Unexpected diversity of Endozoicomonas in deep-sea corals

: The deep ocean hosts a large diversity of azooxanthellate cold-water corals whose associated microbiomes remain to be described. While the bacterial genus Endozoicomonas has been widely identified as a dominant associate of tropical and temperate corals, it has rarely been detected in deep-sea corals. Determining microbial baselines for these cold-water corals is a critical first step to understanding the ecosystem services their microbiomes con-tribute, while providing a benchmark against which to measure responses to environmental change or anthropogenic effects. Samples of Acanthogorgia aspera , A. spissa , Desmophyllum dianthus , and D. pertusum (Lophelia pertusa) were collected from western Atlantic sites off the US east coast and from the northeastern Gulf of Mexico. Microbiomes were characterized by 16S rRNA gene amplicon surveys. Although D. dianthus and D. pertusum have recently been combined into a single genus due to their genetic similarity, their microbiomes were significantly different. The Acanthogorgia spp. were collected from submarine canyons in different regions, but their microbiomes were extremely similar and dominated by Endozoicomonas . This is the first re port of coral microbiomes dominated by Endozoico monas occurring below 1000 m, at temperatures near 4°C. D. pertusum from 2 Atlantic sites were also dominated by distinct Endozoicomonas , unlike D. pertusum from other sites described in previous studies, including the Gulf of Mexico, the Mediterranean Sea and a Norwegian fjord.


INTRODUCTION
There are more species of corals in the deep ocean than there are in shallow waters (Roberts & Hirshfield 2004, Roberts et al. 2009). This incredible diversity of cold-water corals, including both calcifying scleractinians and soft octocorals, serves a fundamental role in creating 3-dimensional structure in the deep sea (Roberts et al. 2006). This structure provides the foundation for biodiversity hot spots that support a large variety of invertebrates as well as economically important fish species (Buhl-Mortensen & Mortensen 2004, Stone 2006, Cordes et al. 2008. However, the technical difficulty and expense ABSTRACT: The deep ocean hosts a large diversity of azooxanthellate cold-water corals whose associated microbiomes remain to be described. While the bacterial genus Endozoicomonas has been widely identified as a dominant associate of tropical and temperate corals, it has rarely been detected in deepsea corals. Determining microbial baselines for these cold-water corals is a critical first step to understanding the ecosystem services their microbiomes contribute, while providing a benchmark against which to measure responses to environmental change or anthropogenic effects. Samples of Acanthogorgia aspera, A. spissa, Desmophyllum dianthus, and D. pertusum (Lophelia pertusa) were collected from western Atlantic sites off the US east coast and from the northeastern Gulf of Mexico. Microbiomes were characterized by 16S rRNA gene amplicon surveys. Although D. dianthus and D. pertusum have recently been combined into a single genus due to their genetic similarity, their microbiomes were significantly different. The Acanthogorgia spp. were collected from submarine canyons in different regions, but their microbiomes were extremely similar and dominated by Endozoicomonas. This is the first report of coral microbiomes dominated by Endozoicomonas occurring below 1000 m, at temperatures near 4°C. D. pertusum from 2 Atlantic sites were also dominated by distinct Endozoicomonas, unlike D. pertusum from other sites described in previous studies, including the Gulf of Mexico, the Mediterranean Sea and a Norwegian fjord. of accessing this environment means that relatively little is known about these cold-water coral ecosystems and the basic biology of their keystone species, including age, growth rates, reproduction, and microbiome compositions.
While there have been increasing numbers of reports describing Endozoicomonas-dominated microbiomes in tropical scleractinian corals (Morrow et al. 2012, Speck & Donachie 2012, Bayer et al. 2013b, Roder et al. 2015, Gignoux-Wolfsohn et al. 2017, Brener-Raffalli et al. 2018, Pogoreutz et al. 2018, Camp et al. 2020) as well as tropical and temperate octocorals (Bayer et al. 2013a, Vezzulli et al. 2013, van de Water et al. 2017, McCauley et al. 2020, Reigel et al. 2020, this bacterial genus has been notably scarce in deep-sea coldwater corals. An analysis that included 6 deep-sea octocoral species found Endozoicomonas to be rare in Anthothela spp. and undetectable in the other species (Kellogg 2019). An exception is M. oculata, whose microbiome is dominated by Endozoicomonas in both the Mediterranean and the Atlantic (Rockall Bank), although not by the same phylotypes (Hansson et al. 2009, Meistertzheim et al. 2016, Galand et al. 2018). These collections occurred over a depth range of 520 to 781 m, with Atlantic site temperatures of 7.6−9.0°C and Mediterranean site temperatures of ~13°C. Further, the 3 studies that identified Endozoicomonas as dominating the M. oculata microbiome also examined D. pertusum collected from the same sites at the same time via the same methods and found few or no Endozoicomonas (Hansson et al. 2009, Meistertzheim et al. 2016, Galand et al. 2018. Prior studies of D. pertusum from the Mediterranean (Yakimov et al. 2006), a Norwegian fjord (Neulinger et al. 2008), Rockall Bank (van Bleijswijk et al. 2015, and Gulf of Mexico/ western Atlantic (Kellogg et al. 2009, Gal k ie wicz et al. 2011) have either not detected Endozoicomonas or found it to be extremely rare (<1.5% relative abundance).
Cataloging microbial baselines for cold-water coral species is a critical first step to understanding and predicting the ecosystem services their microbiomes contribute, as well as providing a benchmark against which to measure changes in response to environmental change or anthropogenic impacts. The specific objectives of this work were to (1) determine if microbiome data would corroborate the taxonomic amalgamation of D. dianthus and D. pertusum into a single genus; (2) compare corals that share a habitat type/depth zone in submarine canyons (D. dianthus, Acanthogorgia aspera, and A. spissa); and (3) establish benchmark bacterial community data for D. dianthus, A. aspera, and A. spissa. This resulted in the unexpected discovery that both Acanthogorgia spp. and some D. pertusum microbiomes were dominated by Endozoicomonas.

Sample sites and collections
Deep-sea coral samples were collected for microbial analysis during 4 research cruises between the years 2013 and 2019 via a combination of crewed undersea vehicles and remotely operated vehicles (ROVs) (Table 1, Fig. 1). Samples from Norfolk Canyon were collected in 2013 on the NOAA ship 'Ronald H. Brown' using the ROV 'Jason II' (RB-13-03-HBH Deepwater Canyons). Samples from the West Florida sites Many Mounds and Okeanos Ridge were collected in 2017 on the NOAA ship 'Nancy Foster' using the ROV 'Odysseus' (NF1708 Southeast Florida Deep Coral Initiative). The 2018 and 2019 collections were part of the DEEP SEARCH program and were conducted on the RV 'Atlantis' using the humanoccupied vehicle (HOV) 'Alvin' (AT41) and the NOAA ship 'Ronald H. Brown' using the ROV 'Jason II' (RB1903), respectively.
Five Acanthogorgia aspera were collected from Norfolk Canyon. These small yellow gorgonians were commonly encountered in this canyon and were collected along a long steep wall. Finding A. aspera in Norfolk Canyon was a northward range ex tension for this species because it was previously only known from the Gulf of Mexico and south of Cape Hatteras ). The A. aspera collected from Cape Lookout was growing on a boulder in an area of scattered boulders with a few coral colonies growing on them, including bamboo and black corals. A. spissa occurring in Pamlico Canyon were collected along the northern steep canyon wall from terrace overhangs commonly also populated by Desmophyllum dianthus. With the exception of the Cape Lookout collection at 944 m depth, Acantho gorgia spp. were collected below 1300 m ( Table 1).
The D. dianthus colony from Blake Escarpment was growing on the skeleton of a dead bamboo coral. The 3 D. dianthus collected at Blake Deep were from a rock overhang at the top of a ridge, also populated by anemones, Anthomastus sp., black coral, and bamboo corals. Pamlico Canyon D. dianthus colonies were also from a ledge underhang community, likewise populated by Solenosmilia variabilis, Acanthogorgia sp., and brisingid sea stars. These solitary cup corals were all collected below 1200 m ( Table 1).
The Richardson Ridge site is part of a D. pertusum mound complex that covers an area over 100 km 2 (Stetson et al. 1962). This large area of linear mounds of D. pertusum exists below the Gulf Stream (Stetson et al. 1962, Legeckis 1979, Popenoe 1994, deeper than most D. pertusum habitats known for this area (Table 1), and is subject to high current speeds. Corals in this area experience a variable temperature range ( Table 1). The Cape Fear coral mound site was also subject to strong currents and can be impacted by Gulf Stream meanders (Bane & Brooks 1979). A single colony of D. pertusum was found growing at the shallow Pea Island seep site on authigenic carbonate substrate. The Gulf of Mexico/West Florida Slope collections of D. pertusum were conducted at 2 sites: Many Mounds and Okeanos Ridge (Fig. 1) (Ross et al. 2017). Western Atlantic and Gulf of Mexico D. pertusum colonies are predominantly the white color morph; however, sample NF1708-10-08 was a rare orange colony. The Okeanos Ridge D. pertusum was collected from a wall slope. Subsamples of coral colonies were collected using the respective vehicle's manipulator arm to remove a branch (or in the case of D. dianthus, the entire cup coral). Samples were placed into individual, thermally insulated containers that had been precleaned (washed with freshwater, interiors wiped with 100% ethanol to remove any biofilms or particulates from prior collections), filled with freshwater, and sent down sealed. When opened to receive a coral collection, the freshwater was replaced by seawater local to the collection site due to density differences. The containers were then sealed at depth to prevent microbial contamination from other sample collections or passing through different water masses during vehicle recovery. At the conclusion of the dive, containers were brought into a cold room or laboratory and samples were removed using ethanol-sterilized forceps. All corals were lightly rinsed with sterile 1× phosphate-buffered saline (PBS) to remove any loosely associated surface microbes. For octocorals (A. aspera, A. spissa), branches that had not been in contact with the manipulator claw were cut off using ethanol-sterilized shears and transferred to sterile tubes. For the stony corals (D. dianthus, D. pertusum), samples were placed into sterile aluminum weigh boats and a flame-sterilized hammer was used to break open the calyces to expose polyp tissue, which was transferred to sterile tubes. All samples were preserved with RNA later (Invitrogen), stored for 24 h at 4°C to allow the preservative to infiltrate the tissues, and then moved to −20°C until processing. Subsamples of the corals collected for microbial analyses were shared with researchers conducting population genetics/taxonomy studies, allowing for accurate host identification of octocorals and the possibility of interpreting microbiome trends against coral genotypes (Goldsmith et al. 2018).

Nucleic acid extraction and sequencing
Prior to extraction, coral samples were rinsed with 0.2 μm-filtered and autoclaved 1× PBS to remove excess salts from RNAlater (Invitrogen). Coral pieces were placed into sterile microcentrifuge tubes using flame-sterilized forceps, 2 ml of the sterile 1× PBS was added to the tube, which was then inverted 3 times and then centrifuged at 4000 × g (30 s). The coral samples were then weighed using a sterile technique and approximately 0.1−0.2 g sample −1 were used for extraction. Extractions were done in duplicate for each coral using the DNeasy PowerBiofilm kit (Qiagen), and replicates were combined after the final step. The manufacturer's protocol was followed with the exception that a FastPrep (MP Biomedicals) was  Table 1 for additional sampling site metadata. Base-map credit: Esri, DigitalGlobe, GeoEye, Earthstar Geographics, CNES/ Airbus DS, USDA, USGS, AeroGRID, IGN, and the GIS User Community used on setting 5 (~3100 rpm) for homogenization in place of a PowerLyzer (Qiagen). Kit blanks (extractions with no sample added) were processed at the same time as the samples. DNA was quantified using a Qubit dsDNA High Sensitivity (HS) assay (Invitrogen), and extractions were diluted to approximately 30 ng μl −1 , and submitted for sequencing.
Polymerase chain reaction (PCR) amplification and sequencing were performed by the RTSF Genomics Core at Michigan State University. The V3-V4 regions of the 16S rRNA gene were amplified with the primers 341F (5'-CCT ACG GGA GGC AGC AG-3') (Herlemann et al. 2011) and 806R (5'-GGA CTA CHV GGG TWT CTA AT-3') (Caporaso et al. 2011). One μl of genomic DNA was added to 7.5 μl of 2× Dream Taq Master Mix (Thermo Scientific) and 6.4 μl of a 0.5 μM primer mix. Amplification conditions were an initial melting period of 2 min at 95°C, followed by 30 cycles of 95°C for 40 s, 50°C for 30 s, and 72°C for 60 s, and a final anneal step at 72°C for 7 min. After PCR, the output of all reactions was batch normalized using an Invitrogen SequalPrep DNA Normalization Plate, and all material recovered from the plate was pooled. The pooled material was cleaned up and concentrated using AmpureXP magnetic beads (Beckman Coulter Life Sciences). The pool was quality controlled and quantified using a combination of Qubit dsDNA HS, Agilent 4200 Tape -Station HS DNA1000, and Kapa Illumina Library Quantification qPCR assays. After quality control, this pool was loaded onto an Illumina MiSeq v2 Standard flow cell, and sequencing was carried out in a 2 × 250 bp paired end format using a MiSeq v2 500 cycle re agent cartridge. Custom sequencing and index primers complementary to the 341F/806R sequences used for preparing the libraries were added to appropriate wells of the reagent cartridge. Base calling was done by Illumina Real Time Analysis (RTA) v1.18.54, and output of RTA was demultiplexed and converted to FastQ format with Illumina Bcl2fastq v2.19.1. Se quence data are available from the NCBI Sequence Read Archive (SRA) under BioProject number PRJNA699458 (submission SUB9027826) and are also available online as a US Geological Survey data release (https:// doi. org/ 10.5066/ P9Z1 HPKR).

Data curation and analysis
Data were imported into QIIME2 (version 2020.8) and sorted into amplicon sequence variants (ASVs) using the DADA2 pipeline with the parameters: --ptrim-left-f 50 --p-trim-left-r 50 --p-trunc-len-f 225 and --p-trunc-len-r 225 (Callahan et al. 2016, Bolyen et al. 2019. Taxonomy was assigned with the SILVA re ference database trained for the V3-V4 region (silva-138-99) (Quast et al. 2013). ASVs identifying as mitochondria, chloroplasts, or not assigned to the domains 'Bacteria' or 'Archaea' were re moved. All samples were rarefied to 11 778 se quences, removing 3 samples from the dataset that did not contain this number of sequences (Table 1). The QIIME2 diversity plugin was used to compute all alpha diversity metrics (observed ASVs, Shannon diversity index, Pielou's evenness, Faith's phylogenetic diversity) and a permutational multivariate ANOVA (PERMANOVA) for each alpha diversity metric (Shannon 1948, Kruskal & Wallis 1952, Pielou 1966, Faith 1992, De Santis et al. 2006). The QIIME2 diversity plugin was also used to calculate all beta diversity metrics (Bray-Curtis dissimilarity, Jaccard similarity index, and weighted and unweighted Uni frac distances) (Jaccard 1908, Bray & Curtis 1957, Lozupone & Knight 2005, Lozupone et al. 2007). For each beta diversity metric, an analysis of similarities (ANOSIM) and a resemblance-based permutation test (PERMDISP) were conducted (Clarke 1993, Anderson 2001, Anderson & Walsh 2013. A second analysis was conducted with only those samples associated with scleractinian corals (D. dianthus and D. pertusum), including all diversity indices and statistical analyses listed above.

Endozoicomonas analysis
All sequences within the genus Endozoicomonas were identified for further analysis. Sequences were aligned in MEGA X (Kumar et al. 2018, Stecher et al. 2020 using ClustalW (Thompson et al. 1994), with the addition of relevant sequences from NCBI's SRA and Nucleotide databases. An evolutionary tree of the 20 most common Endozoicomonas sequences was constructed using the maximum likelihood method (Tamura & Nei 1993), with a bootstrap value of 500.

RESULTS
After quality control, the number of sequences per sample ranged from 494 to 722 149, with an average of 216 210 and a standard deviation of ± 216 213. After rarefaction to 11 778 sequences, 3 samples were removed due to low numbers of sequences (Table 1), as was the extraction kit blank.

Alpha diversity metrics
Desmophyllum dianthus had a significantly higher number of observed ASVs than D. pertusum and Acanthogorgia spissa, but not A. aspera (Fig. 2,  Table 2), likely due to the small number of A. aspera samples that were successfully sequenced (n = 2). Although Shannon diversity and Pielou's evenness were not significantly different between species, Faith's phylogenetic diversity was significantly different when comparing D. dianthus to all other coral species (Table 2). When examined by site location, D.
dianthus demonstrated higher numbers of observed sequences in 2 of the 3 locations (see Fig. S1 in the Supplement at www. int-res. com/ articles/ suppl/ m673 p001 _ supp .pdf).

Beta diversity metrics
In a principal coordinate analysis (PCoA) based upon Bray-Curtis dissimilarities of all corals, there was a clear differentiation between the octocorals and scleractinians, in spite of Acanthogorgia spp.  Fig. S1 for site-specific diversity metrics). Boxplots represent median, 25 to 75% interquartile range and non-outlier range; points beyond the whiskers are outliers D. dianthus having been collected from similar habitats and depths. Both Acanthogorgia species clustered tightly together with the exception of 1 sample from Norfolk Canyon that still clustered relatively close (Fig. 3A). Three groupings of D. pertusum were ob served, one of which clustered tightly with D. di anthus. Similar patterns were seen according to Jaccard similarity indices and unweighted Unifrac distances, but not weighted Unifrac distances (Fig. S2). This indicates that the observed patterns are driven by the equal importance of shared rare and dominant microbial taxa, rather than by abundance (dominance) of specific taxa. While Bray-Curtis does factor in abundance, it weights the abundance of shared species more, which explains the tighter clustering of Acanthogorgia spp. in Fig. 3A compared to the clusters in Fig. S2. Upon the independent analysis of D. pertusum and D. dian thus, each species clustered separately except for 1 sample from each species (Fig. 3B).
In this analysis, D. pertusum samples collected from the Gulf of Mexico sites (Many Mounds and Okeanos Ridge) also clustered apart from other D. pertusum samples.
Microbial community structure (ANOSIM) and dispersion (PERMDISP) were analyzed between all coral species using Bray-Curtis dissimilarities. Dispersion was only significantly different between A. spissa and each scleractinian coral (D. pertusum and D. dianthus; Table 2). Microbial community structure was significantly different between all coral species except for A. aspera and A. spissa, with the high rvalues indicating that the communities were very distinct between species (Table 2).

Community and Endozoicomonas analysis
Community analysis at the class level indicated a substantial percentage of Gammaproteobacteria (Fig. S3). Upon further taxonomic evaluation, the majority of these Gammaproteobacteria were unexpected Endozoicomonas spp. Richardson Ridge and Cape Fear D. pertusum were dominated (13−88%) by ASVs belonging to the genus Endozoicomonas, while those collected from other sites had very few (0−3%) Endozoicomonas (Fig. 4). Microbial communities from D. dianthus did not contain more than 0.13% Endozoicomonas ASVs for any of the sampling sites. Both Acanthogorgia species also contained a large proportion of Endozoicomonas ASVs,  ranging from 11 to 75% (Fig. 4). Interestingly, the 2 Acanthogorgia species shared many Endozoi co monas ASVs, despite being collected from different regions. A phylogenetic tree was created to determine phylogenetic relatedness of the Endozoicomonas ASVs. The Endozoicomonas ASVs clustered into 2 main branches, those from D. pertusum and those from both Acanthogorgia species (Fig. 5).

DISCUSSION
Compared to tropical or even temperate corals, there have been relatively few microbiome studies of cold-water corals from the deep ocean. Deep-sea octocoral microbiomes show variation in their most abundant taxa: Paragorgia arborea is dominated by Tenericutes (Weiler et al. 2018), Anthothela grandiflora is dominated by a combination of Gammaproteobacteria and Spirochaetes (Lawler et al. 2016), Primnoa resedaeformis and Paramuricea placomus are both dominated by Proteobacteria, but vary in whether Alpha-or Gammaproteobacteria are more abundant , Goldsmith et al. 2018. In contrast, deep-sea stony corals including Eguchipsammia fistula in the Red Sea, and Desmophyllum pertusum and Madrepora oculata in the Mediterranean, all are dominated by Gammaproteobacteria (Meistertzheim et al. 2016, Röthig et al. 2017b). We found both Acanthogorgia spp., as well as D. pertusum, to be dominated by Gammaproteobacteria, but D. dianthus had a variable mixture of Alpha-and Gammaproteobacteria (Fig. S3).

Dissimilarity between D. dianthus and D. pertusum microbiomes
The bacterial communities associated with cup coral D. dianthus had significantly more variability and richness compared to the community associated with D. pertusum (Fig. 2, Fig. S1). While a few colonies of the 2 species grouped together (Fig. 3,  Fig. S2), the 2 species hosted significantly different microbiomes (Table 2, Fig. S3). Interestingly, the single D. pertusum sample from the Pea Island Seep site, which might be predicted to have an unusual microbiome due to the shallow depth, relatively high temperature (11.4°C), and influence of seep conditions, was both an outlier from the other D. pertusum samples and closest to D. dianthus samples (Fig. 3)   There was some expectation that D. dianthus and D. pertusum might have similar microbiomes because the 2 corals have been reclassified into a single genus based on genetic similarity (Addamo et al. 2016). Prior work on tropical scleractinians has shown conservation of bacterial communities at the genus level (Littman et al. 2009) or that coral microbiome composition tracks host phylogeny (Sunagawa et al. 2010, Pollock et al. 2018). However, we typically encountered these 2 corals in different depth zones (Table 1) and different habitats, which likely has some influence on their microbiomes (Pantos et al. 2015, Hernandez-Agreda et al. 2016. Moreover, re cent research on D. pertusum has indicated that diet may have a strong influence on the microbiome (Galand et al. 2020) and a comparison between D. pertusum and D. dianthus showed differences in in gestion rates of different sized prey (Tsounis et al. 2010).
An additional hypothesis to consider is that of differing 'microbiome flexibility' between D. pertusum and D. dianthus (Ziegler et al. 2019). A reciprocal transplant experiment with 2 tropical stony corals found that one (Acropora hemprichii) had a flexible microbiome that varied between experimental sites, while the other (Pocillopora verrucosa) maintained a stable microbiome, indicating the existence of different host−microbiome adaptation strategies (Ziegler et al. 2019, Voolstra & Ziegler 2020). In the Mediterranean, microbiome comparisons between D. pertusum and M. oculata revealed that the microbiome of M. oculata was stable across seasons and during a reciprocal transplant experiment, while that of D. pertusum was more clearly influenced by environmental conditions (Meistertzheim et al. 2016, Chapron et al. 2020). Further, multiple prior studies of D. pertusum have provided evidence that the bacterial community of this coral exhibits variability across different geographic sites (Neulinger et al. 2008, Kellogg et al. 2009, Schöttner et al. 2009) and the clustering patterns observed in our current study are consistent with D. pertusum having a more flexible microbiome than D. dianthus (Fig. 3B).
Conserved core ASVs (present in all 6 samples analyzed) included 2 Endozoicomonas phylotypes and 4 Shewanella phylotypes. The 2 A. aspera samples had higher relative abundance of class Bacteroidia compared to A. spissa (Fig. S3). Previous work found that the microbiomes of 2 species of the cold-water octocoral genus Anthothela also clustered together and were statistically indistinguishable (Lawler et al. 2016). The only prior knowledge for Acanthogorgiaassociated bacteria comes from a study that screened a variety of deep-sea corals in the Gulf of Mexico for the presence of chemoautotrophic bacteria from the SUP05 cluster (Vohsen et al. preprint https://www. biorxiv.org/content/10.1101(Vohsen et al. preprint https://www. biorxiv.org/content/10. /2020. That study detected genus-specific SUP05 phylotypes at high abundance (>10% relative abundance) in A. aspera using primers that targeted the V1-V2 region of the 16S rRNA gene (Vohsen et al. preprint https://www.biorxiv.org/content/10.1101(Vohsen et al. preprint https://www.biorxiv.org/content/10. /2020. We detected 2 SUP05 cluster phylotypes at low abundance in our A. spissa samples but not in any of the A. aspera samples. This could be a methodological issue (i.e. differential detection by V3-V4 primers compared to V1-V2) or a biogeographic difference between microbiomes of Atlantic corals and Gulf of Mexico corals. In support of it being a biogeographic difference, we detected 3 SUP05 cluster phylotypes occurring in 4 to 6 D. pertusum samples: 1 phylotype was only present in Gulf of Mexico samples (Many Mounds and Okeanos Ridge), and the other 2 had higher numbers of sequence reads in Gulf of Mexico samples compared to Atlantic samples (Cape Fear and Richardson Ridge).

Dominance of Endozoicomonas in Acanthogorgia spp. and site-specific D. pertusum
There were clear differences between the Endozoicomonas community in D. pertusum compared to Acanthogorgia spp. (Figs. 4 & 5). Host-specificity of Endozoicomonas genotypes has been observed in both shallow-water scleractinians and octocorals (Lee et al. 2012, La Rivière et al. 2015, Neave et al. 2017b, van de Water et al. 2017). There was substantial overlap of Endozoicomonas genotypes between A. aspera and A. spissa (as has been seen in sister species of Eunicella (van de Water et al. 2017)); however, EndoSV 20 was only present in A. spissa samples (Fig. 4). Studies of Mediterranean octocorals found evidence of codivergence between Endozoi co mo nas genotypes and their hosts (La Rivière et al. 2015, van de Water et al. 2017. Examination of rare Endozoico -monas genotypes from D. pertusum and Anthothela spp. found that they clustered with scleractinian and octocoral sequences, respectively, further suggesting influence of host phylogeny (Pollock et al. 2018, Kellogg 2019. In this study, all but one of the major Endozoicomonas genotypes from Acanthogorgia spp. formed a clade that clustered closely with other octocorals, Gorgonia ventalina and An tho thela spp. (Fig. 5). One rare genotype (0.09− 0.42% in A. spissa samples), EndoSV 20, fell outside the main octocoral cluster (Fig. 5). This variation may be evidence of the influence of both the control and local adaptation of the host (Neave et al. 2016).
A biogeographic study of Endozoicomonas genotypes in the widely distributed coral species Stylophora pistillata and P. verrucosa found that S. pistillata had geographically distinct Endozoicomonas communities (Neave et al. 2017b), similar to what we observed here in D. pertusum. The Endozoicomonas genotypes from D. pertusum at Richardson Ridge and Cape Fear sites formed 2 main clades, each including a sequence from D. pertusum from Rockall Bank (van Bleijswijk et al. 2015), on the eastern side of the Atlantic (Fig. 5). These groups were separate from an Endozoicomonas genotype that was detected in D. pertusum samples from multiple Gulf of Mexico sites as well as a western Atlantic site off Cape Canaveral, Florida (Kellogg et al. 2017) and from EndoSV 18 from D. pertusum in the Gulf of Mexico, the Many Mounds site (Fig. 5). In the case of S. pistillata and P. verrucosa, Neave et al. (2017b) hypothesized that the difference could be due to reproductive strategy; as a brooder, S. pistillata could more strictly control its microbiome by vertical transmission, resulting in geographic structuring. This hypo thesis does not hold for D. pertusum, which is a broadcast spawner (Brooke & Järnegren 2013). However, there is prior evidence for differentiation of D. pertusum microbiomes based on geographic region, including differences between the Gulf of Mexico and western Atlantic (Neulinger et al. 2008, Kellogg et al. 2017).

What is driving the unusual D. pertusum microbiomes at Richardson Ridge/Cape Fear sites?
Shallow-water studies of both scleractinians and octocorals suggest that reductions of Endozoico monas abundance in hosts that are typically dominated by them are linked to unfavorable environmental conditions, such as excessive nutrients, temperature stress (hot or cold), or lower pH (Vezzulli et al. 2013, Morrow et al. 2015, Roder et al. 2015, Neave et al. 2016, van de Water et al. 2017, Maher et al. 2019, Ziegler et al. 2019, Shiu et al. 2020). However, we do not know of any prior examples where coral microbiomes that typically do not host large numbers of Endozoicomonas have been found to shift to being dominated by them. In general, environmental disturbances that affect the metabolism and physiology of the coral host also change the microbiome (Vega Thurber et al. 2009, Lee et al. 2017. However, as mentioned in Section 4.1, the amount of 'microbiome flexibility' or restructuring of the bacterial composition under environmental change varies depending on the coral host (Ziegler et al. 2019). This microbiome flexibility is hypothesized to be a rapid adaptation that provides the coral host with a more beneficial bacterial community to improve coral fitness under the new conditions (Reshef et al. 2006, Voolstra & Ziegler 2020. The Richardson Ridge and Cape Fear sites on the Blake Plateau are affected by the Gulf Stream (Stetson et al. 1962, Bane & Brooks 1979, Legeckis 1979, Popenoe 1994, which can translate into extreme variability in water temperature, salinity, nutrients, and current speeds (e.g. Mienis et al. 2014). These factors have been shown to affect coral microbiomes in tropical systems (Guppy & Bythell 2006, Littman et al. 2009, Zaneveld et al. 2016, Lee et al. 2017. The species makeup of a microbial community following disturbance may be explained by the environmental preferences and competitive abilities of the particular microbes (Maher et al. 2019, Ziegler et al. 2019. However, the majority of coral-associated Endozoicomonas appear to be sensitive to temperature stress outside the optimal range of 15−30°C (Kellogg 2019, Shiu et al. 2020, implying that these D. pertusum phylotypes may be particularly unusual in their compatibility with rapid shifts in temperature. The functional diversity revealed from coral-associated Endozoicomonas genomes that have been sequenced underscores the complexity of their codiversification with their hosts (Ding et al. 2016, Neave et al. 2017a, Tandon et al. 2020). This diversity may indicate that different strains of Endozoico monas can provide unique ecosystem services to their coral host, such as carbohydrate cycling, amino acid synthesis, or production of vitamins and cofactors (Neave et al. 2017a).

Conclusions
Two cold-water scleractinian corals, Desmophyllum dianthus and D. pertusum (Lophelia pertusa) that have been combined into a single genus, had significantly different bacterial microbiomes. Benchmark bacterial microbiomes for 2 deep-sea octocorals, Acanthogorgia aspera and A. spissa, were found to be statistically indistinguishable and dominated by Endozoicomonas. This is the first report of coral microbiomes dominated by Endozoicomonas occurring below 1000 m, at temperatures of 3.9−4.5°C. Distinct and diverse genotypes of Endozoicomonas unexpectedly dominated the microbiome of D. pertusum at Richardson Ridge and Cape Fear sites. All prior studies of D. pertusum in other regions found this genus to be rare or absent, even when present in neighboring coral Madrepora oculata. The unusual microbiomes at these sites may be linked to the extreme variability experienced by these corals due to interactions with the Gulf Stream. Future research directions could include further characterization of these unusual Endozoicomonas genotypes by cultivation and comparative genomics to understand their tolerance of cold temperatures and potentially unique metabolic capabilities.