Mass spectrometry proteomics
Digest the proteins into peptides, identify the peptides against a sequence database, and infer which proteins were present. The default workflow for essentially all discovery proteomics, and the origin of both its depth and its interpretive problems.
Our ruleSimulations prepare you for the bench but don't replace hands-on practice.
Bottom-up
Digest the proteins into peptides, identify the peptides against a sequence database, and infer which proteins were present. The default workflow for essentially all discovery proteomics, and the origin of both its depth and its interpretive problems.
How it differs from the standard method
- Sequence coverage is typically a fraction of each protein, so bottom-up data rarely distinguishes isoforms, splice variants or proteoforms.
- Peptides shared between related proteins cannot be assigned, which is why results are reported as protein groups rather than proteins.
- Incomplete digestion splits a peptide's signal across missed-cleavage forms and quietly degrades quantification before any instrument parameter matters.
- Protein-level false discovery rate must be controlled separately; peptide-level control alone leaves a much higher protein-level error.
- Alternative proteases (Lys-C, Asp-N, chymotrypsin) give different peptides and are the standard way to cover a region trypsin leaves blind.
- The instrument identifies peptides, not proteins
- What the two mass measurements are for
- Digestion is part of the measurement
- Two ways to choose what to fragment
- Dynamic range is the real limit
- The false discovery rate holds only where it was calculated
- How quantification goes wrong, by method
- What a proteomics table is actually telling you
- Key terms
- When to use it
- Sub-techniques
1The instrument identifies peptides, not proteins
Almost every proteomics experiment you will read about works like this: proteins are chopped into peptides with trypsin, the peptides are separated on a reversed-phase column, and the mass spectrometer weighs them and then smashes them to read fragments. Nothing in that sequence ever measures a protein.
2What the two mass measurements are for
The first measurement (MS1) is a survey of everything in the ion source at that instant, recorded as mass-to-charge ratios. A peptide appears not as one line but as a cluster of lines spaced by 1/z, because carbon comes with about 1.1% carbon-13: the spacing tells you the charge, and the charge plus the m/z gives the mass.
3Digestion is part of the measurement
Trypsin cuts after lysine and arginine, which happens to give peptides of a convenient length that carry a charge at each end. If the digest is incomplete you do not simply get fewer peptides: you get the same peptide spread across several forms (fully cleaved, one missed cleavage, two missed cleavages) each with its own mass, retention time and ionisation efficiency.
4Two ways to choose what to fragment
Data-dependent acquisition
The instrument surveys, picks the N most intense precursors, fragments them, and repeats. Spectra are clean because one peptide was isolated. But the choice depends on what happened to be most intense in that cycle, so it is partly chance: a peptide identified in one injection can be absent from the next. That is why missing values dominate label-free DDA.
Data-independent acquisition
Instead of picking, the instrument steps a wide isolation window across the mass range and fragments everything inside it. Nothing is skipped, so the data matrix has very few holes, but every spectrum is a mixture, and the software has to pull one peptide's fragments out of a chimeric spectrum.
Isobaric labelling
Tag each sample with a chemical label of identical total mass but a different arrangement of heavy atoms. The samples then co-elute as one peptide and split only at the fragment stage into reporter ions whose intensities give the relative amounts. One injection quantifies up to eighteen samples, with almost no missing values.
Metabolic labelling
Grow cells in medium with heavy arginine and lysine so every peptide exists in a light and a heavy version a fixed mass apart. The ratio is measured at MS1, before anything can go wrong downstream, which makes it the most accurate of the three, and it only works in cells you can feed.
5Dynamic range is the real limit
A mammalian cell's proteins span more than seven orders of magnitude in abundance, and a single injection on a good instrument covers roughly four. The consequence is not that the rest is measured badly; it is that the rest is not measured at all, because a handful of very abundant proteins take most of the ion current and suppress everything co-eluting with them.
6The false discovery rate holds only where it was calculated
Scores from a search engine are not probabilities, so the field estimates error empirically: search the same spectra against a decoy database of reversed or shuffled sequences, where every match is wrong by construction, and use the decoy hits to set a score threshold. Report that threshold as a 1% false discovery rate and you have said something precise, about the level you computed it at.
7How quantification goes wrong, by method
- Label-free DDA: missing values, and they are structured. The peptides that go missing are the low-abundance ones, in every run, so imputing them with a column mean systematically pulls weak proteins towards the middle and invents regulation.
- DIA: interference. A wide window means another peptide's fragment can land on the same m/z as yours, so the extracted trace is partly somebody else's and ratios flatten.
- Isobaric labelling: ratio compression. Whatever else is co-isolated with your precursor contributes its own reporter ions, and since most proteins do not change, that contamination pulls every ratio towards one. A true two-fold change routinely reads as 1.3- to 1.5-fold at MS2.
- All methods: shared peptides. A peptide present in two protein isoforms cannot be assigned to either, and how the software distributes it changes the answer.
- All methods: ionisation efficiency. Peptide intensity is not proportional to peptide amount, so intensities compare the same peptide between samples and never one peptide with another.
8What a proteomics table is actually telling you
A supplementary table with nine thousand rows invites you to read it as a measurement of nine thousand proteins. Read it instead as a list of inferences, each resting on a number of peptides that the table should tell you.
Key terms
Practise these as flashcards →Digest the proteins to peptides, identify the peptides, infer the proteins. The default workflow, and the source of the protein inference problem.
The intact peptide ion selected for fragmentation, and the backbone fragments recorded from it. MS1 gives mass and abundance; MS2 gives sequence.
The two fragment series produced by cleaving the peptide backbone, holding the N-terminal and C-terminal part respectively. Gaps between consecutive ions are residue masses.
A lysine or arginine that trypsin failed to cut, leaving a longer peptide. Splits one peptide's signal across several forms.
Searching spectra against a reversed or shuffled copy of the database alongside the real one, so the number of decoy matches above a score estimates the number of wrong real matches.
A set of protein entries that the identified peptides cannot distinguish between. Reported as one row, which is why isoform claims need checking.
The shrinking of measured fold changes in isobaric labelling, caused by co-isolated peptides contributing their own reporter ions.
A fragment spectrum containing fragments from more than one precursor. The normal case in DIA and commoner in DDA than people assume.
Comparing the same peptide's MS1 intensity, or its summed fragment intensity, between separate injections. Cheap, unlimited sample number, and the most exposed to missing values.
The span of abundances measurable in one experiment, roughly four orders of magnitude per injection against more than seven in a cell.
When to use it
- When you need an unbiased survey of which proteins are present or changed, with no antibody and no prior hypothesis about what to look for.
- To identify the unknown members of a complex or an interactome after a pull-down or proximity-labelling experiment.
- To confirm the identity, sequence coverage and modification state of a purified or recombinant protein.
- To quantify hundreds to thousands of proteins across a set of samples when immunoassays for them do not exist or would be prohibitive.
- To find out which protein a band on a gel actually is, rather than which protein you expected it to be.
Compared with alternatives
Targeted, cheap, and interpretable by eye for one or two proteins you already suspect; mass spectrometry is what you use when you do not know what to blot for.
Far more sensitive and precise for a single known analyte in many samples, and the right choice for absolute concentrations in plasma or supernatant.
The same instrument and workflow aimed at modified peptides, which need enrichment first because they are a tiny fraction of the digest.
Reads an N-terminal sequence directly from a purified protein with no database at all, which is exactly what mass spectrometry cannot do.
Dozens of known analytes in small volumes with better sensitivity than discovery proteomics, but limited to the panel somebody has made antibodies for.
The usual front end: enrich a complex first, then identify its members by mass spectrometry.
Deeper, cheaper and more uniform coverage of expression, but of transcripts; mRNA and protein levels correlate only moderately and say nothing about modification state.
Sub-techniques
Choosing one here switches the whole page, including the protocol and the lab, to it.
Digest the proteins into peptides, identify the peptides against a sequence database, and infer which proteins were present. The default workflow for essentially all discovery proteomics, and the origin of both its depth and its interpretive problems.
- Sequence coverage is typically a fraction of each protein, so bottom-up data rarely distinguishes isoforms, splice variants or proteoforms.
- Peptides shared between related proteins cannot be assigned, which is why results are reported as protein groups rather than proteins.
- Incomplete digestion splits a peptide's signal across missed-cleavage forms and quietly degrades quantification before any instrument parameter matters.
- Protein-level false discovery rate must be controlled separately; peptide-level control alone leaves a much higher protein-level error.
- Alternative proteases (Lys-C, Asp-N, chymotrypsin) give different peptides and are the standard way to cover a region trypsin leaves blind.