Oncofusion

start / graph / concept-a-cd-number-is-not-a-gene

Explainerestablished confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.Q48-surface-antigens

A CD number names a reagent cluster, not a molecule - and four different things can sit behind one

also called CD nomenclature · cluster of differentiation · CD number is not a gene

A CD number like CD19 or CD15 is a label for whatever a group of antibodies all stick to, agreed at a workshop - it is not the name of a gene. Sometimes one gene sits behind it, sometimes a sugar with no gene at all, sometimes two genes, and at least once the antibodies turned out to be sticking to two different things.

The same thing, for a scientist

**Cluster of differentiation** numbers were assigned at the Human Leukocyte Differentiation Antigen workshops to groups of monoclonal antibodies that behaved alike. The number names the **cluster**, and what lies behind it was determined afterwards - sometimes decades afterwards, sometimes wrongly. Four distinct relationships exist and this graph contains examples of all four: | relationship | example | what a symbol lookup returns | |---|---|---| | **one gene, one antigen** | CD19, CD22, CD33, **CD371 -> CLEC12A** | the right gene | | **no gene at all** (a glycan) | **CD15** (Lewis-x), **CD65/CD65s** (glycolipids) | **CD15 -> FUT4, which is WRONG**; CD65 -> nothing | | **more than one gene** (a complex) | **HLA-DR** (HLA-DRA + HLA-DRB1/3/4/5) | nothing, or arbitrarily one chain | | **one number, two antigens** | **CDw65**: VIM-2 binds the sialylated ganglioside, VIM-8/VIM-11 the neutral glycolipid | nothing, and the ambiguity is in the reagent | **The practical rule, and the reason this entry exists:** never let a `CD` prefix license a `gene-` node id, a gene symbol in a query, or a row in a symbol-keyed queue. Check which of the four cases you are in **before** the identifier is written down, because all four produce confident output and only one of them is right.

Go deeper
Why the wrong answer is more dangerous than no answer

Measured on 2026-09-20 against the live HGNC REST API, the same two calls (/search/alias_symbol/<s> and /search/<s>):

stringHGNC returnscorrect?
CD371CLEC12A (alias symbol)yes
CD15FUT4 (previous official symbol, and top free-text hit)no - FUT4 is one enzyme that helps build the epitope, not the epitope
CD65nothing at allhonest blank

12

CD15 is the dangerous row. HGNC once used CD15 as FUT4's official symbol, so the mapping is not a fuzzy-match error - it is a documented nomenclature fact that resolves cleanly, returns a real gene with a real accession, and encodes a biological falsehood. An engineer repairing the named-but-unwritten queue by adding previous symbols would increase the queue's coverage and introduce this error at the same time.

CD65 is the safe row precisely because it fails loudly. This is the general lesson: a blank prompts a human to look; a plausible wrong answer does not.

The four cases, with the sources that establish them

One gene. CD371 is an alias symbol of CLEC12A in HGNC and the CD antigen name in UniProt's Q5QGZ9 record 13. This is the case everyone assumes.

No gene. CD15 is the carbohydrate Gal-beta-1,4-(Fuc-alpha-1,3)-GlcNAc, built by alpha-1,3-fucosyltransferases 4; CD65 and CD65s are glycosphingolipid epitopes 5. A post-translational glycan cannot have a gene, because several enzymes can make it and it sits on many carriers.

More than one gene. HLA-DR is an alpha-beta heterodimer; HLA-DRA (P01903) pairs with HLA-DRB1 (P01911) or with DRB3/4/5 6. The mapping fails in both directions: no single symbol denotes the marker, and the marker does not determine which beta locus contributed.

One number, two antigens. The CDw65 cluster was assigned at the IVth and Vth workshops to antibodies binding a VIM-2 dodecasaccharide; Kniep 1996 then showed VIM-2 binds the sialylated form while VIM-8 and VIM-11 bind the neutral one, with opposite neuraminidase responses, and concluded that they represented two different antigen specificities 5. The cluster itself was wrong, and any pre-1996 CD65 series may be reporting either antigen.

What to do instead - the five-line check

Before writing any identifier for a CD-numbered antigen:

  1. Look the CD number up in HGNC by alias AND by free text, and note whether the hit is an alias (probably right) or a previous symbol (check the biology - this is the CD15 trap).
  2. Ask what class of molecule it is. Protein, glycan, glycolipid, complex. If it is not a single polypeptide, no gene- id is legal.
  3. If it is a complex, name the chain the source measured - or record chain: unstated, which is usually the truth.
  4. Check whether the CD number has ever been split. CDw65 was. The w in a historical CDw designation means workshop provisional, i.e. not yet validated, and is a warning sign.
  5. Record the class in data, not only in prose, so the next instrument can filter on it. Every entry on this beat carries data.encoded_by_a_gene for exactly this reason.
Facts and figures
cases
[
  {
    "case": "one gene, one antigen",
    "examples": [
      "CD371 -> CLEC12A",
      "CD19",
      "CD22",
      "CD33"
    ],
    "lookup_behaviour": "correct"
  },
  {
    "case": "no gene (glycan)",
    "examples": [
      "CD15 (Lewis-x)",
      "CD65",
      "CD65s"
    ],
    "lookup_behaviour": "CD15 resolves WRONGLY to FUT4; CD65 returns an honest blank"
  },
  {
    "case": "more than one gene (complex)",
    "examples": [
      "HLA-DR = HLA-DRA + HLA-DRB1/3/4/5"
    ],
    "lookup_behaviour": "no single symbol; fails in both directions"
  },
  {
    "case": "one number, two antigens",
    "examples": [
      "CDw65: VIM-2 sialylated vs VIM-8/VIM-11 neutral"
    ],
    "lookup_behaviour": "ambiguity is in the reagent cluster, not the database"
  }
]
check
[
  "HGNC by alias AND free text; distinguish alias from previous symbol",
  "determine molecular class: protein / glycan / glycolipid / complex",
  "for a complex, name the measured chain or record chain: unstated",
  "check for a historical CDw designation or a documented cluster split",
  "record the class in data.encoded_by_a_gene, not only in prose"
]
How this could bite us back
  • **The obvious repair introduces a new error.** Adding HGNC previous symbols to the named-but-unwritten queue would make CD15 resolve to FUT4 and produce a confident wrong gene assignment for a two-dozen-entry antigen.
  • **This entry is about nomenclature and cannot make a biological claim.** Knowing that CD15 has no gene does not tell anyone which enzyme makes it on a leukaemic blast - that question is still open [see `target-cd15-lewis-x`].
  • **A clean CD-to-gene mapping is not a guarantee either.** CD371 maps correctly to CLEC12A, and CLEC12A's own alias `CLL1` collides with COLEC10 at twelve residues' difference. Resolving the CD number is the first check, not the last.
Still open
  1. How many other CD numbers in current use resolve to a gene via an HGNC previous symbol while naming a non-protein antigen? CD15 was found by hand; no systematic sweep exists and it would take one afternoon.
  2. How many CD clusters besides CDw65 were later shown to cover more than one antigen? No systematic survey was found on this beat.
  3. Should this graph add a required `data.encoded_by_a_gene` field to every antigen-like entry so the ambiguity is machine-readable rather than prose?
Every source behind this entry (6)
  1. E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    Queried live on 2026-09-20, HGNC returns CLEC12A for the string CD371 (as an alias symbol and as the top free-text hit), returns FUT4 for the string CD15 (as a previous official symbol and as the top free-text hit), and returns no records at all for the string CD65 by either route.

    HGNC record HGNC:4015 (FUT4) · HUGO Gene Nomenclature Committee · HGNC 2026
    What this evidence cannot show

    Measured against the live HGNC REST API on 2026-09-20 using `/search/alias_symbol/<s>` and `/search/<s>` for each string. **The CD371 and CD15 results are the positive controls that make the CD65 blank a result rather than a broken call.** This bounds HGNC's indexing on one date; nomenclature databases change. The claim is about what a lookup returns, and separately about whether that return is biologically correct - only the first half is a database fact.

    Evidence that argues against this (1)
    • HGNC is not wrong to list CD15 as a previous symbol of FUT4 - it is a true historical fact about nomenclature, and HGNC retired the symbol for exactly the reason this entry gives. The error would be made by an implementer reading a previous-symbol list as a CD-number-to-gene map, not by HGNC. HUGO Gene Nomenclature Committee 2026
  2. E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    CD15 is a previous official HGNC symbol of FUT4 (HGNC:4015, 'fucosyltransferase 4', alias name 'alpha (1,3) fucosyltransferase, myeloid-specific', UniProt P22083), while FUT4 encodes an enzyme that transfers fucose rather than encoding the CD15 epitope itself.

    HGNC record HGNC:4015 (FUT4) · HUGO Gene Nomenclature Committee · HGNC 2026
    What this evidence cannot show

    Fetched live from HGNC on 2026-09-20. The second half of the claim - that an enzyme gene does not encode its carbohydrate product - is biochemistry, not a database fact, and rests on the CD15 entry's sources rather than on HGNC.

    Evidence that argues against this (1)
    • FUT4 is nevertheless the single most useful gene-keyed proxy for CD15 in this disease, being a hub gene of the KMT2A-rearranged co-expression module. So 'CD15 is not FUT4' and 'FUT4 is the right gene to query about CD15' are both true, and an entry that states only the first is less useful than one that states both. Zhang H et al. 2020
  3. E3established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    The reviewed human UniProt record Q5QGZ9 gives CLEC12A as the primary gene name with CD antigen CD371, 265 aa, while the CD-numbered strings CD15 and CD65 have no corresponding single UniProt gene record.

    UniProtKB reviewed entry Q5QGZ9 (CL12A_HUMAN) · UniProt Consortium · UniProtKB 2026
    What this evidence cannot show

    Fetched live from the UniProt REST API on 2026-09-20. UniProt records CD antigen designations in the protein-name field for proteins that carry them; the absence of such a field for a glycan epitope is by construction, since UniProt is a protein database, and is therefore weak evidence on its own. It is corroborated here by the HGNC blank [E1].

    Evidence that argues against this (1)
    • Absence from a protein database is not evidence that no gene relates to an antigen - it is evidence that the antigen is not a protein. The distinction matters because CD15 does have related genes (FUT4, FUT9), just not an encoding one. UniProt Consortium 2026
  4. E4established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    Lewis-x (CD15) is a carbohydrate antigen synthesised by alpha-1,3-fucosyltransferases of the CAZy GT10 family, of which human FUT9 generates Lewis-x and Lewis-y, as established by crystal structures with GDP, acceptor glycans and a donor-analogue Michaelis complex.

    Structural basis for Lewis antigen synthesis by the alpha1,3-fucosyltransferase FUT9. · Kadirvelraj R, Boruah BM, Wang S, Chapla D, Huang C, Ramiah A, et al. · Nat Chem Biol 2023
    PMID 37202521DOI
    What this evidence cannot show

    Abstract read in full from raw NCBI efetch XML on 2026-09-20; full text not read. Structural enzymology on recombinant protein; the paper concerns neither leukaemia nor CD nomenclature and is cited for the single biochemical premise that Lewis antigens are enzymatic products of a family of enzymes rather than gene products.

    Evidence that argues against this (1)
    • The paper describes a COLLECTION of GT10 enzymes with modularly evolved specificity, which is the strongest form of the argument here - not merely that CD15 lacks a gene, but that no single-enzyme attribution is safe either. Kadirvelraj R et al. 2023
  5. E5established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    Antibodies given the provisional designation CDw65 at the IVth and Vth Workshops on Human Leukocyte Differentiation Antigens were found to bind the ganglioside X3-NeuAc-VII3Fuc-nLc10Cer; detailed study of VIM-2, VIM-8 and VIM-11 showed VIM-2 binding abolished by neuraminidase while VIM-8 and VIM-11 binding was enhanced, and that VIM-2 bound the sialylated decasaccharide while VIM-8 and VIM-11 bound its asialo form V3FucnLc8Cer - so VIM-2 and the other CDw65 antibodies represented two different antigen specificities.

    The CDw65 monoclonal antibodies VIM-8 and VIM-11 bind to the neutral glycolipid V3FucnLc8Cer. · Kniep B, Peter-Katalinic J, Muthing J, Majdic O, Pickl WF, Knapp W · J Biochem 1996
    PMID 8830039DOI
    What this evidence cannot show

    Abstract read in full from raw NCBI efetch XML on 2026-09-20; full text not read. Three antibodies out of a larger workshop cluster, studied on purified glycolipids. This establishes that one CD designation covered two specificities in this instance; it does not establish how common that is across the CD system, and no systematic survey was found on this beat.

    Evidence that argues against this (1)
    • One documented split does not make the CD system unreliable in general - the overwhelming majority of CD numbers do map cleanly to one protein. The argument of this entry is that you cannot tell which case you are in without checking, not that CD numbers are usually wrong. HUGO Gene Nomenclature Committee 2026
  6. E6established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    HLA-DR is a heterodimer of an alpha chain (HLA-DRA, UniProt P01903, 254 aa) and a beta chain (HLA-DRB1, UniProt P01911, 266 aa, or the products of HLA-DRB3, DRB4 or DRB5), so no single gene symbol denotes the surface marker.

    UniProtKB reviewed entry P01903 (DRA_HUMAN) · UniProt Consortium · UniProtKB 2026
    What this evidence cannot show

    Both records fetched live from the UniProt REST API on 2026-09-20, each returning exactly one reviewed human row whose primary gene name matched the query. That DRB3/4/5 also form DR dimers is standard immunology and was not separately fetched for this entry.

    Evidence that argues against this (1)
    • A great many CD-numbered antigens are multi-subunit (CD3, CD11b/CD18, the T-cell receptor complex) without anyone being confused, because their component genes are routinely named. HLA-DR is a problem in this graph specifically because clinical flow reports write 'HLA-DR' and never a chain. UniProt Consortium 2026
Connected to