Metabolic specialization of denitrifiers in permeable sediments controls N2O emissions

Coastal oceans receive large amounts of anthropogenic fixed nitrogen (N), most of which is denitrified in the sediment before reaching the open ocean. Sandy sediments, which are common in coastal regions, seem to play an important role in catalysing this N-loss. Permeable sediments are characterized by advective porewater transport, which supplies high fluxes of organic matter into the sediment, but also leads to fluctuations in oxygen and nitrate concentrations. Little is known about how the denitrifying communities in these sediments are adapted to such fluctuations. Our combined results indicate that denitrification in eutrophied sandy sediments from the world's largest tidal flat system, the Wadden Sea, is carried out by different groups of microorganisms. This segregation leads to the formation of N2 O which is advectively transported to the overlying waters and thereby emitted to the atmosphere. At the same time, the production of N2 O within the sediment supports a subset of Flavobacteriia which appear to be specialized on N2 O reduction. If the mechanisms shown here are active in other coastal zones, then denitrification in eutrophied sandy sediments may substantially contribute to current marine N2 O emissions.


Introduction
The continental shelves and specifically coastal regions act as a buffer between the land and open ocean, making them vulnerable to anthropogenically induced eutrophication. In these regions, sands cover more than 50% of the seafloor (Emery 1968, Hall 2002, however little is known about the role of sandy sediments in global biogeochemical cycles of carbon and nitrogen. Since the 1860's nitrogen (N) inputs to coastal regions from riverine and land drainage have more than doubled, while atmospheric Ndeposition has risen sixfold, leading to current day inputs of around 90 Tg N yr −1 (Dentener et al. 2006, Gruber andGalloway 2008). In the world's largest intertidal area, the Wadden Sea, which is predominantly comprised of sandy sediments, nitrate inputs have increased eightfold since the 1930's (van Beusekom 2005.
The advection of seawater into permeable sands creates porewater flow, filtering out organic matter and delivering nutrients and electron acceptors (e.g. oxygen and nitrate) at rates orders of magnitude higher than diffusive fluxes (Huettel et al., 2003). However the redox conditions in permeable sands are highly variable over time scales of minutes to hours , Jansen et al, 2009). This is a result of changes in the magnitude of porewater flow, which is controlled by bottom water currents, sediment topography and sediment transport (Ahmerkamp et al., 2015. The combination of high fluxes with microbial abundances of around 10 9 cells ml −1 sediment (Musat et al., 2006), means that permeable sediments act as biocatalysts, enhancing oxygen respiration and denitrification rates (Gao et al. 2012, Janssen et al. 2005, Marchant et al. 2016, Sokoll et al. 2016. Consequently, large amounts of anthropogenic N-inputs that enter coastal regions can be removed within permeable sediments. Still, little is known about how the microbial communities mediating denitrification in permeable sediments cope with the large fluctuations in oxygen and substrate availability that occur on small spatial and temporal scales.
Denitrification is a four-step, modular process, carried out by microorganisms with highly branched respiratory chains. Almost all steps can be catalysed by more than one enzyme, therefore the complete reduction of NO 3 − to N 2 can occur via at least 12 different routes (Schreiber et al., 2012, Simon and Klotz 2013, Kuypers et al., 2018 with each enzyme differing in its affinity, reaction rate and response to environmental factors (Philippot and Hallin 2005). Furthermore, each intermediate reaction of the denitrification pathway can occur independently from the others. While denitrification is often studied in model organisms that are 'complete' denitrifiers, it is becoming apparent that a network of N-cycling microorganisms might be responsible for environmental denitrification, with diverse microbes specialized to respire only one or some of the intermediates (Graf et al., 2014, Kuypers et al., 2018, Lilja and Johnson 2016, Philippot et al., 2013, Roco et al., 2016. It has been recognized for many years that some denitrifiers lack a nitrous oxide (N 2 O) reductase (Henry et al. 2006, Jones et al. 2008, Wood et al. 2001. Therefore in these organisms denitrification is truncated and produces N 2 O, a potent greenhouse gas that is also the most important present day contributor to ozone-depletion (Ravishankara et al. 2009). The presence of truncated denitrifiers has been assumed to play a large role in determining N 2 O emissions. Other factors that are implicated in controlling N 2 O emissions include imbalances in the coupling of individual denitrification steps within single organisms, and the functioning of N 2 O reductase itself (Dendooven and Anderson 1994, Naqvi et al., 2000, Seitzinger et al. 1983, Šimek and Cooper 2002. Recently, sequencing studies have identified a subset of denitrifiers which only possess N 2 O reductase (Sanford et al., 2012) and there is growing evidence that these organisms have a strong influence on the amount of N 2 O released from terrestrial soils (Graf et al. 2014, Hallin et al., 2018, Jones et al. 2012. Theoretically, N 2 O reduction is one of the most thermodynamically favourable steps of the denitrification pathway, with a ΔG 00 of −677 kJ/mol CH 2 O compared to −352 kJ/mol CH 2 O for nitrate reduction to nitrite or − 464 kJ/mol CH 2 O for nitrite reduction to N 2 O (in typical marine sediment conditions of 10 μM NO 3 − , 1 μM NO 2 − , 7.2 nM N 2 O, 10 μM CH 2 O, pH 8.2, salinity 35 PSU). So far however, little is known about the abundance or ecological role of N 2 O reducers, especially in shallow coastal marine sediments where N 2 O dynamics are poorly constrained. Despite the fact that on average 0.01 mol of N 2 O is produced for every mol of N 2 released from marine sediments (Seitzinger et al., 2000, Usui et al., 2001, fluxes of N 2 O out of shallow coastal sediments vary widely, and in some cases are even negative (Foster and Fulweiler 2016, Murray et al., 2015, Schutte et al., 2015. Most previous studies on benthic marine N 2 O emissions have focused on cohesive sediments where N 2 O is largely consumed as it diffuses out of the sediment (Meyer et al., 2008). In the few cases where sandy sediments have been investigated the effect of enhanced transport due to porewater advection has been neglected, although this might lead to greater N 2 O fluxes to the water column. Permeable sandy sediments have high N-loss potentials compared to other marine environments and therefore any imbalances in N 2 O production and N 2 O consumption resulting from community specialization and rapidly fluctuating conditions could have a significant impact on marine N 2 O budgets. Our aim in this study was to investigate the influence of such rapid fluctuations on the occurrence of truncated forms of denitrification and the associated production and consumption of N-oxides. In order to investigate this we sampled at a sand flat in the Wadden Sea, where denitrification removes up to 60% of riverine N inputs (Gao et al., 2012) and the dynamic conditions are representative of coastal permeable sediments. We repeated measurements in different months to ensure the consistency of results, however, we did not sample regularly throughout the year as the current study did not aim to make yearly or seasonal budgets for the sand-flat. Incubations in flow through sediment cores were carried out to examine the production and consumption patterns of denitrification intermediates (NO 2 − , N 2 O). These experiments were repeated under both dynamic and steady state conditions to study how oxygen concentrations influenced denitrification dynamics. Furthermore, we investigated how advection transports N 2 O to the water column, both in laboratory incubations and in situ. To gain insights into whether community structure has an effect on the net production of denitrification intermediates, experimental work was combined with in situ data on the occurrence of functional denitrification genes within the sediment and their transcription over a tidal cycle.

Nitrification
To investigate potential nitrification rates and N 2 O production associated with nitrification, sediments were incubated with oxic water amended with 15 Marchant et al., 2014).

Denitrification under changing conditions
Rates of denitrification and the production and consumption rates of denitrification intermediates ( Net production of N 2 O and N 2 was observed throughout the incubation (Fig. 1C). When the cores were amended with 500 μmol L −1 NO 3 − , the NO 3 − was not entirely consumed by the end of the incubation. Net production of NO 2 − , N 2 O and N 2 was observed throughout the incubation (data not shown). In all three sediment cores, regardless of starting 15 NO 3 − concentration N 2 O and N 2 were produced under oxic conditions; N 2 and N 2 O production rates were calculated from time points when O 2 was still present in the porewater and from time points after O 2 had been consumed. N 2 production rates were substantially lower in the oxic phase than in the anoxic phase (1.26 AE 0.43 mmol N m −3 h −1 and 3.45 AE 1.2 mmol N m −3 h −1 respectively), however N 2 O production rates did not differ significantly between oxic (0.060 AE 0.024 mmol N m −3 h −1 ) and anoxic conditions (0.053 AE 0.023 mmol N m −3 h −1 ) (student t-test, p = >0.05, df = 16). The ratio of N 2 O:N 2 production was therefore higher in the presence of O 2 than in its absence (oxic = 4.5% and anoxic = 1.6%, student t-test p < 0.01).
In three other seasons, the ratio of N 2 O:N 2 under oxic and anoxic conditions was determined in triplicate sediment cores during time series incubations (rates of N 2 production and DNRA for the same incubations are presented in Marchant et al., 2014). Similar ratios were observed in all seasons (N 2 O:N 2 Oxic = 4.3 AE 1.4, 2.1 AE 0.3, 3.1 AE 1.7 and anoxic = 0.7 AE 0.06, 0.5 AE 0.3, 1.2 AE 0.5 SD, in November, April and July respectively).

Transport of N 2 O from the sediment to the water column
To investigate whether advection and the resulting enhancement of porewater transport leads to N 2 O emissions from the sediment, water was pumped from the bottom of a sediment core at velocities comparable to those measured in similar sediments in situ (11 and 21 cm h −1 (Billerbeck et al. 2006b, Precht andHuettel 2004)). Over time, transport of aerated, NO 3 − amended seawater through the column increased the oxygen concentration within the upper layers of the sediment column, and also led to rises in the N 2 O concentration within the sediment (Fig. S6). The N 2 O formed within the sediment was subsequently transported to the water overlying the sediment (Fig. 2). The flux of N 2 O to the overlying water increased with increasing flow velocity.
To investigate whether similar transport of N 2 O from the sediment to the water column is observed in situ we measured water column NO x (nitrate and nitrite) and N 2 O concentrations directly above an intertidal permeable sand flat in the Wadden Sea in May 2011. Water column N 2 O was on average 156% AE 3.9 SD (n = 34) over-saturated and neither NO x nor N 2 O varied significantly with depth or tidal duration. Based on the water column concentration of N 2 O, sea-air fluxes between 1.10 and 1.50 μmol m −2 h −1 were calculated using two different parameterizations of air-sea gas exchange. As such daily export of N 2 O from the back barrier region to the German Bight would be 84 mol day −1 , whereas daily air-sea exchange for the same area was in the range of 1162-2302 mol day −1 . To confirm that the source of N 2 O was within Wadden Sea sediments, we measured N 2 O concentrations at a tidal inlet, where water from the North Sea enters the backbarrier Wadden Sea region as the tide rises. Strong tidal variations in N 2 O over-saturation were observed at the tidal inlet (Fig. 2); water entering from the North Sea at high tide was around 3% over-saturated. At low tide, the N 2 O oversaturation of water leaving the back barrier region was around 10 times higher ($ 27% over-saturated).
Based on a transport model, the experimentally determined volumetric N 2 O and N 2 production rates were used to calculate areal fluxes of N 2 O and N 2 out of the sediment (Table 1). The sediment was always a source of N 2 O to the water column.

Sediment community
Phylogeny. To investigate the sediment microbial community we analysed the taxonomic distribution of all sequences in the 454 metagenomic datasets. Our results indicated that the microbial community in October and March were similar, i.e. the percentage of sequences assigned to each class all fell within 1.5 standard deviations of the mean (Supporting Information  Table S2), with one exception -the Bacillariophyta were most dominant in March, but less so in October. The Bacteroidetes and Proteobacteria were the two dominant microbial phyla present within the sediment (Supporting Information Table S2). Within the Bacteriodetes, the class Flavobacteriia was the most abundant. Within the Proteobacteria, Gammaproteobacteria were most abundant, followed by Alphaproteobacteria, Betaproteobacteria and the Delta/Epsilon divisions.

Taxonomic distribution of denitrification functional genes.
Functional genes for all steps of the denitrification pathway (napA, narG, nirS, nirK, norB, and nosZ) were detected in similar proportions in the 454 metagenomes [independent of detection approach (HMM or ROCker)] and the Illumina metagenome (Fig. 3).
Phylogenetic assignments of functional denitrification genes in the 454 datasets determined using both MEGAN and Kaiju were similar at both the phyla and class level (Supporting Information Fig. S1). Relative abundance and phylogenetic assignment of all reads using ROCker and Kaiju are shown in Fig. 3. The genes for nitrate, nitrite and nitric oxide reduction (napA, nirS, norB) were found to be most similar to known gammaproteobacterial genes. For narG, most reads were associated to the Gammaproteobacteria, Deltaproteobacteria and Alphaproteobacteria. When analysing all functional denitrification genes assigned to the Gammaproteobacteria, the distribution of genes between each step of the pathway was relatively even (Supporting Information Table S3), independent of sequencing technology, detection method and assignment method.
In comparison, the detected genes for nitrous oxide reductase (nosZ) were most similar to genes of the Bacteroidetes, in particular, the Flavobacteriia (139 of 247 total nosZ counts in the 454 metagenome and 4928 of 13471 in the Illumina metagenome). In the 454 data the average odds ratio (the number of Flavobacteriia nosZ reads divided by the number of all other Flavobacteriia denitrification reads over the number of gammaproteobacterial nosZ reads divided by the number of all other Gammaproteobacteria denitrification reads) of all four metagenomes for Flavobacteriia was 7.1 AE 1.2 when based on both total read numbers and 6.9 AE 1.0 based on coverage per base. For the Illumina dataset the odds ratio was 26 based on total read numbers and 39 based on coverage per base. Values substantially greater than one in odds ratio analysis indicate that the ratio of nosZ to other denitrifying genes was much higher in the Flavobacteriia than for the Gammaproteobacteria.
Transcription of functional genes across a tidal cycle. Metatranscriptomes were sampled directly from the sand flat at six time points over a tidal cycle. Transcripts were identified for each of the denitrification marker genes in all metatranscriptomes ( Fig. 4, narG, napA, nirK, nirS, norB, nosZ). napA had the highest relative abundance (when read counts were normalized to gene length and total number of mRNA reads in the metatranscriptome), followed by narG and norB. Counts of nirK, nirS and nosZ were lower (Supporting Information Fig. S3).
Transcripts that were phylogenetically affiliated to the Gammaproteobacteria dominated the relative counts of napA (89 AE 3%), narG (80 AE 7%) and norB (74 AE 12%) at all time points. Reads for nirS and nirK were more homogenously distributed across groups, while reads affiliated to the Bacteroidetes, specifically the Flavobacteriia, dominated the nosZ transcripts [on average 44 AE 12% of nosZ transcripts were associated to the Flavobacteriia, the highest percentage was observed at late low tide (59%)].
Taxonomic distribution of Flavobacteriia denitrification genes and transcripts. Of the total 247 validated nosZ encoding metagenomic reads, all could be aligned via blastx to a reference sequence with an e-value of < e −6 . For 184 reads the e-value was ≤ e −30 , for 104 reads the e-values were < e −49 , and for 16 reads the e-values were < e −100 . All sequences had a top five blastp hit to an annotated nosZ sequence. BLAST analysis followed by assignment using MEGAN or Kaiju revealed that the majority of the Flavobacteriia nosZ reads were related to members of the Flavobacteriaceae. While assignments were more variable at genus level, generally the most common assignments were to the genera Muricauda and Eudoraea (Supporting Information Fig. S2).
Within the metatranscriptomes, transcripts taxonomically assigned to Flavobacteriia were found within the napA, narG, nirK, norB and nosZ reads. However, at all time points, the transcripts taxonomically assigned to the Flavobacteriia were mainly nosZ rather than other denitrification pathway genes ( Fig. 4 and Supporting Information Fig. S4). Of the sequences which could be taxonomically assigned to genera level, most were assigned as either Muricauda or Eudoraea (Supporting Information Fig. S5). Of the 425 Flavobacteriia nosZ transcripts (all of which had an e-value < e −7 ), 124 had a top blastp hit to Eudoraea adriatica and 24 to Muricauda sp. MAR_2010_75. In contrast, no napA, narG, nirS or norB transcripts were assigned to the Muricauda or Eudoraea genera.

Production and consumption of denitrification intermediates indicates metabolic specialization by denitrifiers
Intertidal sandy sediments are highly dynamic environments, where oxygen and nitrate availability differ on cm to mm scales and between high tide and low tide.  Furthermore, redistribution of sand means that microbes attached to sand grains can move rapidly between conditions. Here we used a number of experimental approaches on bulk sediments which integrate the small scale variability observed in these sediments and allow the study of biogeochemical processes.
High rates of denitrification in the presence and absence of oxygen have been observed before in the permeable sediments of the Wadden Sea (Gao et al., 2010, Gao et al. 2012, Marchant et al. 2014. So far however, little is known about the balance between consumption and production of denitrification intermediates in this dynamic intertidal environment. In this study we could confirm that denitrification rates within intertidal sandy sediments from the Wadden Sea were high and were associated with the accumulation of NO 2 − and N 2 O. Nitrification rates were always at least an order of magnitude lower than denitrification, independent of season, and were not associated with detectable N 2 O accumulation. When the sediment was kept anoxic and supplied continuously with replete nitrate, there was a net production of the denitrification pathway intermediates NO 2 − and N 2 O, with net production rates around 380% and 1.4%, respectively, of the N 2 production rate. Such accumulation of denitrification intermediates may have resulted from two factors; 1) poor coupling of individual denitrification steps within single organisms (Bergaust et al., 2008, Lilja and Johnson 2016, Otte et al. 1996, Philippot et al. 2001 or; 2) process partitioning within the denitrifying community, i.e. the presence of truncated forms of Reads were identified using ROCker and taxonomically assigned using Kaiju. See Supporting Information Fig. S1 for a comparison to an HMMbased approach and taxonomic assignment using MEGAN. Coverage per base refers to the sum of all read bases aligning to a functional gene, divided by the average length of that gene. denitrification in different community members (Graf et al. 2014, Kuypers et al. 2018, Roco et al. 2016, Shapleigh 2007. The particularly large accumulation of nitrite provided a first indication that nitrate reduction might be mediated by different members of the community than the rest of the pathway in sandy sediments, similar to other environments where nitrate reduction rates are orders of magnitude higher than N 2 production rates (Ganesh et al. 2015, Kalvelage et al. 2013. Still, steady state conditions provide little insight into these factors as they reduce complex environmental conditions to their most simple form (i.e. constant anoxia or replete nitrate availability). Therefore, we investigated the causes of net NO 2 − and N 2 O production in more detail.
To imitate the fluctuations of oxygen and nitrate that occur in situ (Ahmerkamp et al., 2015, Billerbeck et al., 2006b, Marchant et al, 2017 we investigated the appearance and disappearance of denitrification products in the porewater (NO 2 − , N 2 O and N 2 ) under dynamic conditions. Incubations were started under oxic conditions, with initial 15 NO 3 − concentrations ranging from 15 to 500 μmol L −1 (Fig. 1) and over time oxygen and NO 3 − were consumed. Denitrification occurred in the presence of oxygen, but at a lower rate compared to anoxic conditions. However, there was a significantly higher ratio of N 2 O : N 2 production in comparison to anoxic conditions. This pattern of N 2 O : N 2 production was observed in triplicate sediment cores across different seasons, despite large variations in denitrification rates. Enhanced N 2 O production in the presence of oxygen can be attributed to the inefficient coupling of denitrification enzymes, due to the partial deactivation of nitrous oxide reductase by oxygen (Dendooven and Anderson 1994), or from mismatches in the expression of nitrous oxide reductase relative to nitrate and nitrite reductases (Bakken et al. 2012). We hypothesise that the lower N 2 O : N 2 ratio under anoxic conditions may also have been due to higher activity of specialized N 2 O reducers. Evidence for this is provided by both the changes in intermediate Noxide production in the transient incubations and by the distribution of functional genes observed in the molecular data (see below). When the sediment became anoxic, N 2 production rates rose and there was a net consumption of NO 2 − and N 2 O when NO 3 − and subsequently NO 2 − became limiting within the porewater (Fig. 1). In the incubations to which 500 μmol L −1 15 NO 3 − were added and NO 3 − was never fully depleted, both NO 2 − and N 2 O continued to accumulate steadily even as N 2 was produced. These results indicate truncated forms of denitrification occur within the sediment, where nitrate reduction to nitrite and incomplete denitrification by some community members results in net NO 2 − and N 2 O production in the presence of nitrate. There is then a net consumption of N 2 O when nitrate and nitrite become limiting by specialized N 2 O reducers.
Metagenomic and metatranscriptomic evidence for metabolic specialization within the denitrifying community The incidence of truncated forms of the denitrification pathway in a large proportion of the denitrification community is supported by the uneven taxonomic distribution of denitrification genes within the sediment (Fig. 3). The same patterns were detected independent of sequencing technology or analysis pipeline. Furthermore, the distribution of denitrifying genes was remarkably consistent regardless of the sampling time point (October 2009, March 2010 and April 2013), although we cannot exclude that the community differed during other seasons. For example, there was a higher read abundance and higher coverage of genes from the first three steps of the denitrification pathway (napA, narG, nirK and norB) in gammaproteobacterial groups. Similarly, reads matching the nitrate reductases (narG and napA) and nitric oxide reductase were more common in the Deltaproteobacteria than those for nitrite and nitrous oxide reductase. This suggests that within these two groups, many organisms might carry out only nitrate reduction to nitrite, or truncate their denitrification pathway before the nitrous oxide reduction step.
Recently, metagenomic evidence from uncultured Gammaproteobacteria (Woeseiaceae/JTB255-MBG) abundant at the sampling site and in other sediments worldwide has suggested that these organisms only have a truncated denitrification pathway (i.e. they can only reduce NO 2 − to N 2 O) (Mussmann et al., 2017). The short read length of our sequences and the poor representation of nirS from cultured representatives in the databases makes it difficult to assign our sequences to a phylogenetic level beyond class. However, of the sequences that were recovered, 52% of the validated nirS reads in the 454 metagenomes had a blast hit similar (up to 90%) to Woeseiaceae/ JTB255-MBG nirS sequences. Furthermore, the Woeseiaceae/JTB255-MBG nirS sequences clustered with many of the uncultured bacteria sequences which were the closest blast hits of the validated nirS sequences (Supporting Information Fig. S7). Future investigations will provide insights into the importance of this organism for denitrification within permeable sediments. Even more strikingly, more than half of the nitrous oxide reductases were assigned taxonomically to the Flavobacteriia. Moreover, more than 95% of the denitrification pathway genes assigned to the Flavobacteriia encoded nitrous oxide reductases. We analysed the phylogenetic affiliations of the 92 Flavobacteriia nosZ sequences detected in our 454 metagenomes more closely (Supporting Information Figs. S5 and S8). This revealed that the majority of Flavobacteriia nosZ genes were similar to homologous genes of either Muricauda sp. MAR_2010_75, or Eudoraea adriatica (DSM 19308). No other functional genes of the dissimilatory denitrification pathway were found for these organisms in the metagenomes or for any other organism in the same genera. There is a possibility that the Flavobacteriia within the sediment have obtained other N-oxide reductases via lateral gene transfer, making a simple assignment unreliable. However, the complete genomes which are available for Muricauda sp. MAR_2010_75 and Eudoraea adriatica also only encode nitrous oxide reductase and no other denitrification pathway genes. We confirmed that no further denitrification genes were present in these genomes during our own reanalysis (data not shown). Combined with the relatively high coverage of Flavobacteriia sequences in the metagenomes (Supporting Information Table S2), a more likely explanation for the overabundance of Flavobacteriia N 2 O reductases is that the Flavobacteriia present in the sediment only encoded nitrous oxide reductase.
The metatranscriptomic data also supports metabolic specialization by denitrifiers within the sediment. Few trends could be seen across the tidal cycle when the different genes were examined individually ( Fig. 4 and see Marchant et al., 2017). This suggests that there is no strong response at the transcriptional level to the large fluctuations that occur over a day at the sand flat. What was highly apparent was that nitrate and nitric oxide reductases were more highly transcribed than the nitrite and nitrous oxide reductases (Supporting Information Fig. S3). While transcription does not equate to activity and it seems likely that we could not identify all nitrite reductase transcripts, the general trends in the transcriptomic data agree with the rate-based evidence that there are imbalances in the production and consumption of denitrification intermediates. Furthermore, the taxonomic assignment of nitrous oxide reductase transcripts matched those of the metagenomes (Supporting Information Figs. S2 and S3), i.e. more than half of the nosZ transcripts were assigned to the Flavobacteriia, specifically to the same two genera that were abundant in the metagenomes. Of these, we could find no other denitrification pathway gene transcripts affiliated to genera Eudoraea or Muricauda. Taking into account all of the Flavobacteriia denitrification transcripts, nosZ reads were always the most transcribed gene at all points across the tidal cycle ( Fig. 4 and Supporting Information Fig. S4). Together the metagenonomic and metatranscriptomic data suggest that at the times we sampled, imbalances in the production and consumption of denitrification intermediates are driven largely by community composition and the suite of denitrifying genes that microorganisms encode. Changes in transcription in response to fluctuations in the sand flat seemed to play a more minor role.
Members of the Flavobacteriia are ubiquitous and abundant in the sandy sediments that fringe North-West Europe. Fluorescence in situ hybridization counts made along a 2000 km stretch of the coastline have shown that between 7% and 20% of the sediment community belong to the Flavobacteriia (Rizvi 2014). At the same time, between 5% and 8% of 16S amplicon sequences derived from the upper layers of the sediment in the Wadden Sea are assigned to Flavobacteriia of which the genus Eudoraea were the most abundant single group (Dyksma et al., 2016). Similarly in subtidal sandy sediments from the German Bight, Flavobacteriia were very abundant (Probandt et al., 2017). Furthermore, Muricauda sp. MAR_2010_75 was isolated from the same region. The existence of a subset of denitrifiers which exclusively reduce N 2 O to N 2 was shown recently for terrestrial soil metagenomes (Graf et al. 2014, Jones et al. 2012) and it has been hypothesized that these organisms could be important in mitigating N 2 O releases from soils. The high abundances of a putative N 2 O reducer along the North-West European coast suggest that permeable sandy sediments might enrich for organisms with this trait.

Sandy sediments -N 2 O sink or source?
A previous study suggested that the intertidal sediments of the Wadden Sea were not a source of N 2 O to the atmosphere (Kieskamp et al. 1991). However, Kieskamp et al., (1991) did not take into account the permeable, advection driven nature of sandy sediments which has been shown to lead to severe underestimations of fluxes (Cook et al. 2006, de Beer et al. 2005, Gao et al. 2012). The evidence presented here indicates that the intertidal sandy sediments of the Wadden Sea are a source of N 2 O. We demonstrated this experimentally by pumping water through a sediment core at velocities comparable to those measured in similar sediments in situ, finding that advective flow increases the flux of N 2 O from the sediment to the overlying water. Furthermore, water column measurements made in the Wadden Sea revealed higher oversaturations of N 2 O than have been observed so far in the nearby North Sea (Bange 2006, Law andOwens 1990). The significant enrichment of N 2 O in water leaving the Wadden Sea region (as the tide recedes towards the German Bight) in comparison to water entering the region (Fig. 2) further indicates that the main source of N 2 O is within the Wadden Sea. Similar tidal patterns are observed in the concentrations of NO 3 − , NO 2 − and NH 4 + in the seawater and are attributed to activity and pore water advection within the tidal flat sediment (Grunwald et al., 2010). By combining volumetric rates of N 2 O consumption and production measured in March 2011 with modelled O 2 and NO 3 − and N 2 O penetration depths we estimated that areal fluxes of N 2 O out of the sediment are in the range of 0.11-1.51 μmol N-N 2 O m 2 h −1 . This is in the same range as the air-sea fluxes determined from the in situ measurements of N 2 O concentration in the water column above the sandflat in the same month (1.10 and 1.50 μmol N-N 2 O m −2 h −1 ).

A niche for N 2 O reducing specialists
It is evident that the microbial community within the sediment denitrifies a large proportion of the NO 3 − that enters the Wadden Sea to N 2 . In addition there is a production and emission of N 2 O from the sediments, nevertheless, more than 98% of the N 2 O produced is still reduced to N 2 . The dominance of specialist N 2 O reducing Flavobacteriia in the sediment, indicated by cell counts and metagenome/metatranscriptome nosZ reads, indicates that the Flavobacteriia likely consume a substantial proportion of the N 2 O produced by other microorganisms (Fig. 5). N 2 O respiration would offer the Flavobacteriia a niche within the sediment where they have little competition with other organisms. For example, competition for NO 3 − in this environment is likely to be intense, as there is both a high proportion of nitrate reducers (as seen in the molecular data) and an abundant, nitrate-accumulating microphytobenthos community (Marchant et al., 2014, Stief et al. 2013. N 2 O respiration offers a thermodynamically favourable alternative to NO 3 − respiration, and moreover requires less investment in enzymes than expression of a full denitrification pathway, providing the Flavobacteriia with a niche for what seems to be a rather advantageous lifestyle.

Implications for N 2 O emissions
Denitrification rates consistent with those measured in this study have been observed at different times and locations across the entire Wadden Sea (Gao et al. 2012, Marchant et al. 2014. In fact, denitrification rates have been shown to be generally high in permeable sandy sediments even in subtidal regions (Canion et al. 2014, Evrard et al., 2013, Gihring et al., 2010, Rao et al., 2007, Sokoll et al., 2016 and there is evidence to suggest that N 2 O production rates are also high in subtidal sands (Marchant et al., 2016). The consistency of these studies support a recent estimate that 60% of fixed N input from rivers and atmospheric deposition worldwide is denitrified within coastal sandy sediments (equating to 54 Tg) (Gao et al., 2012). Our areal fluxes indicate that 1 and 2% of this nitrate is denitrified incompletely to N 2 O and released to the water column in the Wadden Sea. If this is the case in other regions, sandy coastal sediments may contribute substantially to N 2 O emissions from the oceans, which are currently estimated to range between 5.4 Tg N yr −1 and 11 Tg N yr −1 (Bange 2008). Our combined results indicate that N 2 O emissions from sands are tightly controlled by the network of N 2 O-producing and N 2 Oreducing microorganisms within the sediment.

Study site and sampling
Sediment sampling was carried out at the Janssand sand flat (53.73515 0 N, 007.69913 0 E), which is located in the backbarrier region of the Wadden Sea, Germany. The flat consists of well sorted silicate sand, with a mean grain size of 176 μm and a porosity of 0.35 (upper 15 cm of sediment) (Billerbeck et al., 2006a, Billerbeck et al., 2006b. The sand flat is inundated with seawater for around 10 h of each tidal cycle. When the tide is out, there is only a thin layer of water covering the sand flat.
Previous studies have shown that oxygen can penetrate more than 5 cm while the tide is in, and while the tide is out only penetrates 1-2 cm (see Marchant et al., 2017, Jansen et al 2009. Sediment was collected from the upper 5 cm of the sand flat and stored in washed plastic containers for transport before use in steady state core incubations, dynamic core incubations and microsensor experiments. Sediment for DNA analyses was immediately frozen at Fig. 5. Specialization of the denitrifying community in permeable sediments. Nitrate is advected into permeable sediments at rates orders of magnitude higher than diffusive fluxes. Nevertheless, competition for NO 3 − is intense, due to high nitrate reduction rates and uptake by diatoms. Intermediate production and release of N 2 O by full denitrifiers is caused by rapid fluctuations in oxygen and nitrate availabilities due to changes in current velocity and sediment movement. Furthermore, partial denitrifiers which do not encode nitrous oxide reductase release N 2 O into the sediment. The pool of N 2 O provides an electron acceptor to specialized Flavobacteriia, however due to the movement of porewater, much of the N 2 O is released to the water column before it is entirely consumed. −20 C, while sediment collected for RNA analyses was immersed in three volumes of RNA preservation solution (LifeGuardTM Soil Preservation Solution, MoBio), and stored at 4 C.

In situ N 2 O concentrations
Sea water sampling to determine in situ N 2 O concentrations was performed directly above the sand flat from a sampling platform on 21 March 2011. Over 12 h (beginning at low tide when the tide was rising and the flat was flooded with seawater until the flat was exposed again) samples were collected as technical triplicates from four water depths (5, 17, 47 and 80 cm above sediment surface) and averaged, leading to n = 14, 11, 6 and 3 respectively. Fewer samples were collected from 80 cm above the sediment surface as water only reached this height at three sampling points. Sampling was carried out using a lance and seawater was filled into 6 ml Exetainers (LabCo, UK) prefilled with 100 μl saturated HgCl 2 leaving no headspace. Sampling for in situ N 2 O concentrations was also carried out from the main tidal inlet between the islands of Langeoog and Spiekeroog (53.44791 0 N, 007.418310 0 E) on 31 May 2011 using the ICBM ship 'Otzum', water was sampled from 3 m depth every 15 min over a 12 h period, using a niskin bottle and filled headspace free into 6 ml Exetainers (LabCo, UK) prefilled with 100 μl saturated HgCl 2 Tidal height data was provided by Thomas Badewien from the long term time series station at 53.450100'N, 007.401630'E.

Rate determinations
A number of experiments were carried out to investigate nitrification and denitrification rates within the sediment, these were comprised of isotope labelling experiments investigating nitrification rates, denitrification rates under steady state conditions, and denitrification rates under dynamic conditions. For all experiments sediment was gently homogenized and carefully filled into sediment cores (I.D. 10.3 cm, height 10 cm) ensuring no air bubbles were trapped. The cores were equipped with two way valves at both the top and bottom to allow the exchange of porewater within the core (Marchant et al., 2014). Before measurements, cores were kept on a simulated tidal cycle consisting of repeated 6 h 'inundation' and 'exposure' periods. During 'inundation' seawater was supplied from the bottom of the core in cycles consisting of a 30 min period where water was pumped in at a rate of 2.5 ml min −1 followed by a 15 min period without pumping. (i.e. aerated seawater was pumped through for 6 h a day and no pumping occurred for 6 h).

Nitrification potential
Nitrification potential within the sediment was determined in triplicate sediment cores November 2011, April and July 2012 (parallel to denitrification measurements). The same incubation procedure was used as in Marchant et al., 2014, where the results from July 2012 are presented. Briefly, the entire porewater volume of the core was exchanged with aerated seawater which had been amended with 15 NH 4 + and 14 NO 2 − . To achieve this 1 l of the amended water was pumped through the core from the bottom using a peristaltic pump. Porewater was then sampled passively from the bottom of the core into a 6 ml Exetainer (LabCo, UK) prefilled with 100 μl HgCl 2 saturated water. On average 12 time points were then taken over the next hour while oxygen was still detectable in the porewater. O 2 , 15 NO 2 − , 15 NO 3 − and 15 N N 2 O concentrations were determined as detailed below.

Denitrification under steady state anoxic conditions
Sediment incubations were carried out under 'steady state' anoxic conditions in August 2015. Briefly, sediment collected from the sand flat was kept on a simulated tidal cycle for 3 days and supplied with filtered North Sea water collected in March 2015 containing 20 μmol L −1 NO 3 − . On the third day, at the end of an 'exposure cycle' i.e. after the sediment had not received any water supply for 6 h, the sediment was carefully packed into three small flow through columns (height 10 cm, I.D. 3.7 cm). North Sea water (collected at the time of sampling, containing < 1 μmol L −1 NO 3 ) was deoxygenated by bubbling with He and amended with 60 μmol L −1 15 NO 3 − . Subsequently, it was pumped through the core at 1 ml min −1 until 15 N N 2 concentrations at the outlet were stable for more than 3 h (confirmed by online MIMS measurements). 3 × 12 ml porewater samples were collected passively from the outlet of each of the three cores in glass syringes and filled into 12 ml exetainers prefilled with 100 μl saturated HgCl 2 . The ratios between the amount of produced NO 2 − , 15 N N 2 O, 15 N N 2 were then determined as below.

Denitrification during changing conditions
Denitrification dynamics i.e. the production and consumption of NO 3 − , NO 2 − , N 2 O and N 2 over time, were determined in sediment cores in March 2011. Three replicates cores were used, which were incubated with aerated 15 NO 3 − amended water using the method described above. The incubation was repeated on each core four times with 15, 50, 100 and 500 μM of added 15 NO 3 − respectively. After adding aerated, 15 NO 3 − labelled water to the core, on average 30 time points were collected over the next 100-150 min, during this time the core became anoxic. At each time point, porewater was sampled passively from the bottom of the core into a 3 ml Exetainer (LabCo, UK) prefilled with 100 μl HgCl 2 saturated water. The same incubations were repeated subsequently in triplicate sediment cores in November 2011, April 2012 and July 2012, except 6 ml exetainers were used over 12 time points and only N 2 and N 2 O production were determined (see Marchant et al., 2014 for N 2 production and DNRA rates from the same incubations).

Chemical and isotopic analysis
Nitrite and nitrate were determined using a chemiluminescence NO x analyser after reduction to NO (Braman and Hendrix 1989). For gas analyses, a helium headspace was added to all exetainers after which they were equilibrated for at least 24 h. N 2 O concentrations from in situ collected samples were determined on a gas chromatograph with a 63 Ni electron capture detector (Shimadzu GC-8A), calibration curves were determined from exetainers spiked with known concentrations of N 2 O and partitioning of N 2 O between the headspace and water was determined according to Weiss and Price (1980). For the denitrification rate determinations, 15 N N 2 and 15 N N 2 O production was determined by injecting gas from the headspace directly into a GC-IRMS (VG Optima, Manchester, UK). The isotope ratios of 28 N 2 , 29 N 2 and 30 N 2 and 44 N 2 O, 45 N 2 O and 46 N 2 O were determined respectively. The concentration of 29 N 2 and 30 N 2 were calculated from the excess of each relative to an air sample, as described in Holtappels et al., (2011), for N 2 O 25 μl of N 2 O was used as a standard (99.5%, Air Liquide, Germany). Rates were determined according to Nielsen (1992), using 'D tot '. For nitrification rate determinations, 15 NO 3 − was converted to 15 N N 2 according to Füssel et al., 2012, and measured as above. N 2 O and N 2 in the steady state incubations were measured on an IsoPrime 100 IRMS coupled to a modified dual trapping tracegas module (Isoprime, Manchester, UK). Areal fluxes across the sediment-water interface were determined from a transport model based on Brooks (1997a, 1997b). The model simulates advective porewater flow in sands with rippled topography as a function of permeability, bedform geometry, and bottom water velocity. For reactive solutes (O 2 , NO 3 − , N 2 O), the model estimates the spatially averaged penetration depth, depending on the initial bottom water concentration and the reaction rate of the solute, assuming zero order kinetics. The same approach was used in Marchant et al., (2016) and Ahmerkamp et al. (2017), except for a modification made to calculate the N 2 O flux out of the sediment (Supporting Information methods). For the transport terms in the model, observed ripple dimensions of 7 cm length and 2 cm height were applied. Permeability was set to 2.11 × 10 −11 m 2 (determined from the measured median grain size of 176 μm using the empirical relationship from Gangi (1985) (see Ahmerkamp et al., 2017). Typical bottom water velocities observed over a tidal cycle were used ranging from 0.15 to 0.35 m s −1 (Gao et al., 2012). For the reaction terms of the model bottom water concentrations of O 2 and NO 3 − were set to 265 μmol L −1 (air saturation at 13 C) and 32 μmol L −1 [average for the Wadden Sea, (Gao et al., 2012)], respectively. The reaction rates for O 2 respiration, denitrification and N 2 O production/consumption were applied from the measurements made under dynamic conditions in March 2011 (see above). The air-sea flux of N 2 O was calculated from in situ measurements of water column N 2 O concentrations, N 2 O solubility and average wind speed using two different parameterizations according to Wanninkhof et al. (2009) and McGillis et al. (2001) (Supporting Information methods).

Microsensor incubations
In March 2010, the effect of porewater flow on N 2 O and O 2 concentrations within the sediment and water overlying the sediment was investigated using microsensors. Porewater flow was changed in order to investigate whether this led to more N 2 O transport to the overlying water. Sediment was filled into a core covered by a 5 cm lense of seawater (I.D. 5.4, sediment height 10 cm, total height 15 cm). The column was allowed to become anoxic, after which aerated seawater (amended with 500 μM NO 3 -) was pumped from the bottom of the core at varying fluid flow velocities (11 cm h −1 and 21 cm h −1 ). O 2 and N 2 O microsensors were used to monitor concentrations at the surface of the sediment on top of the core. That is, once the fluid front had passed through the entire sediment core, repeated porewater profiles in the top 22 mm of the sediment and water column were carried out with N 2 O and O 2 microsensors, constructed and calibrated as previously described (Andersen et al., 2001, Revsbech 1989, Schreiber et al. 2009).

Metagenomics
Two types of metagenomics were carried out on different samples, the first using a 454 GS FLX Titanium platform (500 bp) (Roche Applied Science), and the second using a HiSeq2000 Illumina platform (2 × 150 bp paired end reads) (San Diego, CA, USA).
Sediment samples for 454-based metagenomics were taken in October 2009 from 0-5 cm depth, and in March 2010 from 0-2 cm depth, a detailed description of the samples is given in Marchant et al., (2017). Briefly, three samples were taken in March, two of which were pooled and divided again into two technical replicates, named MarA and MarB, the third sample (the biological replicate) is referred to as MarC. Samples Oct09 and MarA were sequenced in the same run and sample MarB was sequenced together with sample MarC. The four sequence data sets were submitted to the sequence read archive (http://trace.ncbi.nlm.nih.gov/Traces/sra/) under the Bioproject PRJNA174601 (Accession: SRP015924). The sample accession numbers are SRS365699, SRS365698, SRS365700 and SRS365701. A sample for Illumina metagenome sequencing was taken from the upper layer of the sand flat in April 2013, see Dyksma et al., (2018) for further details.

Metagenomic data analysis
The sequence data sets were pre-processed by removing identical reads via a CD-hit based algorithm Godzik 2006, Niu et al. 2010).
Taxonomy analysis -For taxonomy analysis, a reference database was generated by selecting a sequence subset of universally distributed protein orthologs from the FIGfam data base (Meyer et al. 2009) (ftp://ftp. theseed.org/FIGfams/ Release 12), see Supporting Information methods. The resulting sequence collection contained 65 001 sequences. A blastx search (E-value cutoff 1e −10 ) with the metagenomic reads against the generated protein orthologues database was performed, and the reads with a hit were assigned to the respective taxonomic class of the top hit FIGfam sequence (for more details see Supporting Information methods).

Functional gene detection
The 454 data were analysed using two different approaches to both detect and taxonomically assign functional denitrification genes within the metagenomes (narG, napA, nirK, nirS, norB and nosZ). In the first approach, Hidden Markov Models (HMM) were used to detect sequences encoding for functional genes, which were subsequently manually validated. The second approach utilized ROCker (Orellana et al., 2017), which detects sequences in short-read metagenomic data sets by modelling sliding-window bitscores. Reads encoding functional genes were taxonomically assigned using either MEGAN or Kaiju (Huson et al., 2011;Menzel et al. 2016). As these approaches gave similar results (Supporting Information Fig. S1), ROCker + Kaiju were used to analyse the Illumina dataset as the number of sequences precluded manual validation.

HMM-approach
For the detection of functional genes (narG, napA, nirK, nirS, norB and nosZ), protein-coding genes were predicted in the metagenomic sequences and subsequently translated into correct protein sequences using PRODIGAL running in anonymous mode (Hyatt et al. 2010. The HMMsearch function of HMMer software package was used to identify reads of interest in the metagenomes (Eddy 2011). A collection of selected protein sequences (17-30 sequences per collection downloaded from UniProtKB, http://www.uniprot.org/ and Fungene http://fungene.cme.msu.edu/) were used to create profile HMM's. For nirK, additional sequences were used from Decleyre et al., 2016. An automated validation procedure of the reads was carried out against a gene validation database (see Supporting Information methods for further details), followed by manual validation. Sequencing coverage per base of a functional gene was calculated as follows: From the reads validated for each gene, first, the sum was calculated of all read bases aligning to a known functional gene (database sequence). Second, of all gene sequences in the database to which a read (partially) aligned, the average gene length was calculated. To obtain gene coverage per base values, the sum of the aligned read bases was divided by the calculated average gene length. An average coverage per base was calculated for the technical replicates (MarA and MarB).
Taxonomic assignment of functional denitrification genes -To investigate the phylogenetic affiliations of the validated narG, napA, nirK, nirS, norB and nosZ a blastp search (Altschul et al., 1990) was run with default parameters against the NCBI nr database (November, 2015) using the translated and validated reads as query. Phylogenetic affiliations were assigned from the blastp search using LCA algorithm in MEGAN. A second approach utilized Kaiju (Menzel et al., 2016) using the nucleotide sequences (using the standard settings on the webserver: Reference database NCBI nr + euks, seg filter: on, runmode: greedy, minimum match length: 11, minimum match score: 75, allowed mistmatches: 5). For the generation of a phylogenetic tree, narG, napA, nirK, nirS, norB and nosZ genes for known taxa were chosen if they were among the top 10 blast hits of a read, and if the evalue was ≤0.0005. Otherwise the top hit sequence (clone/unknown taxon) was selected. The selected reference sequences for the phylogenetic trees were aligned with ClustalO and consensus trees were generated from 100 maximum likelihood trees generated using PhyML (Guindon et al., 2010, Sievers et al., 2011. The trees were visualized using the iTOL (Letunic and Bork 2016).

Identification of nirK functional genes in the 454 dataset
Of the 141 reads that passed the initial validation step for nirK identification, only 14 had a top 10 blastp hit that was similar to a sequence annotated as nirK. The remaining hits were generally hypothetical proteins, or putative copper oxidases. To gain further insight into the identity of these sequences, the sequences from the metagenome were aligned against a reference library of nirK sequences. If they aligned at Asp62 and His237 (Decleyre et al., 2016) they were validated. For those sequences which did not cover these regions of the reference sequences, then the nearest relative retrieved via a blastp search was aligned to the reference library instead. Only one further sequence was validated during this step. Four out of the 15 reads could not be assigned any taxonomic affiliation.

ROCker-approach
ROCker models were built specifically for either the 454 metagenomes (using a 500 bp average length for the in silico metagenome construction) or the Illumina metagenome (150 bp average length). Models were built for each functional gene using a collection of curated protein sequences and in the case of napA, narG and norB closely related outgroup protein sequences (downloaded from http://enve-omics.ce.gatech.edu/rocker/models and Uni-ProtKB), according to Orellana et al., 2017. Sequences were then taxonomically assigned with MEGAN and Kaiju (454 datasets) or just Kaiju (Illumina dataset) and coverage per base calculated as before.

Odds ratio analysis
For each dataset an odds ratio analysis was carried out to determine whether nosZ was distributed evenly within the Flavobacteriia when compared to all denitrification reads within the Gammaproteobacteria. Briefly, the number of Flavobacteriia nosZ reads was first divided by the number of all other denitrifying reads in the Flavobacteriia, this number was then divided by the number of Gammaproteobacteria nosZ reads over the number of all other denitrifying reads in the Gammaproteobacteria.

Tidal cycle transcriptomics
Sediment sampling for transcriptomics was carried out on March 21 2011as detailed in Marchant et al., 2017. Six metatranscriptome libraries were generated from sediment sampled at six time points over a complete tidal cycle, each of these had between 1.2 and 8 million mRNA sequence reads with a mean length of 180 bp (Supporting Information Table S1). The sample accession numbers are SRS417277, SRS417279, SRS417280, SRS417281, SRS417282 and SRS417283.

Transcriptomic data analysis
The sequence data sets were pre-processed by Casava v1.8.4, and quality trimmed using Trimmomatic (Bolger et al., 2014). SortMeRNA v2.0 (Kopylova et al., 2012) was used to filter out rRNA reads. From the resulting non-rRNA reads identical reads were removed using CDhit (Li and Godzik 2006) and translated into protein sequences using PRODIGAL (Hyatt et al., 2010). The same HMM-approach detailed above was used identify transcripts within the sequence data. Taxonomic identities of transcripts were inferred using MEGAN (Huson et al., 2011). In order to compare the distribution and abundance of Bacteroidetes functional gene transcripts over the tidal cycle, transcript abundances are calculated as read count per gene per kilobase of gene length and then normalized relative to the number of mRNA sequences in the respective metatranscriptome.
Kauji are shown for comparison from the 454 data from the HMM-based approach. For the ROCker based approach data is shown from the 454 dataset and the more deeply sequenced illumina dataset, note that the number of Gammaproteobacteria that could not be classified further is higher in the illumina data due to the shorter readlength. Fig. S1. Comparison of different sequencing technologies, methods to identify functional genes and methods to taxonomically classify reads. Panel a) and b) are the same as in Fig. in the main text. Panel c) and d) show a comparison of the 454 dataset using Hidden Markov Models to detect functional genes and either Kaiju (c) or MEGAN (d) for taxonomic classification. The more common HMM-approach required lengthy manual post validation before the final identification of 780 reads. This was not feasible for the Illumina dataset (in total 34043 reads were identified as napA, narG, nirS, nirK, norB or nosZ). Fig. S2. Comparison of 454 metagenome functional nosZ gene assignments at the genus level using MEGAN (a) or Kaiju (b), and of the Illumina reads using Kaiju. Further Flavobacterial genus could be identified in the Illumina metagenome than the 454 metagenomes, but all had a coverage per base of less than 1 and are note shown. Note that the shorter sequence length in the illumina metagenome means that less reads can be assigned at this level. Kaiju output is from the webserver with the standard settings: Reference database NCBI nr + euks, seg filter: on, runmode: greedy, minimum match length: 11, minimum match score: 75, allowed mistmatches: 5) Fig. S3. a) Relative transcript abundance of all identified denitrification functional genes over a tidal cycle, using phylogenetic assignments from MEGAN. Relative transcript abundance refers to the read count per gene per kilobase of gene length and divided by the total mRNA reads in each respective metatranscriptome. b and c) comparison of nosZ assignments using either MEGAN or Kaiju Fig. S4. Relative transcript abundance of Flavobacteria functional genes over a tidal cycle. Relative transcript abundance refers to the read count per gene per kilobase of gene length and divided by the total mRNA reads in each respective metatranscriptome. No nirS reads were detected. Fig. S5. Taxonomic distribution of nosZ gene transcripts affiliated to the Flavobacteria over a tidal cycle, using Megan (a) or Kaiju (b). Kaiju output is from the webserver with the standard settings: Reference database NCBI nr + euks, seg filter: on, runmode: greedy, minimum match length: 11, minimum match score: 75, allowed mistmatches: 5  Fig. . of the main text. The initial profile was carried out while the core was anoxic and subsequent profiles took place while aerated seawater (amended with 500 μM NO 3 -) was pumped upwards from the bottom of the core. Fluid velocity in the core was 11 cm h −1 during profile 1 and 2 and 21 cm h −1 in profile 3 and 4. Therefore each subsequent profile represents a condition in which nitrate supply and oxygen concentration in the upper core was increased. See Fig. 2 in the main Text for exact timing of sediment profiles. Fig. S7. Phylogenetic tree of nirS relatives from validated metagenome sequences. Maximum likelihood tree (100 bootstraps) constructed of nirS database sequences which were the top blast hit of validated nirS reads within the metagenomes. Highlights show the phylogeny of the reference sequences and the number of reads hitting each reference sequence in each season is shown in the boxes. Highlighted in bold are the full and partial sequences of JTB255-MBG. Fig. S8. Phylogenetic tree of nosZ relatives. Maximum likelihood tree (100 bootstraps) of nosZ database sequences which were the top blast hit of validated nosZ reads within the metagenomes. The cytochrome C oxidase gene of Paracoccus denitrificans is included as an outgroup. Highlights show the phylogeny of the reference sequences and the number of reads hitting each reference sequence in each season is shown in the boxes.