Introduction

Biological nitrogen (N) fixation, or the conversion of inert N2 gas into bioavailable forms of N, is an important process in aquatic environments where it may relieve N limitation of primary productivity. Nitrogen fixation itself may be regulated by the availability of phosphorus (P), iron (Fe), and/or fixed N1,2,3,4,5,6. While phosphorus is an essential macronutrient for the growth of all organisms, Fe is critical due to the high Fe demand of the key enzyme responsible for N2 fixation, nitrogenase. The availability of fixed N, such as nitrate and ammonium, generally slows down N2 fixation but does not necessarily deactivate it6,7,8,9,10,11,12. In rare cases, N2 fixation rates even correlate positively with nitrate concentrations13,14. The supply and interplay of these nutrients (P, Fe, and fixed N) to and in aquatic environments, potentially leads to the large-scale patterns that are currently observed. The stratified, oligotrophic open oceans, such as the subtropical ocean gyres, are typically important sites of N2 fixation, driving up to 50% of organic N and carbon export rates15. However, absolute rates of N2 fixation in these regions can vary by orders of magnitude16, and are possibly linked to varying fluxes of P and Fe into these ecosystems17. Furthermore, the N2-fixing communities can differ from cyanobacteria-dominated regions, such as the tropical North Atlantic18, to regions where non-cyanobacteria may be the more important contributors19.

In contrast to the nutrient-poor surface waters of the open oceans, eastern boundary upwelling systems frequently introduce excess P to surface waters (and potentially Fe20) that have been proposed to foster N2 fixation21. Field studies, however, have found upwelling N2 fixation rates to be overall lower than in much of the open ocean (e.g. Selden et al.22, and references therein). While rates vary widely from below the limit of detection to as high as 100 nM d−1 13,15,22,23,24,25,26,27,28,29,30, they do not re-supply the N lost before/during upwelling as predicted by previous biogeochemical models21. More recent estimates indicate that N2 fixation contributes little as a new N source in upwelling regions, contributing only ~ 1% to the export of organic N and carbon15. Nevertheless, heterotrophic diazotrophs are ubiquitously found in nutrient-rich waters – including the more recently studied coastal and polar systems – which are often but not exclusively dominated by non-cyanobacterial diazotrophs23,24,31,32,33,34,35,36. Despite their prevalence in pelagic environments31,37, heterotrophic diazotrophs may contribute to low rates ( < 1 nM d−1) of N2 fixation37,38.

The large spatial variability in both the rates of N2 fixation and the associated diazotrophic communities, even in environments with similar biogeochemical conditions, indicate a lack of a mechanistic understanding39. A clear causal link between nutrient supply, largely driven by hydrodynamics (e.g. stratification or upwelling), and N2 fixation rates as well as the ambient N-fixing microbial communities has not been established8,9,10,40. Yet, recent models highlight that the parameterization of physical-biological processes is fundamental to understanding the observed distribution of N2 fixation in marine settings15.

The sheer size of the ocean and the complexity of hydrodynamic processes within also further challenge our capacity to identify mechanisms of physical-biological coupling. To this end, Lake Tanganyika (the second deepest lake in the world), which exhibits contrasting hydrodynamic regimes between its north and south basins, represents an ideal testing ground for studying how upwelling and stratification may regulate N2 fixation and the community of diazotrophs. The lake water column is generally depleted in N relative to P and is characteristically structured into oligotrophic surface waters, a nitrate-rich intermediate zone, and a euxinic hypolimnion41. During the dry season, the south basin experiences a strong wind-driven upwelling, while the north/central basins remain permanently stratified42. Previous studies have reported non-diazotrophic phytoplankton blooms in the south during the dry season, whereas massive blooms of filamentous N2-fixing cyanobacteria typically emerge during more intensely stratified periods43,44. A microbial community survey has identified additional putative photoautotrophic and heterotrophic diazotrophs45.

Here, we explored the distribution of N2 fixation and associated diazotrophs under contrasting hydrodynamic regimes across a north-south transect in Lake Tanganyika (Fig. 1a). Using a combination of stable isotope applications and metagenomics, we find that the hydrodynamic regimes selected for particular ecophysiological traits that might enable the key diazotrophs to thrive under either stratified or upwelling conditions.

Fig. 1: Distribution of nutrients, chlorophyll, filamentous nitrogen-fixing cyanobacteria, and rates of nitrogen fixation across a north-south transect in Lake Tanganyika.
figure 1

The transect and profiles shown were taken in April/May 2018. a Aerial surface chlorophyll-a concentrations are represented by the Trophic State Index (TSI), obtained from Sentinel satellite data from 27 April to 7 May 2018. TSI values are proxies of the trophic state: 0–30, indicates clear oligotrophic waters; 40–50, mesotrophic; 60–70, eutrophic (see methods). Panels b–g show the vertical distributions of temperature, nitrate, soluble reactive phosphate (SRP), chlorophyll (Chl), volumetric rates of nitrogen fixation, and abundance of filamentous cyanobacteria. Note that dots indicate discrete sampling depths, while solid vertical lines mark CTD profiles. In panels h–m, depth-integrated values and general north-south patterns are depicted. h Indicates the position of the thermocline (TC) relative to the euphotic depth (Zeu); i surface nitrate inventory (0–25 m) and nitrate flux into the euphotic zone; j surface SRP inventory (0–25 m); k depth-integrated in-situ chlorophyll fluorescence (0–125 m); l depth-integrated values of nitrogen fixation (0–175 m) and the mean δ15N of particulate organic nitrogen in the euphotic zone (0–43 m; δ15N-PONzeu); m depth-integrated values (0–125 m) of filamentous cyanobacteria. Error bars represent standard errors. The main sampling Stations 2 and 7 are marked by red and blue downward triangles, respectively. Stations 1–3, 4–6, and 7–9 represent the North (NB), Center (CB), and South basins (SB), respectively. Data in panels i and l are presented as mean values from duplicate samples (distinct samples), with error bars representing the standard error. The world globe, temperature and chlorophyll-a presented in panels a, b, and e, respectively, were published previously44,47 (licensed under a Creative Commons Attribution 4.0: https://creativecommons.org/licenses/by/4.0/).

Full size image

Results and discussion

Hydrodynamic regimes in Lake Tanganyika

We sampled Lake Tanganyika during the transition from the rainy to the dry season in April/May 2018. This period was characterized by lake-wide stratification with warm surface waters (up to 27.6 °C) overlaying cooler waters (Fig. 1b). Under these stratified conditions, the water column was deficient in oxygen (<10 µM oxygen) below ~110–120 m in the north and below ~130–140 m in the south (Fig. 1b and Supplementary Fig. 1a). The deeper oxycline in the south is a robust feature of Lake Tanganyika46, and develops as a result of deep vertical mixing down to ~150 m42. Even though the entire lake was stratified at the time of sampling, north-south differences were apparent. For instance, the thermocline, separating the warm surface layer from deeper waters, was generally positioned at a much shallower depth (<23 m) in the southern basin compared to most stations sampled in the northern and central basins (45–64 m). The shallower thermocline in the south arose as a consequence of wind-driven upwelling induced by the southeast trade winds42.

Overall, in the upper 50 m of the lake, nitrate concentrations were mostly below the limit of detection (<0.25 µM in 63 % of samples), except for Station 1 and at one depth at Station 3 (Fig. 1c, i). Below ~50 m, nitrate concentrations increased sharply, reaching a maximum of 8–9 µM near 100 m water depth. Nitrate concentrations eventually fell below the detection limit from ∼150 m downward, due to anammox and denitrification activity47 (see also Supplementary Discussion). By contrast, soluble reactive phosphate in the surface waters was below the limit of detection ( < 0.19 µM) only at stations 1 and 2 (Fig. 1d, j), and increased from 50 m downward, reaching up to 1.3–1.5 µM near the nitrate maximum. Ammonium concentrations were below the limit of detection (<0.26 µM) in oxygenated waters and increased below ~150 m (Supplementary Fig. 1b). The low N:P ratios, which were below the Redfield ratio of 16:1 in 95 % of all samples (Supplementary Fig. 1c), indicating that soluble reactive phosphate was available in excess over DIN throughout the water column, consistent with previous studies43,44. The resupply of DIN to the euphotic zone (delimited at a surface irradiance cutoff of 1%; Fig. 1) is controlled by the primary thermocline44, separating the nitrate-depleted surface zone from nitrate-rich intermediate waters. At Stations 2–5, located in the heavily stratified north/center, the thermocline was situated 9 m below the lower boundary of the euphotic zone (Fig. 1h), thereby diminishing the resupply of nitrate to phytoplankton. An additional surface thermocline, observed at Stations 3 and 4, further limited the vertical nutrient transport to the surface mixed layer (upper ~30 m; Fig. 1c). In contrast, at Stations 8 and 9 in the south, the thermocline was situated 10 m above the lower boundary of the euphotic zone, which allows nitrate to be resupplied for assimilation in surface waters more efficiently44. Estimated fluxes of nitrate into the euphotic zone ranged from 140–300 µmol m−2 d−1 in the north/center, and up to 690 µmol m−2 d−1 at Station 9 in the south (Fig. 1i).

Lake Tanganyika also exhibited strong latitudinal trends with regard to chlorophyll, an indicator of primary productivity. In the south, the primary chlorophyll maximum (PCM) was located between the thermocline and the bottom of the euphotic zone, whereas the PCM occurred near the surface in the north and central basins (Fig. 1e). In-situ chlorophyll fluorescence was highest at Stations 7–9 in the south (>1.5 µg L−1; Fig. 1e, k), consistent with previously described basin-wide patterns48. In further support, higher CO2 fixation rates and 13C-enriched particulate organic carbon (indicating higher fractional dissolved CO2 consumption) were reported for the south, compared to the north49. We ascribe the greater surface productivity in the south to the nutrient upwelling; accordingly, we relate the less productive surface PCM in the north to a reduced turbulent diffusive DIN supply due to stronger stratification. Furthermore, the lower surface productivity in the north enabled greater light penetration to deeper depths that (Supplementary Fig. 2), in turn, sustained deep blooms of anoxygenic phototrophs that resulted in an anoxic chlorophyll maximum (ACM) occurring at 140–170 m (Fig. 1e)47. This, combined with the fact that vertical mixing causes a deepening of the oxycline could additionally constrain the available niche for anoxygenic phototrophic sulfur bacteria. The presence of an ACM therefore serves as an indicator of permanently stratified waters and is otherwise absent in the more productive south (Supplementary Fig. 3).

Upwelling impacts filamentous cyanobacteria and associated nitrogen fixation

Based on our 15-15N2 incubation experiments, we find the greatest rates of N2 fixation in the euphotic zone (i.e., above 50 m depth), ranging from below the limit of detection to 22.9 nM N d−1 (Fig. 1f). Our maximum volumetric rates were higher than those measured in the neighboring Lake Malawi (~0.5–5 nM N d−1)50, but comparable to maximum rates measured in the oligotrophic Laurentian Great Lakes of 30 nM N d−1 (with a whole lake mean value between 1–4 nM N d−1;14). It should be noted that in small eutrophic lakes rates of N2 fixation can be 1–2 orders of magnitude greater51,52. Moreover, our rates were higher than in most marine upwelling areas, where N2 fixation typically ranges from below detection to ~10 nM N d−1 8,10,13,23, although exceptional rates above 100 nM N d−1 have been reported9,32. Along our north-south transect, rates of N2 fixation were greatest in the north/central basins, with maximum rates ranging from 7.5–22.9 nM N d−1 (Stations 2, 4, and 6), whereas rates never exceeded 3 nM N d−1 in the south (Stations 7 and 9; Fig. 1f). In accordance with the volumetric rates, the depth-integrated N2 fixation estimates ranged between 192–480 µmol N m−2 d−1 in the north/central basins, which was between 2- to 10-fold higher than values reported in the south basin (46–107 µmol N m−2 d−1; Fig. 1l).

Averaged on a basin scale, N2 fixation in the upper 50 m added 300 ± 70, and 60 ± 40 µmol N m−2 d−1 to the north/central and south basins, respectively (Table 1). The estimated vertical nitrate flux into the euphotic zone amounted to 230 ± 30, and 500 ± 10 µmol N m−2 d−1 in the north/central and south basins, respectively. Atmospheric deposition plus river input may add another 160 and 140 µmol N m−2 d−1, based on annual estimates for the two regions53. These numbers indicate that N2 fixation was the main source of bioavailable N fueling new production in the north and central parts of Lake Tanganyika, whereas the vertical nitrate flux was the major source of bioavailable N in the south.

Table 1 Nitrogen fluxes with standard errors (mmol N m−2 d−1) in Lake Tanganyika during April/May 2018 for the north/central and south basins
Full size table

Our 15N natural abundance analysis of particulate organic N (PON), which provides integrated information on N2 fixation into phytoplankton biomass over longer time scales, showed strong vertical and north-south trends that were consistent with rates determined from short-term 15-15N2 incubation experiments (Fig. 1l). The vertical distribution of δ15N-PON indicated consistently low values in the euphotic zone (0–43 m) attaining a mean (based on all stations) of −0.3‰, indicative of low-δ15N N2 as an important N source sustaining the phytoplankton community (Supplementary Fig. 4)54,55,56. By contrast, below 50 m depth, the higher average δ15N-PON values of 1.4‰, could be ascribed to the assimilation of nitrate (~1.9‰) as the main N source rather than by N2 fixation. In line with the high N2 fixation rates in the north, we find the lowest δ15N-PONZeu signature at Station 3 (northernmost analyzed station) of −1.4‰. In contrast, a significantly higher δ15N-PONZeu value of 1.6‰ was observed in the southern basin at Station 9, while intermediate values were observed for stations in between ranging from −0.8 to −0.1‰. Together the δ15N-PONZeu north-south trend corroborates the differences in surface N2 fixation across the lake.

The north and central basins of Lake Tanganyika were also a hotspot for diazotrophic filamentous cyanobacteria (Stations 2–6; Fig. 1g, m), reaching up to 22 colonies mL−1, whereas the south was largely void of filamentous cyanobacteria. In our parallel study addressing the phytoplankton composition, we found that these blooms were mainly comprised of known diazotrophs Dolichospermum (>99 %), and to a small extent Anabaenopsis (<1 %)44,57. The role of Dolichospermum in N2 fixation was further supported by the presence of a nifHDK operon identified in a recovered Tanganyika Dolichospermum metagenome-assembled genome (MAG) discussed below. Moreover, our sampled Dolichospermum colonies contained up to four heterocyst cells, which are specialized in the catalysis of N2 fixation44.

Depth-integrated abundances of filamentous cyanobacteria compared versus rates of euphotic zone N2 fixation and δ15N-PON were significantly correlated (Spearman, p < 0.05; Fig. 1l, m and Supplementary Fig. 9b, f). Together this suggested that Dolichospermum was a key contributor to the elevated N2 fixation rates in the northern and central basins (Fig. 1f, l). Based on this clear relationship (Supplementary Fig. 9b), and the average number of cells per colony44, we estimate that single-cell N2 fixation rates were roughly 15 fmol N cell−1 d−1, which is within the range of values reported for Dolichospermum in the Baltic Sea (~1–220 fmol N cell−1 d−1)57. Filamentous cyanobacteria were also shown to be responsible for the highest rates of N2 fixation in other large lakes14,50, as well as in many oligotrophic regions of the ocean58,59,60.

Across the lake, the abundance of filamentous cyanobacteria exhibited a significant negative correlation with the nitrate flux from subsurface waters (Spearman, R2 = 0.79, p < 0.05; Supplementary Fig. 9a). Moreover, the rates of N2 fixation showed a trend towards higher activity with a decreasing nitrate flux, albeit this trend was relatively noisy (R2 = 0.50, p < 0.2; Supplementary Fig. 9g). Non-steady state conditions might potentially contribute to a wider margin of error in correlative analyses at some stations. For example, a fluctuating DIN supply is expected for Stations 1 and 2, which are positioned near the river inlets of Rusizi and Malagarasi, respectively. These rivers experience large DIN fluctuations due to rain events during the transition from the wet to the dry season53. Therefore, N2 fixation rates may suffer from a lagged response by the diazotroph community toward a fluctuating nitrate supply44. Nevertheless, based on the δ15N-PONZeu signature (a long-term indicator of N2 fixation) we find a strong correlation with the nitrate flux (R2 = 0.79; p < 0.05; Supplementary Fig. 9d). In further support, the correlation of the nitrate flux with the abundances of filamentous cyanobacteria (Supplementary Fig. 9a), combined with their correlation to N2 fixation activity (Supplementary Fig. 9b), affirm that the nitrate flux is a key moderating factor of cyanobacterial N2 fixation in Lake Tanganyika.

Thus, we attribute the high N2 fixation rates by filamentous cyanobacteria in the north and center basins of Lake Tanganyika to the low resupply of DIN, which provides a competitive advantage to N2-fixing phytoplankton44,61. An exception to this trend is the northernmost Station 1, which was largely devoid of N2-fixing cyanobacteria (Fig. 1g, m). Here, the relatively high availability of DIN in surface waters (Fig. 1c, i), which may have been injected from deeper layers by internal waves62, or may stem from riverine sources53, likely negated the competitive advantage of N2-fixing phytoplankton as discussed elsewhere44. Likewise, in the south basin, the wind-driven upwelling of nutrient-rich waters impeded the proliferation of filamentous cyanobacteria. Instead, these waters were dominated by heterotrophic diazotrophs (discussed below) and were associated with much lower N2 fixation rates compared to the north basin. Interestingly, interseasonal data also support these findings, as blooms of filamentous cyanobacteria tend to be confined to the stratified conditions observed during seasonal transitions as well as the rainy season. So far, there are no reports of blooms occurring during the dry season upwelling period43,63. Likewise, in similarly large lakes50 and marine environments9,59,64, high N2 fixation rates related to free-living filamentous cyanobacteria appear to be most often associated with stratified rather than upwelling conditions. Correspondingly, modeling studies have found that diazotrophs have a niche in nutrient-poor stratified ocean regions15, especially when surface waters lack DIN but P and Fe are available (Fig. 1i, j)7,65.

Nitrogen fixation in the oxygen-deficient zone

We also observed rates of N2 fixation in the oxygen-deficient waters of Lake Tanganyika. Specifically, the greatest deep N2 fixation rates were reported in the anoxic waters coinciding with the ACM at Station 2 in the north, with rates reaching up to 0.75 nM N d−1 at 144 m (Figs. 1f and 2a, b). In the south, which harbors no ACM (Figs. 1e and 2g, h), N2 fixation rates were almost an order of magnitude lower, with a maximum of 0.11 nM N d−1 measured at Station 7. Overall, our range of N2 fixation rates in anoxic waters (0.11- 0.75 nM N d−1) measured in Lake Tanganyika is in the same order of magnitude as determined in anoxic waters of other stratified lakes or basins (0.44–1.01 nM N d−1;37,66,67) and marine oxygen minimum zones23,68,69. However, a potential caveat is that the nominal pore size of glass fiber filters (0.7 µm) used for N2 fixation measurements is larger than some planktonic bacteria. Thus, N2 fixation by smaller diazotrophs, which are characteristic of anoxic depths, may be underestimated70. Moreover, we cannot completely exclude that trace oxygen contamination may have inhibited N2 fixation by strict anaerobes, such as Chlorobium, which dominates in the ACM (discussed below). Therefore, the rates measured in the anoxic zone likely represent the lower limit of deep/anoxic N2 fixation activity in Lake Tanganyika.

Fig. 2: Niche partitioning of N2 fixing microbes identified in Lake Tanganyika.
figure 2

a–f Station 2 is situated in the northern basin, while (g–l) Station 7 is located in the southern basin, and is also marked by red and blue downward triangles, respectively. Shown are the vertical distribution profiles of chlorophyll (based on in-situ fluorescence), oxygen, nutrients, rates of N2 fixation, nifHDK gene abundances (recovered from metagenomes; Supplementary Table 1), and the relative abundance of diazotroph candidate metagenome-assembled genomes. Note that the concentration data for chlorophyll, oxygen, and DOC:DON ratio as well as N2 fixation rates are log-transformed. The upper limit of the anoxic zone is delimited at an O2 cutoff of 1 μM (gray dotted line). Data in panels b and h are presented as mean values for duplicate samples (distinct samples), with error bars representing the standard deviation. DOC dissolved organic carbon, DON dissolved organic nitrogen.

Full size image

The measurable rates of N2 fixation in the anoxic zone raised the possibility of a co-occurrence between N2 fixation and N removal (Supplementary Fig. 7; Supplementary Discussion), which has been previously observed in other stratified lakes and basins37,67. While such a co-occurrence may exist, rates of N2 fixation in the anoxic zone satisfied at most 1 % of the fixed N required to balance N loss (Table 1). Instead, the export of newly-fixed N from filamentous cyanobacteria in surface waters could supply up to 12% of the fixed N for anammox/denitrification in the northern oxygen-deficient zone (Table 1). In addition, depth-integrated rates of N2 fixation and N loss were positively correlated (Spearman, R2 = 0.7, p < 0.05; Supplementary Fig. 8), suggesting that the export of newly-fixed N from filamentous cyanobacteria in surface water could have contributed to the enhanced rates of N loss in the oxygen-deficient zone reported in the northern basin47.

Ecophysiology of diazotrophs under contrasting hydrodynamic regimes

Our analysis of the vertical distribution of the nitrogenase genes nifHDK underpins the two major functional hotspots for N2 fixation: within the euphotic zone and at the top of the anoxic zone (Fig. 2b, h). At Station 2 in the heavily stratified north, the high nitrogenase gene abundances of 70–80 genes per million in the euphotic zone decreased to 10–20 genes per million in underlying oxic waters but experienced a secondary increase to ~250 genes per million at 144 and 156 m within the ACM (Fig. 2b). A similar vertical nifHDK distribution was also observed at Station 7 in the upwelling-driven south, with a peak of ~80 genes per million in the euphotic zone, but a distinctly lower secondary peak of ~50 genes per million in the anoxic zone.

Diversity analysis of the recovered nifH genes in Lake Tanganyika identified 26 nifH-containing metagenome-assembled genomes (MAGs). These MAGs spanned a total of nine putative nitrogen-fixing classes, with the closest relatives indicated in parentheses: Alphaproteobacteria (Methylocystis), Acidimicrobiia (IMCC26256), Bdellovibrionota, Brocadiae, Chloroflexia (Roseiflexaceae), Chlorobia (Chlorobium), Cyanobacteria (Dolichospermum), Gammaproteobacteria (Aquabacterium, Pseudomonas, Methylotetracoccus), Kiritimatiellae (Pontiellaceae), Myxococcota, (Supplementary Fig. 10 and Supplementary Table 2). However, a number of the binned MAGs pertaining to the classes of Alphaproteobacteria, Acidimicrobiia, Bdellovibrionota, Brocadiae, and Chloroflexia did not contain a complete nifHDK operon – either as a result of bin incompleteness, or due to the presence of pseudo-nifH sequences (solitary nifH genes without encoded nifDK;71). Pseudo-nifH sequences, which can comprise a relatively large fraction of curated nifH databases, may not be necessarily representative of bonafide diazotrophs as they are unlikely to encode for a functional nitrogenase71. The MAGs containing complete nifHDK operons were identified within the genera: Aquabacterium, Chlorobium, Dolichospermum, Methylotetracoccus, Pontiellaceae (family), and Pseudomonas (Fig. 3a, Supplementary Fig. 10, and see Supplementary Tables 3–6 for a complete list of accessory N2 fixation genes). All six nifHDK-containing MAGs encoded for the molybdenum-iron (MoFe) nitrogenase, in line with other stratified sulfide-rich lakes72.

Fig. 3: Ecophysiology of key N2 fixing microbes identified in Lake Tanganyika.
figure 3

a Phylogenetic diversity of recovered nifH gene sequences. The unrooted tree was calculated using the maximum likelihood method. Indicated below are the metabolic pathways for the nifH-containing diazotrophs. Marker species, specific for Stations 2 and 7 (shown in red and blue), contained the full array of N2 fixation genes, whereas other members (shown in gray) were missing some key nitrogenase subunits (i.e., contained nifH-only), as described in-text. The complete nifH, nifD, and nifK tree is shown in Supplementary Fig. 10. The size of the bubble reflects the total number of genes identified in a given pathway. b–e Metabolic models summarizing selected key functions of station-specific marker species. The metabolic models are based on near-complete draft metagenome-assembled genomes with the percent completeness indicated in parentheses: Dolichospermum (94.1%), Chlorobium (99.4%), Aquabacterium (90.4%), and Pseudomonas (93.2%) (Supplementary Table 2). For a complete list of genes and a detailed description of metabolic pathways please refer to Supplementary Tables 3–6, and the Supplementary Discussion. G6P glucose-6-phosphate, G1P glucose-1-phosphate, GAP glyceraldehyde 3-phosphate, PEP phosphoenolpyruvate, PolyP polyphosphate, CBB Calvin Bensen-Bassham, rTCA reverse tricarboxylic acid, B-oxid beta oxidation, AA amino acid, 3-OEL 3-Oxodipate-enol-lactone, TBDT TonB-dependent receptors.

Full size image

Some diazotrophs were observed at stations in both the north and the south (including Pontiellaceae and Methylotetracoccus) (Fig. 2f, l), whereas others were unique to either the north (autotrophs such as Dolichospermum and Chlorobium; Fig. 2d, f) or south basin (heterotrophs such as Aquabacterium and Pseudomonas; Fig. 2k, l). Moreover, the diazotroph communities at anoxic depths were distinct from those identified in the oxygenated water column. For example, in the more stratified waters at Station 2, Dolichospermum was most abundant in the euphotic zone, whereas Chlorobium dominated in the ACM zone, comprising 65% and 85% of the total putative diazotrophic community, respectively (Fig. 2d, f). At Station 7, Aquabacterium occurred primarily above the oxic-anoxic transition zone, whereas Pseudomonas dominated below. They represented 53% and 61% of the total putative diazotrophic community in each of these zones, respectively (Fig. 2k, l). Altogether, this shows a high degree of both vertical and basin-wide niche partitioning within the diazotrophic community.

This niche partitioning led to contrasts in ecophysiology. While the cyanobacterium Dolichospermum carries out oxygenic photosynthesis in well-lit surface waters, Chlorobium is an anoxygenic photosynthetic green sulfur bacterium that oxidizes sulfide under extremely low-light conditions, consistent with the light levels in the ACM of Lake Tanganyika (<0.05 µmol photons m−2 s−1,47) and prior observations in deep anoxic waters of many stratified basins and lakes73,74. Typical for these microorganisms, Dolichospermum and Chlorobium MAGs encoded carbon fixation employing the Calvin-Benson-Bassham and the reverse tricarboxylic acid cycles, respectively (Fig. 3b, d and Supplementary Tables 3, 4). Interestingly, they lacked genes for organic matter transport and degradation pathways, suggesting that Dolichospermum and Chlorobium share a general photolithoautotrophic physiology. To exploit the optimum light intensities and nutrient concentrations within stratified layers, Dolichospermum and Chlorobium have the genetic potential to control their buoyancy by gas vesicles (Fig. 3b and Supplementary Table 3) for uplift and/or carbohydrate production/consumption for ballasting. Gas vesicle protein genes were not identified in the Lake Tanganyika Chlorobium, but Chlorobium contained exopolyphosphatases, a polyphosphate kinase, as well as pathways for glycogen synthesis that could moderate buoyancy by metabolizing intracellular polyphosphates and/or glycogen (Fig. 3d and Supplementary Table 4). Competition models have shown that species that control their buoyancy over other non-buoyant microbes have a distinct advantage under more stratified, or less well-mixed, conditions75. The putative buoyancy-based ecophysiology of Dolichospermum and Chlorobium allows them to thrive within the most suitable layer in the heavily stratified water column of the northern basin47.

In contrast, the upwelling waters of Station 7 were dominated by the gammaproteobacterial diazotrophs Aquabacterium and Pseudomonas, which exhibited different strategies of carbon cycling and motility (Fig. 2k, l). The recovered Aquabacterium and Pseudomonas MAGs of Lake Tanganyika included several genes encoding for the uptake and metabolism of high molecular weight compounds (>600 Da), such as those in the dissolved organic matter pool (e.g. TonB-dependent receptors, mono/dioxygenases), as well as genes involved in the uptake and utilization of more labile substrates (e.g., monosaccharides, oligopeptides, and branched-chain amino acids; Supplementary Discussion). The upwelling waters, where Aquabacterium and Pseudomonas prevail, are characteristically depleted in labile organic matter, with an average DOC:DON ratio of 11.5 that exceeds Redfieldean C:N stoichiometry of 6.6 (Fig. 2i). Aquabacterium and Pseudomonas can couple the oxidation of organic matter substrates to oxygen and nitrate respiration. To help navigate toward the patchy sources of DOM/POM, the MAGs encoded for flagella-driven chemotaxis. Chemotaxis is known to be advantageous under turbulent conditions, as it allows microbes to sense and follow diffuse gradients of DOM76. Therefore, the N2-fixing community and ecophysiology of chemoheterotrophic microorganisms in the upwelling waters contrast with the stratified surface waters and the ACM that are dominated by photoautotrophs.

Upwelling and stratification influence the mode of diazotrophy

In the more stratified northern basin, high abundances of diazotrophic filamentous cyanobacteria and elevated rates of N2 fixation coincided with low DIN transport to surface waters44. Given that P and Fe are not drawn down, the DIN availability in surface waters represents the most likely control over surface N2 fixation by filamentous cyanobacteria in Lake Tanganyika. The northern/central basins provide a favorable environment for communities of photoautotrophic diazotrophs in both the euphotic and anoxic zones (Fig. 4). Concurrently, the capacity for vertical migration (via buoyancy regulation) allows such diazotrophs to navigate precisely in the narrow density gradients of stratified waters.

Fig. 4: Conceptual model of nitrogen fixation and removal under contrasting hydrodynamic regimes in Lake Tanganyika.
figure 4

The values indicate the nitrogen fluxes expressed in µmol N m−2 d−1, see Table 1 for associated errors. Rates of nitrogen removal are detailed in the Supplementary Discussion. In addition, the illustration depicts the key diazotroph marker species identified in the stratified north/center and the upwelling-driven south basins, also indicated by the red and blue downward triangles, respectively. ACM anoxic chlorophyll maximum, PCM primary chlorophyll maximum, DOM dissolved organic matter, CBB Calvin-Benson-Bassham Cycle.

Full size image

In contrast, nutrient upwelling established the potential for N2 fixation in the southern basin by adding excess P to surface waters77. As a concomitant effect, however, the upwelling of nutrient-rich waters seemed to locally suppress key diazotrophs, such as filamentous cyanobacteria, resulting in lower rates of N2 fixation in surface waters. While filamentous cyanobacteria were inhibited, these conditions favored heterotrophic gammaproteobacteria Aquabacterium and Pseudomonas which were distinctly reliant on organic matter fluxes. In these productive upwelling waters, their metabolic versatility and chemotaxis make them uniquely suited to a more turbulent and dynamic nutrient landscape, targeting both reactive and semi-reactive sinking organic material. Chemotactic foraging is favored under dynamic nutrient conditions78,79, however, it also comes at a high ATP cost. We suggest that ATP allocation towards flagellated motility and metabolic plasticity – used to scavenge for organic nitrogen – could place constraints on the similarly energy-demanding N2 fixation, explaining the lower rates of N2 fixation reported in Lake Tanganyika.

Overall, our findings demonstrate that the hydrodynamic regimes – upwelling/mixing versus stratification – represent key controls on the nutrient landscape that, in turn, governs the mode of diazotrophy (photoautotrophic vs. chemoheterotrophic) and the rates of N2 fixation in Lake Tanganyika.

The observed physical-biological coupling in Lake Tanganyika, a model system with respect to its hydrodynamics, may help to explain why N2 fixation rates are generally low (<5 nM d−1) in some marine upwelling regions15,22,26,27,28,30,80,81. Many of these upwelling systems are also associated with gammaproteobacterial heterotrophic and often chemotactic diazotrophs79, like Pseudomonas24,31,35,80,82. Moreover, when highly stratified conditions develop in marine upwelling regions (e.g., coastal sulfidic event83), niches for phototrophic green sulfur bacteria become favorable84, consistent with observations in Lake Tanganyika’s Northern basin (Table 1). The predominance of Dolichospermum in the stratified North/Central basins of Lake Tanganyika shares some parallels to the stratified ocean gyre systems that support higher abundances of large-sized cyanobacteria (e.g., Trichodesmium)18. And like Dolichospermum, some Trichodesmium members are also capable of controlling their buoyancy using gas vacuoles85, a strategy that possibly contributes to their success as a major N2 fixing species in gyres. Our conceptual hydrodynamic/ecophysiological model presented in Fig. 4, therefore may broadly hold true for other aquatic systems at large.

Methods

Physicochemical and biological measurements

Vertical profiles of chemical and biological variables were collected along a lake-wide transect (Fig. 1a) during three cruises at the end of the dry season (28 September - 8 October 2017) and the end of the rainy season (27 April–7 May 2018; 26 April–1 May 2019) onboard M/V Maman Benita. The 2018 expedition was the main campaign of this study, combining a high spatial coverage of N process rate measurements, metagenomic analyses of the microbial community, and natural abundance N-isotope tools. During the third campaign in 2019, we collected high-resolution profiles at Station 2 in the north and Station 7 in the south.

Continuous CTD profiles of temperature, photosynthetic active radiation, in-situ chlorophyll fluorescence, turbidity, and oxygen (Sea-Bird SBE 19plus) were followed by discrete water sampling at 5 to 25 m depth intervals using Niskin bottles (20–30 L). Soluble reactive phosphate, ammonium, nitrate, and nitrite concentrations were determined according to standard methods44, with detection limits averaging 0.22, 0.34, 0.20, and 0.03 µM, respectively. Chlorophyll-a concentrations were determined fluorometrically and used to calibrate in-situ chlorophyll fluorescence. The respective samples were extracted in ethanol and calibrated against their respective standards44. Medium- to large-celled phytoplankton was concentrated using a 10 μm plankton net, pooling cells from 2–4 L lake water into a 20 mL sample. Samples were fixed with Lugol solution, and cells were counted and classified by inverted microscopy.

Process rate measurements

Samples for N2 fixation and N turnover incubation experiments were chosen according to the vertical distribution of in-situ chlorophyll fluorescence and oxygen, i.e., targeted at chlorophyll peaks and the top of the anoxic zone. Rates of N2 fixation were determined during incubation experiments at Stations 2 and 9 in September/October 2017 and Stations 2, 4, 6, 7, and 9 in April/May 2018. For determining N2 fixation rates we chose the bubble-removal 15-15N2 tracer method, due to the simple handling and to avoid the risk of trace metal contamination86. After sunset, triplicates from each depth were filled from the Niskin into 4.5 L polycarbonate bottles and capped with polypropylene membranes. We used a gas-tight hose to fill the bottles bubble-free. Anoxic samples were filled with an overflow of approximately the bottle volume to minimize contamination with ambient air. The triplicates comprised one control (no added 15N label) and two duplicate treatments (with amended 15-15N2). Samples from anoxic depths were collected in 5.3 L glass bottles capped headspace-free with butyl rubber stoppers. To each treatment bottle, 8–10 mL 15-15N2 (Cambridge Isotope Laboratories, Lot # I-21065/AR0664729) were added87. The labeled samples were equilibrated for approximately 30 min by rolling/inverting the bottles. Thereafter, the 15-15N2 headspace was released by opening the cap and a 12 mL subsample of water was collected for quantifying the labeling atom percent for each incubation bottle individually, which averaged 4.7% in 2017 and 6.3% in 2018 sampling campaigns (Supplementary Tables 7 and 8). It should be noted that the bottles were opened only briefly, and without any turbulence, N2 does not equilibrate rapidly (N2 is also inherently much less soluble than oxygen). The remaining air headspace was refilled with water from the same depth. The headspace-free bottles were then transferred to on-deck incubators covered with selected light filters (LEE Filters) to simulate the irradiance and light spectrum at the respective sampling depths. Incubators containing samples from the deep anoxic zone were shaded with black aluminum foil allowing only residual light penetration. Incubations were terminated after 24 h by filtration on pre-combusted GF/F filters (nominal pore size 0.7 µm; Whatman). In our samples, the maximum temperature in the on-deck incubators generally did not exceed the in-situ temperature by more than 4 °C; given that differences were small, rates of N2 fixation were not corrected for temperature. The filters were immediately oven-dried (60 °C for 48 h) and, upon return to land-based facilities, fumed under an HCl atmosphere for 48 h to remove inorganic carbon.

The 15N enrichment was measured with an EA-IRMS (vario PYRO cube, Elementar coupled with an IsoPrime IRMS, GV Instruments) and computed according to the following equation (using Ion Vantage Isoprime v1.7.3.0):

$${\delta }^{15}N=\left(\frac{15N/{14N}_{{{{{{\rm{sample}}}}}}}}{{15N/14N}_{{{{{{\rm{standard}}}}}}}}-1\right){{{{{\rm{\times }}}}}}1000$$
(1)

Calculated limits of detection (LOD = δ15Ncontol + 3*standard deviation[δ15Nstandards]), were unusually low with 3*standard deviation [δ15Nstandards] <0.4 ‰ in all batches (Supplementary Tables 7, 8). Thus, we conservatively defined the LOD as δ15Ncontol + 4 ‰ (minimum change; Montoya et al.88). We furthermore defined samples as <LOD, if their δ15N values or calculated N2 fixation rates were negative (Supplementary Tables 7, 8). Standard error was calculated from all samples of one depth. The δ15N values of the standards showed no significant linear trend within all analyzed batches (Spearman rank correlation test, p > 0.05). PN mass on filters ranged from 12–79 µg N for treatments and 11–62 µg N for controls.

Bulk rates of N2 fixation (in nmol L−1 d−1) were calculated according to the following equation:

$${N}_{2}{{{{{\rm{fixation}}}}}}\; {{{{{\rm{rate}}}}}}=\left(\frac{{{{{{{\rm{at}}}}}}\%{{{{{\rm{PON}}}}}}}_{{{{{{\rm{sample}}}}}}}-{{{{{{\rm{at}}}}}}\%{{{{{\rm{PON}}}}}}}_{{{{{{\rm{control}}}}}}}}{{{{{{{\rm{at}}}}}}\%N}_{2}-{{{{{{\rm{at}}}}}}\%{{{{{\rm{PON}}}}}}}_{{{{{{\rm{control}}}}}}}}\right){{\times }}\left(\frac{{{{{{\rm{mass}}}}}}\; {{{{{\rm{of}}}}}}\; {{{{{\rm{PON}}}}}}\; {{{{{\rm{per}}}}}}\; {{{{{\rm{volume}}}}}}}{{{{{{\rm{time}}}}}}}\right)$$
(2)

Whereby the at%PONsample, at%PONcontrol, and at%N2 represent the atomic%15N in the particulate organic nitrogen (PON) of the incubated sample, the natural abundance of the control sample, and the N2 pool, respectively. Notably, the control samples (used for natural abundance measurements) were sampled at the same depth as the 15-15N2 amendment experiments. The mass of PON per volume (i.e., concentration) and time of incubation were also used as inputs in Eq. 2.

It should be further noted, that trace oxygen contamination in anoxic incubations was unavoidable and may have biased N2 fixation rates in those samples. Assuming that enhanced activity of aerobic diazotrophs in anoxic samples through the introduced amounts of oxygen was the dominant effect, we would expect to measure the highest N2 fixation rates in samples from the anoxic zone in the south, where facultative aerobic diazotrophs, such as Aquabacterium or Pseudomonas, were more abundant than in the north (Figs. 2 and 3). By contrast, determined N2 fixation rates in anoxic samples were lower in the south than in the north (Figs. 1f and 2a and Supplementary Fig. 5). Moreover, during the collection of the labeling atom percent subsample the introduction of atmospheric N2 may have lowered the actual 15-15N2 concentration compared to the measured value. Hence, N2 fixation rate estimates should be considered conservative.

Stable isotope 15N-labeling experiments were conducted to quantify rates of denitrification and anammox (Supplementary Discussion). For the incubations, waters were collected from Stations 2–9 (Fig. 1), and processed according to Callbeck et al.47. A total of four 15N addition experiments were carried out, at six depths per station: 15N-NO3−, 15N-NO2− + 14N-NH4+, 15N-NH4+ + 14N-NO2−, and 15N-NH4+. The concentration of 14N15N and 15N15N was then determined in the gas phase by gas-chromatography isotope ratio mass spectrometry (GC-IRMS; VG Optima, Manchester, UK) at land-based facilities. The slope of the linear regression (i.e., 14N15N and 15N15N concentration change with time) was used to calculate the rates of 29N2 and 30N2 production in the different experiments.

POM and nutrient-stable isotopic analyses

Particulate organic matter (POM) samples were collected by filtering 2–4 L lake water onto precombusted GF/F filters (Whatman, UK) and oven-dried at 60 °C for 48 h. At land-based facilities, the POM samples were fumed for 48 h under HCl atmosphere to remove inorganic carbon. The N isotopic composition was analyzed with an EA-IRMS (vario PYRO cube, Elementar coupled with an IsoPrime IRMS, GV Instruments), and calculated according to Eq. 1 (using Ion Vantage Isoprime v1.7.3.0). The final δ15N of the samples was corrected using the standard Acetanilide #1 (Indiana University, CAS # 103-84-4). Standard reproducibility was generally better than 0.5 ‰. Nutrient (DIN) isotope sample treatment and analysis are described in the Supplementary Methods.

Nitrogen flux estimates

N2 fixation rates were integrated over the top 50 m in the euphotic zone, from 125 to 175 m in anoxic waters, or from 0 to 175 m for total N2 fixation rates. Nitrogen removal rates were integrated down to 175 m. The turbulent diffusive fluxes of nitrate and ammonium were calculated with Fick’s law:

$${J}_{i}={K}_{z}{{{{{\rm{\times }}}}}}(\partial C/\partial z)$$
(3)

The measured concentration gradients; the vertical turbulent diffusivity, Kz, was estimated from buoyancy frequency according to Von Rohden et al.89, and parametrized to a lake-wide average Kz of 10−5 m2 s−1 within the thermocline41. The total upward flux of nitrate (Q) in the southern upwelling area, i.e., at Stations 8 and 9, was estimated by adding an advective term:

$$Q={J}_{i}+c\times v$$
(4)

Whereby c is the average concentration at the nitracline and v the upwelling velocity (0.05 and 0.1 m d−1 at Stations 8 and 9, respectively) estimated from the upward tilting of the isotherms between March and April42. Basin-scale estimates were calculated from the average values of Stations 1–6 and 7–9 for the north/central and south basins, respectively. Information regarding the external N sources was extracted from Langenberg et al.53 by downscaling the annual estimates, while information on the regional variability in atmospheric N deposition was not available. Therefore, the lake-wide estimate was used for both regions.

Nitrate flux values were correlated against various parameters including depth-integrated rates of N2 fixation, δ15N-PON, and cyanobacteria cell densities (Supplementary Fig. 9). Correlative analyses were performed using the Spearman test (one-sided). R2 and p-values were evaluated using an 80% and 95% confidence interval.

Sampling and extraction of DNA and metagenomic analysis

Samples for metagenomic analyses were collected at Stations 2 and 7, similar to our previous study44. Briefly, the lake water was filtered onto 0.2 μm cellulose acetate filters, fixed with RNAlater (Sigma Life Science), and stored at 4 °C for 1–2 weeks. Upon return to land-based facilities, the filters were stored at −80 °C. For extraction, filters were thawed and washed with TE buffer (1x) before DNA extraction using the AllPrep DNA/RNA extraction kit (Qiagen) according to the manufacturer’s instructions. The metagenomes were sequenced using the Illumina NextSeq platform to generate 150 bp paired-end reads (averaging ~350 bp in length; Novogene in Hong Kong).

The quality of metagenomic reads was assessed using FastQC (version 0.12.1). Metagenome trimming, quality control, mapping, and taxonomic assignments are detailed in the Supplementary Methods. From the metagenomes, we extracted the genes for N2 fixation (nifHDK) from the gtf file. The number of reads for the detected genes was enumerated and normalized for the gene lengths and sequencing depth to obtain a measure of relative gene abundance. The sequencing depth, filtered reads, mean contig length, and mapped reads for each sample are summarized in Supplementary Table 1.

High-quality trimmed reads from all sampling depths (n = 15) were co-assembled into scaffolds using Megahit (also see Supplementary Methods). Binning and refinement modules (metaWRAP, version 1.3) were applied to the co-assembly to recover high-quality metagenome-assembled genomes (MAGs). Completeness and contamination rates of the final MAGs were assessed using CheckM (version 1.1.6)90. We only used MAGs that passed a threshold for completion of 50 %, and contamination rates less than 10%. MAG abundances were assessed using coverM (version 0.2.0). Here, raw reads were mapped against the putative genomes, and abundance is expressed as the coverage of raw reads on the MAG. The total number of MAGs recovered from our samples, along with bin completeness, and contamination are summarized in Supplementary Table 2.

Remote sensing

The lake-wide surface chlorophyll distribution was inferred from 300m-resolution raster data from Sentinel-3 OLCI data (Copernicus Global Land Service). The datasets are temporal aggregates of valid observations acquired in the intervals between the first, eleventh, 21st and last day of a month. From version 1.3 of this data set, we extracted the Trophic State Index (TSI) for April and May 2018. The TSI increases in discrete increments of ten that correspond to specific chlorophyll-a concentration ranges91.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.