Microbiomes of fish, sediment and seagrass suggest connectivity of coral reef microbial populations

Sample collection

Sediments from the Florida Keys were collected via diving, with locations either in spur and groove sedimentary structures or near sponges (five samples; Table 1). Samples were taken by hand via push coring using 60 ml cutoff sterile syringes and immediately frozen at −20 °C in the field and −80 °C in the laboratory until analysis. Several locations at two different sites within Fiji were sampled (Table 1): sediments from seagrass and corals near the village of Nagigi on the island of Vanua Levu (two samples), a parrotfish (one fish sampled), Calotomus spinidens (Quoy & Gaimard, 1824), from the same seagrass meadows in Nagigi, and surgeonfishes (three fish sampled) (Ctenochaetus striatus (Quoy & Gaimard, 1825); Acanthurus nigricauda, Duncker & Mohr (1929) and Acanthurinus sp. unknown), sediments (four samples) and seagrass (two samples) from a backreef area near the village of Nabukavese on the island of Viti Levu. The Ministry of Education, National Heritage, Culture and Arts (Reference: RA17/13); Fiji Locally Managed Marine Areas (FLMMA) Network, National Trust of Fiji/Conservation International Fiji Office; and the Fiji Immigration Department, Ministry of Defense, National Security and Immigration Office granted approval to access the study site and conduct research. All sampling was conducted with the permission of the community owners of the reefs. All fish were collected under the auspices of the Columbia University Animal Care Board permit to Joshua Drew (ACAAAF6300). The C. spinidens is a strictly herbivorous parrotfish, whereas the surgeonfishes collected are detritivores. Fish and seagrass identifications were made visually. Sediments and seagrass were push cored by hand, and fish were dissected immediately and their gut contents stored separately. These samples were frozen in liquid nitrogen and transported to the University of Delaware, where they were stored at −80 °C. Local waters were unable to be sampled during fieldwork.

During sectioning, seagrass blades and roots were isolated from the two shallow cores and four categories were created: tops of roots (grass surface to 1 cm), bottoms of roots (>1 cm below grass surface), blades with visual epibionts and blades without visual epibionts (Table 1). Epibionts were important to subsample, as it has been shown that their presence dramatically alters organic matter composition of seagrass beds, and may alter microbial composition as well (Spivak et al., 2009).

Sediments were sectioned into 1 cm slices and labeled based on depth (1 is surface to 1 cm, 2 is 1–2 cm and 3 is 2–3 cm). The outside of the core was discarded and 0.5 g of sediment was added to a DNA extraction tube (see below). Sediments were classified during sampling by visual inspection. In general, Florida samples were fine-grained, whereas Fijian sediments ranged from coquina (SD1-2) to mixed fragmented shells and coarse grains (SC1-2) to finer grains/sand (all others).

For fish samples, the intestinal tract and pharyngeal mill/crop were removed intact, and the intestinal length was measured to subsample into foregut, midgut and hindgut based on length when possible. We desired to separate the gut sections as environmental material would be expected to degrade after intake and digestion along the gut. All samples were taken using sterile methods and placed into sterile Eppendorf tubes prior to DNA extraction.

DNA extraction and sequencing

DNA extractions of all samples were done according to the instructions in the PowerSoil DNA Isolation Kit from MoBio (Carlsbad, CA, USA). Samples were measured by Nanodrop (ThermoFisher Scientific, Carlsbad, CA, USA). Successful amplification of the bacterial 16S rRNA gene (primers 8F-1492R) was checked on a 1% agarose gel, alongside an extraction blank. Samples were successful and the extraction blank failed as expected. Full amplicon sequencing was performed at the Molecular Research DNA lab (Shallowater, TX, USA) using bacterial primers 27F-519R, with the barcode on the forward primer. A total of 30 cycles of PCR were performed using HotStarTaq Plus Master Mix Kit (Qiagen, Valencia, CA, USA) under the following conditions: 94 °C for 3 min, followed by 28 cycles of 94 °C for 30 s, 53 °C for 40 s and 72 °C for 1 min, and a final elongation step at 72 °C for 5 min. Samples were pooled in equal proportions and purified using calibrated Ampure XP beads. Then the purified PCR product was prepared for sequencing via Illumina TruSeq DNA library preparation protocol and sequenced on the MiSeq platform (Illumina, San Diego, CA, USA). Sequence reads were joined via the Molecular Research pipeline.

Sequence analysis

The amplicon data was processed with the QIIME software package, where samples were demultiplexed, quality trimmed based on default values (Caporaso et al., 2010). Sequences ranged in number per sample from 41,664 to 226,000 (Table 1), with an average sequence length of 494 nucleotides. Demultiplexed data is available from GenBank accessions: study PRJEB10911; sequence ERS850335. Additional data for ocean water was collected from previous studies that also performed Illumina 16S rRNA bacterial gene sampling: SRR556134, Malaysia (Chan et al., 2013); SRR2015541 IndOce1, SRR2015543 IndOce2, SRR2015540 IndOce3, SRR2015545 IndOce4, SRR2015526 IndOce5 (Jeffries et al., 2015); SRR1693227 MedOce1, SRR1713896 MedOce2, SRR1713895 MedOce3 (Ribes et al., 2015); ERR1103346 AtlaOce1, ERR1103341 AtlaOce2, ERR1103335 AtlaOce3, ERR1103091 AtlaOce3, ERR1103091 AtlaOce4, ERR1103086 AtlaOce5, ERR1103080 AtlaOce6, ERR1103221 AtlaOce7, ERR1103210 AtlaOce9 (Milici et al., 2016). For analysis of samples internal to this study, sequences were clustered into operational taxonomic units (OTUs) by 97% sequence identity using UCLUST (Edgar, 2010). For analyses requiring inclusion of all sequences, sequences were clustered via the closed OTU picking pipeline in QIIME (pick_closed_reference_otus.py) using the default parameters since different primers were used across studies. Representative OTUs were assigned taxonomic classification and aligning to the SILVA database v108 (Quast et al., 2013) using QIIME. The table of representative OTUs was exported into metagenomeSeq software for further analysis (Paulson et al., 2013). Bar graphs of taxonomic abundance at the phylum level were generated from normalized OTU counts. Using sequences seen at least two or more times, or at times sequences seen over 10 times or excluding archaeal reads (see figure legends for details), heat maps and relationships between the different samples were calculated using the Bray–Curtis and Kulczynski dissimilarity indices with second square root transformed data using the vegan R package at three taxonomic levels: phyla, class and OTU level. Archaeal sequences were removed for analyses that only required bacterial sequence comparison. Rarefactions of data from this study were created utilizing the QIIME script with a subsample of 37,000 sequences. Shared (core) microbiome OTUs were calculated based on OTU (97% relatedness) presence in 95% of samples, not including the oceanic water samples, using the compute_core_microbiome.py script in QIIME. Multidimensional scaling plots using a Bray–Curtis statistical metric were generated using the vegan R package.

Analysis of Epulopiscium sequences

Sequence reads identified as Epulopiscium in the QIIME taxonomy table were selected. Heatmaps of the abundances of these reads across samples were produced in R. Sequences were compared via BLAST homology to the nt database at NCBI to find relatives. Relative sequences were aligned by the SILVA SINA aligner (Pruesse, Peplies & Glöckner, 2012) and a maximum likelihood tree was created in FastTree v2.1.4 (Price, Dehal & Arkin, 2010). Read sequences were added to the tree via pplacer since they were so short in comparison to relative sequences (Matsen, Kodner & Armbrust, 2010).

Microbial diversity

Relative abundance at the phylum level suggests a few groups are abundant across samples and that similar sample types host similar communities (Fig. 1). Proteobacteria were numerous in all samples except for two seagrass samples, GA2 and GB2, where Cyanobacteria were dominant. It had been visually noted that these two seagrass samples had fewer epibionts (Table 1). Cyanobacteria were present across sediment and ocean samples, and decreased in abundance along the fish guts. Both fish guts and Florida sediments had appreciable numbers of Planctomycetes, which were less abundant in the other sample types. Chloroflexi were found in sediments and ocean waters, but to note, the water Chloroflexi were from the group SAR202, which is abundant in ocean waters (Morris et al., 2004) and the sediment Chloroflexi were from the Dehalococcoides genus, which is abundant in sediments (Hug et al., 2013). Archaea were only found appreciable numbers in ocean samples, as the primers used for these studies were broader than the ones used for the other samples.

Community relationships

For a broad view of community similarity, and to take into account that our samples come from different environments in size and type (e.g., gut vs. water vs. sediment vs. seagrass blades or roots) and had been sequenced with different primers, we used the closed OTU phylogenetic assignments to cluster the samples via the Bray–Curtis dissimilarity index with square-root transformed data (Fig. 2). This showed the oceanic waters clustering by region, separate from the benthic-related samples of Fiji and Florida. Within this benthic cluster, the host-associated fish gut samples clustered together, Florida and Fiji samples were similar yet distinct, and seagrass samples were distinct. The backreef Fijian samples clustered together, separate from the forereef Fijian samples. A portion of the FL sediments clustered with the backreef Fiji samples, and the other portion clustered with the forereef. Two seagrass root samples clustered with backreef samples, while the others were a distinct group within the benthic clade. Since one major group that differed across these samples was Archaea (Fig. 1), we continued to analyze the data under the closed OTU taxonomy, removing the archaeal signatures and only comparing the bacterial taxa for further analyses.

To further delve into the details of the benthic (sediment/seagrass):water:fish relationship, we used the Kulczynski dissimilarity index with square-root transformed data to assess statistically significant patterns at the phylum level (Fig. 3). The oceanic waters still group together, but now clade with the fish gut samples. The fish gut samples have noticeably higher levels of Proteobacteria, Firmicutes, Bacteriodetes and Actinobacteria than other environments. Combining samples from different diet preferences and species, the physical location within the gut did not determine the clustering arrangement of the gut samples. Just outside the larger sediment group, the seagrasses, along with top roots, form a distinct cluster (Fig. 3). The sediment samples from Florida showed cohesion, within the larger sediment group, which also houses a distinct top root sample, GB3, along with bottom roots. The Fijian forereef and backreef are still clustered separately. Particularly noticeable in this analysis is that the sediment cluster contains a higher number of different taxa at varying abundances, compared to oceanic and gut samples. We removed the ocean waters from further analysis and used open OTU calling methods to assess the diversity within the remaining dataset.

Microbial alpha diversity (i.e., rarefaction) was calculated by randomly sampling a subset of 37,000 sequences from all remaining samples (Table 1). The rarefaction curves suggest that all environments were not sampled to exhaustion, as curve saturation was not reached (Fig. 4). However, a clear pattern of species richness shows that sediment samples were the most diverse when compared to seagrass blade, root and fish samples (Fig. 4). Within this study, Fijian sediments (sediment_Fiji and sediment_Fiji_deep) were more diverse when compared to Floridian sediments (sediment_Florida). Within the Fijian sediments, the sediments collected at a deeper site in the forereef (sediment_Fiji_deep) were more diverse than shallow samples (sediment_Fiji) in the backreef. Seagrass roots and blades have a mid-level of total diversity, between sediments and fish guts. In general, animal-associated habitats had the least diversity; with the herbivorous C. spinidens parrotfish gut having the lowest overall diversity.

Connectivity of environments

Shared microbiome analyses were generated for the Fijian samples in order to better understand the connectivity between the samples based on microbial distribution. While this utilized the “core microbiome” analysis method, we opt to term this a “shared microbiome” considering that multiple habitats are being included in the analysis, whereas a “core microbiome” is typically determined for one type of habitat (Shade & Handelsman, 2011). For this analysis, we used open OTU calling, with 97% similar sequences being called an OTU. This analysis was conducted at a 95% shared threshold, meaning that an OTU was considered to be part of the shared microbiome if it was present in 95% of all samples. From the analysis we found that 34 OTUs are distributed among the shared microbiome (Table S1), including Proteobacteria (17), Cyanobacteria (including chloroplasts) (9), Actinobacteria (2), Bacteroidetes (1) and Planctomycetes (1). While the OTUs within the core microbiome are shared by 95% of the samples, the relative abundance of each OTU per sample is variable (Fig. 5). We highlight six OTUs that have differential distribution in the habitats. OTUs from Cohaesibacteracaea and Pirulleaceae are most abundant in surgeonfish crops and sediments (Fig. 5AB). Xenococcus is found in sediments and a surgeonfish gut (Fig. 5C). Herbaspirillum is most concentrated in parrotfish gut and seagrass blades (Fig. 5D). A Rhodobacteraceae OTU is most abundant in seagrass (Fig. 5E) and a different Rhodobacteraceae OTU is most abundant in surgeonfish (Fig. 5F).

While not detected as a member of the shared microbiome, another taxon that was highly prevalent in the mid and hindguts of the surgeonfish, Acanthurus sp., Acanthurus nigricauda and Ctenochaetus striatus was Epulopiscium (Fig. 6). This taxon known as a symbiont of surgeonfish (Fishelson, Montgomery & Myrberg, 1985) and is one of the largest bacteria ever seen. While it is most abundant in the surgeonfish samples, similar sequences are seen in multiple benthic environments, particularly in the Florida samples. The overall signature of Epulopiscium is most abundant in samples F1, C2, S1, S3 and S4, which are surgeonfish crops (F1, S1), and guts (C2, S3, S4) (Fig. 6; Table 1 shows sample codes). Very few sequences are seen in the parrotfish gut samples (P1-4) (Fig. 6). Five different distinct OTUs are seen, with differing concentrations across the gut regions and fish species, and one of the OTUs, 50432, showing widespread coverage in the Florida sediments in addition to the Fijian fish, suggesting this is a ubiquitous lineage.