding to our field observations), we adapted a published cell metabolism model [37] (Fig. S2). The model simulates changes in intracellular carbon storage (the ratio of carbon storage to biomass or RSto) while sinking linearly in the water column over a wide range of sinking velocities. The model output provides a relationship between the cell’s initial RSto (RSto value before starting to sink) and the depth at which carbon storage depletes (Fig. 4).

Fig. 4: Relationship between the initial carbon storage to biomass ratio (initial RSto) of Trichodesmium cells and the depth at which cell carbon storage is depleted.
figure 4

The x-axis represents the initial RSto, i.e., the carbon storage to biomass ratio of Trichodesmium at 40 m (the bottom of the mixing layer, where cells start to sink). The y-axis represents the depletion depth, i.e., the depth at which carbon storage becomes zero. The blue line represents the threshold depth below which cell carbon storage is depleted (and thus N2 fixation is no longer possible) for various initial RSto values. The shaded area indicates the depletion depth range considering the variability in empirical and theoretical sinking velocities of Trichodesmium (Supplementary Methods; Table S2). The yellow and pink dots indicate the depletion depth for initial RSto values of 1 and 1.3, respectively. The blue line represents the simulation based on the median sinking velocity. The dashed line indicates the depletion depth of 1000 m depth. The depletion depth must be below 1000 m for cells to fix N2 at 1000 m depth and explain our field observations.

Considering that Trichodesmium starts sinking below the mixed layer depth (~40 m; Fig. S2), the cell’s metabolism becomes dependent on intracellular carbon storage when light is extinguished (~170 m; Fig. S2). Below that depth, RSto decreases (Fig. 4). When the initial RSto = 1 (which is at the lower end of the range measured in phytoplankton, see Methods), the cell’s carbon storage is depleted at around 750 m assuming a medium sinking velocity of 375 m d−1 (yellow dot in Fig. 4). With a higher initial RSto of 1.3 and the same sinking velocity, carbon storage is depleted at 1000 m. This suggests that an initial RSto of at least 1.3 is required for Trichodesmium cells to reach 1000 m of depth with enough carbon to sustain N2 fixation (pink dot in Fig. 4) and be consistent with our field data (Fig. 1). We also note that Trichodesmium may obtain carbon from dissolved organic matter available in their surrounding medium [41], which is not modeled here. This could decrease the required initial RSto to sustain N2 fixation at 1000 m depth.

Trichodesmium invests a significant part of its carbon storage in maintaining low intracellular oxygen concentrations through respiratory protection [37, 48]. The decrease in temperature with depth slows down Trichodesmium’s metabolism, decreasing the speed of oxygen diffusion (Walden’s rule [37]), which reduces the level of respiratory protection needed to fix N2 and decreasing and thus carbon storage consumption. While the decrease in temperature with depth also increases the saturated concentration of oxygen, the concentration of oxygen below the mixed layer in our field experiment was under-saturated (Fig. S2), with oxygen deficit particularly pronounced below 600 m. The combined effects of declining temperatures and oxygen under-saturation contribute to the reduction in carbon storage consumption and are reflected as an increasing slope in the non-linear curve in Fig. 4.

The high nitrate concentrations of the mesopelagic ocean could inhibit N2 fixation in Trichodesmium [49]. While no studies have explicitly tested the inhibition of N2 fixation by nitrate in the mesopelagic, we explored this effect by considering two constant levels of inhibition (70% and 50%; Fig. S3), which are in the upper range of previous culture and field studies [50]. The model predicts that under 70% and 50% inhibition the required initial RSto decreases to ~0.5 and ~0.8, respectively (Fig. S3). This indicates that when nitrate inhibits N2 fixation less carbon storage is consumed, allowing Trichodesmium to remain metabolically active at deeper depths (Fig. S3). Previous studies have shown that Trichodesmium can fix N2 in the presence of up to 20 µM nitrate provided phosphate concentrations are high enough to sustain growth [50]. At the depths where traps were deployed during our field experiment, excess phosphate with respect to nitrate according to the Redfield stoichiometry of 16:1 (P* parameter; Fig. S2) likely permitted N2 fixation (Fig. 1). Finally, the increased partial pressure of CO2 with depth could also favor N2 fixation in Trichodesmium as previously shown in culture experiments [51].

Is fast sinking necessary for Trichodesmium to fix N2 in the mesopelagic?

Previous studies have suggested that viable and/or active growing cyanobacteria in the dark ocean are associated with fast-sinking particles, mineral ballasting, or episodic flux events [17, 18, 52]. While the gas vesicles of Trichodesmium may prevent it from sinking, a previous study found that sinking Trichodesmium can contribute importantly to organic matter export during atmospheric dust ballasting events [50]. Three months before our cruise, a volcano erupted over the Tonga-Kermadec volcanic arc [53]. This volcano was particularly close to station S10M (~144 km). Measurements of lithogenic silica in the trap material (Supplementary Methods) showed that the lithogenic to particulate organic matter ratio of particulate matter was maximal at station S10M (Table S4). Scanning electron microscopy images (Supplementary Methods) showed volcanic materials intermingled with Trichodesmium filaments (Fig. S4), particularly at the 1 000 m trap of station S10M where Trichodesmium nifH gene transcripts were more abundant (Fig. 2). Hence, the eruption may have stimulated fast sinking by Trichodesmium, particularly at the S10M site. However, both our laboratory sinking simulations and the results of the cell metabolic model indicate that Trichodesmium can fix N2 within the mesopelagic depth range even when sinking at low velocity (e.g., 20 m d−1).

Potential impact of sinking Trichodesmium on mesopelagic prokaryotic communities

Trichodesmium releases up to 19% of the N2 fixed as ammonium [54], which is the main energy source for dark CO2 fixation by the abundant Thaumarchaeota [55]. Considering the Trichodesmium-specific N2 fixation rates measured here and a ratio of dark CO2 fixed per ammonium oxidized of 0.1 [56, 57], the ammonium released by sinking Trichodesmium may sustain CO2 fixation rates of up to 0.2 µmol C m−3 d−1. This represents ~13% of previous dark CO2 fixation rates measured in the mesopelagic ocean [58]. However, to date very few dark CO2 fixation measurements are available and their magnitude in Trichodesmium-dominated regions (particularly in the South Pacific Ocean) is unknown.

More importantly, Trichodesmium releases ca. 50% of its fixed N2 as dissolved organic nitrogen [59], which mostly composed of labile amino acids [60]. The mycosporine and tryptophan-like compound optical signatures observed at >1000 m in our study are consistent with previous measurements of natural [61,62,63] and cultured Trichodesmium colonies (Supplementary Methods; Fig. S6). These signals were particularly strong at S10M (Fig. S5) coinciding with highest Trichodesmium nifH transcripts (Fig. 2) and supporting the active release of amino acids by sinking Trichodesmium. Amino acids released by active sinking Trichodesmium could add up to the labile compounds released by sinking particles in the dark ocean [64,65,66,67], contributing to the prokaryotic respiration organic matter [68] in the subtropical and tropical regions where Trichodesmium occurs.

Conclusions

This study provides the first cell-specific N2 fixation rates, nitrogenase expression and metabolic mechanistic understanding of Trichodesmium sinking into the mesopelagic zone, representing a step forward from early findings of viable surface ocean phytoplankton at depth. Diazotrophically active Trichodesmium can provide mesopelagic bacteria and archaea with ammonium and labile amino acids, contributing to chemolithoautotrophy and organic matter remineralization in the mesopelagic zone, respectively. The balance between these two processes sets the ultimate role of the dark ocean in carbon sequestration with consequences for global climate. Given the widespread blooms of Trichodesmium in the surface ocean, which are predicted to expand due to climate change [43], we contend that the impact of sinking Trichodesmium on the biogeochemistry of the dark ocean needs to be considered in carbon sequestration models.