Selective feeding of three bivalve species on the phytoplankton community in a marine pond revealed by high-throughput sequencing

The study of the selective feeding of bivalves is necessary in order to improve our understanding of bivalve growth and development, which helps to better define the roles of bivalves in their ecosystems. Little information is currently available on the feeding preferences of bivalves in natural waters, since all diets are provided as single or mixed algae in experiments. In this study, high-throughput sequencing of the 23S rRNA gene was performed to explore differences in the feeding selectivity of Mercenaria mercenaria, Meretrix meretrix and Ruditapes philippinarum during different stages of their culturing to reveal their feeding preferences in natural waters. We found that the three bivalve species had different preferential selection of phytoplankton genera, indicating specific selection and avoidance of particular types of algae during their development in aquaculture. M. mercenaria was the most selective of the bivalves, followed by M. meretrix and then R. philippinarum. With the growth of M. mercenaria and M. meretrix, more kinds of phytoplankton could be ingested. In addition, high-throughput sequencing showed that some picophytoplankton including Synechococcus, Microchloropsis, and Chrysochromulina were dominant in the hepatopancreas samples obtained from these three bivalves. Therefore, the importance of these pico-sized algae in bivalve diets should be reassessed.

Morphological analysis of phytoplankton abundance in aquaculture water. Phytoplankton abundance in the aquaculture water decreased and then increased from June to November, with a minimum of 9.76 × 10 4 cells l -1 on 3 September and a maximum of 1.59 × 10 7 cells l -1 on 5 November (Fig. 3). A total of seven phytoplankton phyla were identified in the aquaculture water by morphological analysis during the study period (Fig. 3). Dinophyta was the most abundant phylum, which contributed more than 51% of the total phytoplankton abundance from June to July. In August and September, the relative abundance of Dinophyta declined, while the relative abundance of Chlorophyta increased to 39.71-78.01% of the total phytoplankton abundance. From October to November, Bacillariophyta was the most dominant, accounting for more than 93% of the total phytoplankton community.
High-throughput sequencing for the identification of phytoplankton community composition. At the phylum level, a total of nine phyla were identified based on high-throughput sequencing (Fig. 4).
In aquaculture water, Cyanophyta were the most dominant phylum, accounting for over 83% of the total phytoplankton from June to July. With the extension of aquaculture, the relative abundance of Cyanophyta declined (36.47%-60.90%) as the relative abundance of Chlorophyta increased (19.64%-36.91%). From late September to early November, the relative abundance of Bacillariophyta increased gradually, with a maximum of 90% on November 5 (Fig. 4a). In the hepatopancreas samples collected from Mercenaria mercenaria and Meretrix meretrix, Cyanophyta was the most dominant phylum (34.29%-98.95% and 55.56%-99.55%), followed by Chrysophyta (0.00%-40.79% and 0.00%-29.21%) and Chlorophyta (0.56%-12.26% and 0.00%-10.38%) (Fig. 4b, c). In the early and middle periods of the experiment, the relative abundance of Cyanophyta was the highest, with a maximum of 98.95% and 99.55% on August 16, respectively. From September to November, the relative abundance of Chrysophyta and Chlorophyta increased, peaking on October 21. In the hepatopancreas samples from M. mercenaria, the relative abundance of Haptophyta was high on November 5 (34.55%). In the hepatopancreas samples from M. meretrix, the relative abundance of Euglenophyta was high on September 3 (15.79%). Only five phyla were identified in the hepatopancreas samples from Ruditapes philippinarum, with Cyanophyta being the most dominant group (> 99%). No representatives of the phyla Cryptophyta, Euglenophyta, Haptophyta, and Xanthophyta were detected in these samples (Fig. 4d).   www.nature.com/scientificreports/   www.nature.com/scientificreports/ A total of 78 phytoplankton genera were identified with high-throughput sequencing. In the aquaculture water, six dominant genera detected were Synechococcus, Cerataulina, Micromonas, Thalassiosira, Heterocapsa, and Heterosigma (Fig. 5a). Synechococcus was the most dominant genus from June to September, whose relative abundance ranged from 34.51% to 94.26%. On July 29, in addition to Synechococcus, Heterosigma was also a dominant genus. In August and September, the relative abundance of Synechococcus was less than that in June and July. The dominant genera also included Micromonas, Thalassiosira, and Heterocapsa. From October to November, Cerataulina became the most dominant genus, with a relative abundance that was over 68%. In the M. mercenaria hepatopancreas samples, the dominant genera detected included Synechococcus, Microchloropsis, Chrysochromulina and Heterocapsa (Fig. 5b). Synechococcus was the most dominant genus from June to September, which had a relative abundance ranging from 71.62% to 96.84%. On July 29, the dominant genus included Heterocapsa in addition to Synechococcus. From October to November, Synechococcus was still dominant, but their relative abundance had declined. In addition to the dominant genus Synechococcus, Microchloropsis was dominant on October 21 (39.66%), while Chrysochromulina was dominant on November 5 (34.55%). In the M. meretrix hepatopancreas samples, the three dominant genera included Synechococcus, Microchloropsis and Eutreptiella (Fig. 5c). Synechococcus was the most dominant genus throughout the experiment, with a relative abundance from 54.85% to 95.64%. On September 3, Eutreptiella was a dominant genus in addition to Synechococcus. From October to November, Synechococcus was still the dominant genus, but their dominance had declined. The relative abundance of Microchloropsis increased over 15%, indicating it was the second most dominant genus in the tested samples (Fig. 5c). In the R. philippinarum hepatopancreas samples, Synechococcus was the only dominant genus, accounting for more than 91% of the total phytoplankton abundance (Fig. 5d).
With respect to the number of phytoplankton groups selected, M. mercenaria selected the most, followed by M. meretrix and R. philippinarum. At the genus level, M. mercenaria showed preferential selection for Microchloropsis and Synechococcus (Fig. 8a). The electivity index of M. mercenaria for Microchloropsis and Synechococcus increased with bivalve growth, while M. mercenaria tended to avoid Thalassiosira and Eutreptiella. Additionally, M. meretrix showed preferential selection for Synechococcus and its selectivity index increased with the growth of M. meretrix (Fig. 8b). At the later stage of the experiment, M. meretrix had high selectivity to Microchloropsis. We also found that M. meretrix avoided Thalassiosira and Micromonas throughout the aquaculture process (Fig. 8b), while R. philippinarum was found to preferentially select Microchloropsis and Synechococcus (Fig. 8c); however, its electivity index was low. Ruditapes philippinarum avoided Thalassiosira, Chrysochromulina and Micromonas throughout the aquaculture process (Fig. 8c). By comparing the number of selected phytoplankton at the genus level, we found that M. mercenaria had the most, followed by M. meretrix and R. philippinarum. With the growth of M. mercenaria and M. meretrix, more types of phytoplankton could be ingested (Fig. 9).

Discussion
Earlier research found that larger particles could be captured more efficiently and removed from suspension faster by most bivalves when compared with smaller particles 9,38-40 , potentially because smaller particles were not efficiently retained by the gills of bivalves 21,41,42 . Conversely, other studies have reported that bivalves can capture smaller particles at a higher efficiency than larger particles 43 . The results from direct sampling methods indicated that mean capture efficiency of 2-4 μm particles ranged from 40 to 80% in mussels, and 20-60% in scallops 40,43,44 . Moreover, M. mercenaria was reported to clear chlorophyte and cyanobacterial picophytoplankton 1-4 μm in diameter from suspension with absorption efficiencies that ranged from 17.6 to 31.1% 45 . Therefore, the contribution of picophytoplankton and small nanophytoplankton as food resources for bivalves might be greater than previously reported. Currently, more studies have investigated the contributions of picophytoplankton and small nanophytoplankton (< 4 μm) to bivalve growth 43,46,47 . In the present study, Synechococcus was detected as the dominant genus in the hepatopancreas samples obtained from three different bivalves (Fig. 5), with diameters of only 0.6-2.1 μm 48 . Microchloropsis salina dominated the hepatopancreas samples from M. mercenaria and M. meretrix in the later stage of this experiment (Fig. 5). A previous study of Microchloropsis salina ultrastructure showed that this species has length and width dimensions of 3-4 μm and 1.5-1.7 μm, respectively 49 . In the hepatopancreas samples from M. mercenaria, Chrysochromulina parva was the dominant species on November 5 (Fig. 5b), whose cell size was 3-7.5 μm × 2-4 μm 50 . This dominant species is a picophytoplankton that was too small to detect using morphological methods. However, they are one of the most abundant phytoplankton components in aquatic ecosystems 51 . Moreover, Synechococcus sp. was found to contribute 5-31% of the carbon necessary to sustain bivalve larval growth and respiration 52 . Some picophytoplankton species such as Microchloropsis sp. (formerly Nannochloropsis sp.) are rich in essential fatty acids necessary for bivalve growth and survival 53 . Consequently, the importance of these small phytoplankton in bivalve diets should be reassessed, especially for some genera of picophytoplankton. Particle selection, including the types of particles chosen for ingestion and the possible mechanisms mediating selection, has been studied extensively and reported in the literature 15,16 . However, no mechanism has been proposed for bivalves to date that would account for the capture of smaller particles in preference to larger sized particles. Recently, studies have reported that the interactions between the physicochemical properties of particles and the mucus covering the pallial organs most likely mediate bivalve food choice [54][55][56] . In particular, hydrophobicity is hypothesized to play a role in the capture of smaller particles 44,57 . Characterization of the surface properties of several planktonic bacteria in the SAR 11 Clade found that they can be more hydrophilic than most other bacteria and microalgal species [57][58][59][60] , which might be one reason why the photosynthetic bacteria Synechococcus, which is in the SAR 11 family, were captured preferentially by three bivalve species compared with other algae reported in this study (Fig. 8).
Studies have shown that different bivalve species have different feeding preferences 15,17,41 . Zhang et al. 23 selected Skeletonema costatum, Chlorella spp., Phaeodactylum tricornutum, Isochrysis galbana, Teraselmis sp., Platymonas subcordiformis, and mixed algae to conduct a laboratory feeding experiment in juvenile M. mercenaria. The results showed that P. subcordiformis and Teraselmis sp. could be used as main baits for juvenile hard clam based on the survival and growth rates of the clams, as well as the degree of difficulty in bait cultivation. In laboratory feeding experiments conducted by Tang et al. 17 , M. meretrix was provided with five different species of marine microalgae (I. galbana, Dunaliella sp., P. tricornutum, P. subcordiformis, and Pavlova viridis), either alone and in various mixtures; they found that the bivalve larvae were characterized by selective feeding behavior, with a preference for I. galbana. Xiao et al. 61 studied the filtration feeding of R. philippinarum juveniles on four species of algae and found that the feeding rate for P. subcordiformis was the highest, followed by Chlorella saccharophila, I. galbana and Chaetoceros muelleri. These feeding experiments of bivalves were conducted in the laboratory, which could not reveal their diet in situ. In the present study, we performed high-throughput sequencing of the 23S rRNA gene to assess the feeding selectivity of M. mercenaria, M. meretrix and R. philippinarum www.nature.com/scientificreports/ on phytoplankton in the natural environment. The results indicated that M. mercenaria showed preferential selection for Microchloropsis and Synechococcus, and tended to avoid Thalassiosira and Eutreptiella (Fig. 8a); M. meretrix showed preferential selection for Synechococcus, and avoided Thalassiosira and Micromonas (Fig. 8b); while R. philippinarum preferentially selected Microchloropsis and Synechococcus, and avoided Thalassiosira, Chrysochromulina and Micromonas (Fig. 8c). By comparing the number of selected phytoplankton at the genus level, M. mercenaria was the most selective of the bivalve species, followed by M. meretrix and R. philippinarum. www.nature.com/scientificreports/ Bivalve feeding selectivity on phytoplankton varied by species and was likely dependent on ctenidial architecture and latero-frontal cilia/cirri microstructure 15 . Following examination with transmission and scanning electron microscopy of the latero-frontal tracts on the bivalve gills, Nuculidae, Mytilidae Semelidae, Petricolidae, Veneridae and Pholadidae were found to have latero-frontal tracts consist of compound eu-latero-frontal cirri coupled with simple pro-latero-frontal cilia; while Anomiidae, Pectinidae, Limidae and Pinnidae had latero-frontal tracts that consisted of simple cilia only [62][63][64] . The latter seemed to be less efficient at entraining particles than former bivalves. Particles are captured by direct interception with the ctenidial filaments, but retention of particles is likely enhanced by mucus present on the frontal surfaces or ordinary filaments 16,65 . Therefore, beyond the differences in ctenidial architecture and laterofrontal cilia/cirri microstructure, the production and composition of the mucus that covers the ctenidia and labial palps varied between species and might be another one reason for why different species had different phytoplankton feeding selectivity.
Studies have shown that the demand for microalgae was different in bivalves at different growth stages 3,10 . In general, the larval stages required high quality algae, while post-larvae could feed on lower quality algae but remained sensitive to the proper biochemical composition 3 . However, the adults were able to feed on a greater variety of microalgae 10 . It is well documented in the literature that filtration rate increases with bivalve body size 66,67 . In this study, we found that with the growth of M. mercenaria and M. meretrix, the electivity index of these two bivalves for Chrysophyta and Cyanophyta also increased (Fig. 8). The kinds of phytoplankton they could ingest increased with the weight of M. mercenaria and M. meretrix (Figs. 2, 8, 9). We hypothesized that the filter feeding organs of juvenile bivalves were not fully developed and resulted in low filtration efficiency on algae in early stages of aquaculture. As the bivalves matured and increased in size, their particle ingestion rates increased 68 .

Conclusions
Using high-throughput sequencing, we identified Synechococcus, Microchloropsis, and Chrysochromulina as the dominant picophytoplankton genera in the hepatopancreas of Mercenaria mercenaria, Meretrix meretrix, and Ruditapes philippinarum, indicating a requirement for reassessing the importance of these pico-sized algae to bivalve diets. The three bivalve species had different preferential selection of phytoplankton genera, indicating specific selection and avoidance of particular types of algae during their development in aquaculture. Therefore, oriented cultivation of the bait microalgae based on the preferences of different bivalve species might be an important way to improve the yield of bivalves in a marine pond.   Sample collection and preparation. Water temperature, salinity, dissolved oxygen (DO) and pH in the aquaculture pond were measured in situ using a Yellow Springs Instrument (YSI) handheld device (YSI Incorporated, USA). Two-liter water samples were collected and prefiltered through a sieve with a pore size of 200 μm to remove large suspended particles, microzooplankton, and other large cells. Then, 1 L of filtrate fixed with 5 mL Lugol's solution for phytoplankton identification and counting; 500 mL filtrate was further filtered through a 0.22-μm Millipore membrane to collect the phytoplankton for DNA analysis. Biological samples of M. mercenaria, M. meretrix, and R. philippinarum were collected simultaneously and temporarily stored at 4 °C prior to dissection. To reduce the diet profile differences among individual bivalves, the hepatopancreases from 15-20 individuals were pooled as a single composite sample, homogenized with a pestle in 5 mL microtubes, and stored at -80 °C until they were used for DNA analysis.

Materials and methods
Morphological analysis of phytoplankton from water samples. Water samples (1 L) containing plankton were fixed with 5 mL Lugol's solution. Each sample was subsequently concentrated to 50 mL in the laboratory, and then stored in the dark at 4 °C until analysis. Phytoplankton species were identified, and cell numbers were counted using a phytoplankton enumeration chamber under an inverted microscope (Olympus CKX41, Olympus Corporation, Tokyo, Japan) 37 .
High-throughput sequencing of phytoplankton and bioinformatics analysis. Total genomic DNA was extracted from all samples using a FastDNA spin kit for soil (MP Biomedicals, OH, USA), following the manufacturer's instructions. The quality of the DNA was verified by gel electrophoresis in 0.5% agarose gels, and DNA concentrations were determined using a NanoDrop spectrophotometer (NanoDrop Technologies, Wilmington, DE, USA). The primers p23SrV_f1 (5ʹ-GGA CAG AAA GAC CCT ATG AA-3ʹ) and p23SrV_r1 (5ʹ-TCA GCC TGT TAT CCC TAG AG-3ʹ) were used to amplify the 23S rDNA gene 31 . Purified PCR products were sequenced at Majorbio Bio-Pharm Technology Co. Ltd. (Shanghai, China) using the MiSeq platform (Illumina, San Diego, USA). The 23S rDNA sequences were processed using the QIIME (version 1.91) software package. The original sequence data were sorted into valid reads after demultiplexing and quality filtering, and the high-quality sequences obtained were clustered into operational taxonomic units (OTUs) with an identity threshold of 97% using UPARSE (version 7.1 http:// drive5. com/ uparse/) with a novel "greedy" algorithm that performs simultaneous de novo-based chimera filtering and OTU clustering. The taxonomy of a representative sequence of each OTU was assigned using the Basic Local Alignment Search Tool (BLAST) in the NCBI database (http:// www. ncbi. nlm. nih. gov). Following the exclusion of bacteria (all non-cyanobacteria) and unclassified sequences, phytoplankton sequences were selected for analysis of community composition based on taxonomic information 37 . The sequencing data used in this study were archived in the Sequence Read Archive (SRA) of the NCBI database under accession number SRP300559.
Feeding selectivity analysis. The selectivity of bivalves with respect to feeding on phytoplankton was evaluated using Ivlev's electivity index (E) 69 , which was calculated using the formula E = (r i − p i ) / (r i + p i ), where r i is the relative abundance of phytoplankton in the bivalve hepatopancreas samples, and p i is the relative abundance of similar items in the water sample. The calculated values of E ranged from −1.0 to + 1.0, with −1.0 ≤ E ≤ -0.3 indicating an avoidance or inaccessibility of a prey item, -−0.3 < E < 0.3 indicating a random selection from the environment, and 0.3 ≤ E ≤ 1.0 indicating active selection 70 www.nature.com/scientificreports/