The study was conducted in a purpose-built bio-secure heated conventional RAS described in Robinson et al. (2015). The experiment was conducted over a 15-day period from 30 January (day -1) to 14 February (day 14) 2014 using juvenile sea cucumbers (Holothuria scabra) imported from a commercial hatchery (Research Institute for Aquaculture III, Vietnam) on 5 September 2013 that were quarantined and acclimated to the experimental system as described in Robinson et al. (2018).

Three experimental treatments were randomly allocated to 15 incubation chambers with five replicates per treatment. The “initial” (In) treatment was included to ensure that there were no significant differences between treatments prior to the start of the experiment and as an initial reference point for evaluating the effect of the treatments. The “no added carbon” treatment (-C) with a C : N of 5 : 1 received aquaculture waste only (215.06 mg day-1 wet weight). The “added carbon” treatment (+C) received aquaculture waste (215.06 mg day-1 wet weight) and carbon in the form of soluble starch (44.50 mg day-1 dry weight) daily to increase the C : N to 20 : 1 (mass ratio) from day 0 (Table 1). The carbon addition treatments (+C) were standardized at a concentration of 400 mmol C m-2 d-1.

Sediment incubation chambers were established by transferring unsieved CaCO3 builder's sand sourced from a commercial dune quarry (SSB Mining, Macassar, South Africa) into Plexiglas^® tubes (25 cm long, 8.4 cm internal diameter) sealed with a polyvinyl chloride (PVC) end cap to a depth of 7.5 cm. The incubation chambers were connected via 4.0 mm air tubing and 4.0 mm variflow valves to a manifold receiving seawater directly from a RAS biofilter (see Robinson et al., 2015, for further details). The water flow rate was 50 mL min-1, equivalent to 16.34 exchanges h-1. The chamber outflows were routed into a main drainage channel and allowed to flow to waste to prevent soluble carbon sources from entering the RAS. Unsieved CaCO3 was preconditioned for 4 weeks in flow-through tanks prior to its transfer into the chambers. The sediment was allowed to condition and stabilize into redox-stratified layers for 14 days prior to the commencement of the experiment. No aeration was provided; however, water was continuously mixed at 60 rpm using a magnetic stirring rod positioned 15 cm above the sediment surface. Stirring rates were just below those causing sediment re-suspension (Ferguson et al., 2004; Gongol and Savage, 2016).

The experimental area was fully shaded from direct sunlight. Light intensity was measured during daylight incubations using a light meter (LX-107, Lutron Electronic Enterprise Co. Ltd, Taipei, Taiwan) positioned 10 cm above each chamber. Additionally, a temperature/light logger (Hobo, UA-002-64, Onset, USA) was placed in an additional chamber positioned in the centre of the experimental treatments. The mean (hours) natural photoperiod was 13.34 : 10.26 (L : D).

The aquaculture waste, used as feed for the sea cucumbers, comprised uneaten abalone (Haliotis midae) feed and faeces. It was collected daily from the backwash of a sand filter in a recirculating abalone grow-out system. Samples were sent for organic-carbon and total nitrogen content analysis (Robinson et al., 2018) and the mean C : N was 5.21 : 1. Soluble starch (Merck Millipore, Pretoria, South Africa) was used as an additional carbon source to increase the C : N to 20 : 1. Additions of waste with (+C) or without (-C) added carbon commenced on day 0. The aquaculture waste was mixed into a wet slurry while the starch was dissolved in seawater and added daily to the respective treatments at 16:00 local time (UTC + 2) from day 0 to day 14.

Baseline data were collected at the start of the experiment (i.e. day -1), with fluxes measured in all 15 chambers under light and dark conditions. All replicates from the In treatment were sacrificed on day 0 and sub-cored for analysis of sediment characteristics.

Animals (n=30) previously acclimated in the RAS were suspended in mesh containers for 24 h to evacuate their guts prior to weighing and photo-identification (Robinson et al., 2015). Three juvenile H. scabra with a mean (± standard deviation) weight of 1.91 ± 0.36 g were added to each of 10 chambers (equivalent to a high stocking density of 1034.00 g m-2) on day 0. They were removed at the end of the experiment (day 14), gut-evacuated for 24 h and reweighed. Wet-weight data were used to calculate the growth rate (g d-1; Robinson et al., 2015).

Benthic flux incubations were conducted on day -1 for all treatments (In, -C and +C) and on alternate days from day 1 to day 13 for the -C and +C treatments, after sacrifice of the In treatment. Light incubations were conducted during daylight hours, commencing after sunrise (08:00 local time), and dark incubations were conducted after sunset (22:00 local time). When data were collected, the flow from each chamber was interrupted, the stirrers were paused (∼ 3 min) and the chambers were uncapped by removing the rubber bung. A portable optical meter (YSI ProODO, YSI Pty Ltd, USA) was inserted through the sampling port to measure temperature (±0.01 ∘C) and dissolved oxygen (DO) concentrations (±0.01 mg L-1). The pH (±0.01 pH units) was measured electrochemically (Eutech Instruments pH 6+ portable meter, Singapore).

Water alkalinity and nutrient concentration (ammonia, nitrate/nitrite, nitrite and phosphate) were recorded at the start and end of each light–dark incubation period. To do this, samples were withdrawn using a 50 mL acid-washed plastic syringe connected to the chamber outflow through 4.0 mm tubing and filtered (Whatman^® glass microfibre filters grade GF/C, Sigma Aldrich, Johannesburg, South Africa) into 15 mL screw-capped polycarbonate vials. All nutrient samples were immediately frozen at -20 ∘C, and alkalinity samples were kept cold at 4 ∘C. The N2 samples were taken on three sampling occasions (days 1, 7 and 13) during dark incubations, as during daylight hours bubbles may form that interfere with the estimation of N2 : Ar and thus lead to an overestimation of N2 production (Eyre et al., 2002). To minimize bubble introduction, N2 samples were collected by allowing the water to flow by gravity from the chamber outflow directly into 7 mL gas-tight glass vials with glass stoppers filled to overflowing. The N2 samples were poisoned with 20 µL of 5 % HgCl2 and stored submerged at 20 ∘C. The N2 samples were collected in duplicate or triplicate; thus, the final values represent the mean value calculated for each replicate (Eyre and Ferguson, 2005).

After withdrawal of all water samples, replacement water was gravity fed into the chamber directly from the manifold and the chambers were recapped and the stirrers restarted. All materials used for sample collection were acid washed, rinsed three times with distilled water and air-dried prior to use. Total oxygen exchange was measured in three randomly selected chambers during incubations (one from each treatment) to ensure that the oxygen concentration did not decrease by more than 20 %. Incubation times were kept short, ranging from 68 to 146 min with an average duration of 104 min, to prevent oxygen depletion and ensure that flux rates were linear (Burford and Longmore, 2001; Glud, 2008).

Dissolved nitrate and nitrite (NOx; 0.01 µM) were determined colourimetrically by flow injection analysis (QuikChem^® 8500 Automated Ion Analyzer, Hach Company, USA) and a commercially available test kit (QuikChem^® method 31-107-04-1-E for the determination of nitrate and nitrite in seawater). All other nutrient samples were analysed manually. Ammonium (0.01 µM) and dissolved inorganic phosphate (0.01 µM) were determined using the methods of Grasshoff (1976) and Grasshoff et al. (1999) respectively, and nitrite (NO2-; 0.01 µM) was determined according to Bendscheider and Robinson (1952).

Alkalinity (0.01 mg L-1) and total dissolved CO2 (0.01 µM) concentrations were determined by potentiometric titration according to Edmond (1970) using an automated titrator system (876 Dosimat plus, Metrohm, USA). Total alkalinity was calculated according to the method of Snoeyink and Jenkins (1980). CO2 concentrations were calculated from alkalinity and pH using the equations given in Almgren et al. (1983). Changes in pH and alkalinity were used to calculate dissolved inorganic carbon (DIC) fluxes.

Dinitrogen gas (N2) was determined from N2 : Ar using membrane inlet mass spectrometry (MIMS) with O2 removal (±0.01 %). Measurement of direct N2 fluxes using this technique represents the net benthic flux of N2 resulting from a combination of processes that produce N2, such as denitrification and anammox, and processes that consume N2 such as nitrogen fixation (Ferguson and Eyre, 2007; Eyre et al., 2013a).

Nutrient and gas fluxes across the sediment–water interface during light and dark incubations were calculated using initial and final concentration data according to Eq. (1). Net flux rates, representing the net result of 13.57 h of dark fluxes and 10.43 h of light fluxes were calculated according to Equation 2 (Veuger et al., 2007). Gross primary production was calculated according to Eq. (3), where light O2 fluxes represent net primary production and dark fluxes represent respiration. Remineralization ratios were calculated according to Eq. (4) (Eyre et al., 2013b). Flux=Cn-C0×VA×t×10000, where Flux is flux (µmol m-2 h-1), C0 is concentration at time zero (µmol L-1), Cn is concentration at time n (µmol L-1), t is incubation time (h), A is area of sediment surface in chamber (cm2) and V is volume of water in chamber (L). Netfluxrates=(hourlydarkrates×hoursofdarkness+hourlylightrates×hoursofdaylight)/24hGrossprimaryproduction=lightO2flux(+ve)-darkO2flux(-ve)Remineralizationratio=darkO2fluxN2+NH4++NOx

On days 0 and 14, three sub-cores (internal diameter 30 mm) were extracted from the In and experimental (-C and +C) chambers. Each sub-core was sectioned into the following five depth intervals: 0.0–0.5, 0.5–1.0, 1.0–2.0, 2.0–4.0 and 4.0–6.0 cm; this was done for the analysis of sediment characteristics. One set of sub-cores was dried at 50 ∘C for 24 h for analysis of total organic carbon and total nitrogen; the second set was frozen in sealed vials in black bags for spectrophotometric analysis of total carbohydrates. Two sets of samples were prepared from the third sub-core: sediment samples were frozen in 2 mL Eppendorf tubes for subsequent deoxyribonucleic acid (DNA) extraction and sequencing. The remaining sediment was added to 15 mL vials filled with 0.2 µm filtered, 1 % buffered paraformaldehyde and refrigerated for the determination of bacterial abundance by flow cytometry.

The organic content measured as particulate organic carbon (OC) and total nitrogen (TN) was determined on an elemental analyser after the removal of carbonates by acid fumigation (Robinson et al., 2015). Total sediment carbohydrates were measured on defrosted samples using the phenol-sulfuric acid method (Underwood et al., 1995).

Aliquots of preserved samples were prepared in duplicate by staining with 4′,6-diamidino-2-phenylindole (DAPI) for 15 min at 4 ∘C in darkness (Marie et al., 1999). Bacterial abundance was analysed with a FACSCalibur flow cytometer (BD Biosciences, Singapore), fitted with a 488 nm, 15 mW laser, using the FL1 detector (λ=530 nm). TruCount beads (BD Biosciences, Singapore) were used as an internal standard. All cytometric data were logged and analysed using Cell Quest (Becton-Dickinson) using Escherichia coli cells as a reference. Cell abundance was converted to cells per gram of dry sediment.

Genomic DNA was extracted from approximately 250 mg of substrate samples using a DNA isolation kit (ZR Soil Microbe DNA MiniPrep, Zymo Research, USA) yielding purified genomic DNA for use in polymerase chain reaction (PCR) amplification. Genomic DNA was stored in sealed, labelled Eppendorf tubes at -20 ∘C prior to being couriered from the Republic of South Africa to the United Kingdom. To comply with the Animal Health Act 1981, the samples were accompanied by a general import license (IMP/GEN/2008/03) for the importation of animal and poultry products, including DNA, from all non-EU countries.

Library preparation was performed using a modified version of the MiSeq WetLab protocol (Kozich et al., 2013). One microlitre of template DNA was arrayed into 96-well plate format with 17 µL of Accuprime Pfx Supermix (Thermofisher, UK), leaving two wells on each plate open for controls. Two microlitres of reconstituted indexed primers at 100 µM were added to the samples to barcode them for identification. To identify any contaminating operational taxonomic units (OTUs), two control samples were included in the sequencing run. The negative control consisted of 1 µL of PCR grade dH2O and the positive control was 1 µL of mock community (HM-278S, BEI Resources, Manassas, USA) at a 1 : 3 dilution. The primer pair 515F/806R was used to amplify the V4 region of the 16S rRNA gene. PCR was performed using the following conditions: initial enzyme activation and DNA denaturation proceeded at 95 ∘C for 2 min followed by cycling parameters of 95 ∘C for 20 s, 55 ∘C for 15 s and 72 ∘C for 5 min for 30 cycles. A final extension was done at 72 ∘C for 10 min. The amplification of the PCR products was checked on a subset of 12 samples using gel electrophoresis on a 1 % agarose gel prior to library clean-up. Samples from all plates were pooled and libraries were subjected to quality control including quantification using a KAPA Biosystems Q-PCR kit, obtaining a bioanalyser trace using the Agilent Technologies HS DNA kit and normalization using the Invitrogen SequalPrep Plate Normalization Kit (Thermofisher, UK). Amplicons were sequenced on an Illumina MiSeq platform by NU-OMICS (Northumbria University, UK).

The raw fastq files were processed using Mothur (version 1.37.0) based on the Schloss MiSeq SOP with modifications. Raw forward and reverse sequence reads were merged to create contigs prior to quality filtering. The sequence reads were trimmed using a sliding window of five base pairs (bp) with an average window quality threshold (Q) of 22 or greater. Sequences containing an ambiguous (N) base, > 8 homo-polymers or that had a sequence length < 275 bp were discarded. Quality-filtered sequences were aligned using a custom alignment created for the variable four (V4) region of the 16S rRNA gene using the Silva database (version 123; July 2015 release). The reads were screened to include only overlapping regions (based on alignment positions), pre-clustered (number of differences: 1) and checked for chimeras using the UCHIME algorithm (Edgar et al., 2011).

Taxons classified as “mitochondria”, “Eukaryota” or “unknown” were specified during the remove.lineage command. The count.groups command was used to determine the minimum number of reads per sample for normalization. To standardize sequencing effort, all samples were subsampled to 550 using the sub.sample command, to ensure that all replicate samples from the experimental treatments (+C and -C) were retained. The subsampled OTU table (shared file) and assigned consensus taxonomy (cons.taxonomy.file) were used in downstream analyses, including alpha and beta diversity, taxonomic composition and meta-genome predictions of the microbial communities.

Environmental (light, temperature, salinity) and flux rate data for nutrients (NH4+, NO2-, NOx and PO43-) and gases (DO, DIC and N2 – night only) collected on day -1 during light and dark incubations were averaged to provide a mean value per replicate chamber for each diurnal period. The data were tested for homogeneity of variance and for the normal distribution of the residuals using Levene and Shapiro–Wilk tests. One-way analysis of variance (ANOVA) tested for differences in the environmental, nutrient and gas flux data between the In, +C and -C treatments on day -1.

The light, water quality and flux rate data (days 1–13) for nutrients and gases were averaged to provide a mean value for each replicate incubation chamber. It was not possible to conduct daytime incubations on day 9 due to lowered O2 concentrations in the chambers; therefore, light incubation data represent a mean of six values (days 1, 3, 5, 7, 11 and 13), while the mean dark incubation data were calculated from the full set of seven incubations. The mean temperature, salinity and mean light, dark and net fluxes of nutrients and gas fluxes, mean remineralization ratios, and mean gross primary production measured during the experimental period (days 1–13) were analysed using a Student t test at alpha < 0.05. Sediment characteristics, including organic carbon, total nitrogen, C : N and bacterial cell abundance, were compared using mixed-model ANOVA with treatment (+C and -C) and sediment depth as fixed factors. When a significant effect was observed, post hoc comparisons of means were conducted with a Tukey's honest significant difference test. Differences in H. scabra growth rate and biomass density were analysed by a Student t test at alpha < 0.05. Data are presented as mean ± standard error unless otherwise stated. All statistical analyses were performed in Statistica v.13.

Alpha (within-sample) diversity metrics for the number of OTUs (observed), richness (Chao 1), abundance-coverage estimator (ACE) and diversity (Shannon, Simpson and Inverse Simpson) were calculated and visualized in the phyloseq package in R (McMurdie and Holmes, 2013). The diversity metrics were generated by the summary.single command by subsampling to the lowest number of reads per sample (n=550) and compared across treatments and sediment depths using mixed-model ANOVA.

Patterns in bacterial community structure between treatments and sediment depths were visualized using principal coordinates analysis (PCoA) based on a Bray–Curtis dissimilarity matrix calculated from the OTU table in R. In addition, a non-parametric multivariate analysis of variance (PERMANOVA) was performed on the community distance matrix based on the Bray–Curtis dissimilarity index to test the null hypothesis that there was no difference in the structure of microbial communities between treatments (In vs. -C vs. +C) and sediment depth using the “adonis” function of the vegan package in R (Oksanen et al., 2016).

Mantel correlation tests were performed on dissimilarity matrices of the community and environmental data to provide an indication of how well microbial community data corresponded to the environmental data. The environmental distance matrix was calculated as Euclidean distances computed from a metadata table containing all of the data describing light, water quality, sediment characteristics and net flux rates for gases and nutrients. The significance of correlation coefficients was assessed using a permutation procedure. In addition, the correlation between environmental data and the sediment microbial communities was determined using the “envfit” function of the “vegan” package in R (Oksanen et al., 2016). Since none of the environmental characteristics were significantly correlated with the microbial community data, the environmental data were not plotted as vectors on the PCoA ordination.

The Tax4Fun package in R was used to predict the metabolic capacities of the microbial communities from the 16S rRNA sequences. The fctProfiling option was set to TRUE (default) to predict the metabolic capacities of the meta-genomes based on pre-computed Kyoto Encyclopedia for Genes and Genomes (KEGG) ortholog reference profiles (Aßhauer et al., 2015). Only KEGG pathways within “nitrogen metabolism” were retained for analysis. The KEGG pathway map 00910 for nitrogen metabolism and associated information was used to extract the KEGG ortholog reference numbers involved in the six fully characterized reactions listed under “nitrogen metabolism” (Table S2 in the Supplement). Anaerobic oxidation of ammonia (anammox) was not included, as although this process is recognized in the KEGG database, it has yet to be assigned to a module or reference profile.

The relative abundance of functional genes predicted from the 16S rRNA sequences within each ortholog reference profile were summed to provide a mean value for each pathway module for each replicate sample from all sediment depths sampled in all treatments (n=45). The relative abundance of functional genes in the In and experiment treatments was illustrated by graphically plotting vertical depth profiles and analysed statistically using a mixed-model ANOVA.

Survival of sea cucumbers was 100 % in the +C treatment; however, one replicate chamber from the -C treatment was terminated on day 9 following a period of water column hypoxia, caused by one animal preventing water exchange by blocking the outflow valve. This resulted in the mortality of all sea cucumbers in this chamber, reducing the overall survival to 80 %. There was no significant difference between the mean sea cucumber wet weight on day 0 or day 14 between treatments; however, despite the short duration of the experiment the sea cucumbers in both treatments lost mass (decreasing from 1.91 ± 0.02 to 1.62 ± 0.03 g; an overall mean growth rate of -0.02 ± 0.00 g day-1). The biomass density decreased from 1,034.00 ± 12.73 to 874.97 ± 18.31 g m-2, although the initial stocking density was comparable to the final densities (1011.46 ± 75.58 g m-2) achieved in previous carbon amended cultures standardized at 200 mmol C m-2 day-1 (Robinson et al., 2018).

Benthic fluxes of dissolved oxygen and DIC can provide an indication of overall benthic metabolism in response to organic enrichment (Eyre et al., 2011). There were no significant differences in the light, dark or net fluxes of DO, DIC or N2 between treatments on day -1 (N2 dark only; Fig. S1 in the Supplement). Sediment oxygen consumption was significantly higher in the +C incubations throughout the experiment in both light and dark incubations (Student's t test; t=-2.87, p=0.006) resulting in a higher net consumption of -2905.84 ± 99.95 µmol O2 m-2 h-1 compared to -2511.31 ± 116.81 µmol O2 m-2 h-1 in the -C treatment (Fig. 1a). Oxygen and DIC fluxes clearly show that the sediment metabolism was net heterotrophic. During the day, DIC release from organic-matter degradation exceeded DIC consumption from primary production (Fig. 1b). There was sediment oxygen consumption during light and dark incubations, indicating that respiration dominated over photosynthesis, supported by the lower gross primary production in the +C treatment (Fig. 1d). There were no significant differences in the light, dark or net fluxes of DIC with a mean net efflux of 12 732.34 ± 2031.69 µmol C m-2 h-1 across the treatments (Fig. 1b). The assumed low rates of photosynthesis may have been due to shading and from turnover of the micro-phytobenthos standing stock due to grazing by sea cucumbers (Glud et al., 2008; MacTavish et al., 2012). In addition, DIC fluxes were 4-fold higher than oxygen fluxes, indicating that the majority of the organic carbon was oxidized by anaerobic pathways (Burford and Longmore, 2001; Eyre et al., 2011).

The mean dark N2 flux on days 7 and 13 was not significantly different between treatments (Student's t test; t=-1.29, p=0.23; Fig. 1c). Carbon supplementation resulted in a net N2 uptake (-142.96 ± 107.90 µmol m-2 h-1), indicating that atmospheric nitrogen fixation dominated over denitrification and anammox during dark incubations. In contrast, the -C treatment had a small but positive net N2 efflux (17.33 ± 36.20 µmol m-2 h-1), indicating that nitrogen removal pathways, such as denitrification or anaerobic ammonium oxidation (anammox), were slightly greater than nitrogen fixation.

Ambient environmental conditions recorded in the incubation chambers at the start of the experiment on day -1, during light and dark periods, are presented in Table S1. There were no significant differences in the dark or net fluxes of any of the nutrients between treatments on day -1, except the NH4+ fluxes during light incubations, which were significantly different (one-way ANOVA; F(2,9)=12.73, p=0.002; Fig. S2). The In chambers had a significantly higher NH4+ efflux of 115.32 ± 11.43 µmol m-2 h-1 compared with an uptake of -9.77 ± 11.82 µmol m-2 h-1 in the -C treatment. The +C treatment had intermediary values with a mean NH4+ efflux of 56.03 ± 25.54 µmol m-2 h-1. NH4+ had the highest flux rate throughout the experiment (Fig. 2b) with mean efflux significantly higher in the +C chambers during light incubations compared with the -C treatment (182.25 ± 120.77 vs. 83.90 ± 26.70 µmol m-2 h-1, t test; t=2.93, p=0.005; Fig. 2b). Sediment–water exchange of NO2-, NOx and PO43- were unaffected by carbon addition. Mean fluxes of NH4+, NO2- and PO43- were positive irrespective of diel cycle, indicating net release from the sediment (Fig. 2a–c); however, NOx fluxes were variable with opposing trends in light, dark and net fluxes between treatments (Fig. 2d). As both dissolved oxygen consumption and NH4+ production were higher in the +C chambers, this indicates an overall increase in benthic metabolism during daylight.

The sediment OC content decreased in the experimental treatments after 14 days compared to the initial treatment (Fig. 3a). The largest decrease was observed at the 1.0–2.0 and 2.0–4.0 cm depth intervals spanning the approximate depth of the oxic–anoxic interface, one of the most active zones of organic-matter mineralization by heterotrophic microorganisms (Reimers et al., 2013). Vertical profiles of TN and the C : N on days 0 and 14 followed a similar trend with the most marked changes occurring at the 1.0–2.0 and 2.0–4.0 cm depth intervals. Carbon addition did not affect the OC or TN, but sediment depth significantly influenced the OC (mixed-model ANOVA; F(4,20)=3.54, p=0.024; Fig. 3a) and TN content (mixed-model ANOVA; F(4,20)=3.37, p=0.029; Fig. 3b), being significantly lower at the 1.0–2.0 cm depth interval with mean values of 0.24 ± 0.02 % (OC) and 0.03 ± 0.00 % (TN). This confirms that the oxic–anoxic interface supported the highest rates of organic-matter mineralization. In contrast, the deepest sectioned interval (4.0–6.0 cm) had significantly higher OC (0.51 ± 0.08 %) and TN content (0.07 ± 0.01 %) than the shallower intervals. Carbon addition did not significantly increase the sediment C : N in the +C treatment (7.90 ± 0.27) compared to the -C treatment (7.12 ± 0.24; mixed-model ANOVA; F(1,20)=4.52, p=0.054; Fig. 3c). However, carbon supplementation resulted in mean remineralization ratios (after the exclusion of outliers) of 15.68 ± 7.43 that were approximately 3-fold higher than chambers receiving aquaculture waste only (5.64 ± 4.50), although the difference was not significant (t test; t=1.08, p=0.32). Remineralization ratios were higher than the sediment C : N in the +C treatment; a trend that is consistent with nitrogen assimilation by heterotrophic bacteria, including nitrogen fixation (Eyre et al., 2013b). Conversely, in the -C treatment receiving raw aquaculture waste at a C : N of 5 : 1, the remineralization ratios were lower than the sediment C : N, indicating net release of nitrogen.

A total of 781 701 16S rRNA reads were generated. Four samples from one replicate of the In treatment were removed during subsampling due to a low abundance of reads and therefore excluded from further analysis. A total of 780 612 sequences in the 41 samples remained subsequent to quality control, primer trimming, size exclusion and the removal of unassigned taxons, mitochondria and Eukaryota.

Neither carbon addition, sediment depth nor the interaction between the factors (treatment × sediment depth) significantly affected the number of sequences, OTUs (observed species), community richness (Chao and ACE), or diversity measured as Simpson and Inverse Simpson indices (mixed-model ANOVA; p<0.05; Fig. 4). Sediment depth significantly influenced Shannon diversity, with the highest diversity of 2.85 recorded in the sediment surface layer (0–0.5 cm) and the lowest (1.54) in the 4–6 cm layer (mixed-model ANOVA; F(4,26)=3.14, p=0.031).

Flow cytometry data compared relatively well with the 16S rRNA amplicon sequencing data. Bacterial abundance (cells per gram; Fig. 3e), the number of sequences and OTUs were higher in the In chambers than the experimental chambers sampled on day 14, presumably in response to grazing by the sea cucumbers. The number of OTUs decreased from 286.81 ± 128.13 in the In chambers to 176 ± 65.15 and 181.20 ± 45.90 in the +C and -C treatments respectively. Overall, the community diversity was low: Shannon diversity: 2.31 ± 0.13; Inverse Simpson: 5.79 ± 0.51. There was a marked increase in community richness at the 1–2 cm depth interval, coinciding with the oxic–anoxic interface. In the In chambers the number of OTUs was 778.00 ± 731.00, compared with 343.33 ± 199.25 and 322.67 ± 307.25 in the +C and -C treatments respectively. The Chao 1 richness indicator also followed this trend (Fig. 4).

The majority of sequences (99.8 %) were assigned to the Bacteria, with only 0.12 % assigned to Archaea. Taxa from three archaeal phyla were present, including Euryarchaeota, Thaumarchaeota and Woesearchaeota. Natronorubrum (Euryarchaeota), a halophilic aerobic chemoorganotroph (Xu et al., 1999), was the most abundant genus, representing 14 of the 27 archaeal reads.

The bacterial community contained a total of 18 phyla, 4 candidate phyla and the candidate division WPS-2. Proteobacteria and Firmicutes were the two dominant phyla, accounting for 47.64 and 34.71 % of the total sequences respectively, with Cyanobacteria accounting for 7.42 %. Planctomycetes (2.45 %), Actinobacteria (2.34 %), unclassified Bacteria (2.12 %) and Bacteroidetes (1.33 %) were minor components. The remainder of the phyla, candidate phyla and the candidate division WPS-2 each represented less than 1 % of the community. Candidate phyla included Hydrogenedentes (formerly NKB19), Latesbacteria (formerly WS3), Parcubacteria (formerly OD1) and Poribacteria.

Taxa within the Oxalobacteraceae and the genus Herbaspirillum were significantly more abundant in the -C treatment (Welch's two-sided t test; p<0.05; Fig. 5). In comparison, the genera Blastopirellula and Litorilinea were significantly enriched in the +C treatment. There were no significant differences in the mean proportion of taxa between experimental treatments at phylum, class or order levels, underscoring the high degree of similarity among the microbial communities between treatments (Fig. 6). Further, there was no correlation between the microbial community and environmental data (Mantel test; r=0.04, p=0.27). The first axis in the PCoA ordination explained 53.4 % of the variation and appeared to be associated with sediment depth, while the second axis (4.7 % of the variation) appeared to be associated with experimental treatment. Treatment did not significantly influence microbial community structure (PERMANOVA; p<0.05; Table 2), which may be a function of the relatively short duration of the experiment. By contrast, there was a significant effect of sediment depth on the microbial community (PERMANOVA; p=0.011; Table 2).

There were no significant differences in the predicted relative abundance of genes involved in the six nitrogen transformation pathways (mixed-model ANOVA; p>0.05; Fig. 7). The relative abundance of predicted nitrification genes peaked at the 0.5–1.0 cm depth interval in the -C treatment, coinciding with the oxic zone. In the +C treatment, the relative abundance of predicted denitrification and DNRA genes were higher in the sediment layers sectioned at 1.0–2.0, 2.0–4.0 and 4.0–6.0 cm. Overall, DNRA was the dominant pathway (20.52 ± 0.01 %) predicted to occur in all treatments and sediment depths, with the exception of the surface layer (0.0–0.5 cm) in the +C treatment, where there was a higher predicted relative abundance of denitrification genes (Fig. 7). Denitrification was the second most abundant predicted pathway (18.02 ± 0.01 %), followed by complete nitrification (8.80 ± 0.43 %), indicating that the potential for coupled nitrification–denitrification was present in all treatments. Genes predicted to be involved in nitrogen fixation represented 2.85 ± 0.32 %.