Proteomics

Mass spectrometry measures peptides; structures explain them. GL1X brings the two together over the MolNFT archive: it digests any protein and tells which peptides are unique to it among every protein of the archive, computes masses, fragment ions and isotope patterns, matches tandem spectra against a record, a FASTA file or the whole archive with a false discovery rate, reads the modifications structures show and predicts the sites motifs suggest, checks crosslinks against structures by surface distance, and paints hydrogen–deuterium exchange or limited-proteolysis values onto a structure. Everything runs in GL1X's own code, validated against pyteomics, IsoSpec and the crosslinking literature (validation).

Masses and peptides · Digestion and proteotypic peptides · Isotope patterns · Spectra · Modifications · Crosslinks · HDX and LiP on structures · Virtual 2D gel · API and MCP · Validation

Masses and peptides #

Atomic masses and isotope abundances are NIST's (AME 2016 masses, IUPAC abundances); monoisotopic masses use each element's most abundant isotope, average masses the abundance-weighted means. A peptide is written with its modifications in brackets: PEPM[Oxidation]TIDE, [Acetyl]-SEQ (N-terminal), SEQ-[Amidated] (C-terminal), PEPC[+57.02146]K (a mass shift). Modifications come from Unimod by name with their compositions and sites: carbamidomethyl, oxidation, phosphorylation, acetylation, deamidation, citrullination, pyro-glutamate, methylations, the GlyGly ubiquitin remnant, amidation, carbamylation, nitration, sulfation, hydroxylation, formylation, HexNAc and Hex, myristoylation, palmitoylation, farnesylation, biotin, the SILAC labels Lys+8 and Arg+10, TMT6plex and iTRAQ4plex.

A peptide's card gives its formula, monoisotopic and average mass, m/z at charges 1 to 4, and its fragment ions: b and y by default, a ions, or c and z• (electron-based fragmentation), each with its m/z at charges 1 to one less than the precursor's (three at most; a singly charged precursor gives singly charged fragments). Peptides of up to 200 residues.

Digestion and proteotypic peptides #

Proteins are cut by 24 enzymes with the rules of the PSI-MS ontology and ExPASy PeptideCutter: trypsin (with and without the proline rule), Lys-C, Lys-N, Arg-C, Asp-N, chymotrypsin (the PSI-MS rule and ExPASy's high-specificity rule), glutamyl endopeptidase (V8), pepsin, proteinase K, thermolysin, elastase, proline endopeptidase, CNBr, formic acid, 2-iodobenzoate and others; with missed cleavages, length and mass windows, removal of the initiator methionine, fixed and variable modifications.

Each peptide is flagged as a targeted-proteomics expert would flag it (S Lange et al. Mol Syst Biol 4, 222, 2008; B Kuster et al. Nat Rev Mol Cell Biol 6, 577–583, 2005): methionine (oxidation), cysteine not alkylated, N-terminal Q, E or carbamidomethylated C (cyclization), DP (acid-labile), NG/NS (deamidation), missed cleavages, length outside 7–25, ragged ends (KK, RR, KR, RK at a cleavage site). For a record's chain, each peptide is also placed on the structure: how many of its residues the structure holds, their secondary structure and mean accessibility — a buried peptide may be slow to digest.

Proteotypic within the archive. With archive, every peptide is looked up in every protein of the archive at once (an Aho–Corasick automaton over the BLAST protein database, leucine and isoleucine equal because they weigh the same): a peptide made by one protein only is unique, the evidence targeted assays and protein inference need; the others list the proteins that share them. A protein is a UniProt accession; a sequence without one belongs to the accession another record gives the identical sequence (the same protein deposited in another collection), and is a protein of its own only when no record names one. With specific, a peptide counts in another protein only where the enzyme would cut it out (its flanking residues obey the rule). Every occurrence counts for uniqueness; the counts given stop at 256 proteins and 5,000 distinct sequences (capped), and up to 2,000 sequences are kept to name the others. Peptides shorter than 6 residues are not looked up: they are too short to tell proteins apart.

Isotope patterns #

The isotope distribution of a formula, a peptide or an averagine protein of a given mass (M Senko et al. J Am Soc Mass Spectrom 6, 229–233, 1995): each element's distribution raised to its count and the elements convolved, the isotopologues of one nominal mass merged into one peak at their mean mass, as an instrument that does not resolve the fine structure sees them, the window of nominal masses that holds the pattern kept wherever it lies (an averagine protein of 2 MDa has its peak more than a thousand nucleons above its lightest isotopologue); peaks down to 10⁻³ of the largest by default (min_rel). The pattern says which peak holds the monoisotopic mass (not the first for selenium or iron proteins) and each peak's m/z at the chosen charge. A composition of up to a million atoms whose pattern spans at most 2,048 nominal masses.

Spectra #

Annotation. A peptide's fragment ions are matched to a spectrum (peaks or an MGF block) within a tolerance in Da or ppm: which ions matched, their errors, the share of intensity they explain.

Peptide–spectrum matching. MGF spectra are searched against a record's proteins, FASTA proteins or every protein of the archive:

• candidates: peptides whose mass falls in the precursor window (ppm or Da, with isotope errors) at each charge the spectrum gives (each once, six at most; else 2+ and 3+), fixed and variable modifications placed in every combination up to a limit; 500,000 candidates per spectrum at most (a spectrum that has more says so, candidates_truncated);
• score: the hyperscore of X!Tandem (D Fenyö, R C Beavis. Anal Chem 75, 768–774, 2003) — the summed intensity of the matched b and y peaks times N_b! N_y!, as a base-10 logarithm — over the 150 most intense peaks, four matched ions at least; an E-value from the survival function of the spectrum's candidate scores;
• false discovery rate by target–decoy competition (J E Elias, S P Gygi. Nat Methods 4, 207–214, 2007): decoys are the target peptides reversed but for their C-terminal residue, FDR = (decoys + 1) / targets above a score, q-values the lowest FDR at which a PSM is accepted; 1% by default;
• protein inference: the accepted peptides grouped by protein (UniProt accession), unique and shared peptides told apart; leucine and isoleucine weigh the same, so a peptide counts for every protein that has it with either letter. In the archive a peptide's first 1,000 places are looked up: one found in more (a constant domain, a common tag) lists those proteins only, is marked proteins_truncated and is never unique.

Searching the archive uses a peptide index built when first needed: every tryptic peptide of every protein (up to two missed cleavages, 7–40 residues, ≤ 5,000 Da, carbamidomethyl cysteine, with and without the initiator methionine), one entry of 12 bytes per distinct peptide, sorted by mass; it is released after 20 minutes without use. For an archive of 235,000 proteins (75 million residues) it holds 14.7 million peptides in 241 MB, built in 7.5 s on two CPUs; 500 spectra are searched against it in 2.4 s.

Modifications #

Observed. A structure shows the modifications its protein carried, as modified residues or as bonds to other groups. GL1X reads them from the atoms: two atoms are bonded when closer than the sum of their covalent radii (B Cordero et al. Dalton Trans 2832–2838, 2008) plus 0.45 Å.

Crystallization additives bonded by accident (a glycerol on a lysine) are kept apart as covalent of the category additive adduct. Every observed modification of the archive is searchable by kind, category, component, residue, accession and collection.

Predicted. PTM motifs mark the sites a sequence suggests: the N-glycosylation sequon N-{P}-[ST]-{P}, PKA, PKC, CK2, CDK and proline-directed phosphorylation sites, SUMOylation, N-myristoylation, CaaX prenylation, the bacterial lipobox, C-terminal amidation and gag attachment. A motif hit is a hypothesis; the structure tells which are plausible: each predicted site comes with its relative accessibility and secondary structure, and can be filtered by them (min_rsa, exclude_ss) — a sequon buried in a β-sheet is unlikely to be glycosylated.

Crosslinks #

Crosslinking mass spectrometry tells which residues a reagent of known length joined; a structure is consistent with a link when the two residues can be bridged. GL1X measures each link two ways:

• the Cα–Cα distance, compared with the linker's limit;
• the solvent-accessible surface distance (SASD), the length of the shortest path between the two residues' Cβ atoms (Cα for glycine) that a linker can take around the protein (as Xwalk and Jwalk, A Kahraman et al. Bioinformatics 27, 2163–2164, 2011; J Bullock et al. Mol Cell Proteomics 15, 2491–2500, 2016): on a 1 Å grid, a point is free when it lies beyond every atom's van der Waals radius plus the 1.4 Å probe, the two residues themselves left out (the linker takes their place); the path leaves each Cβ from the nearest free point within 3 Å (a residue with none is buried) and is found by A* search over the 26 neighbours; ligands block it. Surface distances may take 60 s in one request; links left after that get none (sasd_skipped).

Limits follow the linkers' published distributions (Merkley et al. Protein Sci 23, 747, 2014; Kao et al. 2011; Müller et al. 2010; Leitner et al. 2010, 2014; Belsom et al. 2016). Links are given as text (A 72 B 15, one a line) and each comes back satisfied or violated, with both distances; possible lists every pair of reactive residues the linker could join instead, to plan an experiment.

HDX and LiP on structures #

Hydrogen–deuterium exchange and limited proteolysis measure peptides; their meaning is spatial. POST /api/bio/proteomics/map takes peptides (start, end, value, and optionally their sequence, checked against the protein) or residues (residue, value) in the numbering of the chain's whole sequence or the author's, combines overlapping peptides per residue (mean, max, min, or the shortest peptide's value — the best resolved), and returns each residue's value with its accessibility and secondary structure, ready to colour the structure. The workbench paints them in Mol* on a blue–white–red scale.

Virtual 2D gel #

The theoretical isoelectric point (Bjellqvist pK values, as ExPASy ProtParam) and mass of records' proteins or FASTA proteins, drawn where a 2D gel would place them; with or without the initiator methionine.

In the explorer #

Proteomics in the menu's Bioinformatics onchain section (/bio/proteomics) has four tools: Digest & peptides (a sequence or a record's chain; enzyme, missed cleavages, lengths, fixed and variable modifications; the coverage of the protein, the peptide table with flags and uniqueness in the archive, best targets first, as CSV), Masses & isotopes (a peptide's masses, m/z, fragment ladder and isotope pattern; a formula's or an averagine protein's pattern), Spectra & PSM (an MGF searched against a record, FASTA proteins or the whole archive with target–decoy FDR, any match opened as an annotated spectrum with its b/y ladder; or one spectrum annotated with a peptide) and the Virtual 2D gel. Modifications (/bio/ptms) searches the observed modifications of every structure.

In the research workbench, Modifications draws a structure's modified residues and bonded groups as sticks and its motif sites (filtered by accessibility and secondary structure) in colour; Peptides · HDX digests the chain and paints coverage by uniqueness, or paints HDX or LiP values pasted as peptide or residue rows; Crosslinks checks links or lists every possible one, drawn as rods coloured by verdict.

API and MCP #

Validation #

• Masses and ions of 200 modified peptides equal pyteomics 5.0.1 given the same isotope table to 10⁻⁸ Da (average masses to 10⁻⁶).
• Digests: 2,320 digests (40 proteins × every PSI-MS rule × 0–2 missed cleavages, and the ExPASy rules) identical to pyteomics' cleave.
• Isotope patterns equal IsoSpecPy 2.5.0 with the same table within 10⁻⁹ relative plus the probability IsoSpec leaves out; mean masses to 10⁻⁶ Da.
• Spectrum search: 150 synthetic spectra of oxidized and phosphorylated peptides among 2,000 random proteins: at least 120 identified correctly at 1% FDR, at most 3 wrong; 500 spectra against the whole archive index all identified, no decoy accepted.
• Modifications: the disulfides of 3HSY equal its SSBOND records, its N-glycan bonds its LINK records (one LINK, an O5–ND2 distance of 1.87 Å, is not a bond); lysozyme's four disulfides 6–127, 30–115, 64–80, 76–94.
• Surface distances: exact in free space; around a wall the path is within 8% of the analytic shortest path (the price of a 26-neighbour grid).