Nitrate respiration and diel migration patterns of diatoms are linked in sediments underneath a microbial mat

the of diatoms in phototrophic microbial and underlying of a We found the the and of the deep the to and for the The distribution of DNRA of the We that the for diatoms for in the of of over cation sessile denitri ﬁ ers.


Introduction
In illuminated benthic ecosystems, microphytobenthos (MPB), a general grouping of benthic microalgae, drives primary production via photosynthesis. Due to their high productivity (Kühl et al., 1994), yet inconspicuous appearance, MPB-dominated ecosystems have been coined the 'secret garden' (Cahoon, 1999). Thriving at the thin interface of the water column and the sediment, many ecological aspects of this 'secret garden' derive from the dynamics of the local light climate and mass transport of metabolic substrates and products over a wide range of spatial and temporal scales (Boudreau and Jorgensen, 2001). To understand the controls on benthic primary production and nutrient cycling, it is crucial to link metabolic versatility and microbial behaviour, such as migration, to the spatio-temporal dynamics of physicochemical parameters.
Diatoms are eukaryotic microalgae that often dominate MPB communities (Macintyre et al., 1996;Cahoon, 1999;Guarini et al., 2008;Longphuirt et al., 2009), where they exhibit metabolic flexibility and complex behavioural responses to environmental stimuli. It is well established that due to the interaction between migration of benthic diatoms and changes in the light climate and hydrodynamics, productivity varies with the tidal and diel light cycle (Serôdio and Catarino, 2000;Cartaxana et al., 2016). The ecological drivers for migration are multifaceted, ranging from the escape from grazing (Round and Palmer, 1966;Pinckney and Zingmark, 1991; to light harvesting optimisation (Cartaxana et al., 2016), and are largely shaped by habitat-specific dynamics. Conversely, the migration behaviour of diatoms interacts with a variety of parameters crucial for the activity of other microbial inhabitants of the benthic realm. For instance, the migration rhythm is linked to the assimilation and respiration of inorganic nitrogen (Koho et al., 2011), which can have large effects on the nitrogen (N)-cycle, particularly the balance between denitrification, dissimilatory nitrate/nitrite reduction to ammonium (DNRA) and nitrification (Daims et al., 2015). Intriguingly, diatoms themselves are capable of nitrate (NO 3 − ) respiration via DNRA based on intracellularly stored NO 3 −a rare capability among eukaryotic algae (Kamp et al., 2011;2015). So far, the link between the metabolism of diatoms under dark anoxic conditions and their migration is poorly understood. Instead, the temporal metabolic switch to an anaerobic lifestyle has mainly been studied as a strategy for survival upon burial (Kamp et al., 2011).
To assess the impact of diatoms on the benthic Ncycle, it is crucial to consider their migration behaviour. The typical habitats for benthic diatoms are in coastal and neritic ecosystems, such as intertidal and subtidal mud and sand flats, estuaries, lagoons and sea-grass beds (Macintyre et al., 1996;Cahoon, 1999). Diatoms are also found in lakes, salt flats and thermal springs (Eldridge and Greene, 1994). In several of these environments, MPB communities form laminated microbial assemblagesmicrobial mats. These ecosystems portray highly dynamic microenvironmental conditions, which are primarily shaped by the interaction between local process rates and diffusional mass transfer. As their study is not complicated by advective mass transport dynamics mats represent an ideal natural laboratory to investigate the metabolic performance of diatoms during diel cycles.
Diatom-inhabited photosynthetic mats thrive under a low-O 2 water column in the Middle Island Sinkhole in Lake Huron, Michigan, USA (Ruberg et al., 2005). Highsalinity, anoxic, cold and sulphate-rich groundwater emerges from an underwater source at 24 m depth and cascades into the sinkhole area, creating a chemocline at the base of the Lake Huron water column. Due to the rare co-occurrence of illumination and low-O 2 bottom water, a unique benthic ecosystem flourishes. Thin cyanobacterial mats cover an area of 10,000 m 2 of sediment, enriched in organic carbon originating from Lake Huron's pelagic productivity (Voorhies et al., 2012;Nold et al., 2013). Despite the extremely thin photic zone (<1 mm), the mat is functionally structured according to the light and redox gradients, comparable to other microbial mats (Stal et al., 1985;Gemerden, 1993;Stal and Caumette, 2013). The biogeochemical element cycling is shaped by the interaction of three main functional groups: motile oxygenic phototrophs, namely filamentous cyanobacteria and pennate diatoms, filamentous sulphide oxidizing bacteria, and sulphate reducing bacteria (Biddanda et al., 2006;2015;Voorhies et al., 2012).
In situ observations (personal communication with divers, NOAA Thunder Bay National Marine Sanctuary) suggest that the diel dynamics of photosynthesis in the Middle Island Sinkhole mats can differ substantially from other mats. In some areas of the sinkhole, the mat apperance changes, as expected, from white at night to purple in the night to purple in the morning (Biddanda and Weinke, 2020), while in other areas, this migraton-driven structural shift is delayed to the afternoon. A similar migration lag has been observed in the Frasassi sulphidic springs, where, the cyanobacteria exclusively perform anoxygenic photosynthesis in the morning, and only transition to predominantly oxygenic photosynthesis in the afternoon (Klatt et al., 2020). This delay arises from competition of large sulphur oxidizing bacteria and cyanobacteria for the uppermost position in the mat and results in anoxic conditions in the photic zone for the majority of the illuminated fraction of the day (Klatt et al., 2016). Given that the diatoms are obligate oxygenic phototrophs but do not produce oxygen in this extended exposure time to anoxia, the question arises, how do they compete with the other phototrophs in such environments? We hypothesized that DNRA is the main respiratory pathway of diatoms in the Middle Island Sinkhole mats and that it accounts for a substantial fraction of diel metabolic activity. We used stable isotope incubations and light microscopy to determine how DNRA is linked to diel light dynamics and diatom migration behaviour. As N-loss from ecosystems and therefore eutrophication are largely determined by the balance between denitrification and DNRA (Kuypers et al., 2018), we determined contributions of both NO 3 − respiratory pathways and discuss the importance of intracellular storage and migration by diatoms for the benthic N-cycle.

Results and discussion
Benthos structure and diel variability The microbial mat sampled from the Middle Island Sinkhole was less than 1 mm thick, coloured purple, white and/or brown and covered a fine-grained dark brown sediment. The coherent mat was not firmly attached to the fluid, soft, and organic-rich sediment; it was fragile and could only be sampled with a pipette, spoon or by careful core slicing. Due to microbial migration the colour of the mat surface changed during the day: the mat was white at night and until the early afternoon before transitioning to a purple-brown appearance. The dark brown colour of the sediment did not change during the diel cycle.
DNRA is the main NO 3 − reduction pathway in the benthos The main pathway of NO 3 − reduction in the microbial mat and underlying sediment was determined using a 15 NO 3 − tracer approach in anoxic batch incubations in gastight glass vials containing sediment, mat or mat + sediment.
Most of the 15 NO 3 − could be recovered in the 15 Nammonium ( 15 NH 4 + ) pool and only a small amount was reduced to N 2 , indicating that DNRA was responsible for 98%-99% of the NO 3 − consumption in all of the incubations (Fig. 1). These results show that DNRA, rather than denitrification, was the main NO 3 − reduction pathway in both the mat and sediment. However, dynamics of denitrification and DNRA differed. The production of N 2 by denitrification increased linearly throughout the 24 h time series in all of the incubations. In contrast, the NH 4 + production rate by DNRA decreased after 2-5 h in the mat + sediment incubation (which had highest initial rates of DNRA (Table S1)). This decrease in DNRA occurred when a substantial proportion of the added 15 NO 3 − (30 μM) had been consumed. This suggests that the process of DNRA was limited by NO 3 − availability after 2-5 h while denitrification was not limited by NO 3 − throughout the incubation, even when concentrations became low. Therefore, the denitrifying community must have a lower K m value and thus a higher affinity for NO 3 − compared to the DNRA community, as also observed in other habitats (Behrendt et al., 2014). At the same time, the v max was substantially lower, as evident from the constant rate of denitrification throughout the experiment, which suggests saturation with respect to NO 3 − concentration. Also, this rate was two orders of magnitude lower than the initial areal rates of DNRA before substrate limitation (Table S1). High affinity does therefore not seem to provide sufficient selective advantage in this ecosystem to yield a larger population size and/or cell specific activity than the DNRA performing community. Interestingly, both denitrification and DNRA were lowest in the mat incubation without sedimentthe only layer of the benthos that has diffusion-driven access to NO 3 − from the water column or possibly from nitrification under environmental conditions. Sediment with and without mat had the highest potential for NO 3 − respiration, which seems to contradict the absence of NO 3 − in the porewater under environmental conditions ( Fig. 2A). This could indicate that the experimental conditions artificially stimulated DNRA and are not relevant in the ecosystem. If this were the case, we would have expected a lag period before DNRA commenced. Rather, the immediate and high rates of DNRA (particularly in comparison to the denitrification rates), suggest that microorganisms were − to the water column overlying microbial mat on top of sediment, mat only and sediment slurry only. Note the difference in y-axis scale in (A) and (B). For illustration of the substantial differences in conversion rates we added data from panel (B) to panel (A) (grey asterisk). Lines were obtained by linear regression to estimate rates of denitrification and initial rates of DNRA (Table S1). already actively expressing the enzymes to carry out DNRA despite the apparent lack of NO 3 − . Therefore, NO 3 − is likely available in the sediment under environmental conditions. This is further supported by the observation that a combination of mat and sediment yielded highest rates of DNRA, even though the mat would be expected to hinder NO 3 − supply to the sediment from the water column. NO 3 − must thus be transferred to depth by other mechanisms than diffusion.
15 NO 3 − is actively transported to deeper sediment layers To investigate the depth distribution of DNRA under conditions that reflected those in the mats and sediments in situ, we added 15 NO 3 − to the water column overlying intact sub-cores with mat-covered sediment and followed 15 NH 4 + production over time and depth. DNRA occurred throughout day and night in the mat and sediment ( , where x is the mean diffusion distance of 15 NO 3 − , during diffusion time t and where D is the NO 3 − diffusion coefficient (1.73 10 −5 cm 2 s −1 at 21 C in water (Li and Gregory, 1974)). This means that the 15 NO 3 − must have been transported actively to deeper sediment layers. Typical candidates for rapid NO 3 − storage and transport to the deep are large sulphur oxidisers, such as Beggiatoaceae (Schutte et al., 2018). However, the local Beggiatoaceae are freshwater non-vacuolated species (Sharrar et al., 2017) and therefore not likely to store NO 3 − . We finally excluded them as candidates for NO 3 − transport because intact Beggiatoaceae filaments were not observed in the deeper sediment by microscopy. The microscopic observations revealed the presence of motile diatoms, consistent with highly abundant chloroplast 16S rRNA genes in mat samples throughout the years (Fig. S2), and therefore we focussed on the diatoms. Chloroplast genes belonging to the Bacillariophyta were the most widely observed of identifiable 16S rRNA genes, contributing up to 46% of the reads in each mat sample. Although eukaryotic chloroplast 16S rRNA genes are not representative of eukaryotic abundancedue to the nonlinear relationships between chloroplast 16S rRNA gene copy number, organism, and growth (Green et al., 2011) this observation suggests that diatoms are important members of the mat community.
Using light-and scanning electron microscopy we identified that Craticula cuspidata (Kützing) D.G. Mann, 1990 were the dominant epipelic diatoms in the microbial mat and sediment (Fig. S3). C. cuspidata thrives in brackish, as well as eutrophic, environments and can be differentiated very well within the Craticula genus due to their size (115-120 × 27-29 μm) and characteristic outline (Fig. S3). As the potential for NO 3 − storage of C. cuspidata has not been tested previously, we separated the cells from the sediment, measured intracellular NO 3 − concentration, and found that they stored NO 3 − at 83 ± 25 fmol cell −1 . Assuming a cylindrical diatom volume of 37.7 pl, the intracellular concentration would be 2.2 mM and hence, concentrations within the vacuole must be even higher. We also confirmed that these diatoms instantaneously accumulate 15 NO 3 − by exchange with their internal reservoir (Fig. S4). This process differs from active NO 3 − uptake and can be explained by − was added to the water column above intact mat covered sediment cores in the morning (asterisks) and evening before darkening (black circles). In panels (A-D), symbols differentiate light and dark incubations while in panel (E) all depths from A-D are shown on the same scale with no differentiation of light versus dark. transmembrane counterflow. Briefly, the immediate exchange of the stored 14 N to 15 N in the cells is due to the energetically neutral exchange in steady state (Xie, 2008). Thus, C. cuspidata are able to accumulate 15 NO 3 − instantaneously, given that they already have an internal NO 3 − pool. After this tracer uptake, the diatoms would have had to migrate at least down to 2-3 cm depth within 3 h to explain the production of 15 NH 4 + , which is in line with a migration speed of up to 1.7 cm h −1 as observed in mudflat sediment (Hay et al., 1993). Taken together, the data imply that diatoms with an internal NO 3 − reservoir resided at the surface during 15 NO 3 − addition and accumulated the tracer instantaneously by transmembrane counterflow before transporting it into anoxic sediment layers, whereupon it was respired to NH 4 + .
NO 3 − reduction rates and C. cuspidata migrate over a diel cycle DNRA rates measured in the 'anoxic batch incubations' were highly variable (Fig. 1). To determine if NO 3 − was only supplied to the mat and sediment by diffusion or actively transported into the sediment, the distribution of rates over depth within the sediment was determined in intact, mat-covered sediment cores after addition of 15 NO 3 − to the water column. These incubations on showed that rates in deeper sediment layers were higher when 15 NO 3 − was added in the evening (18:00 h) (Fig. 3). These results suggest a link between the abundance of C. cuspidata at the surface and the occurrence of DNRA, likely because light and/or a circadian migration rhythm shape the NO 3 − uptake at the mat-waterinterphase.
To uncouple DNRA rate measurements from the timing of NO 3 − uptake, we injected 15 NO 3 − through intact subcores over a 24 h period and measured the NH 4 + production rates over depth. While DNRA always occurred, its magnitude varied depending on time of day (Fig. 4A, Table S2). A clear rate maximum moved vertically through the sediment over a diel cycle. In the late afternoon, this NH 4 + production peak detached from the surface and reached its maximum depth of 3-4 cm at 0:00 h. In the early morning, the rate maximum shifted upwards again, reaching 0.5-1 cm before noon (Fig. 4A).
In the late afternoon, NH 4 + production rates were minimal in both the sediment and the mat at all depths. This DNRA minimum coincided with a colour change from white to brown at the surface of the mat. Overall, DNRA rates in the sediment were closely related to the migration pattern of the diatoms (compare Fig. 4A and Fig. 4B). We cannot exclude that a bacterial DNRAperforming community in the deep sediment is fuelled by NO 3 − released from C. cuspidata on their journey.
However, it is well established for other genera of diatoms that they are capable of DNRA (Kamp et al., 2011;. Further, we consider it unlikely that C. cuspidata do not respire during migration through anoxic sediment, and there is no evidence that diatoms use electron acceptors other than O 2 and NO 3 − . Thus, the correlation between DNRA maxima and viable diatom abundance strongly suggests that DNRA is predominantly performed by C. cuspidata. Only 40% of the diatom population migrated deeply into the sediment and tracked the DNRA rate maximum (Fig. 4B, Table S3). The remaining 60% of the population stayed within the uppermost 5 mm, although not directly at the surface. DNRA rates in this uppermost section were highest immediately after the onset of darkness, then decreased and remained highly variable. As anoxic conditions persist in the lower boundary of this layer even in the presence of oxygenic photosynthesis (Fig. 2C), the near-surface residing diatom subpopulation might contribute to the occasionally high DNRA rates. We cannot exclude that a sessile DNRA performing community was driving these rates in the uppermost section. Nevertheless, the correlation of the deep migration and DNRA pattern, together with the observation that a large subpopulation resides at the surface, indicates that DNRA both in the sediment and mat can be explained by diatom activity.

Migration is the key to success
Motile benthic diatoms often undergo vertical migration according to diurnal and tidal cycles in intertidal sediments to escape grazing or hydrodynamic stress (Round and Palmer, 1966;Pinckney and Zingmark, 1991;. Migration patterns are thus usually regulated by the tidal rhythm and even remain active for 3-11 days (Palmer and Round, 1967) after transfer of diatoms to a stagnant environment. In Lake Huron, changes in water level are not dependent on tides, and internal waves do not occur with a fixed rhythm that could explain the C. cuspidata migration pattern. Changes in the light regime, and thus the diel light cycle, can also trigger downward migration of diatoms, yet light in the Middle Island Sinkhole mats only penetrates down to 750 μm (Fig. 2B). C. cuspidata might thus use phototaxis during migration initiation and for micro-cycling in the uppermost 5 mm but it cannot explain behaviour in the deep. Similarly, encountering anoxia is has been shown to stop downwards migration and initiate return of some diatom species to the surface (Kingston, 1999), while C. cuspidata migrate through sulfidic and anoxic sediment. Phosphate and silicate gradients can also guide deep migration (Karen et al., 2019) and gravitaxis (Frankenbach et al., 2014) might also aid this navigation.
However, these do not appear to be the stimuli that trigger descent and ascent of the DNRA-performing C. cuspidata in the Middle Island Sinkhole.
The most straightforward explanation for the unusual migration behaviour of C. cuspidata would be a regulation mechanism based on NO 3 − storage capacity. For marine Beggiatoaceae, the electric potential over the vacuolar membrane changes with the internal NO 3 − concentration (Mussmann et al., 2007;Beutler et al., 2012). It seems likely that such potential change also occurs in diatoms during depletion of NO 3 − , 'signalling' C. cuspidata to resurface to replenish NO 3 − from the water column and thus triggering upward migration. As a single C. cuspidata cell stores 83 ± 25 fmol NO 3 − intracellularly, this internally stored NO 3 − would last for 12.6 h, calculated based on an average production rate of 6.6 fmol NH 4 + cell −1 h −1 (Tables S1 and S2). This capacity is strikingly close to the duration of one migration cycle (18-20 h), suggesting that the amount of NO 3 − in the vacuole could indeed play a regulatory role in migration behaviour and separation of diatom populations. Furthermore, DNRA rates associated with the subpopulation that stays near the surface (Figs 4 and 5) decreased after only a few hours of darkness, suggesting that NO 3 − became limiting. In contrast, the deep migrating population continuously performs DNRA. This indicates that deep migration is only initiated when intracellular NO 3 − concentrations are sufficiently high to guarantee energy supply during the 18 h of deep migration. While separation into subpopulations has previously been linked to reproduction cycles, with cells migrating to depth for cell division (Saburova and Polikarpov, 2003), vacuolar NO 3 − content thus offers an alternative explanation for initialisation of deep migration and separation of populations. While the physical and chemical stimuli guiding C. cuspidata through the dark sediments are at the moment unresolved, it is clear that migration and energy investment into NO 3 − storage for anaerobic respiration is advantageous to the diatoms. C. cuspidata does not reside at the mat surface for most of the day; it performs anaerobic respiration in the deep anoxic sediment instead. This appears paradoxical, as these typical oxygenic phototrophs harvest light for only 5 h. Even the near-surface residing subpopulation remaining in the uppermost 5 mm but not at the mat surfacedoes not have access to light for most of the day as light penetrates only down to <1 mm. This behaviour contradicts any energetic consideration and can only be explained in the context of interaction with other microorganisms and the environment. Thus, migration coupled to DNRA and the resultant circumvention of an interaction with cyanobacteria, must represent an important competitive advantage. The long absence from the surface implies that diatoms are only directly competing with the cyanobacteria during their short phase of oxygenic photosynthesis. Competition is likely based on light access, as the spectral range of light reaching the mat surface is very narrow (400-600 nm, Fig. S1). Both phototrophs possess chlorophyll a which has absorption maxima at 465 nm and 665 nm (Stomp et al., 2007). The local cyanobacteria are additionally equipped with phycocyanin, phycoerythrin and phycoerythrocyanin (Voorhies et al., 2012) to absorb light at between 500-600 nm, whereas diatoms possess fucoxanthin, also providing absorbance in this spectral range (Stomp et al., 2007;Kuczynska et al., 2015). Thus, the spectral niche of diatoms and cyanobacteria overlap. Their ecological niches are, however, separated for the restthe majorityof the day. The cyanobacteria perform anoxygenic photosynthesis in the morning (Klatt et al., 2020) and might perform fermentation, aerobic or anaerobic S 0 -based respiration at night (Stal and Moezelaar, 1997). Diatoms meanwhile switch to DNRA, which implies niche separation and thus facilitation of coexistence (Gause et al., 1936).
In addition to internal competition within the mats, avoidance of predation, e.g. by occasionally observed gastropods, might be an important factor selecting for migration (Consalvey et al., 2004). Furthermore, the flow velocity of ground water can vary substantially over days to seasons, which can lead to irregularly occurring mat disruption events, leaving uncovered sediment behind (pers. comm. with divers who sampled, NOAA Thunder Bay National Marine Sanctuary). Diatoms hidden in the deep sediment would likely be the first colonizers after such events. This is consistent with observations after mat subsampling in the lab: C. cuspidata were the first organism to repopulate the sediment surface (within minutes to hours), while the filamentous cyanobacteria had to migrate horizontally from the peripheral, intact mat. Hence, the ability to migrate overall enhances the chance of survival during disturbance events such as grazing or mat disruption, and allows rapid recolonisation.

Implications for the nitrogen cycle
The controlling factors that determine whether DNRA or denitrification is dominant in a habitat are a subject of debate. In the past studies have focussed on the impact of kinetics (King and Nedwell, 1985;Tiedje, 1988;Behrendt et al., 2014;Kraft et al., 2014;McTigue et al., 2016;Murphy et al., 2020), inhibition and/or thermodynamics (Dong et al., 2011;Kraft et al., 2014;Bonaglia et al., 2017;Dolfing and Hubert, 2017 (Consalvey et al., 2004). Craticula cuspidata perform photosynthesises (PS) in the afternoon and accumulate NO 3 − (turquoise ellipses = stored NO 3 − ) from the water column. In the early evening deep vertical migration of a subpopulation is initiated (deep migration = white dotted line), and the intracellular NO 3 − reservoir is respired via DNRA. This subpopulation only returns to the surface in the early afternoon of the next day, which coincides with a change of mat structure: The white Beggiatoaceae-dominated surface layer disappears, and purple cyanobacteria and diatoms are visible. DNRA rates are negligible at this time of day because the diatoms are at the mat surface performing oxygenic photosynthesis. The other C. cuspidata subpopulation resides within the first 5 mm of the mat and sediment for the complete diel cycle and shows a micro-cycling behaviour (Micro-cycling = grey dashed line) performing oxygenic photosynthesis or DNRA, depending on the time of day and position in the mat. respiration may be favourable for other microorganisms, potentially equipped with the denitrification pathway instead, as observed in non-illuminated sediments, such as off the coast of California (Prokopenko et al., 2011) or in the deep sea (Schutte et al., 2018).
In the Middle Island Sinkhole, DNRA is daily routine for diatoms and is intimately linked to their migration to deeper sediments. C. cuspidata usually inhabit finegrained freshwater sediments with elevated electrolyte content but are cosmopolitan and might therefore show similar behaviour in many other ecosystems. The potential for NO 3 − storage and respiration via DNRA, and migration, is not limited to specific illuminated freshwater systems but is spread across phyla and ecosystems Kamp et al., 2015;Schutte et al., 2018). For instance, in intertidal sediment conditions similar to the Middle Island Sinkholeillumination and regular anoxiaare met and 'secret gardens' (Cahoon, 1999) flourish. Also, eukaryotic NO 3 − storage has been confirmed and migratory organisms are abundant . This raises the question if DNRA in redox-dynamic ecosystems has been underestimated, as most studies on DNRA do not state specifically that they took diel or tidal rhythms into account. For instance, patchy DNRA rates at the same sampling site could be explained by the lack of diatoms in the sample, destruction of gradients guiding migration or by sampling during times when the diatoms have not accumulated NO 3 − (Sundbäck et al., 2000;Risgaard-Petersen, 2003;Giblin et al., 2013;Decleyre et al., 2015). The role of DNRA in the biogeochemical Ncycle can therefore only be assessed by considering the importance of migrating organism and the corresponding adjustment of sampling timing. C. cuspidata is a key player in the 'secret garden' (Cahoon, 1999) studied here because it can flexibly transition between energy sources, namely between using light during a short fraction of the day and chemical energy for most of their daily life (Fig. 5). This flexibility coupled their ability to store the electron-acceptor NO 3 − enables C. cuspidata to migrate into anoxic sediments, removing it from direct competition to other microorganisms and allowing it to escape predation and hydrodynamic stress. This ecological strategy makes the diatoms one of the most abundant taxa in the Middle Island Sinkhole mat as well as underlying sediment, and fundamentally shapes the benthic N-cycle by retaining N in the system.

Experimental procedures
Sampling Sediment cores (25 cm diameter, 20-25 cm sediment sampling depth) were taken by SCUBA divers at approximately 24 m depth in the Middle Island Sinkhole (45 11.941 N, 83 19.671 W). Sampling took place in 2012-2017 for 16S sequencing; and in May 2017, June 2018, and July 2019 for activity measurements. Mat sampling sites were chosen based on brownish appearance of the mat surface, indicative of diatom presence. Bottom water was retrieved using 30 m of oxygen impermeable tubing (Masterflex, Merck) and a peristaltic pump (Masterflex) placed on deck. Sediment cores and water samples were transported cooled and in the dark to the laboratory in Ann Arbor, Michigan. In the laboratory, the cores were kept at in situ temperature (12 C) in a thermostated water bath. The mats were illuminated at 90 μmol photons m −2 s −1 by LED light sources, which is in the higher range of values measured in situ in proximity of the mat surface (Fig. S1a). To mimic environmental light quality, optical filter foil (cut-off λ ≈ 600 nm; Fig. S1b and c) was used. Cores were kept on a 12 h:12 h light: dark cycle to approximate the natural light cycle.

16S sequencing and analysis
Samples for 16S sequencing were collected by carefully removing the mat layer from freshly retrieved cores and immediate flash-freezing on-board. DNA extractions were performed using the MPBio Fast DNA Spin Kit for Soil (MP Biomedical, USA) with modifications (see Supplementary Information). DNA samples were submitted to the University of Michigan Host Microbiome Core for Illumina library preparation and sequencing (Seekatz et al., 2015;Kozich et al., 2013) (see Supplementary  Information).

O 2 microprofiles
O 2 microsensors for in situ profiling were built and calibrated as described by (Revsbech, 1989). Depth profiles were acquired in July 2016 using a microprofiler as previously described (Wenzhöfer et al., 2000;De Beer et al., 2006).

In situ hyperspectral light profiles
Seven in situ spectral profiles were collected in the sinkhole and in open water in 2015-2016. Profiles were collected using a Sea-Bird HyperPro II profiler equipped with up-and down-facing HyperOCR radiometers measuring wavelengths 348-801 nm (bin size = 3.3 nm), plus an identical fixed surface radiometer to record sky conditions (Bosse et al., 2019). The profiler was deployed on the sunny side of the vessel and allowed to free-fall through the water column to avoid the vessel shadow. A duplicate cast was collected as soon as the profiler was returned to the surface to capture identical light conditions. Data were processed using ProSoft (Sea-Bird) proprietary software.

Ex-situ light micro-profiles
Light penetration depth was determined ex-situ in a matcovered sediment core using a scalar irradiance probe with a 100 μm tip (Zenzor, Denmark) on a motorized vertical positioner, after (Kühl and Jørgensen, 1992). The light sensor was connected to a USB400-FL Spectrometer (Ocean Optics, USA). A halogen light source (Schott, KL 1500) was used for illumination of the mat during measurements.

Porewater
Porewater was collected from freshly collected sediment cores on-board in 0.5-1 cm-intervals down to 5 cm using 2 cm long Rhizons (Rhizosphere Research Products, Netherlands) in July 2019. Per depth, only 500 μl were extracted including volume of the Rhizon and tubing to minimize smearing effects between depths. Porewater was flash frozen on board, immediately after sample collection. NO x concentration ( P NO 2 − , NO 3 − ) was subsequently determined using a chemiluminescence detector after reduction to NO with 90 C acidic Vanadium(III)chloride (Braman and Hendrix, 1989). NH 4 + in the porewater samples was determined colorimetrically according to (Holmes et al., 1999).

Stable isotope incubations
To determine the contribution of denitrification and DNRA to NO 3 − reduction rates, a Na 15 NO 3 − (99% CP, CAS:31432-45-8, Sigma-Aldrich) tracer method was used. Batch incubations were performed of (i) a mixture of mat and sediment, (ii) sediment and (iii) only microbial mat on agarose. First, sediment was homogenized and 6 ml were transferred into 12 ml gas-tight glass vials with a septum cap (Exetainer, Labco). The vials were filled with He-degassed, filtered bottom water (0.45 μm PES membrane) without headspace. After settling of sediment, we placed pieces of mat on the sediment surface, and briefly purged the water column with He again to ensure that oxygen in the vials was exclusively introduced by photosynthetic production. A subset of the vials was left with bare sediment. An additional set of exetainers was filled with 6 ml agarose (1.5%) instead of sediment. Filtered bottom water was added and purged with He for 30 min every few hours over several days until the agarose was anoxic down to at least 2.5 cm, as confirmed with O 2 microsensor measurements. Subsequently, mat pieces were placed on the agarose, following the same procedure as for the sediment incubations. The agarose was used to ensure comparable water column volume and distance to the light source among the different treatments. The vials were exposed for 1-2 h to darkness to deplete remaining NO 3 − as well as residual O 2 in the water column. The absence of oxygen was confirmed in three of the exetainers dedicated to microsensor measurements. 15 NO 3 − was injected into the water column from a He-purged stock solution using gas-tight glass syringes (Hamilton, Australia). To achieve approximately homogenous distribution of the label in the water column, injections were done by holding the plunger in the fixed position and slowly pulling up the syringe, starting at the mat or sediment surface. Incubations were stopped in 1.5-4 h intervals over 24 h by injecting 50 μl saturated HgCl 2 . To prevent precipitation of Hg 2+ with sedimentary sulphide, we simultaneously added 200 μl 10% ZnCl 2 through the septum. Injections were followed by rigorous mixing. The isotopic composition of N 2 and NH 4 + at each time point was determined after replacing 2 ml of the water in the incubation vial with a helium headspace and transferring the subsampled 2 ml into a 6 ml glass vials prefilled with 2 ml 0.35% NaCl solution.
Gas from the headspace of the incubation vial was injected directly into a GC-IRMS (isoprime precisION, elementar, UK) and the isotope ratios of 28 N 2 : 29 N 2 and 28 N 2 : 30 N 2 were determined (Holtappels et al., 2011). The concentration of 29 N 2 and 30 N 2 were subsequently calculated from the excess of each relative to an air sample. 15 NH 4 + concentrations were determined in the same way from the subsampled liquid after rigorous purging with He and oxidation of NH 4 + with hypobromite to N 2 . 15 NH 4 + and 15 N 2 production rates were calculated based on the increase of 15 NH 4 + and 29 N 2 + 30 N 2 concentration over time, and subsequently converted into areal rates across the sediment/mat surface based on the area exposed to the tracer (Warembourg, 1993;Preisler et al., 2007;Marchant et al., 2016). To distinguish between NO 3 − diffusion or active transport into deeper sediment, the DNRA rates over depth within the sediment was determined in intact, matcovered sediment cores after addition of 15 NO 3 − to the water column. First, 3 large sampler cores (25 cm diameter) were subsampled using smaller cores (7 mm diameter) that were pushed 6 cm deep into the sediment. These sub-cores were darkened with black tape up to the mat surface to ensure illumination only from above. Using a gas-tight syringe (Hamilton, Australia) 15 NO 3 − was added to the water column of each sub-core at either 6.00 h or 18.00 h, reaching a final concentration of 200 μM 15 NO 3 − in the bottom water. Subsets of the cores were sectioned in 1 cm depth intervals every 3 h. The slices were immediately submerged in 1 ml ultrapure water spiked with 50 μl saturated HgCl 2 and 200 μl 10% ZnCl 2 solutions to stop the incubation. 15 NH 4 + concentration in these slurries was determined by transferring the supernatant into septum vials and following the conversion procedure described above.
In a third stable isotope experiment, we assessed the temporal and spatial dynamics of DNRA rates dependent on time of day by injecting 15 NO 3 − into mat and sediment of the 7 mm sub-cores. To ensure a vertically homogenous distribution of the tracer, the injection was done by slowly pushing up a syringe filled with 15 NO 3 − stock solution while keeping the plunger in a fixed position. Core slicing, sample preparation and mass spectrometry were conducted as described above.

Depth profiles of diatom abundance determined by microscopy
The depth distribution of motile diatom cells over a lightdark cycle (12:12 h) was determined by sectioning subcores (see above) in regular time intervals, separation of viable diatom cells from sediment and cell counting. Specifically, sediment slices were transferred into a conical centrifuge tube filled with filtered in situ water. The tube was covered with black tape except for an uncovered area (3 mm 2 ) on the side of the tube, which was illuminated by a Schott lamp (300 μmol photons m 2 s −1 ). We harvested the cells that had migrated towards the light and accumulated on the side of the tube, and suspended them in 1 ml filtered in situ water. Diatoms were counted using a light microscope and a counting chamber (Neubauer improved, Germany).

Scanning electron microscopy
Subsamples of the diatoms harvested based on their phototaxis (see above) were conserved by flash freezing for Scanning Electron Microscopy (SEM) imaging. Samples were prepared on small chips of Si-and GaAs-wafer material. To preserve the surface structure of all cells in the sample the material was dehydrated using critical point drying before SEM imaging, after water removal by an ethanol series with increasing ethanol concentrations (30%, 50%, 70%, 80% and 96%). The ethanol in the sample was removed by critical point drying (Leica EM CPD300 Wetzlar, Germany). Secondary electron (SE) micrograph images were taken using a Quanta 250 FEG (FEI, Eindhoven, The Netherlands) with an accelerating voltage of 2 and 10 kV.

NO 3 − storage and tracer uptake by diatoms
To quantify intracellular NO 3 − storage capacity, diatoms were harvested from the top of the mat by gentle elution with water from a Pasteur pipette and collection of water column with the suspended cells. We then enriched the viable cells as described above based on their phototaxis and suspended them in filtered in situ water. NO 3 − concentration in the in situ water with and without diatoms was measured using a chemiluminescence detector as described in the 'porewater' section. Assuming that the diatoms lysed, releasing all intracellular NO 3 − when injected into the 90 C acidic Vanadium(III)chloride solution, we calculated the concentration of stored NO 3 − based on the difference to the cell-free control. A subsample of the injected diatom suspension was used for cell counting. Internally stored NO 3 − per diatom and intracellular concentration were estimated by assuming a cylindrical diatom shape.
To determine the rate of exchange of stored unlabelled NO 3 − with externally supplied 15 NO 3 − , diatoms pre-fed for 24 h with 14 NO 3 − , were transferred to 15 NO 3 − -amended in situ water. Specifically, after a washing step in filtered in situ water without NO 3 − , aliquots of 1 ml cell suspension were added to 10 ml of filtered, He-degassed bottom water amended with 20 μM 15 NO 3 − and incubated in the dark.
The incubation was stopped after 0, 10, 20 and 30 min by filtering the tube contents. The filter was washed with 10 ml deionized water and immediately frozen in liquid N 2 .
To release internally stored NO 3 − , the filters were submerged in ultrapure water and exposed to thaw-freeze cycles . 15 NO 3 − concentration in these cracked diatom samples was measured after 15 NO 2 − removal using sulfamic acid. Spongy cadmium was applied to reduce 15 NO 3 − to 15 NO 2 − . To reduce 15 NO 2 − to N 2 an additional sulfamic acid treatment was conducted (Füssel et al., 2012). The isotope ratio in the resultant N 2 pool was measured from the headspace using GC-IRMS as described above (Marchant et al., 2016).