Purple sulfur bacteria fix N2 via molybdenum-nitrogenase in a low molybdenum Proterozoic ocean analogue

Sampling

Samples were collected on 28 August 2018 during a field campaign to Lake Cadagno²⁹, Switzerland. In situ measurements and water collection was performed at the deepest part of the lake (21 m). Water was collected using a pump CTD system as described in Di Nezio et al.³⁶. Online in situ data were obtained during a continuous downcast of the CTD-system from the water surface down to ~17.5 m depth. During the upcast, discrete water samples were collected from a total of 20 depths (between 12 m and 17 m) above, in, and below the chemocline for chemical analyses and from 3 depths for incubation experiments (13.7 m, 14 m, and 15.5 m).

In Lake Cadagno, wind-driven internal waves lead to vertical shifts of the water masses and their corresponding physicochemical parameters⁵⁶. While sampling, it was apparent that the depths of the individual water masses had slightly shifted between the down- and the upcast. Therefore, we corrected the water depths of the samples collected during the upcast so that the physicochemical parameters during sampling best matched those of the continuous downcast, to ensure that samples were assigned to the respective water mass that they originated from. A custom R script was employed for the depth correction. In brief, all parameters measured by the CTD-system during the upcast and the downcast were normalized to percent (with 100% as the maximum observed value, and 0% the minimum observed value). Per individual sampling depth (during the upcast, where the pump cast CTD remained stationary for some time), average values of conductivity, temperature, and pressure were calculated and converted to percent values. Then, the depth from the downcast profile was identified that best matched all calculated percent values. This was achieved by subtracting the percent values per parameter from all respective data points of the downcast profile. Absolute values of the calculated differences per data row were summed. The depth with the lowest resulting sum, i.e., with the most similar physicochemical parameters, was then chosen as the corrected depth.

Chemical analyses, flux calculations, and rate determinations

For chemical analyses, lake water from the individual sampling depths was sterile-filtered (0.2 µm, cellulose acetate filter) and frozen at −20 °C until analysis. Samples were analyzed with a QuAAtro39 autoanalyzer (Seal Analytical) using the methods described in Strickland and Parsons⁵⁷ to determine concentrations of dissolved inorganic phosphorus (PO₄³⁻), nitrite (NO₂⁻), nitrate (NO₃⁻), and reactive silica (Si). Ammonium concentrations were determined from the same filtered samples using the colorimetric analysis described in Kempers et al.⁵⁸. Molybdenum concentrations were determined from filtered samples after acidification with 1% HNO₃ (69%, ROTIPURAN^®, Roth) using an ICP-MS 7900 (Agilent, Santa Clara, USA). Molybdenum was analyzed on mass 95 in He-mode using a multi-element calibration SRM (21 elements, Bernd Kraft). The SRM NIST 1643f was analyzed in parallel to guarantee the quality of analyses. Concentrations of sulfide were determined colorimetrically from unfiltered Lake water samples, following Cline⁵⁹.

To calculate the turbulent flux (J) of ammonium into the chemocline, we assumed a steady-state using Fick’s first law: J = −D∂C/∂x. A turbulent diffusion coefficient (D) of 1.6 × 10⁻⁶ m² s⁻¹ was used, corresponding to turbulence at the Lake Cadagno chemocline boundaries⁶⁰. The change in concentration (∂C) was calculated over 14.25 m to 14.77 m depth, where the steepest ammonium gradient was observed. Ammonium uptake rates were calculated for the chemocline by integrating this flux over the chemocline from 13.45 m to 14.45 m depth.

To quantify N₂ fixation and primary production (i.e., CO₂ fixation) rates, stable isotope incubations with ¹⁵N₂ and ¹³CO₂ were performed using established protocols⁶¹. Briefly, lake water from three different depths of the chemocline was sampled directly from the CTD pump system into five 250 ml serum bottles per depth. Water was filled into the bottles from bottom to top, allowing 1–2 bottle volumes to overflow to minimize oxygen contamination before crimp-sealing the bottles headspace-free with butyl rubber stoppers. Back in the field laboratory, no more than 8 h after sampling, one bottle per depth was filtered onto pre-combusted (460 °C, 6 h) glass microfiber filters (GF/F, Whatman^®, UK) for in situ natural abundance of C and N. ¹³C-labeled sodium bicarbonate (NaH¹³CO_3, 98 atom% ¹³C, dissolved in autoclaved MilliQ water; Sigma-Aldrich) was injected (320 µL) into three bottles per depth, to achieve a final concentration of 160 µmol L⁻¹. Then, a volume of 5 ml ¹⁵N₂ gas (Cambridge Isotope Laboratories, >98 atom% ¹⁵N, Lot #: I-19197/AR0586172) was injected as a bubble into the same bottles and shaken for 20 min to equilibrate the ¹⁵N₂ gas. Sulfide solution was injected aiming for a final concentration of approximately 2 µM to remove trace oxygen contamination in the incubation bottles. Finally, the ¹⁵N₂ gas bubble was replaced by anoxic in situ lake water from the respective depth. The bottles, together with one untreated control bottle per depth (containing unamended lake water), were incubated for a full light-dark cycle (13 h light, 11 h dark) under natural light conditions (0–8267 Lux, average: 247 Lux, median: 10.8 Lux, as determined by a HOBO pendant data logger, Onset Computer Corporation, Bourne, USA) in a water bath kept at ~12 °C.

After incubation, samples were filtered onto pre-combusted GF/F filters. The filters were dried at room temperature and frozen at −20 °C for transport and storage. In addition, subsamples for nanoscale secondary ion mass spectrometry (nanoSIMS) analysis and for the determination of ¹³C and ¹⁵N enrichments in the substrate pools were taken from all bottles amended with ¹³C and ¹⁵N. NanoSIMS samples were fixed with 2% (final w/v) formaldehyde solution for 1 h at room temperature, prior to filtration onto gold-sputtered 0.22 µm polycarbonate membrane filters (GTTP Isopore^TM, Merck Millipore, USA). Subsamples for label% determinations were taken in gas-tight glass vials (Exetainer Labco, UK) and biological activity was terminated with HgCl₂.

Samples on GF/F filters were analyzed for C and N content and the respective isotopic composition by an elemental analyzer (Thermo Flash EA, 1112 Series) coupled to a continuous-flow isotope ratio mass spectrometer (Delta Plus XP IRMS; Thermo Finnigan, Dreieich, Germany). Enrichment of ¹⁵N in the N₂ pool was determined using a membrane inlet mass spectrometer (MIMS; GAM200, IPI). Enrichment of ¹³C in the dissolved inorganic carbon pool was determined from ¹³C/¹²C-CO₂ ratios after sample acidification with phosphoric acid using cavity ring-down spectroscopy (G2201-I coupled to a Liaison A0301, Picarro Inc., connected to an AutoMate Prep Device, Bushnell, USA). In addition, we tested the used ¹⁵N₂ gas bottle for contamination with ¹⁵N-ammonia⁶². Briefly, a 2 ml subsample of the used ¹⁵N₂ gas was injected into a 12 ml gas-tight glass vial (Exetainer) filled with MilliQ (pH < 6). Any potentially present ammonia/ammonium was dissolved into the MilliQ via shaking and incubation overnight. Next, the liquid sample was repeatedly bubbled with helium to remove residual ¹⁵N₂ gas. 40 µmol L^-1 of ¹⁴N-ammonium (as (NH₄)₂SO₄) was added to the sample as a carrier for the gas chromatography isotope ratio mass spectrometry (GC-IRMS) analysis. Total dissolved ammonium was oxidized to N₂ using alkaline hypobromite iodine⁶³, which combines two ammonia/ammonium molecules to N₂. Due to the added ¹⁴N-ammonium, any ¹⁵N-ammonium contamination would result in the formation of ²⁹N₂, thus enabling us to differentiate ¹⁵N-ammonium contamination from potential residues of the tested ³⁰N₂ gas. After hypobromite conversion, N₂ isotopes were analyzed by GC-IRMS on a customized TraceGas coupled to a multi collector IsoPrime100. Together with the samples, we analyzed hypobromite treated ¹⁵N-ammonium standards⁶⁴. We detected no ¹⁵N-ammonia contamination in the used ¹⁵N₂ bottle (detection limit: 5 pmol in 2 ml gas, equaling 0.056 ppm).

Bulk N₂ and CO₂ fixation rates were calculated from the incorporation of ¹⁵N and ¹³C, respectively, into biomass as described in Mohr et al.⁶⁵, according to Eq. 1

With at%PON_sample, at%PON_NA, and at%N₂ representing the atomic%¹⁵N in the particulate organic nitrogen (PON) of the incubated sample, the incubated natural abundance sample and the N₂ pool, respectively. [PON] is the concentration of PON in the incubated sample per L, time is the incubation time in days. CO₂ fixation rates were calculated accordingly, using atomic%¹³C, and particulate organic carbon (POC).

We further calculated the amount of N required to sustain the measured autotrophic C fixation, based on the depth-specific biomass C/N ratio (determined for the in situ bulk biomass in the respective water depth using an elemental analyzer), according to Eq. 2

Calculations were performed using Microsoft Excel 2016.

Re-analysis of metagenomes from 2014

Raw metagenomic sequence data from the chemocline of Lake Cadagno (SAMEA4666021) and an enrichment culture from the lake (SAMEA4666022) previously published by Berg et al.²⁹, was downloaded from NCBI. Reads were adapter- and quality-trimmed using BBDuk⁶⁶ v37.24 (phred score 10, minimum length 50 bp). Reads from both metagenomes were co-assembled using metaSPAdes⁶⁷ version 3.14.0. Assembly statistics are shown in Table S6. Scaffolds were renamed to match the anvi’o workflow requirements, using anvi-script-reformat-fasta with anvi’o⁶⁸ version 6.2. Reads from both chemocline and enrichment metagenomes were mapped to the assembled and renamed scaffolds using Bowtie 2⁶⁹ version 2.3.5.1. Sorted and indexed bam files were created using samtools⁷⁰ 1.10 and anvi-init-bam. We followed the anvi’o metagenomics workflow, including gene prediction with Prodigal⁷¹ V2.6.3, functional annotations with NCBI COGs and GhostKOALA/KEGG⁷², and taxonomic gene annotation with Centrifuge⁷³ version 1.0.4. Predicted gene sequences annotated as either nifH, nifD, or nifK were manually validated by a blastx search⁷⁴ to the NCBI non-redundant protein sequences (nr) database. Only sequences whose best blastx hits matched their respective functional annotation were retained as nif genes.

A merged anvi’o profile database was created using sequence coverage information from both metagenomes and a minimum contig length of 1000 bp. External binning of scaffolds was performed using MetaBAT⁷⁵ version 2.12.1, using a minimum scaffold length of 1000 bp and minimum mean coverage of a scaffold in each library of 0. We additionally binned the scaffolds with concoct⁷⁶ 1.1.0 and used DAS Tool⁷⁷ version 1.1.2 to obtain an optimal, non-redundant set of bins that is based on the results of the two individual binning methods. The resulting bins were imported into anvi’o. A manual binning of as yet unbinned scaffolds was performed in anvio. Briefly, the unbinned fraction of the anvi’o database was visualized with anvi-refine, and several additional bins were compiled based on the anvi’o hierarchical-clustering. All bins containing at least one of the structural nitrogenase genes (MoFe nitrogenase: nifH, nifD, or nifK; VFe nitrogenase: vnfH, vnfD, vnfK, vnfG; FeFe nitrogenase: anfH, anfD, anfK, anfG) were identified, manually refined using anvi-refine and saved as an independent collection (in the following termed nif-MAGs). Nif-MAGs were further analyzed for completeness and contamination using CheckM⁷⁸ v1.0.18 and taxonomically classified using GTDB-Tk⁷⁹ v1.1.0.

Metatranscriptome sequencing and analysis

Samples for nucleic acid extraction (2018 campaign) were obtained by filtration of 100 ml lake water per incubation depth onto 0.22 µm GVWP Durapore^® membrane filters (Merck Millipore, USA). Samples were immediately frozen and transported at -20 °C. The samples were then stored at −80 °C until further processing. Total RNA was extracted using the RNeasy PowerWater Kit (Qiagen, Germany) according to the manufacturer’s protocol, including an on-column DNase digestion. Total RNA was sequenced (2x250bp) using Illumina HiSeq2500. Sequencing, including library preparation, was performed by the Max Planck-Genome-Centre Cologne (https://mpgc.mpipz.mpg.de/home/).

Raw reads were adapter- and quality-trimmed using BBDuk⁶⁶ v37.82 (phred score 10, minimum length 50 bp, tbo and tpe flag). Read-pairs were merged using BBmerge⁸⁰ v37.82. Mapping of the merged transcriptome reads to the metagenome assembly (see above) was performed using Bowtie 2⁶⁹ version 2.3.5.1 as described above. featureCounts⁸¹ version v2.0.1 was used to extract read counts per predicted gene from the bam files.

To check for transcription of additional nifH genes and genes encoding for alternative nitrogenases that were not included in the metagenome assembly (and therefore not captured by the mapping approach), ROCker⁸², and HMMER (hmmer.org) searches were performed. First, the merged reads were sorted into rRNA and mRNA datasets using SortMeRNA⁸³ version 4.2.0 (Table S4). ROCker searches were performed on the mRNA datasets using three custom nifH ROCker models (read lengths 250, 450, and 500, respectively). For hmmsearch, mRNA read sequences were translated into amino acid sequences using prodigal (meta mode) and searched with the TIGRFAMs models TIGR01287, TIGR01861, TIGR02929, TIGR02931, TIGR01860, TIGR02930, and TIGR02932 for nif, anf, and vnf genes using a bit score cutoff of 100. From all hits obtained through ROCker and HMMER, only reads that had not mapped against the metagenome assembly were kept. The sequences were then manually validated as described above (blastx search).

The relative abundance of SSU rRNA transcripts was used as a combined proxy for cell abundance and cellular ribosome content (reflecting activity). We used phyloFlash⁸⁴ with the SILVA 138 SSU database and a read limit of 900,000 per sample to determine the taxonomic composition of the SSU rRNA reads. The results were visualized in R⁸⁵ using ggplot2⁸⁶.

Nif amino acid trees

A NifH amino acid sequence dataset was compiled that contained NifH sequences originating from the Lake Cadagno metagenome assembly and NifH sequences originating from the three metatranscriptomes. A blastp search⁷⁴ of the Cadagno NifH sequences to the NCBI nr database was performed to identify closely related, full-length NifH reference sequences from taxonomically classified organisms. Manually selected, representative reference sequences were downloaded from NCBI. A multiple amino acid sequence alignment of all sequences, including Cadagno and reference NifH sequences, was obtained with MAFFT⁸⁷ version 6.717b. RAxML⁸⁸ version 8.2.12 with the PROTGAMMAWAGF substitution model was used to calculate a maximum likelihood tree based on the alignment, but only including full-length NifH sequences. Partial NifH sequences which were excluded from initial tree calculation were then added to the tree using Taxtastic v0.9.0, pplacer, and guppy⁸⁹ v1.1.alpha19-0-g807f6f3.

In addition, we constructed NifD and NifK phylogenetic trees, including the NifD/NifK sequences retrieved from the MAGs. Reference sequences sharing >95% sequence identity to any of the MAG NifD/NifK sequences were identified with a blastp search⁷⁴ to the NCBI nr database. Multiple sequence alignments were obtained with MAFFT⁸⁷. All full-length sequences were used to construct base trees with RAxML⁸⁸ and 100 bootstraps in ARB⁹⁰. The ARB Parsimony function was employed to add partial sequences to the base trees.

The resulting trees were visualized in iTOL⁹¹.

FISH, cell counts, and cell sizes

From each incubation depth, 10–30 ml lake water was filtered onto 0.22 µm polycarbonate membrane filters (GTTP Isopore^TM, Merck Millipore, USA). The filters were fixed in 2% formaldehyde solution in sterile-filtered lake water for 10–12 h at 4 °C and then washed with MilliQ water. The filters were frozen and stored at −20 °C until further processing.

The 16S rRNA FISH probe “Thiosyn459” (Table S7), exclusively targeting T. syntrophicum Cad16, was designed in ARB⁹⁰. In addition, two competitor probes and four helper probes⁹² were designed (Table S7) to ensure efficient and specific binding of the probe to the target. All FISH probes and respective formamide concentrations are listed in Table S7. Probes, but not helpers and competitors, were double-labeled with either Atto488 or Atto594 fluorophores. Samples were embedded in 0.05% low melting point agarose. Cells were permeabilized with lysozyme (1.5 mg ml⁻¹) for 30 min at 37 °C. Hybridization was performed for 2–4.5 h at 46 °C. Washing included 15 min in washing buffer at 48 °C and 20 min in 1× PBS buffer at room temperature. We used the hybridization and washing buffers described in Barrero-Canosa et al.⁹³ to reduce background fluorescence. Cells were counterstained with DAPI.

Samples were analyzed using a Zeiss Axio Imager.M2 microscope equipped with a Zeiss Axiocam 506 mono camera. Z-stack images were taken and the number of fluorescently labeled cells per image was counted for the individual probes. For each PSB population, we analyzed ≥38 randomly selected fields of view and ≥54 target cells, on one filter replicate each (see Supplementary File S1). Total cell counts were obtained in triplicates through flow cytometry as described in Danza et al.⁹⁴.

For cluster-forming organisms (Thiodictyon syntrophicum, Lamprocystis purpurea, Lamprocystis roseopersicina, and Lamprocystis spp.), the cell size (length and width, for biovolume and C-content calculations, see section below) of 100 cells per population was determined from the maximum-intensity projection of the z-stack images using the Zeiss Zen blue software 3.2.

Single-cell analysis with nanoSIMS

For nanoSIMS analyses, we chose the replicate sample from 13.7 m depth that exhibited the highest bulk N₂ fixation rate. Random spots were marked with a laser microdissection microscope (6000 B, Leica) on the gold-sputtered GTTP filter covered with cells incubated with ¹⁵N₂ and ¹³CO₂. After laser marking, FISH was performed as described above. For analysis of Thiodictyon cells, no permeabilization was performed, while for analysis of the other population’s permeabilization was reduced to 15 min at 37 °C using 2 mg ml⁻¹ Lysozyme. Within one hybridization reaction, we simultaneously applied Apur453 with S453D and Laro453 with Cmok453, each probe double labeled with different fluorescent dyes (Atto488 and Atto594).

Single-cell ¹⁵N- and ¹³C-assimilation from incubation experiments with ¹⁵N₂ and ¹³CO₂ was measured using a nanoSIMS 50 L instrument (CAMECA), as described in Martínez-Pérez et al.⁵³. Briefly, instrument precision was monitored regularly on graphite planchet. Samples were pre-sputtered with a Cs⁺ beam (~300 pA) before the measurements with a beam current of around 1.5 pA. The diameter of the primary beam was tuned <100 nm. Measurements were carried out with a dwelling time of 1 ms per pixel and a raster size of 6 × 6 to 48 × 48 µm. The pixel resolution for all measurements was 256 × 256. Between 20 and 70 planes were recorded for every measurement.

Measurements were analyzed using the Look@NanoSIMS software⁹⁵ version 2015-10-20 as described in Martínez-Pérez et al.⁵³. Briefly, the recorded secondary ion images were drift corrected and accumulated. Using the corresponding epifluorescence microscopic images, regions of interest containing the target cells were defined. Ratios of ¹⁵N/(¹⁵N + ¹⁴N) and ¹³C/(¹³C + ¹²C) were used to calculate cell-specific N₂ and CO₂ fixation rate⁹⁶ only when the overall enrichment Poisson error across all planes of a given cell was <5% (using ¹⁵N/¹⁴N and ¹³C/¹²C ratios) according to Eq. 3

Where at%excess¹⁵N_cell and at%excess ¹⁵N_medium is the excess ¹⁵N atom percent enrichment above natural abundance measured for a given cell (from nanoSIMS measurements) and the incubation medium (¹⁵N₂ added to lake water, from MIMS measurements), N_cell is the amount of N per cell in fmol (see Eqs. 4 and 5 and text below) and time is incubation time in days. Cell-specific C fixation rates were calculated accordingly, using atomic%¹³C, and cellular carbon content (C_cell, see Eqs. 4 and 5, and text below).

One Lamprocystis spp. cell that did not fix N₂ and CO₂ during the incubation is not shown in the figures (with the exception of Fig. S5) but was included for all calculations. For single-cell rate calculations, we did not account for isotope dilution effects, which have been reported for FISH-treated samples e.g.⁹⁷. To assess a possible dilution effect from FISH on the isotopic composition (¹⁵N/(¹⁵N + ¹⁴N) and ¹³C/(¹³C + ¹²C)) of PSB in our experiments, we performed nanoSIMS analysis of C. okenii cells that did not undergo FISH but were identified based on morphology. Using the same replicate sample as for all other nanoSIMS analyses (but untreated by FISH or DAPI staining), we found no significant difference in ¹⁵N enrichment of FISH-treated and untreated C. okenii cells, and slightly lower ¹³C enrichment in untreated compared to FISH-treated cells (Supplementary Fig. S9). The variability in ¹⁵N/(¹⁵N + ¹⁴N) ratios across measured cells was calculated following Svedén et al.⁹⁸ (Fig. S10), which confirmed that sufficient cells of all populations have been analyzed. The cellular carbon content of the comparably large PSB cells was estimated according to Verity et al.⁹⁹ according to Eq. 4

BV is the cellular biovolume in µm³ and C_{molar mass} is the molar mass of carbon (12 g mol⁻¹). Cellular nitrogen content (N_cell) was calculated based on the cellular carbon content (C_cell), and assuming a C:N ratio of 8.6:1, as determined for the in situ bulk biomass in the respective water depth using an elemental analyzer.

Biovolumes (BV) of PSB were calculated using the formula for a prolate sphere¹⁰⁰, according to Eq. 5

Cell width (i.e., cross-section), and length are given in µm.

Biovolumes of the non-cluster forming C. okenii cells were calculated from cell (region of interest) dimensions obtained from nanoSIMS data and used for the calculation of a cell-specific carbon content (C_cell). For the cluster-forming PSB, biovolumes of 100 cells per population were calculated using the cell dimensions determined from FISH images (see above). The median biovolume was used for the calculation of the population-specific cellular carbon content.

Reporting summary

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

Source: Ecology - nature.com