Microbiomes of stony and soft deep-sea corals share rare core bacteria

Sample data

Datasets of 16S rRNA amplicons and associated environmental data from four prior publications covering seven species of deep-sea corals were combined and reanalyzed in aggregate (Additional file 1; [11, 27, 28, 30]). These raw sequence data are available from the NCBI Sequence Read Archive under BioProjects PRJNA296835, PRJNA297333, PRJNA305617, and PRJNA348705 as well as USGS data releases for each of the original papers [31–34].

DNA extraction and 16S rRNA gene amplicon sequencing

Uniform methods were used across these four studies to preserve, extract, and sequence bacterial amplicons associated with the corals. Briefly, coral samples were preserved in the field using RNAlater and stored at − 80 °C until processed. DNA was extracted using the MO BIO Powerplant DNA Isolation Kit with the modifications suggested by Sunagawa et al. [4, 35]. DNA samples were amplified with primers 563F (5′-AYTGGGYDTAAAGNG) and 926R (5′-CCGT CAATTYYTTTRAGTTT) which target the V4-V5 hypervariable region of the 16S rRNA gene [36]. Samples were pyrosequenced using Roche 454 GS FLX Titanium chemistry.

Sequence analysis of amplicon datasets

Bioinformatic analysis was conducted using QIIME 1.9.1 [37] and DADA2 1.9.2 [38] following the workflow of specific scripts and parameters for each step listed in Additional file 2. In brief, for QIIME, individual datasets had been previously denoised and trimmed of primers, and the resulting files were combined into a single fasta file for an open-reference operational taxonomic unit (OTU) picking [39]. Greengenes release 13_8 was used as the reference database [40, 41], and chimeras were removed by usearch61 [42]. Non-bacterial sequences and singletons were filtered out. A non-rarefied OTU table (Additional file 3) was used to determine the core OTUs shared across multiple coral species. Samples were then randomly rarefied to the size of the smallest library (4287 sequences) before the diversity metrics were calculated [43]. In brief, for DADA2 (run in RStudio using R version 3.5.1 [44]), the Roche 454 ssf files were converted to individual fastq files using QIIME scripts, and primers were removed using Biostrings [45] and Cutadapt [46]. Reads were trimmed to 325 bp based on quality profiles. Sequences were processed through the filter and inference modules of DADA2 in groups corresponding to original 454 runs (6 separate runs total) in order to more accurately estimate the error rates. The data were then merged for the chimera removal and taxonomy assignment. Greengenes release 13_8 was used as the reference database to keep the consistency with the OTU taxonomy. Amplicon sequence variants (ASVs) that were unclassified below the domain Bacteria level were removed. The resulting ASV table (Additional file 4) contains 4299 ASVs.

Sequence analysis of Endozoicomonas sequences

Operational taxonomic units that were classified by Greengenes as belonging to Endozoicomonadaceae were identified from the OTU table (Additional file 3). The Endozoicomonadaceae OTU sequences and those of reference clone library sequences previously derived from corals were aligned using Clustal X [47]. Stony corals that had comparable reference Endozoicomonas sequences included Acropora humilis (Genbank Accession KC668469.1) [22], Montipora aequituberculata (FJ347758.1) [48], Pocillopora damicornis (KC668770.1) [22], and Stylophora pistillata (KC669131.1) [22]. Soft corals that had comparable reference Endozoicomonas sequences included Eunicella cavolini (JQ691583.1) [21] and Gorgonia ventalina (GU118516.1) [4]. An online phylogenetic tree viewer was used to visualize the relationships revealed by the alignment [49].

Statistical analyses

Alpha and beta diversity metrics and relative abundance summaries were calculated within QIIME 1.9.1 [37]. Community similarity was assessed by principal coordinate analysis (PCoA) using weighted and unweighted Unifrac, Bray-Curtis, and Binary Sorenson Dice to determine the importance of taxonomic and phylogenetic relationships and sequence abundance. PERMANOVA analyses were conducted using PRIMER v7 software [50] on the Binary Sorenson Dice distance matrix from QIIME. A one-factor test design was based on the partial sum of squares type III, 9999 permutations of residuals under an unrestricted permutation of raw data. A two-factor test was based on the same parameters but with the factor “species” nested under the factor “genus.” Core microbiomes were identified based on the presence of an OTU in > 50% of the coral samples (n = 51) or the presence of an ASV in > 30% of the coral samples (n = 51) with the additional caveat of being present in > 5 coral hosts. This additional caveat was necessary to avoid detection of sequences that were conserved only at the level of species or genus.

Core bacterial associates

A Propionibacterium sequence was present in all 7 coral hosts, across 92% of the samples based on an OTU and 63% of the samples based on an ASV (Tables 1 and 2). This finding required some sleuthing, since the sequence appears under two different OTU numbers. In Lo. pertusa, P. placomus, and Anthothela sp., this sequence appears as OTU 4447394, whereas in the 2 Primnoa species, it shows up as OTU 4462014; however, the only difference between the OTUs is two additional bases (GG) at the beginning of OTU 4462014. There is a 100% sequence identity between OTU 4462014 and ASV_50. The Propionibacterium sequences from Primnoa, Paramuricea, and Anthothela have 100% identity across 331 base pairs (bp). The sequence from Lo. pertusa is 99% similar, sharing 217/220 bp with the other coral sequences, but containing 3 single bp insertions.

The second most commonly conserved OTU (245163, Pirellulaceae, found in 84% of samples and identical to a conserved ASV found in 53% of samples) was present in 6 coral species. A Bradyrhizobiaceae sequence (OTU4475561 and ASV_53) was found in 61% and 53% of the samples depending on the pipeline. The other broadly conserved sequences include a number of Pirellulaceae (as OTUs or ASVs), plus orders Campylobacterales, Kiloniellales, and Phycisphaerales; families Enterobacteriaceae, Methylobacteriaceae, and Vibrionaceae; and the genera Acidovorax, Acinetobacter, Bacillus, Lysobacter, Pseudomonas, and Sphingobium (Tables 1 and 2).

Endozoicomonas

There were 7 OTUs and 6 ASVs that were classified as belonging to the family Endozoicomonaceae (Additional files 3 and 4). These rare sequences were typically found in 1–7 samples, with low numbers of reads. The exceptions were OTU 743665 (ASV_56) which was found in 6 Lo. pertusa samples with over 100 reads each for a total of 2574 sequences, and New.ReferenceOTU28 (ASV_45) which was found in 7 Anthothela spp. samples with a total of 3446 sequences. When aligned with known coral-associated Endozoicomonas sequences derived from shallow-water stony and soft corals, Endozoicomonaceae OTU 743665 (Lo. pertusa) grouped with other stony corals and New.ReferenceOTU28 (Anthothela spp.) grouped closely with other soft corals (Fig. 1).

Bacterial community composition

For comparisons of the entire bacterial community composition between the 7 corals, the most informative principal coordinate analysis was that based on the Binary Sorenson Dice distance matrix (Fig. 2). The samples appear to cluster at the level of host genus, with the two Anthothela spp. grouping together and the two Primnoa spp. grouping together. To test the robustness of this pattern, I performed a PERMANOVA across corals using genus as a factor. There were significant differences between the coral genera (P_PERMANOVA = 0.0001, pseudo-F = 8.066, 9787 unique permutations). Subsequent pair-wise tests showed all coral genera were significantly different from each other with the exception of pairs that included L. grandiflora, since there is only one replicate of that coral genus and therefore the test did not have sufficient statistical power. Nesting the factor of “species” within “genus,” there were significant differences between Pr. pacifica and Pr. resedaeformis (P_PERMANOVA = 0.0005, t = 1.9316, 5645 unique permutations) in spite of their close clustering. Conversely, the Anthothela spp. group was not significantly different: A. grandiflora vs. Anthothela sp. (P_PERMANOVA = 0.3426, t = 1.0376, 1808 unique permutations), Anthothela sp. vs. Anthothela ND (P_PERMANOVA = 0.3186, t = 1.0464, 35 unique permutations), and A. grandiflora vs. Anthothela ND (P_PERMANOVA = 0.0238, t = 1.3249, 455 unique permutations).

Conserved vs. contaminants

It is problematic that acquisition of these sequence datasets predated the 2014 publication by Salter et al. [51] that triggered widespread recognition of the need to run kit blank controls to detect possible contamination from DNA extraction kits. Such controls are becoming a common practice and are particularly critical for Illumina datasets which are an order of magnitude deeper than 454 datasets such as the ones compiled in this study. Not having kit controls for these datasets means having to interpret the findings with caution; however, it would be an oversimplification to automatically assume that the presence of certain bacterial groups that have been detected as contaminants in some kits indicates contamination [51, 52]. Even studies that have identified kit contaminants acknowledge that many of the contaminants detected can be indistinguishable from bacteria genuinely present in samples and that arbitrarily discarding low prevalence microbes to “correct” for contamination could hinder the identification of relevant minor components of microbiomes [52]. Bearing this in mind, I have marked the 8 OTUs in Table 1 and 4 ASVs in Table 2 that represent bacterial families/genera also commonly found as kit contaminants. All of these bacterial groups (Acidovorax, Acinetobacter, Bacillus, Bradyrhizobiaceae, Enterobacteriaceae, Propionibacterium, Pseudomonas, Methylobacteriaceae, Sphingobium) have commonly been found in coral microbiomes by recent high-throughput sequencing studies [10, 21, 24, 29, 53–62]; however, those studies also did not run kit blanks to assess the possible contamination. More convincing are studies that have cultured these groups from corals [23, 63–72] (e.g., see matching sequences listed in Table 3). The fact that so much of the coral microbiome literature is overshadowed by the uncertainty of whether these associates are real or contaminants makes it imperative to include kit blanks in future sequence surveys, as well as to focus future work on cultivation and confirmative microscopy of these bacterial taxa.

Core bacterial associates

The Propionibacterium OTU was found in 92% of the 51 coral samples and was present across all 7 coral hosts (Table 1). This OTU was included in the 100% conserved core of individual species Anthothela sp., Lo. pertusa, P. placomus, Pr. resedaformis, and Pr. pacifica. Of the 9 Propionibacterium ASVs, the most common was ASV_50 which was 100% identical to the OTU sequence and was found in 63% of the samples (Table 2). The presence of Propionibacterium has previously been noted in the conserved cores (90–100% sample inclusion) of a number of stony corals: Acropora hycinthus, Acropora muricata, Acropora rosaria, Coelastrea aspera, and Porites lutea [16, 54, 56]. Using less stringent requirements to define the conserved core, Propionibacterium has also been described as a rare but consistent endosymbiont in stony corals Acropora granulosa, Leptosiris spp., and Montipora capitata and has been detected at low relative abundance in a number of other stony coral species [29]. Similar sequences were found from a number of coral studies, including one cultured isolate (Table 3). The fact that this Propionibacterium sequence was found in both stony corals (order Scleratinia) and soft corals (order Alcyonacea) raised the question of whether this putative symbiont might also be found in other members of class Anthozoa–like anemones (order Actiniaria) or zoanthids (order Zoantharia). Propionibacterium was a minor member of the core microbiome of the anemone Aiptasia [73] and was present in anemones Actinia equina, Anemonia viridis [74], and Edwardsiella andrillae [75]. Propionibacterium sequences were also detected in the microbiomes of zoanthids Palythoa caribaeorum [55] and Palythoa australiae [76]. Based on the literature, Propionibacterium appears to be conserved not just in corals but across class Anthozoa. However, given the recent finding that Propionibacterium is also a dominant contaminant in extraction kits [52] and most of these studies did not include kit controls, this finding should be confirmed by non-sequence-based methods, including cultivation and microscopy.

A single Betaproteobacteria OTU from the family Comamonadaceae was conserved in almost 70% of the samples. Greengenes [40, 41] identified the sequence as Limnohabitans, a relatively new genus [77], but RDP Classifier [78] identified the OTU as Acidovorax. Aligning the OTU sequence against representatives from both genera revealed that for the 328-bp length of the OTU, there were only 3 bp differences between Limnohabitans and Acidovorax; however, in all cases, OTU 4483490 shared the same base as Acidovorax. Further, OTU 4483490 shared 99% identity over 306 bp with an Acidovorax clone from coral Stephanocoenia intersepta (Table 3; [79]). Acidovorax has been detected in association with tropical stony corals, often as the most abundant OTU [60, 79–81] and as a rare associate of a temperate soft coral [21]. Some strains of Acidovorax are capable of heterotrophic denitrification of nitrate, so this bacterium may play a role in coral nitrogen cycling [82].

The possibility of complete nitrogen cycling within corals by bacterial associates has been previously hypothesized based on individual coral core microbiome compositions [11, 28]. A similar hypothesis could be sketched based on several of the conserved OTUs identified in this study. The Bradyrhizobiaceae OTU had over 100 identical matches to nitrogen-fixing Bradyrhizobium strains from root nodules. Methylobacterium sequences have been noted as abundant or core members of other tropical stony coral microbiomes [56, 60] and similar to Bradyrhizobiaceae; some Methylobacteriaceae have been found to form root nodules and act as nitrogen-fixing symbionts [83]. Fixed nitrogen in the form of ammonia could then be converted to nitrate by the Pirellulaceae [84]. From there, nitrate could undergo ammonification by the Campylobacterales [85] or alternately undergo denitrification by Kiloniellales [86] or Acidovorax [82]. This combination of conserved bacterial OTUs could indicate the importance of nitrogen cycling to deep-sea corals.

Gammaproteobacteria are common members of coral microbiomes [2, 87–89], and all 7 conserved gammaproteobacterial OTUs were similar to the previously identified coral associates (Table 3). Acinetobacter, Pseudomonas, and Vibro have been hypothesized to play a role in oil degradation to benefit the holobiont [10, 90]. Supporting this idea is the finding that flocculent material coating deep-sea soft corals impacted by an oil spill was dominated by Acinetobacter and Pseudomonas [91]. Acinetobacter, Pseudomonas, and Vibrio have also been implicated in the degradation of dimethylsulfoniopropionate (DMSP)/dimethylsulfide (DMS) [92], suggesting a role in sulfur cycling. Previously, zooxanthellae were credited with the production of DMSP in corals; however, recent research has shown that aposymbiotic corals also are capable of making DMSP, indicating relevance to all corals not just those in the photic zone [93]. Enterobacteriaceae are found in healthy tropical corals [24, 94], and their conserved presence in deep-sea corals which are not directly impacted by sewage suggests that they have a role other than as an indicator of poor water quality [95]. The conservation of specific OTUs from gammaproteobacterial families Enterobacteriaceae, Moraxellaceae, Pseudomonadaceae, Vibrionaceae, and Xanthomonadaceae and their common occurrence in shallow-water corals (Table 3) implies functional importance to the coral holobiont that requires further study.

Three OTUs belong to the order Phycisphaerales (phylum Planctomycetes) and had 99–100% identity with uncultured bacteria collected from deep ocean waters [96, 97]. The family Phycisphaeraceae within this same order was found to be associated with tropical stony corals and more abundant under highly variable environmental conditions [98]. However, no similar sequences to these Phycisphaerales OTUs appear to be associated with tropical corals, possibly suggesting a role specific to cold-water corals.

Endozoicomonas

The rare Endozoicomonaceae sequences (more properly, Hahellaceae [99]) found associated with Lo. pertusa and Anthothela spp. are very similar to the sequences previously found associated with shallow-water corals and are clearly partitioned by whether the host is a stony coral (order Scleractinia) or a soft coral (order Alcyonacea) (Fig. 1). This is supported by a recent study on five Mediterranean soft corals that also found Endozoicomonas-affiliated sequences formed a phylogenetic relationship that mirrored that of the hosts’ systematic classification [5]. A recent genomic comparison of functional capabilities of seven strains of coral-associated Endozoicomonas indicated that stains had differing abilities, and therefore, divergent genotypes are expected to have different specializations [100]. However, the enrichment of particular functional genes across all strains suggested potential roles centering on carbohydrate cycling and amino acid synthesis, with some strains also contributing vitamins, cofactors, and pigments [100].

Endozoicomonas have been found to dominate coral microbiomes in shallow waters, regardless of the host being tropical or temperate, stony or soft [18, 21–23, 25, 101]. Is the rarity in deep-sea corals an artifact of methodology or a reflection of environmental restriction? Acropora millepora nearer to natural CO₂ seeps (pH range 7.28–8.01) had 50% less Endozoicomonas than corals at a control site (pH range 7.91–8.09) [102]. The Lo. pertusa sites had pH values 7.79–7.86 [13], and the deep-sea soft corals came from sites with pH ranges of 7.94–8.15 [103]. Three of the four cultured type strains of Endozoicomonas that have been isolated from shallow corals (E. montiporae, E. euniceicola, and E. gorgoniicola) all have an optimal growth pH of 8.0 [48, 104], while the fourth, E. acroporae, has an optimum of pH 7.0 [105]. A second environmental factor could be temperature: the cultured type strains have a minimum growth temperature of 15–20 °C and an optimal range of 22–30 °C; the deep-sea coral habitats sampled in this study had temperatures ranging from 5 to 11 °C (Additional file 1). It is possible that separately or in combination, the lower pH and lower temperatures found in deep-sea coral environments limit the growth of coral-associated Endozoicomonas.

Bacterial community composition

The Binary Sorenson Dice index is unweighted and does not take phylogenetic relationships into account, so separation of coral microbiomes into groups based on host genus is driven by the presence/absence of OTUs (Fig. 2). This makes sense since the presence/absence analyses better reflect the importance of rare OTUs, which make up the majority of these coral-associated bacterial communities. The previous finding by Lawler et al. [11] that the two Anthothela spp. microbiomes were not significantly different suggested that conservation of bacterial communities at the host genus rather than species level might be a trend for deep-sea corals. However, in spite of the two Primnoa species clustering together, there were significant differences between their microbiomes (P_PERMANOVA = 0.0005, t = 1.9316, 5645 unique permutations). When examined using a weighted Unifrac dissimilarity matrix which takes into account both the abundance of taxa and phylogenetic relationships, Pr. pacifica and Pr. resedaeformis microbiomes separate by species [30]. The two Anthothela species had a very low sequence divergence from each other [11], suggesting the similarity in their microbiomes could be due to how recently the host species diverged, in contrast to the two Primnoa species that inhabit the Atlantic and Pacific Oceans respectively [106]. Alternately, it may be that limitations of molecular markers for discriminating soft coral taxonomy may be complicating the host resolution in Anthothela spp. [107]. Regardless, the host is consistently the dominant driver of deep-sea coral microbiome structure rather than the environment.