Filter processing, DNA extraction, quantitation and sequencing

All filter samples were subsequently processed for DNA extraction, quantitation, qPCR, metagenomic sequencing and computational analysis as described in our previous study²². In brief, the filter samples were first washed 3 times using 2 mL of phosphate-buffered saline (pH 7.2) with 0.1% (v/v) Triton X-100 assisted with water-bath sonication at room temperature for 1 min. After washing, the suspension liquid was concentrated onto a 0.02 µm Anodisc filter (Whatman, UK) using a vacuum manifold (DHI, Denmark). DNA was then extracted from the Anodisc with the DNeasy PowerWater kit (Qiagen, USA) following the manufacturer’s standard protocol with modifications to increase DNA yield²⁶.

Final DNA solution was subjected for fluorometer quantification, qPCR and shotgun metagenomic sequencing. Fluorometer quantitation was measured with Qubit 2.0 (Invitrogen, USA) using the High Sensitivity double stranded DNA (HS dsDNA) kit. Taqman qPCR assays with universal bacterial (16S rRNA gene)⁴¹ and fungal (18S rRNA gene)⁴² primer set and probes were used to quantify the copy numbers of bacteria and fungi, respectively. The complete list of primers can be found in Table ​Table22.

List of primers and probes applied in the study.

For direct metagenomic sequencing, libraries were prepared using Swift Biosciences’ Accel-NGS 2S Plus DNA Library kit following the standard protocol. All libraries were subsequently dual-barcoded with Swift Biosciences’ 2S Dual Indexing kit. PCR amplification selectively enriches for library fragments that have adapters ligated on both ends. The number of cycles were adjusted based on the starting amount of DNA (8–15 cycles). Upon pooling at equal volumes, libraries were sequenced on Illumina HiSeq2500 Rapid runs at a final concentration of 10–11 pM and a read-length of 251 bp paired-end (Illumina V2 Rapid sequencing reagents). Each ultra-low biomass sample was sequenced to a depth of at least two million paired-reads.

Raw reads from the sequencer were first trimmed from adapter sequences, low quality bases (<20 score) and short reads (<30 bp) using Cutadapt (v.1.8.1)⁴³. The processed reads were then aligned against the NCBI’s NR database (v.25-02-2016) using RAPSearch2 (v.2.15)⁴⁴. Results from the RAPSearch2 alignment were finally converted to read-match archive (rma) to be visualised with MEGAN5 software³³.