SweetSpheres

Abstract

Characteristics of polysaccharide particles

Custom polysaccharide particles consisting of alginate (CAS #9005-38-3) or pectin (CAS #9000-69-5), approximately 200 μm in diameter, were fabricated by geniaLab (Braunschweig, Germany) by immersing polysaccharide in calcium chloride solution containing metallic beads (Supplementary Methods). Four different versions were produced: magnetic alginate and pectin particles (polysaccharide coated on magnetite core) as well as non-magnetic alginate and pectin particles (polysaccharide coated on ferrous iron core). Particles contained on average 4% polysaccharide (Supplementary Table 6), approximating the 1:100 solid:solvent ratio in natural hydrogels (Verdugo et al., 2004). Magnetic and non-magnetic particles of each polysaccharide allowed co-incubation of both polysaccharides and hence similar selection pressures per treatment (i.e. always two available particle types, only varying in magnetism). Applying external magnetic force (Supplementary File 1) subsequently allowed the targeted sampling of particle types.

Seawater sampling and experimental set-up

Seawater was sampled from approx. 1 m depth above macroalgal forests at Helgoland Island (54.190556 N, 7.866667 E) in June 2017. Seawater was filtered through a 100 μm mesh, brought to the lab within 2 h, and filtered again through a 20 μm mesh to remove larger particles and organisms. Each 12 L of filtered seawater were distributed into 20 L Clearboy bottles (Nalgene, Rochester, NY) previously rinsed with the same seawater. Per bottle, 10 μM NaNO3 and 1 μM NaH2PO4 (wt./vol.) were added as additional nitrogen and phosphorous source to avoid limitation. Three experiments were set up in triplicate: (i) magnetic alginate particles and non-magnetic pectin particles, (ii) non-magnetic alginate particles and magnetic pectin particles, and (iii) control without particles (Supplementary Fig. 1). Each particle type was added at 3500 L−1, resulting in ~42 000 particles per bottle. Bottles were incubated statically at 15°C (approx. in situ temperature) in the dark.

Sampling and nucleic acid extraction of particle-associated and free-living cells

250 ml of the original seawater were filtered onto 0.2 μm polycarbonate filters for determination of the ambient in situ community (start). Filters were flash-frozen in liquid nitrogen and stored at −80°C. Incubations were sampled after 24, 48 and 60 h. At each sampling point, bottles were mixed by inversion and ca. 550 ml withdrawn into rinsed measuring cylinders. Each sample was distributed (2 × 250 ml) into sterile RNase-free Nunc tubes (cat. no. 376814; Thermo Fisher Scientific, Waltham, MA). Particle-associated communities on magnetic AlgP or PecP were sampled by holding a neodymium magnet (cat. no. Q-40-10-10-N; Supermagnete, Gottmadingen, Germany) next to the tube. AlgP and PecP were washed with sterile seawater (filtered through 100 and 20 μm; mixed 3:1 with ddH2O to prevent salt precipitation during autoclavation for 20 min at 121°C) and transferred to 2 ml RNase-free microcentrifuge tubes. The supernatant was transferred to a separate tube and non-magnetic particles were removed by filtration through 5 μm polycarbonate filters. The flow-through was captured on 0.2 μm polycarbonate filters to obtain the FL community. All samples were directly flash-frozen in liquid nitrogen and stored at −80°C. Simultaneous extraction of DNA and RNA was done using a modification of Schneider et al. (2017). Purified DNA and RNA were sent on dry ice to DNASense (Aalborg, Denmark) for quality control and sequencing. For particles, several subsamples per replicate were pooled to obtain sufficient DNA and RNA (Supplementary Table 1).

16S rRNA gene amplicon sequencing

Briefly, the V3–V4 region of bacterial 16S rRNA genes was sequenced using primers 341F-806R (Sundberg et al., 2013) with MiSeq technology (Illumina, San Diego, CA). Internal company standards worked as expected (Supplementary Methods). Reads were classified into ASVs using DADA2 (Callahan et al., 2016) and taxonomically assigned using SILVA v132 (Quast et al., 2013). Rarefaction analysis showed that diversity was reasonably covered (Supplementary Fig. 9). Replicates were congruent per treatment and time, without significant differences in Bray–Curtis dissimilarities (PERMANOVA; p = 0.72 to 0.98). Furthermore, FL communities from AlgP and PecP were congruent as expected, and FL data combined in subsequent analyses. Alpha-diversity indices (richness, Shannon and inverse Simpson) were calculated using R package iNEXT (Hsieh et al., 2016).

Metagenomics

As amplicon data confirmed the consistency of replicates, DNA from the three AlgP, PecP and FL replicates at 24 and 60 h were pooled respectively. DNA was quantified using Qubit (Thermo Fisher Scientific) and fragmented to ~550 bp using M220 using microTUBE AFA fibre screw tubes (Covaris, Woburn, MA) for 45 s at 20°C with duty factor 20%, peak/displayed power 50 W, and cycles/burst 200. Libraries were prepared using the NEB Next Ultra II kit (New England Biotech, Ipswich, MA) and paired-end sequenced (2 × 150 bp) on a NextSeq system (Illumina). Adaptors were removed using cutadapt v1.10 (Martin, 2011) and reads assembled using SPAdes v3.7.1 (Bankevich et al., 2012). Genes were predicted using Prokka (Seemann, 2014) and assigned to KEGG categories using KAAS-SBH-GhostX (Moriya et al., 2007). CAZymes were predicted using dbCAN2 with CAZyDB v8 (Zhang et al., 2018), only considering hits with >80% coverage. Ammonium transporters were predicted by BLASTp of AmtB (P69681) in the Transporter Classification Database (Saier et al., 2016).

Metatranscriptomics

RNA was quantified in duplicate per sample using the Qubit BR RNA assay (Thermo Fisher Scientific). RNA quality and integrity were confirmed using TapeStation with RNA ScreenTape (Agilent, Santa Clara, CA). rRNA was depleted using the Ribo-Zero Magnetic kit (Illumina) and residual DNA removed using the DNase MAX kit (Qiagen, Hilden, Germany). Following sample cleaning and concentrating using the RNeasy MinElute Cleanup kit (Qiagen), rRNA removal was confirmed using TapeStation HS RNA ScreenTapes (Agilent). Sequencing libraries were prepared using the TruSeq Stranded Total RNA kit (Illumina), quantified using the Qubit HS DNA assay (Thermo Fisher Scientific) and size-estimated using TapeStation D1000 ScreenTapes (Agilent). For RNA from particle samples, four to five subsamples per replicate were pooled in equimolar concentrations and sequenced on a HiSeq2500 in a 1 × 50 bp Rapid Run (Illumina). As the first sequencing run did not deliver sufficient data for seven metatranscriptomes, a second run was performed on the same library. PCA confirmed consistent sequencing runs (data not shown), and read counts were subsequently aggregated. Raw fastq sequence reads were trimmed using USEARCH v10.0.2132 (Edgar, 2010) using -fastq_filter and settings -fastq_minlen 45 -fastq_truncqual 20. rRNA reads were removed using BBDuk (http://jgi.doe.gov/data-and-tools/bb-tools) using the SILVA database as reference (Quast et al., 2013). Reads were mapped to the predicted genes using Minimap2, discarding reads with sequence identities <0.98. Relative transcript abundances were obtained by dividing raw counts by the length of each gene (RPK) and normalized by per-million scaling factors. Resulting transcripts per million (TPM) were summed per gene annotation (Supplementary Table 2). Differential transcript abundances were calculated on raw read counts using the default DESeq2 workflow in R v3.6 (Love et al., 2014; R Core Team, 2018) in RStudio (https://rstudio.com), only considering log2-fold changes >2 with padj < 0.001 (Supplementary Table 3).

MAG binning and analysis

MAGs were binned using MetaBat2 and mmgenome2 (Karst et al., 2016; Kang et al., 2019). Reads were mapped back to the assembly using Minimap2 v2.5 (Li, 2018), and the average coverage of each bin was calculated using mmgenome2 (weighted by scaffold sizes). Based on results from CheckM and GTDB-Tk (Parks et al., 2015; Chaumeil et al., 2020), we selected five near-complete MAGs (≥90% estimated genome completeness and <5% genome contamination) representing the major taxa in amplicon data (Table 1, Supplementary Table 4). Whole-genome comparison with type strains was carried out using the MiGA web application (Rodriguez-R et al., 2018). Normalized coverage in metagenomic data was calculated following Poghosyan et al. (2020) after multiplying the coverage of each MAG in every sample with a normalization factor (sequencing depth of the largest sample divided by the sequencing depth of each individual sample). A maximum-likelihood phylogeny based on 92 core genes identified using the UBCG pipeline (Na et al., 2018), including medium-quality MAGs (>70% estimated completeness/<10% contamination) assigned to the same genus plus related genomes from public databases, was calculated using RaxMLHPC-Hybrid v8.2.12 with the GTRGAMMA substitution model and 1000 bootstrap replicates on the CIPRES Science Gateway v3.3 (Miller et al., 2010; Stamatakis, 2014). Genes were assigned to KEGG categories using KAAS, and pathways reconstructed from these predictions using KEGG Pathway Mapper (Moriya et al., 2007; Kanehisa and Sato, 2020). Gene annotations were refined using UniProtKB/Swiss-Prot (Boutet et al., 2016) by custom-BLAST in Geneious v7 (https://www.geneious.com). Genes for processing alginate and pectin monomers were predicted based on the fully reconstructed pathways in A. macleodii and Gramella forsetii (Kabisch et al., 2014; Koch et al., 2019a). For comparative purposes, CAZymes of strains B3M02 and E3R01 (Enke et al., 2019) were re-annotated with dbCAN2 v8.0 and compared to MAGs Ten-26 and Psym-73 using custom-BLAST in Geneious, only considering hits with >30% query coverage and >40% amino acid identity. ANIs between genomes were calculated using enveomics (Rodriguez-R and Konstantinidis, 2016). Prophages and