Bloom of a denitrifying methanotroph, ‘Candidatus Methylomirabilis limnetica’, in a deep stratified lake

Methanotrophic bacteria represent an important biological filter regulating methane emissions into the atmosphere. Planktonic methanotrophic communities in freshwater lakes are typically dominated by aerobic gamma-proteobacteria, with a contribution from alpha-proteobacterial methanotrophs and the NC10 bacteria. The NC10 clade encompasses methanotrophs related to 'Candidatus Methylomirabilis oxyfera', which oxidize methane using a unique pathway of denitrification that tentatively produces N2 and O2 from nitric oxide (NO). Here, we describe a new species of the NC10 clade, 'Ca. Methylomirabilis limnetica', which dominated the planktonic microbial community in the anoxic depths of the deep stratified Lake Zug in two consecutive years, comprising up to 27% of the total bacterial population. Gene transcripts assigned to 'Ca. M. limnetica' constituted up to one third of all metatranscriptomic sequences in situ. The reconstructed genome encoded a complete pathway for methane oxidation, and an incomplete denitrification pathway, including two putative nitric oxide dismutase genes. The genome of 'Ca. M. limnetica' exhibited features possibly related to genome streamlining (i.e. less redundancy of key metabolic genes) and adaptation to its planktonic habitat (i.e. gas vesicle genes). We speculate that 'Ca. M. limnetica' temporarily bloomed in the lake during non-steady-state conditions suggesting a niche for NC10 bacteria in the lacustrine methane and nitrogen cycle.


Introduction
Temperate lakes are environments with intense methane cycling. Methane, a potent greenhouse gas, is abundantly produced in lake sediments from buried organic matter. Due to the comparably low sulfate concentrations, sulfate-dependent anaerobic methane oxidation often fails to completely consume the upward methane flux, in contrast to marine sediments. Therefore, large amounts of methane tend to enter the bottom waters of lakes. Lakes with oxic water columns, in which aerobic methane oxidation is constrained to a thin layer at the sediment surface, significantly contribute to atmospheric methane emissions (Bastviken et al., 2004). In contrast, methane is often completely consumed at the lake oxycline by aerobic methane oxidation in lakes that develop hypoxic and anoxic bottom waters.
Aerobic methane-oxidizing bacteria have long been recognized to play an important role in the regulation of methane emissions to the atmosphere (Oremland and Culbertson, 1992;Reeburgh, 2003). Major taxa of gammaproteobacterial methane-oxidizing bacteria (gamma-MOB) in lakes and other aquatic habitats include Methylomonas, Methylobacter, Methylosoma and Methylosarcina (Bowman, 2006). It has emerged recently that some of the gamma-MOB also thrive in apparently anoxic waters and sediments, where their activity can be sustained by oxygen production and transport (Blees et al., 2014;Milucka et al., 2015;Oswald et al., 2016), fermentation (Kalyuzhnaya et al., 2013) or denitrification (Kits et al., 2015a(Kits et al., , 2015bOswald et al., 2017;Padilla et al., 2017). Interestingly, dedicated anaerobic methane oxidizers belonging or related to the ANME archaea (Knittel and Boetius, 2009;Haroon et al., 2013;Ettwig et al., 2016) seem to be constrained to lake sediments (Schubert et al., 2011;Weber et al., 2017) and play a comparably minor role in methane removal even in fully anoxic water columns.
A group of methanotrophs, whose role in the environmental methane cycle is yet to be fully assessed, are the bacteria of the NC10 phylum (Raghoebarsing et al., 2006). These organisms oxidize methane using nitrite as an electron acceptor. The first described representative of this clade, 'Candidatus Methylomirabilis oxyfera' has been proposed to have a unique capacity to disproportionate nitric oxide intracellularly and produce molecular oxygen, which is used for methane oxidation (Ettwig et al., 2010;Ettwig et al., 2012). This unique pathway allows NC10 to thrive in oxygen-depleted habitats, despite the obligate need for oxygen to activate and oxidize methane (Raghoebarsing et al., 2006;López-Archilla et al., 2007;Zhu et al., 2012;He et al., 2016;Padilla et al., 2016;Shen et al., 2016). Recent studies have demonstrated that NC10-related methanotrophs are present in the anoxic water column of a freshwater reservoir and sediments of deep freshwater lakes such as Lake Constance (Deutzmann and Schink, 2011;Deutzmann et al., 2014) and Lake Biwa (Kojima et al., 2012). They were proposed to significantly contribute to methane removal in these lakes even though direct activity of NC10 in situ has not been demonstrated to date and 'Ca. M. oxyfera' rarely dominates bacterial, or specifically methanotrophic communities, particularly in planktonic habitats.
So far, bacteria of the NC10 phylum have not been found in Lake Zug (Oswald et al., 2016) or the other wellstudied temperate lakes of Switzerland. Here, we report the finding of a 'Ca. M. oxyfera'-related bacterium that dominated the bacterial community in the deep anoxic methane-rich hypolimnion in two consecutive years. We give a genomic description of this new putative species 'Candidatus Methylomirabilis limnetica', infer its in situ activity from metatranscriptomics and describe the biogeochemical conditions during the sampling period that presumably led to the bloom of this bacterium.

Results and discussion
Biogeochemistry of Lake Zug Lake Zug is a deep eutrophic freshwater lake located in Central Switzerland. The lake is permanently stratified and has reportedly not turned over since 1950 (Müller, 1993). During the sampling campaign in September 2016, several interesting features in the chemical profiles were noted ( Fig. 1A and B).
The oxycline was located at about 106 m depth, well above the usual depth (140-150 m) that was measured in 2012, 2013 and 2014 (Oswald et al., 2016). No oxygen was detected with the trace optode (detection limit approximately 20 nmol l −1 ) below this depth on two consecutive sampling days. In comparison to the years before ( , Oswald et al., 2016, methane concentrations at the given depths were ca. twofold higher and methane was depleted ca. 10 m below and not directly at the oxycline. During the sampling campaign in 2016, the flux of oxygen and methane across their respective zones of consumption was 7.8 and 2.2 mmol m −2 d −1 , respectively, and thus lower than the years before. Most interestingly, the concentration profile of NO x (nitrate + nitrite) was concave, decreasing from ca. 22 μmol l −1 at the oxycline to a minimum of ca. 6 μmol l −1 at 150 m depth and then increasing again to a maximum of ca. 45 μmol l −1 at 180 m depth, as opposed to decreasing towards the sediment in the previous years (Oswald et al., 2016). Ammonium concentrations in the bottom waters were ca. threefold higher than in previous years and ammonium disappeared at ca. 115 m, the same depth where methane disappeared ( Fig. 1B and Supporting Information Figure 1b). The flux of ammonium into this zone was 1.8 mmol m −2 d −1 . The profiles of both methane and ammonium indicated an additional zone of consumption between 150 and 160 m. The flux of methane and ammonium in the deeper depths was 2.6 and 4.9 mmol m −2 d −1 respectively.
We speculate that higher deposition rates of organic matter into the hypolimnion (through higher primary productivity or enhanced transport) resulted in a concomitant upward shift of the oxycline. This might have resulted in increased fluxes of methane and ammonium out of the sediment. The most parsimonious explanation for the obviously non-steady state NO x profile is that, prior to our sampling campaign, vertical mixing or inflow of oxygenrich, high density water (hyperpycnal flow) resulted in oxygen intrusion into the bottom waters, which in turn led to the oxidation of ammonium and production of nitrogen oxides.
The biogeochemical profiles recorded during a sampling campaign in October 2015 were more similar to the profiles described previously (Oswald et al., 2016). Oxygen was fully consumed at ca. 124 m depth, coinciding with the depletion of methane and ammonium (Supporting Information Figure 1a ,b). NO x concentrations decreased starting from 21 μmol l −1 at 118 m to 11 μmol l −1 at 140 m, the deepest depth measured in 2015 (Supporting Information Figure 1b). Between 122 m and 128 m nitrite was detected at low nanomolar concentrations but was otherwise below detection limit.
NC10 bacteria were abundant in the profundal anoxic waters of Lake Zug In previous years, a large portion of upwards-diffusing methane was shown to be oxidized near the oxycline by abundant gamma-proteobacterial methane-oxidizing bacteria (Oswald et al., 2016). Additionally, it was shown that filamentous gamma-proteobacterial Crenothrix bacteria were major methane consumers in Lake Zug (Oswald et al., 2017).
To obtain a quantitative overview of the methaneoxidizing community in Lake Zug in 2016, we first classified and quantified 16S rRNA gene sequences in the unassembled metagenomic sequences from all three depths; near the oxycline (110 m), below it (120 m) and in the middle of the anoxic hypolimnion (160 m; Fig. 1A). For all three depths, a metagenome (Illumina HiSeq2500 2 × 250 bp) and a metatranscriptome (Illumina HiSeq3000 1 × 150 bp) were generated (Supporting Information Table 2). Methylococcales were stable members of the microbial community at all three investigated depths. Up to 10% of all 16S rRNA gene sequences were classified as Methylococcales; the majority of these belonged to genera Methylobacter, Crenothrix, Methylomicrobium and the CABC2E06 clade. Sequences classified as verrucomicrobial methanotrophs (mainly 'Ca. Methylacidiphilum') were also detected, albeit at low abundance (0.1%-0.3%). Although verrucomicrobial methanotrophs related to 'Ca. Methylacidiphilum' are known to thrive under thermoacidophilic conditions (Op den Camp et al., 2009), several studies have now detected 'Ca. Methylacidiphilum'-related bacteria in more 'conventional' habitats (i.e. paddy soils, lake sediments; Sinclair et al., 2017;Vaksmaa et al., 2017).
Known alphaproteobacterial methanotrophs (e.g., Methylocystaceae and Beijerinckiaceae) were not detected. This is consistent with the proteobacterial methanotrophic community analysed in this lake previously by catalysed reporter deposition fluorescence in situ hybridization (CARD-FISH; Oswald et al., 2016Oswald et al., , 2017. Interestingly, in the metagenomic sequences from 120 m and 160 m depth we found conspicuously high abundances of 16S rRNA gene sequences putatively assigned to the NC10 phylum. In these metagenomes, NC10-related sequences constituted approximately 10% and 19% of all classified metagenomic 16S rRNA gene sequences, respectively, and thus were up to two-fold more abundant than gamma-proteobacterial methanotrophs. At 110 m depth, which was nearest to the oxycline, only 0.7% of all classified 16S rRNA gene sequences were assigned to NC10. Somewhat lower abundances of NC10 were detected in Lake Zug in 2015, when NC10 abundance detected by 16S rRNA gene amplicon sequencing increased with depth to approximately 7% of all 16S rRNA gene sequences at 136 m depth (Supporting Information Figure 1b).
The DBACT-0193 probe has 1 nucleotide mismatch to the assembled 16S rRNA gene sequence belonging to NC10 retrieved from Lake Zug (Supporting Information  Table 1). A similarly high proportion of cells was hybridized with the DBACT-1027 probe (0 mismatches; data not shown), whereas no hybridized cells were found with the DBACT-447 probe (5 mismatches).
Our CARD-FISH and metagenomic analyses showed that planktonic NC10 bacteria were the dominant methanotrophic microorganisms in the profundal, anoxic waters of Lake Zug in two consecutive years. To our knowledge, the abundance of NC10 found in 2016 is the highest relative abundance of NC10 that has been so far reported from any environment. The highest previous report was from the Feitsui reservoir where up to 16% of all cells were identified as NC10 using CARD-FISH (DBACT-1027 probe; Kojima et al., 2014). Interestingly, apart from being eutrophic, Lake Zug and Feitsui reservoir share few similarities. Whereas Lake Zug is a deep and permanently stratified temperate lake, Feitsui reservoir is a comparably shallow (mean depth of 40 m) and monomictic subtropical reservoir. It is thus not immediately obvious which conditions might favour the growth of NC10 to such high abundances.
One possible explanation for the high abundance of NC10 might be a transient formation of nitrite in the anoxic bottom waters that could have then subsequently stimulated growth of NC10 bacteria and lead to the observed bloom. For example, nitrite could have been formed by: (i) ammonium-oxidizing microorganisms after intrusion of oxygenated water into the ammonium-rich bottom waters (as indicated by the unusual NO x profile) or by (ii) nitrate-reducing microorganisms reducing nitrate to nitrite. Several taxa which are known to contain denitrifying bacteria, such as Methylococcaceae, Methylophilaceae and Comamonadaceae were indeed found in the metagenomes from all three depths. The 'Ca. Methanoperedens'-like archaea which have been repeatedly observed to co-occur with NC10 in bioreactor enrichment cultures (Raghoebarsing et al., 2006;Haroon et al., 2013;Ettwig et al., 2016) were not detected in our metagenomes. Intriguingly, sequences assigned to known ammonium-oxidizing genera such as 'Ca. Nitrosoarchaeum' and 'Ca. Nitrosopumilus' were detected even in the 'anoxic' metagenome from 160 m (1.2% of all 16S rRNA gene sequence, data not shown), which might indicate a past oxygenation event. In any case, the high abundance of NC10 bacteria in Lake Zug also suggests that these microorganisms were able to successfully compete for nitrite with anaerobic ammonium-oxidizing anammox bacteria, which were also detected in our Lake Zug metagenomes and are generally assumed to outcompete NC10 when ammonium is not limiting (Luesken et al., 2011a;Haroon et al., 2013). Anammox bacteria (Ca. Brocadia) were detected in all three investigated depths at abundances of ca. 0.1% -0.6% of the total bacterial community in situ.
Genome reconstruction and phylogenetic assignment of 'Ca. M. limnetica' The high abundance of NC10 bacteria in the samples enabled us to assemble a putative NC10 genomic bin. The binning process was based on guanine-cytosine content as well as differential contig coverage between metagenomes from 120 m and 160 m. A putative NC10 genomic bin was obtained from a co-assembly of all three depths. The contigs within this bin had the highest average coverage in the metagenomes from 120 m and 160 m (average contig coverage 451-fold and 851-fold) but only comparatively low coverage (30-fold) in the 110 m metagenome. The average contig coverage of the metagenomic bin matched well with the abundance of NC10 bacteria previously estimated in our 16S rRNA read survey and CARD-FISH analysis. Furthermore, we could not retrieve any other genomic bin related to 'Ca. Methylomirabilis' from the co-assembled metagenomes from all three depths suggesting that only one dominant 'Ca. Methylomirabilis' strain was present at all three depths. Summary statistics of the genomic bin (after targeted re-assembly) and comparison to the closed genome of 'Ca. M. oxyfera' is shown in Supporting Information Table 3. Analysis by CheckM (Parks et al., 2015) suggested that the genomic bin was of high quality with similar estimates of completeness (96.2%) and contamination (1.7%) as the closed genome of 'Ca. M. oxyfera' (accession FP565575.1, Supporting Information Table 3). Furthermore, analysis of the genomic bin by CheckM did not indicate strain heterogeneity.
Next we used the assembled full-length 16S rRNA gene sequence (1549 bp) retrieved from the NC10 genomic bin for taxonomic classification. Comparative analysis of the 16S rRNA gene sequence showed 95.1% identity to 'Ca. M. sinica'  and 96.3% identity to 'Ca. M. oxyfera' (Raghoebarsing et al., 2006;Ettwig et al., 2010). These values are higher than the threshold for genus definition (95%) but are below the species cutoff value (98.6%; Yarza et al., 2014;Konstantinidis et al., 2017). A whole genome analysis further showed that the pairwise average nucleotide identity (ANI) between our retrieved NC10 genome and the genome of 'Ca. M. oxyfera' was 81.8%. This is well below the accepted ANI species boundary (95%-96%; Goris et al., 2007;Richter and Rosselló-Móra, 2009). The V3-V4 region of the assembled 16S rRNA gene sequence was identical to the sequence of the dominant NC10 OTU (428 bp) identified by 16S rRNA gene amplicon sequencing from samples retrieved in 2015. Taken together these data suggest that the NC10 population present in Lake Zug qualifies as a new species within the genus 'Ca. Methylomirabilis', which we named 'Candidatus Methylomirabilis limnetica' [lim.ne 0 ti.ca. N.L. fem. adj. limnetica pertaining to lakes].
Phylogenetic analysis of the 16S rRNA gene sequences showed that the sequence of 'Ca. M. limnetica' clustered within a subgroup of NC10 ( Fig. 2A). The sequences within this subgroup were nearly identical (>99% sequence identity) and likely represented the same species. Interestingly, the sequences were retrieved from geographically distant freshwater lakes (Lake Constance, Germany; Lake Biwa, Japan; Deutzmann and Schink, 2011;Kojima et al., 2012), the Feitsui freshwater reservoir (Taiwan; Kojima et al., 2014) and a minerotrophic peatland (Brunssummerheide, The Netherlands; Zhu et al., 2012). Both currently described species of the genus 'Ca. Methylomirabilis', 'Ca. M. oxyfera' and 'Ca. M. sinica', clustered in a different, more divergent branch of the 16S rRNA gene tree ( Fig. 2A).

Genome-inferred central C1 and energy metabolism
The high-quality genome of 'Ca. M. limnetica' allowed for reconstruction of pathways involved in carbon and energy metabolism (Fig. 3). 'Ca. M. limnetica' encoded the pathway for complete aerobic oxidation of methane (Table 1), including particulate methane monooxygenase (pMMO; pmoCAB) and one Xox-type methanol dehydrogenase (MDH; xoxFJG). Genes encoding for soluble methane monooxygenase (sMMO) and MxaF-type methanol dehydrogenase were not encoded in 'Ca. M. limnetica' genome. Downstream conversion of formaldehyde to formate could either proceed via tetrahydromethanopterin (H 4 MPT)-dependent or tetrahydrofolate (H 4 F)-dependent C 1 transfer pathway. Formate dehydrogenase (fdhA), catalysing the oxidation of formate to CO 2 , was also encoded in the genome of 'Ca. M. limnetica'.

Respiratory complexes of 'Ca. M. limnetica'
It has been proposed that NC10 bacteria, specifically 'Ca. M. oxyfera', produce O 2 from nitrogen oxides by a unique intra-aerobic denitrification pathway involving a nitrite reductase and a putative nitric oxide dismutase (Ettwig et al., 2010;Wu et al., 2011a;Ettwig et al., 2012). Also the genome of 'Ca. M. limnetica' encoded for a partial denitrification pathway ( Fig. 3; Table 1) including periplasmic nitrate reductase (napAB), cd 1type nitrite reductase (nirS), two putative nitric oxide dismutases and one gene encoding for quinone-interacting nitric oxide reductase (qNOR; norB). Membrane-bound nitrate reductase (narGHI) and nitrous oxide reductase (nosZ) were not found in the 'Ca. M. limnetica' genome. The amino-acid sequences of two putative nitric oxide dismutases of 'Ca. M. limnetica' (encoded by tandem genes CLG94_00135 and CLG94_00140, Table 1) featured nearly all modified residues of the quinol-binding and catalytic site that have been identified in two putative nitric oxide dismutases of 'Ca. M. oxyfera' and marine NC10 bacteria (Supporting Information Figure 5; Ettwig et al., 2012;Padilla et al., 2016). These putative nitric oxide dismutases, which are related to qNOR, have been suggested to catalyse the dismutation of two molecules of nitric oxide into N 2 and O 2 thus allowing NC10 bacteria to oxidize methane using pMMO in the absence of exogenous O 2 (Ettwig et al., 2010;Ettwig et al., 2012). Like 'Ca. M. oxyfera', 'Ca. M. limnetica' also encoded for a third, most likely genuine nitric oxide-reducing qNOR that contained the same conserved residues of canonical qNORs of 'Ca. M. oxyfera' and other microorganisms (Supporting Information Figure 5).
In addition to nitrogen oxides, 'Ca. M. limnetica' also has the genomic potential to use O 2 as terminal electron acceptor. We identified genes encoding for two types of terminal oxidases (Table 1); a heme copper oxidase (A1-type HCO; Pereira et al., 2001) and an alternative oxidase (AOX) that belongs to the di-iron carboxylate group of proteins (Berthold and Stenmark, 2003). Although NC10 bacteria grow anaerobically, it has been speculated that O 2 from nitric oxide dismutation could be respired by these terminal oxidases (Wu et al., 2011b). The genome of 'Ca. M. oxyfera' also encoded for two additional heme-copper terminal oxidases (bo-and ba 3 -type, labelled HCO 2 and 3; Wu et al., 2011b) which were however not found in the genome of 'Ca. M. limnetica'. Alignment of Lake Zug metagenomic reads (160 m) to the genome of 'Ca. M. oxyfera' further confirmed that these HCOs were encoded in genomic regions not present in our metagenomic dataset (Supporting Information Figure 3).
Intriguingly, several genes encoding for a cytochrome bc 1 complex, which have been identified in 'Ca. M. oxyfera' (Ettwig et al., 2010;Wu et al., 2011b), were not found in the genome of 'Ca. M. limnetica'. Instead we identified genes encoding for two cytochrome bc-like complexes, homologs of which were also encoded in the genome of 'Ca. M. oxyfera' (Table 1). Both cytochrome bc-like complexes featured tandem genes encoding for a cytochrome b and a Rieske iron-sulfur protein in addition to multi-heme cytochromes c (Supporting Information Figure 4). The cytochromes b were either encoded as single, long gene (CLG94_00770) or as two separate, shorter genes (CLG94_01010-15) that appeared to constitute two different, evolutionary-related subfamilies of cytochromes b (Dibrova et al., 2013;Dibrova et al., 2017). As 'Ca. M. limnetica' did not encode for a canonical cytochrome bc 1 complex, we suggest that either one or both cytochrome bc-like complexes, which were transcribed (Table 1), act as a quinol:cytochrome c oxidoreductase. The importance of Rieske/cytochrome bc complex for growth on nitrate and N 2 O has recently been demonstrated in Wolinella succinogenes (Hein et al., 2017).
In addition to the cytochrome bc 1 complex, several other key metabolic genes encoding for nitrogen metabolism and the respiratory chain of 'Ca. M. oxyfera' appeared to be absent from the genome of 'Ca. M. limnetica' (Supporting Information Table 4). These were genes encoding for membrane-bound nitrate reductase (Nar), two heme copper oxidases (previously assigned HCO2 and HCO3; Wu et al., 2011b), MxaFtype as well as one additional XoxF-type methanol dehydrogenase and hydroxylamine oxidoreductase (Hao, also known as hydroxylamine dehydrogenase). The absence of hao genes from the 'Ca. M. limnetica' genome is intriguing as Hao has been suggested to play a role in the detoxification of hydroxylamine in methanotrophs (Nyerges and Stein, 2009;Campbell et al., 2011). Hydroxylamine is formed via the co-metabolism of ammonia by methane monooxygenase-a process that is likely also relevant in Lake Zug as methane and ammonium were present in almost equimolar concentrations in situ ( Fig. 1A and B). Other genes encoding for enzymes known to be involved in hydroxylamine detoxification, such as cytochrome P460 (cytL; Bergmann et al., 1998), were also absent from the 'Ca. M. limnetica' genome thus raising the question of how 'Ca. M. limnetica' disposes of this toxic intermediate. Detoxification of potentially formed hydroxylamine could either be performed by 'Ca. M. limnetica' periplasmically via a yet unidentified non-canonical hydroxylamine oxidoreductase or could alternatively involve hydroxylamine export and degradation by other members of the microbial community.
To confirm that genes known to mediate hydroxylamine detoxification (hao, cytL) in methanotrophs were indeed absent from the genome of 'Ca. M. limnetica' we searched the whole metagenomic assembly for genes encoding the aforementioned enzymes but could not identify highly covered contigs encoding for close homologs. Additionally, by mapping the sequences of the 160 m metagenome to the genome of 'Ca. M. oxyfera' we found that whereas genomic regions with gene homologs shared between both species were well covered (per-base coverage >100), the average coverage was close to zero for all genomic regions containing genes, Table 1. In situ transcription of functional genes involved in methane oxidation and respiration of 'Ca. M. limnetica'. Listed are functional genes encoding for dissimilatory nitrogen metabolism and respiratory complexes of 'Ca. M. limnetica' as well as genes encoding for the complete methane oxidation. Transcription was quantified as RPKM (reads per 1 kb gene length and per million mapped transcripts) in metatranscriptomes obtained from 110 m, 120 m and 160 m depth. Gene homologs of 'Ca. M. oxyfera' (FP565575.1) have been identified using amino-acid sequences BLASTP; only the top hit (by e-value or % sequence identity) is shown (alignment coverage >93%; except NapA 83%).

Protein
Gene which included hao, that appear exclusive to 'Ca. M. oxyfera' (Supporting Information Figure 3).

In situ gene transcription of 'Ca. M. limnetica'
To investigate whether 'Ca. M. limnetica' was transcriptionally active in situ during our sampling campaign in 2016, we aligned the metatranscriptomic reads obtained from 110 m, 120 m and 160 m depth to the 'Ca. M. limnetica' genome. We found that nearly one third of all non-rRNA metatranscriptomic sequences from 120 m and 160 m (28.4% and 32% respectively) aligned to the genome of 'Ca. M. limnetica'. The overall alignment rate of the metatranscriptome from 110 m (2.8%) was much lower, in line with the much lower abundance of 'Ca. M. limnetica' at this depth. 16S rRNA amplicon sequencing of RNA samples obtained in 2015 also showed that NC10 bacteria were actively transcribing 16S rRNA genes. In water samples from 136 m and 140m, the relative abundance of 16S rRNA assigned to NC10 bacteria constituted approximately 22% of all 16S rRNA sequences. A comparison of the 100 most transcribed genes of 'Ca. M. limnetica' in our metatranscriptomic datasets from 2016 showed a clear difference between metatranscriptomes originating from near the oxycline (110 m) and from below (120 and 160 m). We found that 94 out of the top 100 transcribed genes were shared between the two metatranscriptomes from 120 m and 160 m depth. This was not the case for the 110 m metatranscriptome, where only about half of the top 100 transcribed genes were shared with the metatranscriptomes from 120 m and 160 m. We found several genes encoding for toxinantitoxin systems and proteases exclusively transcribed among the top 100 genes by 'Ca. M. limnetica' at 110 m. Toxin-antitoxin systems appear to have an important role in bacterial stress physiology and growth control (Buts et al., 2005;Hayes and Low, 2009;Blower et al., 2011). Hence, the increased transcription of these genes might reflect a response of 'Ca. M. limnetica' to hypoxic conditions close to the oxycline.
Transcription of functional genes involved in methane oxidation and denitrification was in accord with the proposed anaerobic, methanotrophic and denitrifying lifestyle of NC10 bacteria. Among the 100 most transcribed genes of 'Ca. M. limnetica' shared in all three metatranscriptomes were the genes encoding for methane oxidation and denitrification (Table 1); in particular genes encoding for particulate methane monooxygenase, nitrite reductase and one gene copy of the putative nitric oxide dismutase (CLG94_00135) were highly transcribed. At 120 m and 160 m, transcription of the second gene copy (CLG94_00140) was three orders of magnitude lower. Interestingly, at 110 m both putative nitric oxide dismutase genes were highly transcribed (Table 1). This observation is in line with the suggestion of Luesken and colleagues (2012) who proposed that 'Ca. M. oxyfera' might only transcribe both copies of nitric oxide dismutase when exposed to oxygen. Transcription of the 'canonical' nitric oxide reductase was detected at all depths, implying that 'Ca. M. limnetica' might also reduce nitric oxide to N 2 O in situ. However, the transcription levels of the 'canonical' nitric oxide reductase were significantly lower than those of the predominantly transcribed putative nitric oxide dismutase (84 to 146-fold). 'Ca. M. limnetica' also encoded a periplasmic nitrate reductase (NapAB) but its transcription was much lower than that of, for example, the nitrite reductase (NirS, Table 1). It thus remains unclear whether 'Ca. M. limnetica' can use nitrate as an electron acceptor in situ. 'Ca. M. oxyfera', which also possesses NapAB, was incapable of using nitrate as electron acceptor for methane oxidation (Ettwig et al., 2010).
Besides genes involved in methane oxidation and dissimilatory nitrogen metabolism, many genes encoding for proteins involved in transcription and translation (i.e. RNA polymerase, translation initiation factor) as well as numerous ribosomal proteins were among the highest transcribed genes at all depths. Furthermore, we identified a well transcribed gene cluster encoding for several gas vesiclerelated proteins including the main structural gas vesicle protein (GvpA) and associated proteins (GvpL/F, GvpN and GvpK). In fact, one gvpA gene was among the highest transcribed genes at all depths (30,000 RPKM at 120 m and 160 m). The presence and transcription of genes encoding for gas vesicles suggests that 'Ca. M. limnetica' might be capable of adjusting or maintaining its position in the water column. Interestingly, 'Ca. M. oxyfera', which was isolated from freshwater sediment, appears not to encode homologs of these Gvp-associated proteins.
Concluding remarks 'Ca. Methylomirabilis limnetica', a new species of the candidate genus Methylomirabilis, dominated the methanotrophic and even overall microbial community in the hypolimnion of a deep stratified lake. We speculate that high organic matter deposition and a possible oxygenation of the hypolimnion prior to the sampling campaign might have triggered the bloom of 'Ca. M. limnetica'. The large proportion of highly transcribed genes in all three in situ transcriptomes suggests that at the time of sampling the 'Ca. M. limnetica' population was still transcriptionally active, even though it is not clear whether this activity was accompanied by methane oxidation. Our phylogenetic analyses show that 'Ca. M. limnetica' occurs in a variety of freshwater habitats and might possibly be adapted to a planktonic lifestyle, as suggested by the unique presence of gas vesicle-encoding genes in its genome.

Geochemical profiling and sample collection
Sampling was carried out in September 2016 and October 2015 at a single station located in the deep, southern lake basin of Lake Zug (200 m water depth; 47 06 0 00.8 00 N, 8 29 0 35.0 00 E). A multi-parameter probe was used to measure conductivity, turbidity, depth (pressure), temperature and pH (XRX 620, RBR, Ottawa, ON, Canada). Dissolved oxygen was monitored online with normal and trace micro-optodes (types PSt1 and TOS7, Presens, Regensburg, Germany) with detection limits of 125 and 20 nM, respectively, and a response time of 7 s.
In 2016, water samples for measurements of methane and nitrous oxide concentrations were retrieved from distinct depths with a syringe sampler. Water from individual 50 ml syringes was filled through a gas-tight rubber tubing into serum bottles (120 ml), allowing water to overflow. Solid copper chloride (CuCl 2 ) was immediately added in excess to the water samples and the bottles were closed with a butyl rubber stopper (head-space free) and crimped. Before analysis, a 30 ml headspace was set with N 2 and after overnight equilibration methane and nitrous oxide concentrations were measured in the headspace. In 2015, 20 ml of lake water were taken for methane analysis with a syringe sampler and filled into closed 40 ml serum bottles containing 4g NaOH pellets and 600 mbar N 2 gas. The NaOH pellets were found to release methane, and this contamination was subtracted from the total methane measured in each sample. Headspace methane samples from 2016 and 2015 were measured with a gas chromatograph (Agilent 6890 N, Agilent Technologies, Santa Clara, CA, USA) equipped with a Carboxen 1010 column (30m × 0.53 mm, Supelco, Bellefonte, PA, USA) and a flame ionization detector. Methane concentrations in the water phase were back-calculated according to Wiesenburg and Guinasso (1979).
Concomitantly with dissolved gases, water samples for ammonium and NO x measurements (only samples from 2016) were collected from the same depths using the same syringe sampler. Forty millilitre of water was directly injected into a 50 ml Falcon tube containing 10 ml of OPA reagent for fluorometric ammonium quantification according to Holmes and colleagues (1999). For NO x quantification, 1.5 ml of water was added to an eppendorf cup prefilled with 15 μl HgCl 2 and combined nitrate and nitrite concentration was determined by commercial chemiluminescence NO x analyser after reduction to nitric oxide with acidic Vanadium (II) chloride (Braman and Hendrix, 1989). After reduction to nitric oxide, nitrite was determined with acidic potassium iodide and nitrate was then calculated as the difference between NO x and nitrite. For ammonium and NO x analysis of samples collected in 2015, syringe samples were pre-filtered with a cellulose acetate filter (0.2 μm pore size). Ammonium was quantified according to Parsons and colleagues (1984) and nitrate and nitrite were measured by flowinjection analysis (FIA; SAN++, Skalar) following ISO 13395.
In 2016, samples were taken for CARD-FISH analysis. For that, 3 ml of water were sampled from 120 m and 160 m depth into a 15 ml Falcon tube containing 0.3 ml 20% formaldehyde.
In 2016, water for DNA/RNA analyses was collected with a Niskin bottle from 110 m, 120 m and 160 m water depth. For each depth, 2 × 1 L water was immediately filtered onboard (0.2 μm GTTP filter; Merck Millipore, Darmstadt, Germany); filters for DNA extraction were air dried and filters for RNA extraction were immediately immersed in RNAlater preservation solution (Life Technologies, Carlsbad, CA, USA). DNA and RNA filters were stored at -20 C until further processing. In 2015, 105-220 ml water for DNA/RNA analyses were collected with a syringe sampler, immediately filtered on board onto a 0.2 μm pore size cellulose acetate filter (Sartorius, Göttingen, Germany) and immediately frozen on dry ice and stored at -80 C until further processing.

Diffusive flux calculation
Diffusive fluxes (J) of oxygen, ammonium and methane were calculated assuming steady state using Fick's first law: where ∂C=∂z is the concentration gradient. Following Oswald and colleagues (2016), we used a turbulent diffusion coefficient D of 2.7 × 10 −5 m 2 s −1 . This value was determined in Lake Zug by the heat budget method (Powell and Jassby, 1974) using temperature data from September 2012 to June 2013. In order to calculate fluxes towards the chemocline, a depth interval between 60 and 106 m was used for O 2 and a depth interval between 120 and 117.5 m for CH 4 and NH + 4 . CH 4 and NH + 4 flux in deeper depths was calculated over a depth interval between 150 and 180 m and 170 and 180 m respectively.

Catalysed reporter deposition-fluorescence in situ hybridization (CARD-FISH)
Waters samples (3 ml) were fixed with formaldehyde (final concentration 2% [w/v]) for up to 12 h at 4 C before being filtered onto polycarbonate GTTP filters (0.2 μm pore size, effective filter diameter = 20 mm; Merck Millipore, Darmstadt, Germany). Permeabilization with lysozyme, peroxidase inactivation, hybridization with specific oligonucleotide probes labelled with horseradish peroxidase (for details see Supporting Information Table 1; Biomers, Ulm, Germany), tyramide reporter deposition (Oregon Green 488) and 4 0 ,6-diamidino-2-phenylindole (DAPI) counter staining was performed according to Pernthaler and colleagues (2002). Filters were embedded in a mix of Citifluor/Vectashield (4:1) and mounted onto glass slides. Cell counting was performed with a Eclipse Ci microscope (Nikon Instruments, Germany) in randomly selected fields of view until 1,000 DAPI-stained cells were counted.
Nucleic acid extraction, 16S rRNA (gene) sequencing and whole metagenome and metatranscriptome sequencing For nucleic acid extraction from samples collected in 2015, DNA and RNA were extracted simultaneously with the AllPrep DNA/RNA Mini Kit (Qiagen). Residual DNA in the RNA extract was digested with TURBO DNA-free kit (Invitrogen).
For samples collected in 2016, DNA was extracted from cut-up filters using Powersoil DNA isolation kit. For RNA extraction, filters were briefly rinsed with nucleasefree water and RNA was extracted from cut-up filters using PowerWater RNA isolation kit (including removal of genomic DNA by DNase I digestion). Both nucleic acid extraction kits (MoBio Laboratories, Carlsbad, CA, USA) were used according to manufacturer's instructions. DNA and RNA were quantified using the Qubit dsDNA HS or RNA HS Assay kits and the Qubit 2.0 Fluorometer (Invitrogen, Carlsbad, CA, USA).
For metagenomic sequencing, DNA was fragmented by sonication (500 nt) using a Covaris S2 sonicator (Covaris, Woburn, MA, USA) and library preparation was done according to manufacturer's instructions using NEBNext Ultra II DNA Library Prep Kit for Illumina (New England Biolabs, Ipswich, MA, USA). Paired-end sequencing (2 × 250 bp) was performed using the Illumina HiSeq2500 platform (Illumina, San Diego, CA, USA) in rapid mode with SBS chemistry v2.
For metatranscriptomic sequencing, total RNA was first concentrated using the RNA Clean and Concentrator kit (Zymo Research Corp., Irvine, CA, USA) according to manufacturer's instructions. Depletion of rRNAs was done with the Ribo-Zero rRNA Removal Kit (Bacteria) for Illumina (Epicentre, Madison, WI, USA) with a protocol adaptation for low input amounts. cDNA library preparation was done with the NEBNext Ultra Directional RNA Library Prep Kit for Illumina (New England Biolabs) according to protocol and sequencing (1 × 150 bp) was performed using the Illumina HiSeq3000 platform (Illumina) with SBS chemistry. Library preparation and sequencing was performed by the Max Planck-Genome-centre Cologne, Germany (http://mpgc.mpipz.mpg.de/ home/). Detailed information for each metagenomic and metatranscriptomic dataset can be found in Supporting Information Table 2.
For 16S rRNA (gene) sequencing (only 2015 samples), RNA was reverse-transcribed with random hexamers into cDNA with SuperScript™ IV First-Strand Synthesis System (Invitrogen). The library preparation of 16S rRNA and rDNA for Illumina MiSeq sequencing was done with a two-step PCR using NEBNext Q5® Hot Start HiFi PCR Master Mix (New England Biolabs). The first PCR (17 cycles, initial denaturation 30 s, 98 C, denaturation 10 s, 98 C, annealing 35 s, 54 C, extension 35 s, 65 C and final extension 5 min, 65 C) was done in triplicates with tailed forward primer S-D-Bact-0341-b-S-17 and tailed reverse primer S-D-Bact-0785-a-A-21 (Klindworth et al., 2013). The products were pooled and cleaned with Agencourt AMPure XP kit (BeckmanCoulter, Brea, CA, USA) followed by a second PCR (8 cycles, annealing at 55 C) to attach Illumina barcodes and adapters using Nextera XT Index Kit set A and D (Illumina). Libraries were cleaned again, quantified by fluorometry with Qubit DNA BR reagents (Invitrogen) and measured on a microplate reader (Spark M10; Tecan, Männedorf, Switzerland). Libraries were pooled and quality was controlled with Tape Station 2200 (Agilent). Sequencing was performed using the Illumina MiSeq platform (Illumina) with 600-cycle MiSeq reagent kit v3 (Illumina) and 10% PhiX. Amplicon sequencing was done at the Genetic Diversity Centre (GDC) of ETH Zurich.
The Whole Genome Shotgun project of 'Ca. Methylomirabilis limnetica' has been deposited at DDBJ/ENA/ GenBank under the accession NVQC00000000 and Bio-Project PRJNA401219. The version described in this article is version NVQC01000000. Raw metatranscriptomic sequencing data has been deposited at the NCBI Sequence Read Archive under accession SRP132411 (SRR6684601, SRR6684602, SRR6684603).
Pairwise average nucleotide identity (ANI) values between the genomes of 'Ca. M. limnetica' and 'Ca. M. oxyfera' were calculated using BLAST (ANIb) and the JSpeciesWS online service (Richter et al., 2016). Relative genome sequence coverage of 'Ca. M. oxyfera' (Supporting Information Figure 3) was calculated by mapping trimmed metagenomic sequences from Lake Zug (160 m) to the genome of 'Ca. M. oxyfera' (retrieved from GenBank; accession FP565575) using BBmap 35.43 (Bushnell, 2016) and standard settings. Average genome coverage (500 bp interval) was calculated and visualized using BLAST Ring Image Generator (Alikhan et al., 2011). Gene coordinates of selected genes were imported from GenBank and Refs. (Wu et al., 2011b;Luesken et al., 2012). Homologs shared between 'Ca. M. limnetica' and 'Ca. M. oxyfera' were identified using BLASTP (Camacho et al., 2009) with protein-coding CDS of 'Ca. M. limnetica' as queries and all protein-coding CDS of 'Ca. M. limnetica' as subject database. The homologs reported in Table 1 represents the top BLASTP hit (by e-value and % sequence identity) and were manually inspected to assure that the alignment coverage was sufficient (typically >90%).
Multiple sequence alignment of amino-acid sequences of nitric oxide reductase was done following the previous alignment by Ettwig and colleagues (2012). Sequences were retrieved from GenBank, imported into JalView 2.10.1 (Waterhouse et al., 2009) and aligned using Clus-talOmega 1.0.2 (Sievers et al., 2011) web service implemented in JalView.

Phylogenetic analyses
Full length 16S rRNA gene sequence was retrieved from the genome of 'Ca. M. limnetica' using RNAmmer 1.2 (Lagesen et al., 2007), aligned using the SILVA incremental aligner (SINA) (Pruesse et al., 2012) and imported to the SILVA SSU NR99 database [release 123; ] using ARB 6.1 (Ludwig et al., 2004). Additional NC10 16S rRNA gene sequences originating from Lake Constance (Deutzmann and Schink, 2011), Brunssummerheide (Zhu et al., 2012) and the Eastern Tropical North Pacific (Padilla et al., 2016) were also added to this dataset. Maximum likelihood phylogenetic trees of 16S rRNA gene sequences were calculated using RAxML 7.7.2 (Stamatakis, 2006) integrated in ARB with the GAMMA model of rate heterogeneity and the GTR substitution model with 500 bootstraps.