On the same day as sampling, we quantified denitrification rates via acetylene blockage/inhibition assays. This indirect method has been successfully applied to investigate coral reef associated denitrification activities ⁵⁴. Acetylene blocks the activity of the enzyme nitrous oxide (N₂O) reductase within the denitrification pathway, leading to the accumulation of N₂O which can be used as a proxy for the relative activity of denitrification of the coral holobiont ^54–57. The efficacy of acetylene assays has been validated in previous work which used both acetylene assays and nirS gene copy numbers to quantify denitrification ³⁶. The observed patterns of nirS gene abundance corresponded closely with the denitrification rates measured using the acetylene method, providing evidence that the acetylene method is accurate and can be relied upon ³⁶.

To set up the acetylene assays, we secured each coral fragment to a stand using rubber bands and placed them inside a gas-tight glass beaker (Figure S1a). We specially adapted the beakers for acetylene assays by having an 8 mm hole drilled into the lid. The hole was sealed with a gas-tight rubber stopper, and a hypodermic needle (hypodermic needle with polypropylene hub 30G x 3/4", Tyco Healthcare group, MonojectTM) was permanently inserted through the stopper into the jar. On the outside of the jar, the needle hub was attached to a 2-way stopcock with a Luer lock connection (two-way stopcock, BraunTM DiscofixTM) where a gas syringe (50 ml gastight syringe model 1050 TLL PTFE Luer Lock, Hamilton) could be later fitted when required, for gas samples to be taken (Figure S1a). We filled each beaker with seawater (taken from the sampling site the same morning) to 80% of its capacity, leaving a 20% headspace. We then replaced ten percent of the seawater volume with acetylene enriched seawater, and likewise, replaced 10% of the beaker headspace with acetylene gas (Yoshinari & Knowles, 1976) (Figure S1a). We made acetylene gas freshly on the same day prior to the incubations (detailed in the Supplementary). For the incubations, we secured corals to a stand in gas-tight glass beakers filled with site-collected temperature-controlled seawater. We placed beakers into water baths equipped with thermostats (3613 aquarium heater. 75W 220–240 V; EHEIM GmbH and Co.KG) and temperature controllers (Schego Temperature Controller TRD, max. 1000W) that heated the water to the corresponding in situ temperature per sampling month. The water bath was then placed on top of magnetic stir plates which powered magnetic stir bars in the base of each beaker to ensure adequate water circulation (~ 220 rpm). we incubated corals for 12 h in the dark followed by 12 h in the light. During the light incubation, we supplied light at the same intensity as the in situ conditions of the respective sampling month. However, we used new fragments for the following light incubation to minimise the potential impact of stress on the coral’s denitrification rates. We included four control beakers containing no corals in each incubation run to account for potential background denitrification activity in the seawater. We took gas samples (3 ml) from the beaker headspace at the beginning (T0) and end (T12) of each incubation, using a gas syringe (50 ml gastight syringe model 1050 TLL PTFE Luer Lock, Hamilton) and stored the samples in gas tight vials until further measurement.

Gas samples generated from the acetylene assays are typically measured using a gas chromatograph (GC) fitted with an electron capture detector (ECD). However, in our study, we measured the gas samples using a N₂O microsensor (custom-made, Unisense). Although a GC with ECD was tested, it was not used as microsensors outperformed it in several aspects. For example, the electrochemical microsensor has high sensitivity (detection limit: 25 nM) and higher throughput compared to GC-based methods. We connected the microsensor to a multi-channel (fx-6 UniAmp multi-channel 110394, Unisense) and calibrated it every day prior to usage with a two-point calibration curve consisting of a low point (ambient air: 0.009 µmol L^− 1) and a high point (a known standard: 0.575 µmol L^− 1) at a consistent room temperature (21°C) and pressure (1 bar). To record and visualise measurements, we synced the N₂O microsensor with the Sensor Trace Suite software (v.1.13) on a computer desktop. Further technical details about the microsensor and the steps to use the microsensor and normalise the data can be found in the supplementary material.

While the acetylene inhibition technique has historically been popular to quantify denitrification rates ^36,38,41, it has several limitations to be aware of ^54,58. Sometimes incomplete inhibition of N₂O reductase occurs, meaning that N₂ is produced as normal instead of accumulating as N₂O, causing an underestimation of denitrification rates ⁵⁴. Further underestimation of denitrification may arise from the inhibition effect of acetylene on the nitrification pathway ⁵⁹. Nitrification is the preceding step in the N-cycle that provides nitrate as a substrate for denitrification. Therefore, denitrification rates may be underestimated due to substrate limitation ^58–60. Another common technique is to run 15N tracer incubations ⁴⁰ which measures the actual N₂ production directly. The benefits of this technique include the ability to distinguish between pathways and the lack of an inhibitory effect on related pathways. However, in our case, using the acetylene inhibition technique was more suitable, as it allowed us to measure denitrification activity under ambient conditions, without artificially enriching with 15N-labelled substrates.

Firstly, we blasted the tissue off the coral skeleton. To do so, we held the fragment within a sterile clear sampling bag (Whirl-pack sample bag) and used an airbrush (model S68 with dual action siphon feed, Master Airbrush) to remove the tissue with high pressure air and MilliQ water. We always sterilised the airbrush with 70% ethanol and rinsed it with MilliQ water between samples. We then stored the tissue slurries in Falcon tubes at − 20°C and defrosted them in prior to the next stage. Once samples had defrosted, we separated the tissue slurry into host and symbiont fractions via centrifugation (2.5 minutes at 500 x g) and then washed and resuspended the symbiont pellet in 3 ml of MilliQ until it was clean (i.e., free from white residue), which typically required three washes. Following this, we filtered each fraction through 0.7 µm GF/F filters (fitted to a vacuum assembly), treated to remove skeletal contaminants (2 ml x1N HCl), rinsed with MilliQ, and immediately dried the samples at 60°C for 48 h. We then scraped the dried mass into tin cups and weighed and analysed the samples for δ13C, δ15N and mass percent C and N via EA-IRMS at the Natural History Museum, Berlin, using a Flash 1112 EA and Thermo Scientific Delta V IRMS (Berlin, Germany). We report the isotopic data using the conventional delta notations (δ) and express in ‰ relative to the international standards (Vienna Pee Dee Belemnite) for δ13C (0.01118) and atmospheric N2 for δ15N (0.00368) ⁶¹. Within-run standard deviations (SD) of the standards were < 0.15 per mil (‰) for δ13C and δ15N and the SD of replicate measurements of the lab standard are < 3% of the concentration analysed.

Twenty four hours before sampling for nutrients, we acid-washed all water sampling equipment in a 4% HCl bath to minimise potential contaminations. On the same day as coral fragments were collected, we e collected water samples from the average study site depth (~ 3–5 m depth) using 2 x 5 L Niskin bottles. On the boat, we aliquoted water for separate nutrient measurements into Falcon tubes. Immediately on the boat, water aliquots for nitrite and nitrate were filtered (0.22 µM Millex®-GV), yet we performed no immediate filtration steps for ammonium, DOC and chl a (stored in opaque bottles). We kept all water samples on ice during transport and then stored samples for nitrite, nitrate and ammonium at − 20°C, and stored samples for chl a and DOC analysis at + 4°C.

For the measurement of nitrate and nitrite, we analysed samples with a segmented flow analyser (Model AA3 HR, SEAL Analytical IC). We performed a calibration prior to every run and accepted upon the criteria that R² > 0.99. To prepare the calibration standards, we used a ready-made stock standard of 1000 ppm nitrite and nitrate to create low (5, 20, 50 and 100 ppb) and high (100, 200, 500, 1000 ppb) standards. The instrument detection limit for nitrate was 0.0322 µmol L^− 1 and 0.0217 µmol L^− 1 for nitrite. For the measurement of ammonium, we analysed samples using a fluorometer (Turner Designs, Trilogy Fluorometer) via Orthophthaldialdehyde (OPA) derivatisation. We performed a calibration prior to every run and accepted upon the criteria that R² > 0.99. We prepared a mother solution of ammonium chloride to make standards of 0.0, 0.03, 0.05, 0.1, 0.15, 0.2, 0.25, 0.3 µmol L^− 1. The detection limit of the instrument was 0.058 µmol L^− 1. For DOC analysis, we filtered 500 ml of water through 0.7 µm GF/F filters (pre-combusted at 450°C for 4.5 h). We divided the filtrate into sterile amber glass vials, that we then acidified with 0.1 ml of 85% phosphoric acid to prevent bacterial activity and analysed them on a TOC Analyser (TOC-L, Total Organic Carbon Analyser, Shimadzu, Kyoto, Japan). We performed a calibration before each run using a standard addition curve of Potassium Hydrogen Phthalate (0; 33; 25; 50; 62; 83; 100; 125; 167; 250; 500 µmol C L^− 1). We prepared internal controls using Consensus Reference Materials (CRM; Batch 12; 2012; DOC: 42–45 µmol L − 1) provided by DA Hansell and W Chen (University of Miami). The average analytical variation of the instrument was < 3.5% for DOC based on 5–7 injections per sample.

On the same day as collection, we filtered water samples for chl a (2 L in duplicates) through 0.7 µm GF/F filters and then stored them at − 80°C until further processing. We prepared samples for measurement adapting a protocol from Erar & Collins, (1997) ⁶² and a protocol used in the California Operative Oceanic Fisheries Investigations. In brief, we soaked the filters in 10 ml of 90% acetone, vortexed, sonicated in an ice bath and stored them at 4°C overnight in the dark. The following morning, we repeated the sonication and vortexing steps twice more, and then centrifuged the samples at 2500 rpm at 4°C for 10 min. Next, we measured the samples fluorometrically (Turner Designs, Trilogy Fluorometer) on a fluorometer fitted with a chl a module (Turner Designs, Wide-Chlorophyll a Acidification Module). We calibrated the instrument prior to use with standards of 0.5, 1.0, 2.5, 5.0, 10.0, 20.0, 100.0 and 200.0 µg L^− 1 which generated a calibration curve of R² > 0.99. Following this, we validated the instrument calibration with two solid secondary standards adjusted to 2.5 and 20 µg L^− 1 (Turner Designs, Adjustable Solid Secondary Standard - Red) and ran “blanks” of 90% acetone at the beginning, and after every few samples.

Lastly, we measured seawater temperature continuously throughout the sampling period using an Onset Hobo pendant temperature logger deployed at the reef. We deployed the logger at 1–2 m depth, and measured temperature at 10 minute intervals throughout the sampling period April 2022 – February 2023. We wrapped the logger in white electrical tape to minimise solar bias and achieve better measurement accuracy ⁶³. For light intensity, we extracted data from Copernicus ERA5. The data are hourly measurements under direct clear sky radiation from the sampling site between April 2022 – February 2023, that we converted into µmol photons m^− 2 s^− 1.

We assessed denitrification data for statistically significant differences between species, between months per species and between light and dark incubations per month and species. All denitrification rate data were tested for normality via Shapiro-wilk tests, revealing that data were not normally distributed, even following transformations. Therefore, the non-parametric Kruskal-Wallis test was used followed by a Dunn’s test with Bonferroni p-value adjustment for post-hoc analysis.

Secondly, we employed a random forest model to identify key environmental variables that may influence denitrification rates across different coral species. The random forest regression model consisted of 500 trees, with 3 variables tried at each split. The variable importance for predicting denitrification was determined by the mean squared error (%MSE), where a high %MSE indicates that the variable is important, as if it were removed, the model’s error would significantly increase. As a follow-up to the random forest analysis, we plotted partial dependence plots (PDPs) for the top 3 most influential parameters per species. This provided insight into how these parameters interacted with denitrification, by displaying how changes in one parameter influences the predicted outcome, while averaging the influence of all other parameters.

Next, we correlated the denitrification rates of each species with its measured biogeochemical signatures, including isotope (δ15N and δ13C) and elemental data (C:N) of both the host and symbiont. For this, we used a non-parametric Spearman rank correlation, as data were not normally distributed following Shapiro-wilk testing. Data were pooled across the sampling year and cleaned prior to analysis. This included removing samples below the instrument’s detection limit of 0.015 mg, followed by identifying and excluding statistical outliers.

Lastly, the baseline trophic strategies and niche widths for the three zooxanthellate corals (S. pistillata, Acropora sp., and M. dichotoma) were estimated following the approach of Jackson et al., (2011) ⁶⁴ adapted for corals by Conti-Jerpe et al., (2020) ⁴⁷ and Fox et al., (2023) ⁶⁵. The isotope data (δ15N and δ13C) were pooled and cleaned in the same way as the correlation analysis (above), with the added criterion that only paired δ15N and δ13C values for both host and symbiont fractions were included. This resulted in sample sizes of n = 18 for S. pistillata, n = 14 for Acropora sp., and n = 16 for M. dichotoma. Following this, we visualised the isotopic niches of each species, by plotting the standard ellipse areas of the coral host and symbiont fractions. Next, standard ellipse areas (SEAb) were calculated directly from posterior sampling of theSIBER model, and niche widths were summarised using the posterior mode along with 50% and 95% credible intervals. Thirdly, we used bootstrapped estimates (n = 10,000) of SEAc overlap between the host and symbiont fractions as a proxy for trophic strategy (Conti-Jerpe et al., 2020). In addition, we calculated Layman’s metrics ⁶⁶ which serve as useful descriptions of data dispersion, offering insight into trophic diversity.

The software R (version 4.3.2) (R Core Team, 2023) was used to generate figures using packages ‘ggplot2’ ⁶⁷, ‘ggpubr’ ⁶⁸, ‘dplyr’ ⁶⁹, ‘RColorBrewer’ ⁷⁰, ‘gridExtra’ ⁷¹, ‘cowplot’ ⁷². Likewise, statistics were also computed in R, using packages ‘rstatix’ ⁷³, ‘dunn.test’ ⁷⁴, ‘tidyverse’ ⁷⁵, ‘randomForest’ ⁷⁶, ‘pdp’ ⁷⁷, ‘SIBER’ ⁷⁸ and ‘rjags’ ⁷⁹.

Denitrification was detected in all four Red Sea coral species (Fig. 1a). Averaged over the year, denitrification rates of the three zooxanthellate species (S. pistillata, Acropora sp. and M. dichotoma) were similar at 0.09 ± 0.16, 0.10 ± 0.16 and 0.08 ± 0.10 nmol N cm^− 2 h^− 1 respectively and therefore did not significantly differ (p > 0.05; Fig. 1a). However, the average denitrification rate of the azooxanthellate species T. coccinea was significantly higher than all three zooxanthellate species (p < 0.05), being 5-fold higher than both S. pistillata and M. dichotoma and 4-fold higher than Acropora sp. at 0.43 ± 0.61nmol N cm^− 2 h^− 1 (Fig. 1a).

Monthly denitrification rates significantly differed across the year in all four species (Kruskal-Wallis; S. pistillata: H = 18.2, p < 0.01; Acropora sp.: H = 14.7, p < 0.05; M. dichotoma: H = 13.2, p < 0.05; T. coccinea: H = 14.6, p < 0.01), with a trend of higher denitrification activity during the spring and summer months compared to the autumn and winter months (Fig. 1b - e). In S. pistillata, denitrification rates were significantly higher in April (0.20 ± 0.20 nmol N cm^− 2 h^− 1) and June (0.23 ± 0.26 nmol N cm^− 2 h^− 1) compared to August (0.02 ± 0.04 nmol N cm^− 2 h^− 1), October (0.03 ± 0.05 nmol N cm^− 2 h^− 1), December (0.01 ± 0.02 nmol N cm^− 2 h^− 1) and February (0.05 ± 0.10 nmol N cm^− 2 h^− 1), all p < 0.05 (Fig. 1b). In Acropora sp., rates were significantly higher in June (0.17 ± 0.10 nmol N cm^− 2 h^− 1) than in October (0.02 ± 0.03 nmol N cm^− 2 h^− 1) and December (0.03 ± 0.04 nmol N cm^− 2 h^− 1), all p < 0.05 (Fig. 1c). In M. dichotoma, denitrification rates were significantly higher in June (0.18 ± 0.10 nmol N cm^− 2 h^− 1), than in April (0.05 ± 0.07 nmol N cm^− 2 h^− 1) and August (0.03 ± 0.06 nmol N cm^− 2 h^− 1), all p < 0.05 (Fig. 1d). Lastly, in T. coccinea, denitrification rates were significantly higher in April (0.89 ± 0.86 nmol N cm^− 2 h^− 1) and June (0.69 ± 0.71 nmol N cm^− 2 h^− 1) than in December (0.02 ± 0.06 nmol N cm^− 2 h^− 1), all p < 0.05 (Fig. 1e).

Significant differences between denitrification rates measured in light and dark incubations per month were observed in all four species (Fig. 1b - e). Mostly, denitrification rates were significantly higher in the dark than in the light incubations, with this significant trend observed in 11 out of 13 of the significant pairings identified. More specifically, in S. pistillata denitrification rates were 23-fold higher in April in the dark versus the light (Kruskal-Wallis, H = 6.00, p < 0.01) and 0.06 in the dark versus 0 nmol N cm^− 2 h^− 1 in the light in October (Kruskal-Wallis, 5.54, p < 0.05; Fig. 1b). In Acropora sp., denitrification rates were 2-fold higher in June (Kruskal-Wallis, H = 3.94, p < 0.05) and 39-fold higher in August in the dark compared to the light (Kruskal-Wallis, H = 6.99, p < 0.01). In October, denitrification rates were 0.05 in the dark versus 0 nmol N cm^− 2 h^− 1 in the light (Kruskal-Wallis, H = 5.54, p < 0.05; Fig. 1c). In M. dichtoma, denitrification rates were 2-fold higher in June (Kruskal-Wallis, H = 4.81, p < 0.05) and 8-fold higher in October in the dark compared to the light incubations (Kruskal-Wallis, H = 6.90, p < 0.01; Fig. 1d). Lastly, in T. coccinea, rates were 17-fold higher in April (Kruskal-Wallis, H = 6.99, p < 0.01), 7-fold higher in June (Kruskal-Wallis, H = 5.77, p < 0.05) and 53-fold higher in October in the dark compared to the light (Kruskal-Wallis, H = 6.21, p < 0.05), while in August rates were 0.06 in the dark versus 0 nmol N cm^− 2 h^− 1 in the light (Kruskal-Wallis, H = 7.76, p < 0.01; Fig. 1e). However, the reverse trend was observed in 2 out of 13 significant pairings, where denitrification was higher in the light incubation compared to the dark incubation. This was observed only during February in Acropora sp. where rates were 382-fold higher in the light versus the dark (Kruskal-Wallis, H = 7.26, p < 0.01), and T. coccinea where rates were 0.3 in the light versus 0 nmol N cm^− 2 h^− 1 in the dark (Kruskal-Wallis, H = 7.20, p < 0.01;Figure 1c & e).

The measured environmental variables (nitrate, ammonium, DOC, water chl a and temperature) were used in a random forest model, to determine the most influential variables over denitrification rates, for each species (Fig. 2). The amount of denitrifying variation that the environmental variables explained, varied between species. For example, the environmental variables explained 12% of denitrifying variation in S. pistillata, 57% in Acropora sp., 38% in M. dichtoma and 50% in T. coccinea.

In S. pistillata, DOC, nitrate and ammonium were identified as the top three most influential variables over denitrification rates (Fig. 2a). DOC availability was the strongest predictor of denitrification in S. pistillata (15% IncMSE), with higher levels promoting denitrification (Fig. 2b). Low nitrate availability was the second most influential over denitrification rates in S. pistillata (12.5% IncMSE; Fig. 2a, 2c). Moderate ammonium availability also influenced denitrification rates of S. pistillata, yet to a much lesser extent (3.6% IncMSE; Fig. 2a, 2d). In Acropora sp., temperature, ammonium and water chl a were most influential over denitrification rates and to similar extents (Fig. 2e). High temperature was the top predictor of denitrification in Acropora sp. (14.4% IncMSE, Fig. 2e, 2f), followed by moderate ammonium availability (13.5% IncMSE, Fig. 2e, 2g) and water chl a (13.3% IncMSE; Fig. 2e, 2h). In M. dichotoma, denitrification rates were evenly influenced by moderate temperature (13.6% IncMSE, Fig. 2i, 2j), high DOC availability (13.3% IncMSE, Fig. 2i, 2k) and moderate water chl a concentrations (13.3% IncMSE; Fig. 2i, 2l). Lastly, in T. coccinea, low nitrate availability was the most influential variable over denitrification rates (21.1% IncMSE, Fig. 2m, 2n), followed by high DOC availability (12.7% IncMSE; Fig. 2m, 2o). High water chl a concentration (5.8% IncMSE, Fig. 2m, 2p) was ranked third but influenced denitrification to a lesser extent than the top two variables. Overall, each species was influenced by a unique combination of environmental variables. However, we also identified environmental drivers that are common across multiple species, such as DOC availability and temperature.

In S. pistillata, we found a significant positive correlation between denitrification rates (0.09 ± 0.16 nmol N cm^− 2 h^− 1) and symbiont δ13C (-14.16 ± 0.77‰; rho: 0.44; p < 0.05), as well as a significant negative correlation with symbiont δ15N (2.55 ± 0.54‰; rho: − 0.45; p < 0.05; Fig. 3). In Acropora sp., we found significant positive correlations between denitrification rates (0.10 ± 0.16 nmol N cm^− 2 h^− 1) and host δ13C (-15.34 ± 1.03‰; rho: 0.44; p < 0.05) and symbiont δ13C (-14.56 ± 0.48‰; rho: 0.51; p < 0.05; Fig. 3). In M. dichotoma, no significant relationships were detected between denitrification rates and biogeochemical signatures (p > 0.05; Fig. 3). Additionally, in T. coccinea, no significant relationships were detected between denitrification rates and biogeochemical signatures (p > 0.05; Fig. 3). Yet only one relationship (denitrification and host δ13C ) could be analysed since symbiont-related parameters were not applicable for this azooxanthellate species, and the other parameters had too few samples for a reliable analysis (Fig. 3).

We used Bayesian analysis of isotopic niches ⁶⁴ adapted for corals ^47,65 (Fig. 4) combined with Layman metrics of trophic diversity ⁶⁶ (Table 1) to quantify and compare the isotopic niches of the three zooxanthellate corals S. pistillata, Acropora sp. and M. dichotoma. The three species showed similarly broad isotopic niche areas (Fig. 4a). However, M. dichotoma exhibited the largest Bayesian standard ellipse area (SEAb) of the host fraction (2.87‰², 95% CI: 1.67–4.70), compared to S. pistillata (2.15‰², 95% CI: 1.31–3.51) and Acropora sp. (1.34‰², 95% CI: 0.86–2.54, as well as the largest Bayesian standard ellipse area (SEAb) of the symbiont fraction (2.34‰², 95% CI: 1.14–3.18), compared to S. pistillata (1.05‰², 0.66–1.73) and Acropora sp. (0.76‰², CI: 0.38–1.18; Fig. 4b). This was also supported by Layman’s metrics, where M. dichotoma had the largest NR, CR and TA (Table 1). However, M. dichotoma also had the highest NND and SDNND (Table 1), indicating higher variability between samples. Lastly, when we quantified the relative reliance of heterotrophy versus autotrophy as the percentage overlap between host and symbiont ellipse areas (corrected for sample size (SEAc), the three species all exhibited a mixotrophic feeding strategy, with no marked differences in mean SEAc (Fig. 4c). Although sample sizes were limited, the resampling errors stabilised and converged, rather than spanning the full possible range (0–100%), and patterns were consistent across coral species. Therefore, we believe our interpretations are conservative and robust, despite these constraints.

The trophic strategy of T. coccinea could not be quantified because this analysis requires isotope data from both host and symbiont fractions, and since T. coccinea lacks symbionts, the method was not applicable. Though, as an azooxanthellate coral, its feeding mode is already known to be entirely heterotrophic. We were, however, able to measure host parameters for T. coccinea and the mean host δ13C was – 24.01 ± 2.08‰, which was higher than the mean host δ13C of S. pistillata (− 14.72 ± 0.75‰), Acropora sp. (− 15.21 ± 0.60‰), and M. dichotoma (-15.3 ± 0.80‰; Figure S2). Unfortunately, host δ15N could not be reliably quantified because too few samples remained following data cleaning steps (detailed above in section 3.5 Data analyses).

The denitrification rates measured here are comparable to the ranges presented in previous studies on Red Sea corals, once appropriate conversions are applied (Fig. 1a) ³⁶. Rates also significantly varied throughout the year for all four species, with a general seasonal trend of higher rates in the spring/summer compared to the autumn/winter months of the year (Fig. 1b). The effect of seasonality on denitrification has not been investigated before in corals, but studies on other systems such as estuaries and marshlands have reported similar seasonal patterns to our study, finding higher denitrification rates and, in addition, higher denitrifier diversity in spring ^80,81. The seasonal trend observed in these former studies and our own can be explained by the sensitivity of the denitrification pathway to environmental conditions that fluctuate over a year (Figure S3).

Temperature emerged as the primary driver of denitrification in Acropora sp., and M. dichotoma, though its influence differed between the two species. In Acropora sp., denitrification rates increased linearly with temperature. This relationship reflects the general principle of higher temperatures stimulating microbial metabolism and activity ⁵¹. The same pattern was observed in Red Sea seagrass sediments ⁵². Additionally, among corals, elevated temperatures increase the activity of diazotrophs that govern N₂ fixation ⁸² introducing more in hospite N available for denitrification. Furthermore, a recent study by Rädecker et al. (2021) ⁸³ demonstrated that under high temperatures, the coral catabolises amino acids, again introducing more in hospite N available for denitrification. For M. dichotoma, however, denitrification rates peaked at moderate temperatures and decreased at higher temperatures These differences may reflect the response (and presence of) species-specific denitrifying microbiomes, as has been found between corals in previous work ⁸⁴. While there is no literature directly comparing the denitrifying microbial communities between Acropora sp. and M. dichotoma, studies reveal that Acropora sp. hosts a greater bacterial diversity than M. dichotoma ⁸⁵, indicating a potential for differences.

High DOC availability was identified as the top driver of denitrification in S. pistillata, and the second driver in M. dichotoma and T. coccinea. The positive relationship between DOC and denitrification has been widely observed in other environments i.e., sediments and freshwater systems ^86–89 and is attributed to the role of DOC as a key element supporting heterotrophic microbial growth and activity ⁹⁰. Therefore, our study provides evidence that denitrifiers in coral holobionts are also capable of utilising environmental DOC as a source of C, and may not solely rely on symbiont derived C. One study investigating the influence of DOC on octocoral denitrification reported contrasting results, where excess DOC (supplied as glucose) reduced denitrifier abundance by an order of magnitude in Xenia umbellata, but had no effect on Pinnigorgia flava ⁹¹. This discrepancy may be due to variations in the composition and concentration of DOC ⁹¹.

Against expectations, low nitrate availability was another key driver of denitrification. Nitrate is essential to the denitrification process, acting as an electron acceptor for denitrifying bacteria, which sequentially reduces it to dinitrogen gas ⁹². Consequently, one may expect that the more available nitrate, the more is taken up by the coral symbionts (among zooxanthellate corals) and the higher the denitrification rates. However, the opposite relationship was apparent in our study, with denitrification rates linked to low nitrate availability (Fig. 2). This may simply be because denitrification activity on an ecosystem scale is sufficiently high to reduce nitrate concentrations in the surrounding water column. Alternatively, a study by El-Khaled et al. (2020) ³⁵ demonstrated that N cycling is nuanced, with opposing pathways like N₂ fixation and denitrification increasing together under higher environmental N. In the oligotrophic Red Sea, where nitrate stays low year-round (0.2–1.3 µmol L-1; Figure S3), N₂ fixation may rise in response to low N while denitrification simultaneously increases. This suggests denitrification becomes dominant only at higher nitrate concentrations, whereas at lower concentrations it may co-occur with N₂ fixation as previously found ³⁵. To fully explain this finding, however, further investigation into denitrification and N₂ fixation activity in response to a range of DIN concentrations is required.

While the random forest analysis highlighted key environmental drivers of denitrification, a portion of denitrifying variability remained unexplained by the environmental parameters in our study. For example, 12% of denitrifying variation was explained by environmental conditions for S. pistillata, 57% for Acropora sp., 38% for M. dichotoma and 50% for T. coccinea (Fig. 2), emphasising the potential influence of additional factors beyond the scope of our study.

Furthermore, denitrification rates were significantly higher during dark incubations than under light conditions in the acetylene assays (Fig. 1b - e). This likely reflects reduced oxygen availability in the dark where the absence of photosynthesis and continued respiration creates conditions that favour denitrification by anaerobic microbes ⁹³. Therefore, our study also confirms that denitrification is a more active pathway under low oxygen conditions, aligning with findings of former studies ⁴¹.

By correlating denitrification rates with species-specific biogeochemical signatures, we found a significant negative correlation between denitrification rates and symbiont δ15N in S. pistillata (Fig. 3). This is an indication that denitrification may enhance or maintain internal N-limitation within the coral holobiont, as expected ³. However, this relationship was not observed for the other zooxanthellate species Acropora sp., and M. dichotoma, challenging the functional role of denitrification in these species, which we expand upon later. Furthermore, we found a significant positive relationship between denitrification rates and symbiont δ13C in Acropora sp. and S. pistillata, and an additional significant positive correlation between denitrification and host δ13C in Acropora sp. (Fig. 3). These findings provide evidence that denitrifiers utilise autotrophically derived-C, as seen in former studies ^36,38,91. However, we of course also provide that evidence that denitrifiers utilise environmental-DOC (Fig. 2). Naturally, this raises the question of how host trophic strategy might further influence denitrification activity.

We also examined the influence of host trophic strategy on denitrification rates. Unexpectedly, denitrification rates of the fully heterotrophic species T. coccinea were significantly higher than those of all other species in our study, being 5-fold higher than both S. pistillata and M. dichotoma and 4-fold higher than Acropora sp. (Fig. 1a). This result contradicts our initial hypothesis where we predicted higher denitrification rates in the more autotrophic species based on findings from former studies ^36,38,91. Therefore, our findings suggest that environmental C may even fuel denitrification at a faster rate than photosynthetic C, given the exceptionally high rates measured in T. coccinea. However, such high rates may also be attributed to other factors beyond the C source. For example, T. coccinea may host a more diverse and efficient denitrifying community ⁸⁴. We also need to measure the denitrification activity of additional heterotrophic species to see whether this finding is unique to T. coccinea or applies broadly to heterotrophic species. Furthermore, since our species were all mixotrophic (Fig. 4), we could not assess the denitrification activity in more autotrophic species and would suggest future work to include a species that exhibits a greater reliance on autotrophy. However, since we had to pool the isotopic data over the year due to limited sample sizes, we may have missed seasonal shifts towards greater autotrophy in our species. Therefore, we would recommend higher sample sizes at each seasonal time point.

Furthermore, by mitigating excess N, denitrification is proposed to sustain N-limitation within the coral holobiont, which is critical to the stability of the coral-algal symbiosis ³. However, the occurrence of denitrification in an azooxanthellate coral, challenges this proposed functional role. Our findings suggest that denitrification is not an adaptive trait among azooxanthellate corals, but rather a passive, opportunistic response to a suite of environmental conditions that favour its activity. This interpretation aligns with previous research on denitrification in octocorals ⁹¹ and tropical scleractinian corals ⁴⁰.

Having addressed the specific research questions of our study, it is important to interpret these results in the context of the Red Sea’s unique characteristics for a broader perspective of their ecological relevance. The Red Sea is one of the warmest and saltiest seas on Earth, exhibiting strong temporal and spatial gradients ⁴³. Spatially, the temperature, nutrient availability and chl a concentration differs between the north and south of the Red Sea, with the highest temperature, DIN and chl a occurring in the south, and decreasing northwards ^43,94, the nutrient availability in the shallow zone is very low and increases with depth. Our study was conducted in shallow waters of the central Red Sea where environmental conditions were found to have a strong influence over denitrification activity. Broadly, temperature and nutrient availability emerged as key drivers of denitrification. Based on our findings, we anticipate that corals of the same species may exhibit spatial variation in denitrification activity across the Red Sea in response to environmental conditions. For example, denitrification activity in Acropora sp. may increase southwards as temperature and DIN increases. However, caution should be taken when extrapolating our seasonal findings beyond the Red Sea, as seasonal dynamics differ among coral reef regions worldwide. Some regions experience weak seasonality such as the equatorial Indo-Pacific. Therefore, it is likely that the same spike in denitrification activity in the spring/summer seasons as measured in our study, may not be seen globally.

Our study shows that the denitrification pathway is seasonally variable in corals, exhibiting generally higher rates in the spring and summer months compared to the autumn and winter months (Fig. 1b), which can be explained by the high sensitivity of denitrification to environmental conditions (Fig. 2). Our study identified species-specific environmental drivers of denitrification (Fig. 2). The top driver in S. pistillata was DOC, providing evidence that denitrifiers do not exclusively rely on photosynthetic C for energy (Fig. 2). For Acropora sp., and M. dichotoma, the top driver was temperature, although its influence differed between the two species (Fig. 2). In Acropora sp., we found a linear relationship between denitrification and temperature, whereas in M. dichotoma, denitrification activity increased with temperature until an upper threshold, beyond which it declined (Fig. 2). For T. coccinea, we unexpectedly identified low nitrate availability as a driver of denitrification (Fig. 2), yet this is likely the effect not the cause of high denitrification activity, or this is due to the co-occurrence of denitrification with N₂ fixation when nitrate levels are low. Our study also showed that for all four species, denitrification activity was higher under dark conditions where oxygen levels are reduced (Fig. 1b), thus favouring anaerobic microbial activity ⁹³ as previously found for other reef substrates ⁴¹. Furthermore, our study demonstrated that denitrification is also influenced by the internal nutrient dynamics of the coral holobiont, as the positive relationship between denitrification and host ∂13C in S. pistillata, and both host and symbiont ∂13C in Acropora sp. (Fig. 3), suggest that denitrifiers utilise symbiont-derived C as shown in former studies ⁹⁵. Lastly, we found that denitrification rates were affected by host trophic strategy, finding significantly higher denitrification in the azooxanthellate, fully heterotrophic coral T. coccinea, being 5-fold higher than both S. pistillata and M. dichotoma and 4-fold higher than Acropora sp. (Fig. 1a) that we determined were all mixotrophic (Fig. 4). This suggests that environmental C may even fuel denitrification at a faster rate than photosynthetic C, or points to the influence of distinct denitrifying communities that may exhibit different rates. However, the exceptionally high denitrification rates measured in the azooxanthellate species challenge the functional significance of N removal, since no coral-algal symbiosis is present to necessitate such regulation. Therefore, we suspect that denitrification may play a more passive role than previously suspected among Red Sea corals, reinforcing findings from coral-associated denitrification research in other geographic regions ⁴⁰ and on octocorals ⁹¹. Overall, our study established a valuable baseline for seasonal denitrification rates in the Red Sea, and an understanding of its environmental drivers across four species. This foundation offers a springboard for future research to explore the influences of environmental change on N cycling in greater detail.