Label-free versus labelled quantification

Label-free and labelled quantification answer the same English question, which sample has more of this protein, with different weaknesses. When the numbers disagree with each other or with a blot, the useful move is to name which weakness you are looking at. Bottom-up proteomics in plain language separates identification from abundance. Commissioning a sequencing or proteomics study is the place to see how that choice belongs in a study plan before the first vial is labelled. This page is the troubleshooting version of the choice.

What label-free is actually comparing

Label-free work measures peptides in separate liquid-chromatography runs. The usual signal is the precursor intensity in the first stage of mass spectrometry, integrated across the peak. Spectral counts, the number of times a peptide was chosen for fragmentation, are an older proxy. Counts saturate, they favour long abundant proteins, and they are a blunt instrument for small fold changes. Intensity-based label-free values are sharper and still hostage to the run.

Because each sample has its own spray, its own column ageing and its own retention times, technical variance is not shared. Run order becomes part of the biology if you let it. A column that drifts after sample twelve will draw a false difference between the group you ran first and the group you ran last. Randomise, interleave a common quality-control digest, and record the batch boundary when a column, an emitter or a calibration changes. Missing values are the other label-free habit. A peptide near the sampling limit is identified in some runs and absent in others. Filling those absences with zero, or dropping every protein that is not seen everywhere, changes the answer. The analysis plan should say which rule you used.

What a label shares, and what it still leaves free

Metabolic labelling, classically SILAC, grows cells in medium with heavy lysine and arginine, or a light counterpart. Peptides from the two populations co-elute and appear in one spectrum as a pair of precursors. The ratio is inside that spectrum, so the liquid chromatography is shared. The cells must incorporate the label to completion. Incomplete incorporation looks like a ratio even when the biology did not change. SILAC does not apply, without a special design, to tissue slices and body fluids that were never grown in labelled medium.

Chemical labels at the precursor level, such as reductive dimethylation, put a small mass difference on free amines. Light, intermediate and heavy forms can be mixed after digestion and read in one run, again at the precursor. The chemistry has to go to completion on every sample or the ratio records labelling efficiency. Isobaric tags, including the TMT and iTRAQ reagent classes, are different again: the precursors overlap completely, and the ratio appears as reporter ions after fragmentation. They are the subject of their own comparison. Here the point is only that they share one injection among several samples, and they pay for that sharing with interference.

A label does not cancel a bad digest, a lost pellet or a pipette error before the tag was added. Shared chromatography removes one layer of variance. Upstream sample preparation remains private to each tube.

When the fold change will not match the blot

Start with the protein identity. A western blot often depends on one antibody and one band. The mass spectrometry table may be a protein group: several sequences that share peptides, collapsed because the data cannot tell them apart. If the blot sees an isoform the table has merged with its relatives, the numbers are allowed to disagree. NCBI Protein records are a place to check how many sequences sit under a name you have been using loosely.

Next look at missing values. A label-free protein that is "down" because its peptides were not sampled in the treated runs is not the same observation as a protein whose peaks are present and smaller. Open the peptide table. Presence, absence and a real intensity change are three results.

Then look at batch. If all treated samples were prepared on Monday and all controls on Thursday, the label-free comparison confounds biology with that gap. A labelled design would have put both in one spectrum only if the labelling itself was balanced. A blot run with both samples on one gel is, in that limited sense, more like a labelled experiment than like a long label-free queue. The disagreement is a design clue. HUPO community discussions of quantification keep returning to this point: the ratio is only as comparable as the handling. That is a standards conversation, not a verdict on your blot.

Shared peptides cause a quieter mismatch. A peptide used for quantification that also belongs to a second protein will move when either protein moves. Intensity-based methods try to prefer unique peptides. If the protein has few unique peptides, the reported change can be the neighbour's change. Ask for the unique-peptide evidence before you rebuild a pathway around the fold change.

Compressed ratios are a physical limit

In an isobaric experiment the isolation window is rarely pure. Other precursors co-elute, fragment together, and throw their reporter ions into the same low-mass peaks. Extreme ratios are pulled toward one. The file is behaving as the method behaves. Calling it fraud because a blot shows a larger swing misunderstands both assays. The blot has its own compression and its own saturation.

Design responses exist as method classes. Narrower isolation, an additional fragmentation stage on instruments that support it, and limiting how much ratio you try to read from a very complex mixture, all reduce interference. They do not abolish it. Carrier or boost channels, loaded far more heavily than the analytical channels, change the experiment again and can distort the channels you care about. If the ratio must be believed at a large fold change, consider whether label-free measurement or a metabolic label on a simpler mixture is the honest tool. Do not expect a published performance figure from this page. The size of compression depends on the sample, the gradient and the isolation width.

A comparison you can use when the numbers fight

PeptideAtlas and ProteomeXchange are public places to see how deposited studies name the quantification class. Use them to recognise a design, not to borrow a fold change into your own report.

A short path when a result will not repeat

If label-free differences vanish when you re-randomise the queue, the first result was order. If they vanish when you require unique peptides, the first result was a protein group. If they vanish when you stop imputing missing values, the first result was the imputation. Each of those is a successful troubleshooting outcome. The biology may still be real and simply smaller than the technical effect you just removed.

If a labelled ratio is stable and small, and an orthogonal assay shows a large change, believe that the assays can both be "right" about different molecules or different interferences. Then decide which molecule the claim needs. Remeasuring the same file with a more aggressive normalisation does not create orthogonal evidence.

Research limits

Quantification here is a research comparison of peptide signals. It is not a diagnostic concentration, and it is not a licence to report a clinical cut-off. Biosafety of the material you labelled is an institutional decision. A ratio without replicates, without a stated missing-value rule and without a stated false discovery approach on the protein list is a sketch. Differential abundance analysis is a method reference for how that sketch becomes a designed comparison.

Shared instrument time, and where a batch begins

Facilities often run label-free queues across days, and a power cut or a column change splits the queue into two experiments. Write that split down before anyone calculates fold changes. In a specification for a shared instrument, state whether samples will be randomised inside one uninterrupted queue, what quality-control digest marks a new batch, and what happens if the queue stops overnight. Heat and humidity matter less here than the habit of treating Tuesday's files and Thursday's files as one distribution. Labelled designs shrink that exposure only when every sample that must be compared was actually mixed into the same injection. A multiplex prepared after a gap is still two chemical batches joined by a name.

What to send when you are choosing a class

State the sample number, the groups, whether the cells can be metabolically labelled, the protein amount class, and whether multiplexing is required to keep the samples inside one batch. Say if the result must be compared with a blot, and whether isoforms have to stay separate. Those facts decide between label-free measurement, a metabolic or chemical precursor label, and an isobaric multiplex.

The choice can be discussed against the shotgun discovery proteomics reference and the protein identification by LC-MS/MS reference, then written into the quote request. Ask for a method discussion that names the quantification class and the batch rule. A preference for "quantitative proteomics" is too vague to troubleshoot later.