Mining viral peptides from MS/MS spectra

Metaproteomes were obtained from the same soil samples as for the metatranscriptomes, resulting in 12 matched metaproteomes (three replicates each for wet and dry incubations and two locations). The LC-MS/MS were performed using two different approaches as described previously⁸. Briefly, all of the samples were analyzed individually with one-dimensional LC-MS/MS. In addition, the pooled replicate samples went through off-line liquid-chromatography-based fractionation followed by LC-MS/MS analysis⁸.

We first curated a viral protein database including the proteins predicted from contigs in DNA viral database mentioned above (Prodigal, v2.6.3), plus a set of unique reference viral structural proteins collected from NCBI Virus (a total of 63,088 capsid proteins, 25 envelope and 15,817 tail proteins). The LC-MS/MS spectra were searched against the metaproteome database using MS-GF+ search engine⁶². The spectrum level peptide confidence score of the peptide-spectrum match (PSM; i.e., MSGFDB_SpecProb in MS-GF+) and mass error (in ppm) of the precursor peptide ion (i.e., DelM_PPM in MS-GF+) were optimized to achieve the highest number of peptide identification within each dataset while keeping the target-decoy-based FDR of peptide identification below 5%. In order to obtain the most confident viral identification, any redundant peptide identifications between the viral metaproteome and the rest of the metaproteome were excluded. The remaining peptide identification were from the viral metaproteome only. The data also went through a manual quality control process to remove any PSM with low confidence based on MS/MS fragmentation coverage of the peptide and MS peak quality. Final spectra are shown in Supplementary Data 5, and their qualification are shown in Supplementary Data 4.