Terminology: antigen, epitope, and what a receptor recognises#

This page exists because confounding antigen with epitope is the most common imprecision in this field, and reviewers are right to flag it. VDJdb’s own column names carry the loose usage for historical reasons, so the distinction has to be written down somewhere rather than inferred from them.

The two words#

Antigen - in a T-cell context, the peptide-MHC complex. That is what the receptor engages: a peptide held in the groove of a particular allele. Neither half alone is the antigen.

The word itself is used three ways in print, so the section below states which one this page means and what the field’s own counts are. What no source disputes is the second half of the distinction, which is the part the database is keyed on: the presenting allele is not optional, and the peptide’s origin is not the antigen.

Epitope - the part of the antigen that is specifically recognised. The parallel with the antibody field is exact: there, the antigen is the molecule and the epitope is the surface the antibody binds. Here the antigen is the pMHC and the epitope is the portion of it that determines recognition.

So a receptor does recognise an antigen. What it does not recognise is a pathogen, a gene, a protein, or a species. Those are the provenance of the peptide - where it came from.

Provenance is a real axis, and most questions start there#

“Which T cells were found against SARS-CoV-2 epitopes?” and “which against human neoantigens?” are exactly the questions antigen.species and antigen.gene exist to answer, and they are good questions. Grouping epitopes by where they came from is how anyone approaches this data, and the database would be far less useful without it.

What such a group is, precisely, is a union over pMHCs: every receptor reported against some epitope of that origin, under whatever restriction each study used. That is well defined, and it is not one specificity.

The error is only in collapsing the union:

Statement	Status
“this TCR recognises `GILGFVFTL` presented by `HLA-A*02:01`”	precise
“receptors in VDJdb reported against SARS-CoV-2 epitopes”	precise, and a union over many pMHCs
“T cells recognising human neoantigens”	precise as a set; the members were shown different antigens
“this TCR is specific for the influenza M1 protein”	loose: M1 yields many epitopes under many restrictions, and the receptor was shown one
“this TCR is influenza-specific”	a property claimed of the receptor that the data does not carry
“this TCR recognises the NY-ESO-1 antigen”	conflates provenance with the antigen. The antigen was a peptide on an allele; `NY-ESO-1` is the gene it came from

The practical consequence is about aggregates, not about asking. A motif, a lift or an enrichment computed over a species group pools receptors that were shown different antigens, so the number describes the group and cannot be read as a motif for that pathogen. vdjdb.corpus documents a: on exactly those terms: reach for it to find the papers and the receptors, and condition on e:<epitope> - or on that plus a restriction - when the claim is about recognition.

How VDJdb’s columns map onto this#

The antigen. prefix predates the distinction and is a published contract, so the names stay. What each column actually holds:

Column	What it is	Under the strict terms
`antigen.epitope`	the peptide sequence	half of the antigen; the epitope is the recognised part of the whole complex
`mhc.a`, `mhc.b`, `mhc.class`	the presenting molecule	the other half
`antigen.gene`	the gene the peptide came from	provenance, not the antigen
`antigen.species`	the organism the peptide came from	provenance, not the antigen

The database is keyed accordingly, and this is the part to trust over the naming:

restriction is keyed on (antigen.epitope, antigen.species, mhc.a, mhc.b) and pmhc_id identifies one peptide as presented. Those are the antigen.
antigen.gene and antigen.species are deliberately absent from the row identity vdjdb diff keys on, because they annotate the peptide rather than identify the record. A correction to either is a changed cell on the same record, never a record retired and another allocated.
There is no identifier for a gene, a protein or a pathogen anywhere in the schema, and that is not an omission.

What the field’s own words are, measured#

Checked against PubMed on 2026-09-30, because a page about precision should not assert a convention without knowing whether the field shares it. It largely does not, on this one word:

Phrase, title and abstract, singular and plural summed	Records
`peptide-MHC complex`	956
`peptide-MHC ligand`	63
`peptide-MHC antigen`	12

So the field’s name for the composite is complex, or ligand, and “antigen” most often means the peptide alone. All three senses are in print: antigen as the peptide (the majority), antigen as the peptide with the epitope being peptide-plus-allele (the inverse of this page), and antigen as the source protein, which is how IEDB labels its own results.

Two consequences for how to write, and neither weakens the distinction above:

Where a sentence has to be unassailable, name the object rather than the word: “the receptor engages a peptide-MHC complex” is supported by every source checked and cannot be read three ways. Reserve cognate for the pair actually measured, cross-reactive for further ligands.
Say which sense you mean the first time “antigen” appears in a document. This page means the complex. A paper that means the peptide is not wrong, it is the more common usage.

The provenance half needs no hedging, and has an institutional precedent to point at. IEDB names this axis Epitope Source, carries a protein tree with a parent protein per isoform, takes species from NCBI Taxonomy at an Organism level, and is adding a gene layer so epitopes can be searched by gene. That is antigen.gene and antigen.species under better names, built by the group that curates the same literature, and it is never called the receptor’s antigen.

One thing to carry from the cross-reactivity literature, because it bounds the claim in both directions. A receptor recognises many peptides, and they usually share a recognition motif rather than being unrelated: five mouse and human receptors each selected hundreds of reactive peptides whose motifs resembled the known antigen, which the authors present as surveillance without degenerate recognition of non-homologous peptides. So “one clonotype, one specificity” is wrong, and so is treating a receptor as unboundedly promiscuous. The assumption that a receptor has exactly one cognate epitope is named in Nature Reviews Immunology as still embedded in the pre-processing of many prediction models, which is the same error as collapsing a provenance union onto a receptor.

References#

Retrieved from PubMed, 2026-09-30.

Sewell AK. Why must T cells be cross-reactive? Nat Rev Immunol 2012;12(9):669-77. 10.1038/nri3279
Hudson D, Fernandes RA, Basham M, Ogg G, Koohy H. Can we predict T cell specificity with digital biology and machine learning? Nat Rev Immunol 2023;23(8):511-521. 10.1038/s41577-023-00835-3
Birnbaum ME, Mendoza JL, Sethi DK, et al. Deconstructing the peptide-MHC specificity of T cell recognition. Cell 2014;157(5):1073-87. 10.1016/j.cell.2014.03.047
Vita R, Blazeska N, Marrama D, et al. The Immune Epitope Database (IEDB): 2024 update. Nucleic Acids Res 2025;53(D1):D436-D443. 10.1093/nar/gkae1092

Shorthand is fine once the precise form is written down#

Prose in this repository asks “is the CAS motif specific to HIV-1, or to its TRBV?” That question is understandable and short, and it is also imprecise on both halves. Stated properly it is:

Is the CAS motif specific to any of HIV-1’s epitopes, or to its TRBV germline?

Two corrections, and they are the two this page is about. HIV-1 is provenance: no receptor is specific for a species, so what the group can be specific to is some epitope of that species, which is why the word is “any of”. And the alternative to the motif is the germline the V gene templates, not the gene as a label, because that is the thing that would produce CAS without any epitope being involved.

The short form may stand where it already does. What must not happen is the short form being the only record, because then nobody can tell a shorthand from a claim.

The take-home#

Be precise, and the precision costs nothing: ask the provenance question freely, then say what the answer is a set of. Write “epitope-specific” or name the pMHC when the claim is about recognition; reserve “antigen” for the complex; call a gene or a species what it is, which is where the peptide came from. proofreading/epitope_proteome.tsv exists to make that provenance checkable - and, because a reference proteome is one genome and a patient cohort is not, a peptide differing from it is a description of the peptide and never a verdict on the record.