Acropora corals from Sekisei Lagoon, Ishigaki Island, Okinawa Prefecture, Japan were identified and ~20-cm-diameter fragments were collected in May 2016. In an exception, sampling of corals was permitted by the Okinawa Prefectural Government for research use (No. 27-73). In total, we collected 19 species (63 colonies) and put 16 shallow-water (5–10 m) Acropora species in Tank 1 (A. acuminata, A. austera, A. cytherea, A. carduus, A. digitifera, A. florida, A. grandis, A. hyacinthus, A. intermedia, Acropora sp., A. microphthalma, A. muricata, A. nasta, A. selago, A. tenuis) and three deeper-water species (≥15 m) in Tank 2 (A. echinata, Acropora sp., A. awi) respectively at Seikai National Fisheries Research Institute at Ishigaki Island, and kept for 2 days with running seawater (Figure 1, Table 1). Numbers of colonies and coral masses determined in seawater differed among species in order to test detection limits of eDNA in seawater (Table 1). Tank sizes and water flow of the two overflow tanks are as follows; Tank 1: 300 cm length × 80 cm width × 50 cm height (water depth: ~20 cm) and a flow rate of 35 L of seawater/min, Tank 2: 200 cm diameter and 100 cm depth (water depth: ~80 cm) and 10 L/min. Small fragments of each coral colony were isolated and stored in the guanidinium reagent, CHAOS (Fukami et al., 2000), for DNA extraction.

DNA from each colony was isolated from CHAOS solution using a Maxwell 16 Automated Purification System (Promega). We used 200 ng of DNA from each sample for PCR-free library preparation with KAPA HyperPlus Kits (NIPPON Genetics Co, Ltd) following the manufacturer's instructions. All libraries were multiplexed and sequenced (250-bp paired-end, using a HiSeq 2500 in Rapid mode; Illumina). Raw sequence data were submitted to the DNA Databank of Japan (DDBJ) Sequence Read Archive (DRA) under accession number DRA005695 (Table S1, BioProject PRJDB5633). Sequencing data from each library were assembled with IDBA_UD assembler version 1.1.1 (Peng et al., 2012) with different kmer lengths (60, 80, and 100). Identification of complete circular mitochondrial genomes from assembled contigs was performed by (1) comparing them with the Acropora tenuis complete mitochondrial genome (NCBI accession: AF338425.1) (BLASTN e-value ≤ 1e⁻¹⁰⁰), and (2) confirming that 100 bp of both head and tail DNA sequences of a contig were identical, indicating that the sequence was circular.

After 2 days of incubation, 3 L of raw seawater were collected from outlet pipes of the two tanks (Figure 1) and were filtered through a 0.8 μm pore size polycarbonate filter (47-mmΦ, Advantec Toyo Kaisha, Ltd., Tokyo Japan). In addition, seawater from inlet pipes was also collected to provide control samples (before the seawater was passed over the corals) (Figure 1). Filters were frozen at −20°C for several days, and then boiled in 500 μL of TE buffer. Details of the TE boiling method for DNA extraction are provided in Koike et al. (2007) and Yamashita et al. (2011). A part of the Acropora mitochondrial control region (mtCR, ~800–1,000 bp in length) has been widely used as a molecular marker for Acropora species identification (van Oppen et al., 2002; Suzuki et al., 2008). Thus, eDNA was subjected to PCR amplification using RNS2 (5′-CAGAGTAAGTCGTAACATAG-3′) and GR (5′-AATTCCGGTGTGTGTTCTCT-3′) primers for amplifying mtCR (van Oppen et al., 2002; Suzuki et al., 2008) together with genomic DNA isolated from Acropora echinata as a positive PCR control (expected amplicon size: 935 bp). PCR cycling conditions were 15 min at 95°C, followed by 32 cycles of 30 s at 94°C, 90 s at 58°C, and 60 s at 72°C, with an extension of 30 min at 60°C in the final cycle. To amplify ~600–700 bp of Symbiodinium internal-transcribed spacer regions (both ITS-1 and ITS-2) r18Sf primer (5′-GAAAGTTTCATGAACCTTAT-3′), which binds to an 18S region, and Sym28Sr primer (5′-CTTGTRTGACTTCATGCTA-3′), which binds to the ITS2-flanking 28S region, were used (Yamashita and Koike, 2013), together with genomic DNA isolated from clade C Symbiodinium culture strain CCMP2466 (clade C, type C1; purchased from the Provasoli–Guillard National Center for Culture of Marine Algae and Microbiota, East Boothbay, Maine, USA) as a positive PCR control (expected amplicon size: 676 bp). PCR cycling conditions were 35 cycles of 45 s at 94°C, 45 s at 51°C, and 60 s at 72°C. To check major symbiotic Symbiodinium types in each Acropora colony, DNAs isolated from each colony were also used for PCR amplification using r18Sf and Sym28Sr primers under the same PCR conditions mentioned above. All PCR products were cleaned with a QIAquick PCR Purification Kit (Qiagen).

Sequencing libraries of cleaned PCR products (ten samples, DNA amounts used from each sample are shown in Figure S1) were prepared using a KAPA Hyper Prep Kit (NIPPON Genetics Co, Ltd) without fragmentation. Libraries were multiplexed and 300-bp paired-end reads were sequenced with a MiSeq platform (Illumina) with PhiX control (Illumina). Raw sequence data were submitted to the DDBJ DRA under accession number DRA005680 (BioProject PRJDB5633). Read pairs that both PCR primers of Acropora (RNS2 and GR) or Symbiodinium (r18Sf and Sym28Sr) recognized were selected using custom perl scripts (Supplementary Materials) and subjected to quality filtering. Low quality bases (Phred quality score <30) were trimmed using SolexaQA (Cox et al., 2010) and high-quality sequences longer than 200 bp were retained for subsequent analyses.

To identify Acropora species from eDNA, we used mtCR sequences based on the complete, assembled, mitochondrial genomes of corals used in this study (see above) and on A. digitifera sequencing data from five individuals previously collected at Kabira, Ishigaki, Japan (Shinzato et al., 2015). Although mtCR is highly variable within Acropora mitochondrial genomes (Figure S2), percent identities of mtCR sequences between species are high (88.5–100%, Table S2). Lengths of mtCR exceeding paired-end sequencing length (300-bp paired-end) and 300-bp sequences flanking the RNS2 primer are more variable than those of the GR primer (Tables S3, S4). Accordingly, 300 bp flanking the RNS2 primer sequence in the mtCR from each colony were extracted and exactly the same sequences of clustered using CDHIT-EST (Li and Godzik, 2006), and the clustered sequences (14 types) comprised a mtCR database in this study (Table S3). Quality trimmed reads having the RNS2 primer sequence were identified and chimera filtering based on the mtCR database was performed using the UCHIME2 algorithm (Edgar, in review). We aligned these filtered sequences to the database using BLASTN (Camacho et al., 2009) with e-value cutoff of 1e⁻¹⁰⁰. Output files were stringently filtered and numbers of valid alignments that ensure all of the following criteria were counted (details are available on Supplementary Materials); (1) percent identity = 100% as mitochondrial genome sequences of all colonies were in the database; (2) the BLASTN bit score of the best alignment must be greater than the second-best alignment. In order to determine the correlation between coral wet masses and the number of valid alignments, regression analysis was performed using R, version 3.1.3 (R Core Team, 2015).

To develop a Symbiodinium ITS-2 database, we first clustered exactly the same DNA sequences in a database of ITS-2 gene sequences of most reported Symbiodinium types, GeoSymbio (410 ITS-2 sequences) (Franklin et al., 2012) using CDHIT-EST (Li and Godzik, 2006) with option “–c 1.” This resulted in 391 sequences (different ITS-2 sequences sharing same DNA sequences in the original GeoSymbio database were annotated using a concatenation of the original GeoSymbio names with “_”). Then we downloaded Symbiodinium DNA sequences containing the term “internal transcribed spacer 2” from NCBI in October 2017, and Blasted (Camacho et al., 2009) these against the clustered GeoSymbio (BLASTN, e-value < 1e⁻⁵). Aligned sequences longer than 200 bp were extracted and annotated with the names of BLASTN top-hit sequences in the clustered GeoSymbio together with their NCBI accession numbers. Finally, exactly the same DNA sequences were clustered using CDHIT-EST (Li and Godzik, 2006) with option “–c 1.” For identification of Symbiodinium types from eDNA and each Acropora colony, ITS-2 gene sequences flanking the Sym28Sr primer were subjected to chimera filtering (Edgar, in review) using the database and aligned to the Symbiodinium ITS-2 database using BLASTN (Camacho et al., 2009) with an e-value cutoff of 1e⁻¹⁰⁰. Output files were stringently filtered and the number of valid alignments that ensure all of the following criteria were counted (details are available on Supplementary Materials); (1) percent identity ≥ 98% as there might be Symbiodinium types that are not included in the NCBI database; (2) the BLASTN bit score of the best alignment must be greater than the second-best alignment.

We isolated genomic DNA from tissues taken from each of 63 colonies, containing DNA from both Acropora and Symbiodinium, except for A. digitifera, from which we failed to isolate useful DNA. We successfully assembled complete (circular) mitochondrial genomes of all species (accession numbers: LC201813—LC201870) (Table S1). We identified putative mtCR in the assembled mitochondrial genomes and used these to create a database comprised 14 mtCR types (Table S3).

For development of a Symbiodinium ITS-2 database, 5,962 DNA sequences of possible Symbiodinium ITS-2 were downloaded from NCBI, annotated using GeoSymbio database, and exactly the same DNA sequences were clustered. Then, we prepared a Symbiodinium ITS-2 database containing 2,352 non-redundant sequences. Using the latter database, we identified major types of Symbiodinium from each colony (Figure 2). Most colonies exclusively harbored clade C (99.96% of total valid alignments, 4533872 out of 4535626, Figure 2 and Supplementary Material), but clade D was also a major symbiont in some colonies (e.g., D1 and D1a in A. tenuis ten3 and ten12 colonies, Figure 2). Interestingly, C50 is the most major in most colonies in shallow-water Acropora in Tank 1, although C3d_C21 occurred in half of Tank 2 Acropora colonies (Figure 2).

After culturing corals for 2 days, we isolated eDNA from both inlet and outlet seawater. We performed PCR to amplify Acropora mtCR and the Symbiodinium ITS-2 region (Figure S1). Although all PCR products using Symbiodinium primers could be identified, some PCR products of Acropora mtCR were not visible by electrophoresis. However, we did isolate DNA from all Acropora mtCR PCR products (Tank 1 outlet: 8.4 ng/μL, Tank 2 outlet: 4.3 ng/μL, Tank 1 inlet: 4.6 ng/μL, Tank 2 inlet: 4.5 ng/μL, respectively) and successfully prepared Illumina sequencing libraries for all samples (Figure S1). These ten amplicon samples, including two positive controls (A. echinata and Symbiodinium culture strain CCMP2466), were multiplexed and sequenced (Table 2).

As mtCR sequences are highly conserved among Acropora corals (Tables S2–S4), most sequences mapped to more than two sequences in the databases. Thus, we subsequently filtered mapping output so as to count only valid alignments to the database. In the positive control of A. echinata DNA PCR amplicon, 99.1% of valid alignments (35339 of 35662) were correctly assigned to Acropora sp. (Figure 3), indicating that our filtering method worked properly. As expected, few valid alignments from inlet water of Tank 1 and Tank 2 were detected (Table S5). We identified all 14 mtCR types in the database, although only a single valid alignment for car1 (A. carduus) was detected (Figure 3A). The most abundant mtCR types in Tank 1 included A. tenuis (Figure 3A), probably reflecting 15 A. tenuis colonies in the tank (Table 1). In Tank 2, mtCR sequences from Acropora sp. and A. echinata were detected more frequently than A. awi (Figure 3). However, although coral species in Tank 1 and Tank 2 were different, a small proportion of alignments to some mtCR types of coral species that did not exist in each Tank were detected in outlet water of both tanks (Tank 1: 0.05%; 40 out of 67971, Tank 2: 0.8%; 440 out of 52449, Figure 3 in red colors). In inlet water of Tanks 1 and 2, few valid alignments were observed (Figure 3). The proportions of valid alignments of each mtCR types to the total numbers of valid alignments and proportions of weights of corals (Figure 3B, Table S5) were positively correlated in both tanks (regression analysis, R² = 0.7793, p-value < 0.001), in Tank 1 (regression analysis, R² = 0.9807, p-value < 0.001) and in Tank 2 (regression analysis, R² = 0.5985, p-value < 0.01), respectively, suggesting that proportions of DNA sequences reflect the proportions of coral types in surrounding water. In addition, the number of sequences and weights of corals (Figure 3C, Table S5) were positively, but not strongly correlated in both tanks (regression analysis, R² = 0.6371, p-value < 0.01), in Tank 1 (regression analysis, R² = 0.6923, p-value < 0.01), and in Tank 2 (regression analysis, R² = 0.7109, p-value < 0.01), respectively. We detected coral colonies heavier than 0.04 kg (e.g., A. microphthalma deeper-morph: 0.04 kg in Tank 1, Table 1). Overall, the calculated average number of valid alignments per kg (Table S5) was 6145 ± 7058 (R² = 0.7962, p < 0.001), while for Acropora species in Tank 1 it was 3389 ± 2895 (R² = 0.987, p < 0.001) and in Tank 2 it was 13492 ± 10411 (R² = 0.5658, p < 0.05).

In the positive control of Symbiodinium CCMP2466 (clade C, type C1) PCR amplicon, 97.23% of valid aliments were correctly assigned to C1 types (C1 and C1ca_C1b_C1e, Figure 4), indicating that our bioinformatics filtering worked properly for Symbiodinium as well. Clade C, the major symbiotic Symbiodinium clade in the corals used (Figure 2), was detected in both tanks (Figure 4). The most abundant clade C type detected was C50 in Tank 1 and C3d_C21 in Tank 2, respectively, possibly reflecting abundant clade C types of the Acropora corals in each tank (Tank 1: C50, Tank 2: C3d_C21, Figure 2). We were not able to detect some clade C types that were identified from the corals; however, the total number of valid alignments of these types comprised only ~1% of total alignments (46314 out of 4535626) of symbiotic Symbiodinium in the Acropora corals (Figure 2, Supplementary Materials), indicating that these were minor types of clade C symbiont, which could not be detected in seawater. Although clade C types were exclusively detected from Acropora colonies, those in outlet water were low (Tank 1: 35.6%; 44178 out of 36748, Tank 2: 89.3%; 111107 out of 124375). We detected considerable number of clade D types in outlet water of both Tanks 1 and 2 (Figure 4). Within clade D, D1 was the most abundant in both tanks, possibly reflecting more D1 symbionts in the corals (Figure 2). We also detected other clade D and clade A types, which were not detected in the Acropora corals used in this study (Figure 2), from both outlet and inlet seawater. In outlet water, only small numbers of valid alignments originated from ITS-2 types that were not detected in Acropora corals (Figure 2 and Supplementary Materials, Tank 1: 1%; 1003 out of 124223, Tank 2: 3.2%; 4026 out of 124375), albeit in inlet water, large proportions of valid alignments originated from ITS-2 types that were not detected in the corals (Figure 2 and Supplementary Materials, Tank 1: 51.7%; 19005 out of 36748, Tank 2: 99.8%; 38821 out of 38918). In inlet water detected ITS-2 types were considerably different between Tanks 1 and 2 (Figure 4). For example, some abundant types in Tank 1 inlet water (A2-ISS-C2-Sy and D5) were not detected in Tank 2 inlet water, whereas A3 and F4.4b were abundant types in Tank 2 inlet water (Figure 4).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.