Dataset: Shotgun Metaproteomics of Bering Strait surface water and Chukchi Sea bottom water.

All computation was performed on a Univa Grid Engine cluster with 1.90 GHz AMD Opteron processors. The MOCAT pipeline was used to assemble a metagenome and predict genes as follows. Trimmed and filtered reads from both BSt and CS samples were aligned to the human hg19 reference using SOAPaligner v2.21, and aligned reads were removed. The remaining reads were assembled into contigs and scaftigs with SOAPdenovo v1.06. The assembly was revised, correcting for indels and chimeric regions, with SOAPdenovo v1.06 and BWA v0.7.5a-r16. Genes were predicted using Prodigal v2.60. We used three well-established gene fragment prediction tools to predict gene fragments directly from shotgun metagenomic sequencing reads from each sample: MetaGeneAnnotator (in multiple species mode), FragGeneScan version 1.2.0 (illumina_10 model parameters), and Orphelia (with Net300 prediction model). Separate metapeptide databases were constructed from the BSt and CS sequencing runs, from either predicted gene fragments or raw read sequences. When starting from raw read sequences, each read was translated in all six reading frames, and reading frames containing a stop codon were discarded. The results described in section 3 were obtained by starting with predicted gene fragments from MetaGeneAnnotator. Whether starting from gene fragments or raw read sequences, amino acid sequences from each nucleotide sequence were trimmed to the first and last tryptic cleavage site (or discarded if fewer than two sites), and the remaining ends were discarded. This was done in order to remove partial tryptic peptide sequences that are unlikely to be detected by LC-MS/MS of a trypsinized metaproteome. The resulting candidate sequences were discarded if they were less than 10 amino acids long, if they contained no tryptic peptides with seven or more amino acids, or if the minimum Phred quality score over the length of the sequence was less than 30. Finally, metapeptide candidates meeting all the above criteria were discarded if they were represented by fewer than two reads. A FASTA database was constructed from the remaining metapeptides. For purposes of comparison, we also made use of a metagenome-derived database of translated genes from the metagenome described above and the NCBI nonredundant database of protein sequences from large environmental sequencing projects (‘env_nr’, downloaded from ftp://ftp.ncbi.nlm.nih.gov/blast/db/FASTA/env_nr.gz on December 1, 2015). All database searches were performed using Comet version 2015.01 rev. 2, using a concatenated decoy database in which peptide sequences were reversed but C-terminal amino acids were left in place. Search parameters included a static modification for cysteine carbamidomethylation (57.021464) and a variable modification for methionine oxidation (15.9949). Enzyme specificity was trypsin, with one missed cleavage allowed. Parent ion mass tolerance was set to 10 ppm around five isotopic peaks, and fragment ion binning was 0.02, with offset 0.0. Peptide-spectrum matches (PSMs) from all technical replicates were combined into a single data set. As described previously, after each unique peptide was associated with its top-scoring spectrum, irrespective of charge state, we used the widely used target–decoy search strategy of estimating the false discovery rate (FDR) associated with a given set of accepted peptides. In this context, the FDR is defined as the proportion of the accepted peptides that are not responsible for generating observed spectra. We then empirically examined the trade-off between FDR and the number of accepted peptides, since in practice the mass spectrometrist is typically interested in accepting as many peptides as possible while maintaining an acceptable FDR. Note that this trade-off is similar to the distinction between precision (1 – FDR) and recall or sensitivity. Results of searches of individual samples against multiple databases were integrated as follows. PSMs from searches against all databases were combined into a single tab-delimited file of features for input to Percolator. For each database, a new binary feature was added to the combined feature file indicating whether the PSM was derived from a search against that database. Percolator was then used to analyze the combined set, thereby computing a discriminant score for each PSM. For each scan with multiple PSMs (from multiple databases), all but the highest-scoring PSM were removed. Peptide-level FDR was then calculated as described above.

May, D.H., Timmins-Schiffman, E., Mikan, M.P., Harvey, H.R., Borenstein, E., Nunn, B.L., Noble, W.S. (2016) An alignment-free "metapeptide" strategy for metaproteomic characterization of microbiome samples using shotgun metagenomic sequencing. Journal of Proteome Research 15, 2697-2705.