Deoxygenation lowers the thermal threshold of coral bleaching

Exposure to deoxygenation from climate warming and pollution is emerging as a contributing factor of coral bleaching and mortality. However, the combined effects of heating and deoxygenation on bleaching susceptibility remain unknown. Here, we employed short-term thermal stress assays to show that deoxygenated seawater can lower the thermal limit of an Acropora coral by as much as 1 °C or 0.4 °C based on bleaching index scores or dark-acclimated photosynthetic efficiencies, respectively. Using RNA-Seq, we show similar stress responses to heat with and without deoxygenated seawater, both activating putative key genes of the hypoxia-inducible factor response system indicative of cellular hypoxia. We also detect distinct deoxygenation responses, including a disruption of O2-dependent photo-reception/-protection, redox status, and activation of an immune response prior to the onset of bleaching. Thus, corals are even more vulnerable when faced with heat stress in deoxygenated waters. This highlights the need to integrate dissolved O2 measurements into global monitoring programs of coral reefs.

Staghorn' undetermined species of Acropora (hereafter, Acropora sp.), which had been propagated in a mesocosm aquarium tank (at B&B Aquakultur, Konstanz, Germany) under consistent parameters for > 10 years (Table S1) to conceive a zero ecological footprint coral experiment. Fragments can therefore be considered genetically identical and appropriate for experimental examination of putative interactive heat and deoxygenation stress effects for this colony and hence bleaching phenotype (sensu 25 ). The undetermined species was originally sourced off Bali, Indonesia in 2010, where the maximum monthly mean (MMM) sea surface temperature (SST) is ~ 30 °C. Experimental coral fragments were fragmented from the parent colony in two subsequent batches (a first CBASS run without deoxygenation H0 and then a second run with deoxygenation Hd) a week prior to CBASS assaying. A total of 32 fragments ranging from 2 to 4 cm in length were used for each experimental CBASS run and positioned upright on ceramic plugs ( Figure S1). Coral nubbins were positioned in a rack in seawater in a polystyrene box for the short transportation from aquaria rearing facilities to the University of Konstanz for CBASS assaying (31st May 2021, heat only, H0; 4th June 2021, heat with deoxygenation, Hd).
Experimental setup. We used the Coral Bleaching Automated Stress System (CBASS) to conduct shortterm acute heat stress assays (as described in Voolstra et al. 29,47 ), modified to incorporate a seawater reservoir to finely regulate dissolved O 2 (DO) and pH that enters the CBASS (see below). The system was set up in a wet laboratory (University of Konstanz) near the aquarium holding tanks to minimise coral transport time. Briefly, the core CBASS consisted of four replicate 10L flow-through tanks equipped each with a temperature controller, chillers, a heater, and a 165 W full spectrum LED aquarium light ( Figure S1). Temperature profiles for each tank were programmed by a temperature controller (InkBird ITC-310T-B) connected to 2 thermoelectric chillers (IceProbe T, Nova Tec) and 1 aquarium heater (Titanium, 200W; Schego). For each tank, photon irradiance was delivered by a dimmable 165 W full spectrum LED aquarium light (Galaxyhydro) to match the rearing aquaria light fields (~ 240 μmol photons m −2 s −1 using half blue and half white LED spectral range intensities), as quantified using an Apogee MQ-510 Underwater Quantum meter. In each tank, a HOBO pendant temperature logger was positioned on the opposite side to the heater to provide consistent temperature recordings throughout the experiment ( Figure S2). Regulation of O 2 and pH levels was through a 140L sealed seawater reservoir connected to CO 2 , compressed air, and O 2 gas cylinders. Customised gas regulators were fitted to adjust the DO and CO 2 entering the seawater through aquarium air stones. DO and pH probes (WTW Multi 3630 IDS) were introduced into a pipe of the flowing reservoir water to constantly read O 2 and pH to finely regulate gas contributions. The reservoir was first flushed with N 2 to lower DO to < 2 mg L −1 , where after CO 2 was flushed through the deoxygenated seawater to offset changes in pH, as described in Alderdice et al. (2021). DO and pH in the reservoir and experimental tanks were also measured hourly during the 6 h heat stress phase ( Figure S3) using robust O 2 and pH probes (OXROB10 and PHROB-PK8, Pyroscience) connected to a FireStingPro fibreoptic multi-meter (Pyroscience). We note that the starting levels of pH between each CBASS run (H0 and Hd) were largely identical (± < 0.05). Similarly, inherent lowering of pH levels by 0.2 units under heating is within the natural diel range experienced by corals on reefs 30,51 as well as within the diel range in the holding aquarium (Table S1). Water in the reservoir was pre-heated to the control/baseline temperature with aquarium heaters (Titanium, 200W; Schego) linked to a programmable temperature controller (InkBird ITC-310T-B). Before the start of the experiment, the four CBASS tanks were slowly filled with pre-heated deoxygenated seawater to avoid re-oxygenation until starting of the actual CBASS assays. A transparent lid was placed over experimental tanks to reduce the air space above the water surface, and any potential air gaps were sealed using Sanitop-Wingenroth www.nature.com/scientificreports/ 'plastic-fermit' ( Figure S1). Seawater from the reservoir was delivered during the CBASS assays by the flowthrough system at a flow rate of 25 mL min −1 to achieve seawater turnover (1.5 L/hour) for each experimental tank. Seawater for long-term aquaria rearing and in the CBASS assays had a salinity of 35 PSU and was prepared using the same salt (Fauna Marin, Germany). Out of the 4 CBASS tanks, a control tank was maintained at 30 °C for the duration of the experiment to exhibit the maximum monthly mean (MMM) temperature of the parent colony origin (Bali, Indonesia; MMM approximately 28.5-29.5 °C pending on exact sampling location) which was determined from the NOAA Coral Reef Watch 5 km database 52 . MMM was used as the baseline/control temperature rather than the long-term aquarium rearing temperature of 27 °C. This is in line with the notion that corals exhibit locally adapted thermal thresholds 50,53,54 and following recently established CBASS protocols in which short term assays were shown to be representative of longer-term experiments 48,49 . In support of this notion, we trialled an initial CBASS run where the baseline temperature was set to a lower temperature of 27 °C and observed a projected ED50 temperature threshold (ED50 = 36.05 °C) that exceeded the highest treatment temperature (36 °C); conventionally, the ED50 should ideally lie below the highest treatment temperature for prediction accuracy ( Figure S4). Our trial emphasizes how corals seem to retain their evolutionarily acquired thermal threshold, even under long-term aquaria rearing conditions; this further corroborates previous CBASS studies that use MMM as the baseline/ control temperature 47,49,50 . Using the MMM temperature (30 °C) as the baseline/control, we then applied heat treatments to the other 3 tanks accordingly (sensu 47,49 ) at 33 °C (MMM +3 °C), 36 °C (MMM +6 °C), and 39 °C (MMM +9 °C). All temperature treatments were heated over a 3 h period, held for another 3 h at target temperature, and then decreased back to baseline (30 °C) over 1 h, where they were retained for the remainder of the 18 h experiment ( Fig. 1a; Figure S2). For the combined heat and deoxygenation CBASS run, we used deoxygenated seawater of ~ 2 mg L −1 O 2 during the heating ramp and heat-hold but allowed for re-oxygenation during the overnight recovery phase ( Figure S3). Fragments from all CBASS assays were sampled for RNA-Seq by flashfreezing in liquid N 2 after the 6 h heat stress phase to capture the transcriptional response to either only heating (H0) or to heating and deoxygenation (Hd). Dark-acclimated photosynthetic efficiencies, i.e. the maximum PSII quantum yield (F v /F m ), were measured on the remaining fragments following 1 h dark-acclimation at the start of the overnight recovery phase. As such, for both CBASS runs (H0 and Hd), a total of 7 clonal ramets were assayed in each temperature tank, 4 used for RNA-Seq analysis (sampled after 6 h of the CBASS profiles) and the remaining 3 used for measuring dark-acclimated F v /F m (measured after 7 h, Fig. 1).

Coral bleaching assessment.
Measurements of the dark-acclimated maximum photosynthetic efficiency of photosystem II (F v /F m ) were used to determine ED50-based thermal tolerance limits as a proxy for coral bleaching (sensu 29,49 ). Photosynthetic efficiency was previously found to be most indicative of coral thermal tolerance when comparing coral whitening, chlorophyll a, host protein, algal symbiont counts, and algal type association 47 . For all F v /F m measurements, coral fragments were retained under darkness for at least 1 h prior to measurements using a pulse-amplitude modulated fluorometer (Mini PAM II; Walz). One measurement per fragment was taken to avoid PAM saturating light pulses potentially inducing light artefacts. Measurements were consistently taken on the mid side of the upright fragment to avoid the tip or base of the fragment that would have experienced different light exposure (e.g., 55 ), putatively influencing F v /F m values. Optimized PAM settings (following 56 ) were used as follows: Signal Gain and Damping of 2, Measuring Light Curve Intensity (Mi) of 5, Actinic Light Factor of 1, Saturation Intensity and Width of 3 and 0.8, respectively. Of note, recorded readings were taken when the minimum fluorescence (F 0 ) was > 200 (instrument units) on fragments. Temperature tolerance thresholds were determined for both H0 and Hd as the mean temperature (across all ramets) at which F v /F m dropped to 50% of the value measured at baseline temperatures, here defined as the Effective Dose 50 or ED50 49 using the DRC package in R 57 . Statistical differences among treatment-specific ED50s were assessed via a Welch's unequal variances (one-tailed) t-test with replicate-based treatment ED50s as the response variable and treatment as the respective factor. The script and data are available at https:// github. com/ reefg enomi cs/ CBASS_ hypox ia. Since all fragments were sourced from a single parent colony, we are testing for a significant treatment effect between ramets of a genet. Hence, the true effect size might be larger if response is measured across biological population replicates.
Coral fragments were photographed using an Olympus TG-6 digital camera alongside a 'Coral Watch Coral Health Chart' colour reference for visual bleaching assessments, before being flash-frozen in liquid N 2 and stored at -80 °C until further processing. Photographs were taken after the 6 h stress phase (T1) and also prior to the CBASS experiment at T0 using the same camera settings, working distance, and illumination level (all images are available at https:// zenodo. org/ record/ 64972 21; 58 ). We devised a bleaching index score (BIS) based on the previously established 'colour score' 59 and 'relative bleaching score' 60 . For this, colour saturation of the tissue for each coral fragment was ranked on a 6-point scale via visual assessment to the colorimetric reference card with 6 representing maximum saturation and 1 representing no pigment (i.e., completely white), respectively. Bleaching was considered to have occurred when a decrease in colour saturation of two or more values was observed prior or after the stress exposure 59 . Instances where we found inconsistent coloration across the fragment, two scores (one for the paler part and one for the more colourful part) were determined and averaged. Bleaching Index Scores (BIS) were determined by 5 independent assessors, who did not have prior knowledge of the experiment and who scored the coral fragments in a randomised manner. We chose to include BIS, as BIS accounts for a coral holobiont compound phenotype (i.e. colour), whereas F v /F m relates to the photochemical efficiency of the algal symbionts and may not encompass host-specific effects to deoxygenation. Standardized thermal temperature thresholds were determined as for F v /F m , but this time ED50 scores designated the inferred temperature at which the BIS dropped to or below 50% of the value at baseline temperatures (ED50; 29 RNA extraction and sequencing. Total RNA was extracted using the Qiagen RNeasy mini kit using the QIAcube Connect. Frozen coral fragments were immersed in RLT buffer (Qiagen) within zip-lock bags and tissue was air-picked from the skeleton using airflow from a sterile, 1,000 µL pipette tip connected via a rubber hose to a benchtop air pressure valve for a maximum of 3 min. A 400 µL aliquot of the resulting tissue slurry was centrifuged at full speed for 3 min and then 350 µL of the supernatant of each respective sample were loaded into 2 mL tubes and inserted into the QIAcube Connect to run the RNeasy extraction protocol. RNA concentrations were assessed using the Qubit RNA Broad-Range assay kit on the Qubit 4 fluorometer (Invitrogen). RNA quality was evaluated by capillary electrophoresis via the presence of intact 18S and 28S ribosomal RNA bands using the QIAxcel RNA quality control kit v2.0 on the QIAxcel Advanced system (Qiagen De novo transcriptome assembly. Demultiplexed reads were quality-checked using FASTQC 61 before and after read trimming with trimmomatic v0.38 62 to remove Illumina adapters, low quality reads, and reads www.nature.com/scientificreports/ shorter than 50 bp. Each sample retained > 96% of the paired-end read counts (Table S2). For an initial analysis, a de novo transcriptome was assembled using SOAPdenovo-Trans using the default 23-kmer length 63 with reads from all samples. To gauge the proportion of assembled contigs across taxa of the coral holobiont (e.g., Cnidaria, Dinophyceae, bacteria, virus, and fungi), the transcriptome was queried against the BLASTn database. Given the relatively high proportion of contigs assigned to bacteria (Table S3), the transcriptome was re-assembled using only reads of non-bleached samples (i.e. excluding 36 °C and 39 °C samples), which lowered the proportion of bacteria contigs from 16 to 5%. The transcriptome was further filtered to only consider contigs with a length ≥ 500 bp, comprising a total of 20,115 contigs of which 17,975 could be assigned to Cnidaria 58 . To assess the number of distinct cnidarian loci, we clustered the 17,975 contigs assigned to Cnidaria with a similarity threshold of 90% using CD-HIT-EST 64 . This procedure returned 17,960 transcripts, suggesting that the majority of cnidarian contigs reflects distinct genes. The assembly comprising only non-bleached samples also had a higher scaffold N50 of 1,946 compared to 1,442 for all samples (considering contigs ≥ 500 bp (Table S4). We consider the transcriptome of 20,115 contigs the reference transcriptome on which all expression analyses were based on, accessible at https:// zenodo. org/ record/ 64972 21 58 ; we did not remove putative non-cnidarian loci to account for cross-mappings of RNA-Seq reads. For gene expression analyses, trimmed paired-end reads from all samples were mapped to the reference transcriptome using Bowtie 2 v2.3.5.1 65 . Resultant mapping files were processed with SAMtools 66 to generate bam files to quantify mapped contigs (i.e., transcripts) using Salmon 67 .
All samples had at least 5 million mapped reads for read 1 and so were included in the downstream analysis (Table S2).
RNA-Seq analysis. Tximport 68 was used to import count data from Salmon into the R environment for transcript-level count estimation. To visualise general patterns of gene expression, variance-stabilising transformed counts were used for the Principal Component Analysis (PCA), plotted using the R package ggplot2 69 . Count data was then filtered to include only cnidarian transcripts for differential expression analysis using the package DESeq2 70 . The number of reads assigned to Dinophyceae was insufficient to perform differential expression analysis. Normalization for sequencing depth was applied through the DESeq2 dispersion function. Wald testing for significance difference of coefficients with a negative binomial general linear model (GLM) was applied in DESeq2. P values were corrected using Benjamini-Hochberg (BH) at a default false discovery rate (FDR) cut-off of 0.05 (Table S5, Table S6). Venn diagrams of the common and unique differentially expressed (DE) transcripts between comparisons were created using the R package ggVennDiagram 71 . The EggNOG v5.0 ortholog database by EMBL was used to annotate the reference transcriptome 72 . To assess DE transcripts annotated to genes commonly involved in the coral heat stress response, DE transcripts were screened for annotations associated with the following processes: calcium homeostasis, heat shock proteins (protein homeostasis), cytoskeleton rearrangement, cell death, mitochondria activity suppression (or glycolysis promotion), and increased reactive O 2 and N 2 species (summarised by 17 . Shared DE transcripts across conditions (determined by Venn diagrams) were screened against the list of heat stress associated genes of interest (see above) and the log2 fold change was visualized across comparisons in a heat map in the R package ggplot2. Genes also reported from coral hypoxia stress responses were indicated using the symbol "*" in the heat map 25,34 . FPKMs (i.e., fragments per kilobase per million mapped reads) were estimated using DESeq2 for genes known to stabilise or suppress HIFα protein. These genes are key to activating the hypoxia stress-mitigating gene network and were plotted using the R package ggplot2 to assess expression patterns across temperatures under normoxic (H0) or deoxygenated (Hd) conditions. FPKM expression estimates were summed up for transcripts that shared the same annotation (e.g., HSP90) sensu 25,34 . Gene ontology (GO) enrichment analyses for all DE transcript lists were performed using the R package TopGO 73 with a recommended weighted-Fisher P value cut-off of < 0.001. All scripts can be accessed on GitHub at https:// github. com/ reefg enomi cs/ CBASS_ hypox ia. www.nature.com/scientificreports/ sequences across samples and within the SymPortal database. These profiles were characterized by unique combinations of defining intragenomic variants (DIVs). Output data from SymPortal was then plotted in R. SymPortal output files are available at https:// zenodo. org/ record/ 64972 21 58 .

Results
Deoxygenation lowers the thermal threshold of coral bleaching. To assess the effect of deoxygenation on coral bleaching, we ran short-term heat stress assays using the CBASS under normoxic (6 mg L −1 O 2 ) and deoxygenated (2 mg L −1 O 2 ) conditions. We subsequently determined the effective dose 50 (ED50) thermal threshold to obtain a standardised measure of coral thermal tolerance based on either BIS (termed ED50 BIS ) or measured photosynthetic efficiencies (termed ED50 Fv/Fm ). Based on BIS, coral fragments exhibited a greater extent of paling at lower temperatures when exposed to combined heating & deoxygenation conditions (Hd) compared to heating alone (H0), whereby ED50 BIS was 1.09 °C lower under Hd compared to H0 alone (ED50 BIS Hd = 36.02 °C ± 0.39 vs. H0 = 37.11 °C ± 0.1 [mean ± SE], respectively; Figure S5). This difference was significant despite the small sample size and temporal constraints of short-term assays to manifest treatment differences (t = 2.41, df = 4.9, P value = 0.03 . It is important to highlight that these ED50s are not an indicator of in situ thermal thresholds, but rather a diagnostic proxy of thermal limits based on 18-h thermal profiles. ITS2-based Symbiodiniaceae profiling using the SymPortal analytical framework 77 confirmed uniform symbiont algal assemblage across samples, with samples associated with the two genera-Cladocopium (C21), which comprised the majority of sequences, and Symbiodinium (A1; Figure S6). To explore for any consistencies or differences between normoxic and deoxygenated heat stress, we evaluated the expression profiles of 20,115 assembled transcripts using RNA-Seq. Principal component analysis revealed that samples were largely separated by the heat stress temperature, with PC1 explaining 74% of the variation (Fig. 1d). In addition, we found sub-clustering of samples at 30 °C and 33 °C versus 36 °C and 39 °C (Fig. 1 c, d). Further, samples at 39 °C under deoxygenated heat stress conditions were most distinct and separated from 39 °C heat stress samples suggesting that deoxygenation exerts a more pronounced effect under high heat stress. Lastly, greater dispersion of samples, i.e. a higher variance, was consistently demonstrated by heat and deoxygenation samples compared to heat only samples across all temperatures.

Corals exhibit a similar response to heat stress with and without deoxygenation.
To elucidate how the heat stress response of corals exposed to heating (H0) compares to heating & deoxygenation (Hd), we assessed differential expression between each heating temperature (33 °C, 36 °C, 39 °C) and the baseline temperature of 30 °C for the normoxic and deoxygenated CBASS runs. Most differentially expressed (DE) transcripts (FDR < 0.05) were common between H0 and Hd at all heating temperatures with 62%, 80%, and 74% being shared at 33 °C, 36 °C, and 39 °C, respectively (Fig. 2a). Both the common and the total number of DE transcripts highlighted similar magnitudes of expression response for H0 and Hd, reflecting the PCA clustering predominantly by temperature irrespective of oxygen levels (Fig. 1d).
Among the commonly expressed EggNOG-annotated transcripts, many were involved in processes typically associated with coral heat stress responses including calcium homeostasis, heat shock proteins (protein homeostasis), cytoskeleton rearrangement, cell death, mitochondrial activity suppression, and increased RNS and ROS 17 . Of these, we found similar log 2 -fold changes for H0 and Hd at each heating temperature; for example, heat shock protein 70 (HSPA4L) exhibited greater log 2 -fold changes with higher temperatures (Fig. 2). We note that approximately half of the examined genes typically expressed under heat stress are shown to be activated under deoxygenation stress alone for other species of Acropora 25,34 . This includes the HIF-target genes pyruvate dehydrogenase kinase (PDK2) that promotes anaerobic respiration via the glucose to lactate pathway as well as heme oxygenase (HMOX1) that assists antioxidant activities (indicated by asterisks in Fig. 2b). Further, fragments per kilobase per million mapped reads (FPKM)-based expression estimates showed that the O 2 -sensitive hypoxia-inducible factor subunit (HIFα) and genes which either suppress (prolyl hydroxylase domain 2/4, EGLN1/P4HTM; HIFα inhibitor, HIFAN) or stabilise (heat shock protein 90; HSP90B1) HIFα proteins all followed similar expression dynamics under H0 and Hd conditions, except for P4HTM, which was significantly higher in Hd than H0 at 33 °C (log 2 -fold change (FC log2 ) = 1.66, FDR < 0.05). Although not significantly differentially expressed, the other two suppressors, EGLN1 and HIFAN, exhibited greater expression under Hd at 33 °C and 36 °C. Notably, compared to the baseline temperature, EGLN1 expression was significantly higher for both 33

Discussion
Coral bleaching is a highly networked biological process, which reflects the outcome of multi-level and -scale stress exposures 22,38 . However, how the coral heat stress response is affected by O 2 availability is not well understood, despite reefs becoming increasingly subject to deoxygenation under ocean warming-including water column stratification-and elevated biological O 2 demands 1,2,8 . The Coral Bleaching Automated Stress System (CBASS) is a standardised system 47 that allows for explicit testing of how individual and/or combined environmental factors affect thermal stress tolerance (i.e., bleaching) by assessing coral functioning under heat stress alone and in combination with an additional stressor, in this case deoxygenation. In line with recent CBASS studies 29,[47][48][49]79 , we extended the application of this approach to detect subtle, yet important differences in thermal sensitivity under differing O 2 levels. Specifically, we found deoxygenation to lower the thermal bleaching threshold in Acropora coral by 1 °C and 0.4 °C according to bleaching index score-or photosynthetic efficiency-based ED50 thermal threshold modelling, respectively. Such difference is remarkable in the context of the constraints of a short-acute thermal stress assay based on 18-h thermal profiles and suggest that physiological impacts from altered O 2 availability can manifest rather rapidly under heat stress. Whilst hypoxia stress-induced genes key to the HIF-HRS are differentially expressed under heat stress alone, a distinct deoxygenation response is also evident.

Deoxygenation lowers thermal thresholds of coral bleaching. Previous studies have demonstrated
that deoxygenation alone can drive coral bleaching and subsequent mortality 25,26 , suggesting that accelerating ocean deoxygenation under climate change and eutrophication could drastically impact the ability of corals -as aerobic metazoans -to respond to heat stress 7,22,24,25 . Based on established coral bleaching proxies of BIS 59,60 and F v /F m (e.g., 80 , the Effective Dose 50 (ED50; 47,49 ) metric affirms that deoxygenation lowers thermal thresholds for the species of Acropora examined here. Under 36 °C heating stress and deoxygenation, coral samples (Hd) exhibited greater visual paling with a lower BIS and a reduction in the ED50 BIS and ED50 Fv/Fm by 1 °C and 0.4 °C, respectively, compared to normoxic conditions (Fig. 1, Figure S5). With current climate models predicting at least a + 1.5 °C warming 81,82 , the lowering of the coral thermal threshold by a whole 1 °C (ED50 BIS ) under deoxygenation is a drastic result that warrants careful consideration in projecting the future survival of coral reefs. It is important to note that the ED50 Fv/Fm (in contrast to the ED50 BIS ) was not significantly different between O 2 treatments, which might be partially attributed to the low replicate number and the inherent constraints of short-term assays, where any measured response variable has to exhibit differences within 6 h. However, significance levels were much lower for ED50 Fv/Fm compared to ED50 BIS , likely also reflecting that BIS accounts for a coral holobiont compound phenotype (i.e. colour), whereas F v /F m relates to the photochemical efficiency of the algal symbiont and may not encompass host-specific effects to deoxygenation. Whilst temperature was the main driver of the transcriptional response, we found a subtle yet equally important difference in response to environmental O 2 levels that needs to be accounted for when researching factors determining coral bleaching susceptibility (Fig. 1d). However, more experiments will be needed to ensure that our observations on the impact of deoxygenation on coral stress thermal thresholds are robustly drawn out. In particular, whether the observed relative minor difference in ED50 may be related to a relatively "mild", i.e. sublethal rather than lethal, deoxygenation stress of 2 mg L −1 O 2 . Here, dose response experiments of both O 2 and heat need to consider recent evidence for variable 'oxy-regulation' capacity amongst coral species that inherently determines the poise between sublethal and lethal physiological responses to reduced O 2 availability 23 .
Under heating, the projected ED50 Fv/Fm thermal threshold values for our Acropora sp. were about 6 °C higher than the MMM temperature of the source of origin (ED50 H0 = 36.12 °C, MMM ~ 30 °C). This is similar to previous observations for Acropora cervicornis (Florida; ED50 Fv/Fm = 35.88 °C, MMM = 30 °C; 79 ), but lower than for Stylophora pistillata from the Central Red Sea (ED50 Fv/Fm = 38.27 °C, MMM = 30.75 °C; 29 ), considered to be among the most thermally resilient corals 83 . Our results therefore add to the growing evidence that CBASS can resolve differences in thermal susceptibility across different taxa 29,47,50 or environments 29,47,49, Notably, caution needs to be exercised that the ED50 thermal tolerance thresholds do not equate to long-term in situ thermal resilience, but rather to thermal limits within the framework of 18-h short term acute heat stress assays.
Corals exhibit a similar response to heat stress with and without deoxygenation. In agreement with the overall expression patterns (Fig. 1d), a large percentage (62-80%) of differentially expressed genes was consistently observed under normoxic and deoxygenated heat stress (Fig. 2). Such similar magnitudes of expression difference likely highlight how the response to heating is associated with the response to hypoxia stress [35][36][37] . Notably, a common gene repertoire of stress response genes was previously proposed, where corals subjected to deoxygenation alone expressed genes typical of heat stress, such as different heat shock proteins and those www.nature.com/scientificreports/ involved in altering mitochondrial activity, lipid uptake, immunity, structural reorganisation, ROS-handling proteins, and cell apoptosis 17,25,34 . In line with this, we found the key hypoxia-induced transcriptional factor, HIFα, and its suppressor, EGLN1/PHD2, to be expressed under normoxic heat stress at lower temperatures (30 °C and 33 °C), which followed a similar expression pattern under deoxygenation heat stress. Consistent with our previous studies of deoxygenation stress alone 25,34 , Acropora sp. exhibited an increased expression of EGLN1 (PHD2) with increasing temperature, supporting the notion that overexpression of EGLN1/PHD2 could signal for greater susceptibility to hypoxia or heat stress-induced bleaching. By comparison, an increased PHD expression under prolonged hypoxia stress is known to reactivate proteasomal degradation of HIFα by using intracellular O 2 that is no longer being consumed in mitochondrial aerobic respiration 84 . Thus, the ability of HIFα and PHD2 to register cellular hypoxia is desensitised, acting as a negative feedback mechanism of the HIF-HRS to protect cells against excessive cell death in an attempt to adjust to the hypoxic state 84,85 . As such, previous heat stress studies have likely also inherently reported a conflated response to deoxygenation, where corals shifted into a highly O 2 -demanding stress response exceeding the O 2 supply. Tolerance to hypoxia stress in corals may thus ultimately both contribute and correspond to tolerance to heat stress, an outcome recently suggested where corals commonly ascribed to have greater heat stress tolerance (e.g., Porites lutea) possess a greater gene copy number of hypoxia stress-associated genes 32 . Coral samples from our analysis were associated with a Symbiodiniacae commonly found in species of Acropora (Cladocopium-dominated; C21), although the relevance of this is currently unknown. A key aspect still missing is the contribution of other coral holobiont members to deoxygenation. For instance, associated microbes may be consuming large amounts of O 2 or shifting to a community of anaerobes, both of which could alter the overall O 2 budget of the coral holobiont [86][87][88] . Further, high microbial O 2 consumption associated with excessive algae production on coral reefs can lead to deoxygenation of the reef and consequential changes in the microbial and benthic communities 10,89 .
Deoxygenation induces coral vulnerability to light-and heat-stress. Assessing differential expression of the coral host under normoxic and deoxygenation conditions at the baseline temperature enabled identification of deoxygenation-induced stress processes that could otherwise be disguised under heating. Under deoxygenation (2 mg L −1 O 2 ) and prior to the visible onset of bleaching, several genes involved in photoreception, such as rhodopsin, retinal pigment (RPE65), and retinol dehydrogenase, were downregulated. Rhodopsin downregulation has been observed for corals exposed to only deoxygenation 25 or heating 35 , but how this potentially feeds into an altered capacity to sense and respond to light stimuli is currently unconsidered. In other animals, photoreceptors including rhodopsin, are known to be sensitive to hypoxia, as photoreceptor metabolism involves high levels of O 2 consumption 90 . Consequently, rhodopsin activity and regeneration can be hindered by oxidative stress under hypoxia 91 , and such stress was evident in our deoxygenation treatment at the baseline temperature via significant upregulation of numerous antioxidants (e.g., sulfite oxidase and glutathione peroxidase) and pro-oxidants, such as D-amino acid oxidase, which generates hydrogen peroxide 92 . Consequently, it is plausible to hypothesise that an inability to trigger photoprotective mechanisms under high irradiance levels and limited O 2 supply may be key in driving metabolic bleaching cascades. Specifically, irrespective of whether triggered by reduced external O 2 availability or by an enhanced intra-tissue hypoxic state from high O 2 demands under heat stress, a lack of photoprotection would presumably influence 'optical feedback loops' (e.g., loss of light capture by Symbiodiniaceae) that affect coral bleaching [93][94][95] . Evidence of oxidative stress under deoxygenation as observed here, even before the onset of visual bleaching, is consistent with the oxidative stress theory of coral bleaching 17,43,96 , although oxidative stress itself might be the consequence of further upstream dysbiotic events 22,97 . Regardless of ROS being the cause or a consequence of bleaching, it highlights how hypoxia stress may be an overlooked but fundamental stressor to heat-induced coral bleaching (summarised in Fig. 4).

Figure 4.
Incorporating hypoxia stress into the coral bleaching response. ROS accumulation is commonly observed in heat stressed corals, although it is unclear whether it is the actual trigger or a 'smoking gun' , i.e. downstream of further upstream events that initiate the symbiotic breakdown 22,97 . Despite such uncertainties, hypoxia stress during heating may play a pivotal role in determining coral bleaching susceptibility. Specifically, hypoxia-and heat-driven excessive oxidative stress could hamper signalling of the Hypoxia-inducible Factor (HIF) hypoxia response system (HRS), enhance photodamage, and leave the coral immune-compromised. Increasing deoxygenation may thus increase coral bleaching susceptibility.

Summary
We employed the Coral Bleaching Automated Stress System (CBASS) to show that deoxygenation can lower thermal stress thresholds of corals. Subsequent RNA-Seq analysis suggests a high degree of consistency in the coral stress response to heating with and without deoxygenation and key hypoxia-induced genes were differentially expressed under heating alone. This highlights the inherent link between heat stress and hypoxia at the molecular level. At the same time, we observed a specific gene expression response associated with deoxygenation that corroborates how heat stress is further exacerbated under low O 2 levels, such as compromising photoprotective mechanisms. Thus, our results suggest that coral with a high tolerance to low O 2 are likely more tolerant to heat stress. This study supports hypoxia as an overlooked factor in coral bleaching and further flags the need to establish routine O 2 monitoring on coral reefs.

Data availability
Sequence data determined in this study are available under NCBI BioProject PRJNA808230 at https:// www. ncbi. nlm. nih. gov/ biopr oject/? term= PRJNA 808230. Scripts and input data used for RNA-Seq data analysis and for DRC models are available at GitHub https:// github. com/ reefg enomi cs/ CBASS_ hypox ia.