Oncofusion

start / graph / gene-c2cd3

Genemoderate confidence 3 of 5: One strong study, or several smaller consistent ones.Q96-sarcoma-mbm-partners

C2CD3 - the centriole-elongation regulator fused to KMT2A exon 8 in an MPNST, predicted OUT OF FRAME, and the partner whose RefSeq transcript is 390 residues shorter than UniProt's canonical

also called C2CD3 · C2 domain-containing protein 3 · C2 calcium dependent domain containing 3 · Q4AC94

C2CD3 builds the tip of the centriole, the little barrel a cell needs in order to grow a cilium and to read the Hedgehog signal. In one patient's malignant peripheral nerve sheath tumour KMT2A broke at exon 8, keeping the whole leukaemia-type front end, and joined C2CD3 at a point that is predicted to be out of frame.

The same thing, for a scientist

C2CD3 is a **positive regulator of centriole elongation** and the gene of **orofaciodigital syndrome 14**. It localises to the **distal end of the centriole**, associates physically with OFD1, and the two act antagonistically: OFD1 loss causes centriole hyperelongation while C2CD3 loss gives **short centrioles with no subdistal or distal appendages** [E1]. It is required for assembly of the distal appendage - and therefore for recruiting CEP83, SCLT1, CEP89, FBF1 and CEP164 - for **primary cilium formation**, for **sonic hedgehog signalling** and for **proteolytic processing of GLI3** [E2, E3]. Its architecture is **six C2 domains** in a large, heavily disordered protein with no catalytic module [E3]. **In case S14 the fusion is `KMT2A exon 8 :: C2CD3 exon 24`** [E4]. KMT2A supplies aa 1-1362, the complete leukaemic N-terminal module - MBM, IBM1, IBM2, all three AT-hooks, CXXC - with the break **inside the leukaemic breakpoint cluster region** [E5]. This is the second of the three cases in the cohort that carry a leukaemia-type KMT2A portion, and the diagnosis is the same as case S12: malignant peripheral nerve sheath tumour. **But the junction is predicted OUT OF FRAME.** C2CD3 exon 24 in NM_015531 begins 4,603 nucleotides after the ATG - not a multiple of three - while every KMT2A donor exon in this cohort ends at a complete codon. So the predicted product is a **KMT2A 1-1362 fragment with a short frameshifted tail**, not a chimeric protein [E6]. Two of the seven cases are predicted frameshifted (this one and RALGAPA2) and one joins an untranslated exon (AHDC1), so **three of seven supply essentially no partner protein**. **Identifier trap, and it is the one this beat would most like recorded:** the transcript the source cites, **NM_015531, encodes 1,963 aa, while UniProt's canonical Q4AC94 is 2,353 aa** - a 390-residue, 20% difference *within the same gene*. Every C2 domain coordinate above is in the 2,353 frame, so **partner-domain retention cannot be read off an exon number without aligning the two** [E7].

What it looks like
C2CD32,353 aa
C2 dC2 domain 1 · residues 521–678C2 dC2 domain 2 · residues 787–919C2 dC2 domain 3 · residues 985–1147C2 domain 4C2 domain 4 · residues 1171–1339C2 dC2 domain 5 · residues 1403–1533C2 dC2 domain 6 · residues 1617–174505001,0001,5002,000

No three-dimensional model is available for this protein, so the map above is the whole picture. That is itself informative: AlphaFold does not model chains this long, and the parts that have been crystallised are held under the entries for the individual domains.

Go deeper
What C2CD3 normally does

C2CD3 sets centriole length. It was found in a genome-wide siRNA screen of centriole biogenesis as a gene required for distal centriole formation, with the authors' reading that it functions in the basal body to template primary cilia 2; and independently as the gene of a new subtype of orofaciodigital syndrome with severe microcephaly and cerebral malformations, where it colocalises with OFD1 at the distal end of centrioles, physically associates with OFD1, and the two oppose each other - OFD1 deletion causes hyperelongation, C2CD3 loss gives short centrioles without subdistal or distal appendages, and C2CD3 overexpression drives hyperelongation which OFD1 antagonises 1.

Downstream of that structure, C2CD3 is required for assembly of the distal appendage - the anchor of the cilium - and hence for recruiting CEP83, SCLT1, CEP89, FBF1 and CEP164; it is required for primary cilium formation, for sonic hedgehog signalling and for proteolytic processing of GLI3; and it is not required for centriolar satellite integrity or RAB8 activation, which is a useful negative 3. It associates with the BBSome and interacts with IFT88, BBS4 and PCM1 (all ECO:0000250, by similarity) 3.

The GLI link is the functionally interesting one for a cancer graph: a ciliopathic C2CD3 mutant, the avian talpid2, is one of the models used to dissect GLI-mediated transcription in craniofacial development 8. C2CD3 is therefore upstream of a transcription-factor family, via a structure rather than via chromatin.

What the fusion joins - and why the frame matters here
sidegeneexonsresiduesmodules supplied
5-primeKMT2A1-8aa 1-1362MBM 6-25, IBM1 123-134, IBM2 147-152, AT-hooks 169-309, CXXC 1147-1195
3-primeC2CD324-31aa 1535-1963 in the NM_015531 frame - but predicted out of framenone predicted

The KMT2A side is one of the three strongest in the cohort: exon 8 is inside the leukaemic breakpoint cluster region (exons 8-14) and the retained module is complete, so nothing about it distinguishes this MPNST from an infant-leukaemia fusion 5.

The partner side is predicted not to be translated in its own frame. In NM_015531 (CDS 211..6102), exon 24 spans nt 4814-5161, which begins 4,603 nt after the ATG - and 4,603 is not divisible by three. KMT2A exon 8 ends at a complete codon. So the junction is predicted to shift C2CD3's reading frame, and the product would be a KMT2A 1-1362 protein terminating shortly after the junction 6. That is a KMT2A truncation, not a chimera - a different lesion class, with a different therapeutic logic and a different set of things worth measuring.

How much of this beat rests on the frame calculation is worth stating plainly. Of the seven cases, four are predicted in frame (CPT1A, DOCK10, ELP4, TMPRSS13), two frameshifted (C2CD3, RALGAPA2) and one joins an untranslated exon (AHDC1). The calculation is a derivation from RefSeq exon coordinates and can be wrong if the commercial panel's exon numbering differs from RefSeq's, or if a breakpoint is intra-exonic. The KMT2A-side conclusions do not depend on it.

The isoform trap - and why length readback does not catch it

The transcript the source cites and the protein UniProt shows you are 390 residues apart, in the same gene.

recordlengthstatus
UniProt Q4AC94 (canonical, isoform 5)2,353 aadisplayed
RefSeq NM_015531.6 / NP_056346.31,963 aaone of four described isoforms

Every C2-domain boundary in this entry - 521-678, 787-919, 985-1147, 1171-1339, 1403-1533, 1617-1745 - is annotated on the 2,353 frame. The exon-to-residue mapping used for the breakpoint is on the 1,963 frame. The two cannot be combined without an alignment, and this entry does not attempt one: it reports C2CD3 exon 24 as aa 1535-1651 in the NM_015531 frame and makes no claim about which C2 domains a hypothetical in-frame product would retain 7.

This is the family of defects this project already documents for PDB numbering - ASH2L papers numbering from the 534-residue isoform while UniProt is 628, a constant offset of 94 - and it has a specific consequence for the corpus's own defences. verify-accessions.py asserts accession plus length and so catches a wrong gene. It cannot catch a wrong isoform of the right gene, and here the two differ by 20%. The same trap sits one case over in this beat, more subtly: TMPRSS13's cited NM_001077263 encodes 567 aa against UniProt's 586, a 19-residue offset, which is small enough to look like a rounding error and large enough to move an active-site residue.

Epigenetics (PI dimension 3) - inbound only, and the writer is a histone methyltransferase

C2CD3 writes, erases and reads no mark. Its architecture is six C2 domains - calcium- and phospholipid-binding modules - plus long disordered stretches, with no SET, JmjC, HAT, HDAC, bromodomain, chromodomain, Tudor, PHD or CXXC module anywhere in 2,353 residues 3. It is a cytoplasmic, centriole-associated structural protein. That is the NOT-APPLICABLE statement, justified on architecture in the gene-supt6h form.

Inbound, there is exactly one result and it is a good one. C2CD3[tiab] AND histone[tiab] returns 1 record, and it reports that SMYD3 - a histone/lysine methyltransferase - binds the promoter region of C2cd3 and regulates the transcription of cilia genes, alongside Cep164, Ttbk2, Dync2h1 and Cp110; SMYD3 is itself a novel component of the distal appendage and is required for centriolar appendage assembly, and its loss reduces ciliation and gives stumpy cilia 9. So the C2CD3 promoter is a direct SMYD3 target, which makes C2CD3 an epigenetically regulated gene even though it is not an epigenetic protein.

And note the convergence inside this beat. SMYD3 is also the histone methyltransferase that drives fatty-acid oxidation through the FABP5/PPARD/CPT1A axis in CML leukaemia stem cells (see gene-cpt1a). Two of these seven unrelated KMT2A sarcoma partners are downstream of the same histone lysine methyltransferase, by two entirely different routes. That is almost certainly coincidence at n = 2 out of 7 - SMYD3 has a large and heterogeneous target literature and nothing was pre-specified - and it is recorded as an observation with its null stated rather than as a pattern. SMYD3 has no entry in this graph.

C2CD3[tiab] AND chromatin[tiab] returns 1, and it is the talpid2 GLI-transcription review, which is about transcription rather than about chromatin 8. Control: KMT2A[tiab] AND histone[tiab] = 218 through the identical shape.

Localization, cell type and cell state (PI dimensions 1 and 2)

Subcellular: the distal end of the centriole - and this is the most precise localisation of the seven partners. UniProt gives Cytoplasm, cytoskeleton, microtubule organizing center, centrosome, centriole with two ECO:0000269s (PMIDs 23769972 and 24997988), and separately Cytoplasm, cytoskeleton, cilium basal body as ECO:0000250 - by similarity, an inference rather than a human measurement 3.

Rule 91, checked: the OFD paper does support a distal-centriole assignment - it reports that C2CD3 colocalises with OFD1 at the distal end of centrioles 1. The other coded citation is a genome-wide siRNA screen whose own wording is that the result "suggests that it functions in the basal body to template primary cilia" 2 - a functional screen with a suggestion, which is weaker than the annotation implies. So the centriole call is sound and the basal-body call is an inference twice over.

Subnuclear: not applicable, correctly. C2CD3 is a centriolar protein with no reported nuclear pool. As with CPT1A in this cohort, the fusion would join a nuclear-addressed KMT2A module to a cytoplasmic, cytoskeletal protein - except that here the partner is predicted not to be translated at all.

Cell type: no lineage resolution whatever. C2CD3's evidence is in human cell lines in a centriole screen, in patient fibroblasts and families with orofaciodigital syndrome, and in the avian talpid2 mutant. There is no haematopoietic and no mesenchymal or sarcoma measurement, and C2CD3 has essentially no cancer literature at all: C2CD3[tiab] AND cancer[tiab] returns 2 and both are off topic (a salivary-microbiome in-silico study and a vitamin-A paper) 10. Cell state: ciliated. The only state that C2CD3's biology distinguishes is whether a cell has built a primary cilium - which is a state most tumour cells lose, and which nobody has examined in a KMT2A-rearranged tumour.

Analogous biology in other fusion families (PI dimension 5)

C2CD3 is not a fusion partner anywhere, and this is the cleanest negative of the seven. C2CD3[tiab] AND fusion[tiab] = 0 and C2CD3[tiab] AND translocation[tiab] = 0; C2CD3[tiab] AND KMT2A[tiab] = 0 and AND MLL[tiab] = 0. Controls through the identical shape: EWSR1 + fusion 1,208, EWSR1 + translocation 462, AFF1 + KMT2A 104. The whole C2CD3[tiab] literature is 42 records 10.

So the analogous biology has to be looked for somewhere other than fusion genetics, and there is one place worth naming: the cilium as a signalling organelle is a recurring route into transcription-factor processing, and C2CD3 is required for the proteolytic processing of GLI3 38. A structural protein that controls whether a transcription factor is cleaved into its repressor form is mechanistically the same kind of thing as a scaffold that controls whether a methyltransferase reaches chromatin - and in Ewing sarcoma, one of these seven partners (AHDC1) turns out to be exactly such a scaffold for EWS-FLI1. There is no evidence connecting C2CD3 to any fusion oncoprotein, and this paragraph is an analogy, not a finding.

Status: what this case is and is not
  • One patient. Case S14, malignant peripheral nerve sheath tumour - one of the three cases breaking inside the leukaemic breakpoint cluster region 4.
  • The partner is predicted frameshifted, so the product is most likely a KMT2A N-terminal fragment rather than a chimera 6.
  • The cited transcript is not UniProt's canonical isoform, and the two differ by 390 residues, so no partner-domain retention claim is made here 7.
  • No fusion protein, no transcript sequence, no reciprocal product, no functional assay.
  • DepMap: positive in every arm. Mean Chronos +0.064 in 11 KMT2A-rearranged lines, +0.029 in 43 other leukaemia lines, pan-essentiality +0.020 over 1,208 lines 11. Not required anywhere, so no dependency claim is possible in either direction.
  • Not a treatment rationale. See risk-a-menin-inhibitor-prescribed-off-a-kmt2a-gene-name-in-a-solid-tumour.
Facts and figures
uniprotQ4AC94
gene symbolC2CD3
length AA2353
accession resolutiongene_exact:C2CD3 AND organism_id:9606 AND reviewed:true returned exactly one row, primary gene name C2CD3, entry C2CD3_HUMAN, canonical length 2,353 aa (isoform 5, displayed). Primary name read back and length asserted - BUT SEE isoform_frame_trap: the length asserted is NOT the length of the transcript the fusion source cites.
refseq transcriptNM_015531.6 / NP_056346.3 (31 exons, CDS 211..6102, 1,963 aa; exon 24 = nt 4814-5161 = aa 1535-1651 IN THAT FRAME)
isoform frame trapTHE CITED TRANSCRIPT AND THE CANONICAL PROTEIN DIFFER BY 390 RESIDUES IN THE SAME GENE. UniProt Q4AC94 canonical (isoform 5) = 2,353 aa; RefSeq NM_015531.6 / NP_056346.3 = 1,963 aa, one of five listed isoforms. All six C2-domain boundaries below are annotated on the 2,353 frame while the exon-to-residue mapping used for the breakpoint is on the 1,963 frame, so THE TWO CANNOT BE COMBINED WITHOUT AN ALIGNMENT and no alignment was done here: this entry deliberately makes NO claim about which C2 domains an in-frame product would retain. Same family as the ASH2L 534-versus-628 offset this project already records for PDB numbering. AND NOTE WHAT DOES NOT CATCH IT: verify-accessions.py asserts accession plus length, which catches a wrong GENE and not a wrong ISOFORM of the right gene. The same trap sits on TMPRSS13 in this beat at a smaller and therefore more dangerous magnitude - NM_001077263 encodes 567 aa against UniProt's 586.
domain coordinate frameUniProt Q4AC94 canonical, 2,353 aa - NOT the 1,963-aa NM_015531 frame used for the breakpoint
epigenetic readoutWRITES, ERASES AND READS NO HISTONE MARK, justified on domain architecture in the gene-supt6h form: C2CD3's modules are SIX C2 DOMAINS - calcium- and phospholipid-binding - plus extensive disordered regions, with no SET, JmjC, HAT, HDAC, bromodomain, chromodomain, Tudor, PHD or CXXC module anywhere in 2,353 residues, and it is a cytoplasmic centriole-associated structural protein with no reported nuclear pool. BUT NOT 'NOT APPLICABLE', BECAUSE THE INBOUND LINK IS REAL AND SPECIFIC: SMYD3, a histone/lysine methyltransferase, BINDS THE PROMOTER REGION OF C2cd3 and regulates the transcription of cilia genes including C2cd3, Cep164, Ttbk2, Dync2h1 and Cp110; SMYD3 is itself a novel distal-appendage component required for centriolar appendage assembly, and its loss reduces ciliation and gives stumpy cilia (PMID 38892227 - gene names written in mouse style and the cell system is not stated in the abstract, so treat the species as unresolved). So the C2CD3 PROMOTER IS A DIRECT TARGET of a histone lysine methyltransferase, which makes C2CD3 an epigenetically regulated gene without being an epigenetic protein. CONVERGENCE WORTH RECORDING WITH ITS NULL: SMYD3 is ALSO the methyltransferase driving fatty-acid oxidation through the FABP5/PPARD/CPT1A axis in CML leukaemia stem cells, so TWO of these seven unrelated partners are downstream of the same enzyme by two different routes - almost certainly coincidence at 2 of 7 with nothing pre-specified and a large heterogeneous SMYD3 target literature, and recorded as an observation rather than a pattern. SMYD3 HAS NO ENTRY IN THIS GRAPH. BOUNDED 2026-09-22, one tag per term: C2CD3[tiab] AND histone[tiab] = 1 (that paper), C2CD3[tiab] AND chromatin[tiab] = 1 (PMID 27799912, the talpid2 GLI-transcription review - about TRANSCRIPTION, not chromatin, and recorded so a later beat does not count it), against KMT2A[tiab] AND histone[tiab] = 218 through the identical shape.
non histone substratenone - C2CD3 is a structural protein, not a transferase, and modifies nothing covalently.
subnuclear localizationNOT APPLICABLE, and correctly so rather than as a gap: C2CD3 is a centriolar and basal-body protein with no reported nuclear pool in any system. Worth stating because of what the fusion would do - join a NUCLEAR-ADDRESSED KMT2A module, CXXC included, to a CYTOPLASMIC CYTOSKELETAL protein - except that the junction here is predicted frameshifted, so the partner is probably not translated at all.
cell typeNO LINEAGE RESOLUTION WHATEVER. C2CD3's evidence comes from human cell lines in a centriole biogenesis screen, from patients and families with orofaciodigital syndrome 14, and from the avian talpid2 mutant. There is no haematopoietic, mesenchymal or sarcoma measurement, and the gene has essentially no cancer literature: C2CD3[tiab] = 42 records in total and C2CD3[tiab] AND cancer[tiab] = 2, both off topic.
cell stateCILIATED versus UNCILIATED is the only state C2CD3's biology distinguishes: loss gives short centrioles without appendages and no primary cilium, with downstream failure of sonic hedgehog signalling and of GLI3 processing. Most cultured tumour cells have lost their primary cilium, and nobody has examined ciliation in a KMT2A-rearranged tumour of any lineage.
localization shown for fusionNO - 'nobody has looked in any fusion' (rule 86). C2CD3[tiab] AND KMT2A[tiab] = 0 and AND MLL[tiab] = 0, and no sarcoma KMT2A fusion protein has been visualised by any method.
fusion partner in other tumoursNONE REPORTED, and this is the cleanest negative of the seven: C2CD3[tiab] AND fusion[tiab] = 0, AND translocation[tiab] = 0, AND KMT2A[tiab] = 0, AND MLL[tiab] = 0, with EWSR1 controls of 1,208 and 462 through the identical shape. It is also the weakest negative in terms of what it licenses, because the entire C2CD3[tiab] literature is 42 records - a gene nobody studies will return zero for almost any conjunction. THE ANALOGY, EXPLICITLY LABELLED AS ANALOGY AND NOT A FINDING: C2CD3 is required for the PROTEOLYTIC PROCESSING OF GLI3, so it is a structural protein that controls whether a transcription factor is converted to its repressor form. That is mechanistically the same KIND of thing as a scaffold controlling whether a methyltransferase reaches chromatin - and in Ewing sarcoma one of these same seven partners, AHDC1, turns out to be exactly such a scaffold for EWS-FLI1. No evidence connects C2CD3 to any fusion oncoprotein.
absence classnever-measured
domains
C2 domain 1521.0C2 domain 1: 521C2 domain 2787.0C2 domain 2: 787C2 domain 3985.0C2 domain 3: 985C2 domain 41,171C2 domain 4: 1171C2 domain 51,403C2 domain 5: 1403C2 domain 61,617C2 domain 6: 1617
aliases and retired symbols searched
[
  "C2CD3",
  "C2 domain-containing protein 3",
  "C2 calcium dependent domain containing 3"
]
domains — the numbers behind the chart
[
  {
    "name": "C2 domain 1",
    "start": 521,
    "end": 678
  },
  {
    "name": "C2 domain 2",
    "start": 787,
    "end": 919
  },
  {
    "name": "C2 domain 3",
    "start": 985,
    "end": 1147
  },
  {
    "name": "C2 domain 4",
    "start": 1171,
    "end": 1339
  },
  {
    "name": "C2 domain 5",
    "start": 1403,
    "end": 1533
  },
  {
    "name": "C2 domain 6",
    "start": 1617,
    "end": 1745
  }
]
functions
[
  "positive regulator of centriole elongation",
  "assembly of the centriolar distal appendage and recruitment of CEP83, SCLT1, CEP89, FBF1 and CEP164",
  "primary cilium formation",
  "sonic hedgehog signalling",
  "proteolytic processing of GLI3",
  "physical association with and antagonism by OFD1",
  "NOT required for centriolar satellite integrity or RAB8 activation"
]
sarcoma case
{
  "patient": "S14",
  "diagnosis": "malignant peripheral nerve sheath tumour",
  "fusion": "KMT2A exon 8 :: C2CD3 exon 24 (NM_015531)",
  "kmt2a_polarity": "5-prime - the LEUKAEMIC orientation, opposite to the recurrent sarcoma arms",
  "kmt2a_exons_retained": "1-8",
  "kmt2a_residues_retained": "aa 1-1362",
  "kmt2a_modules_retained": [
    "MBM 6-25",
    "IBM1 123-134",
    "IBM2 147-152",
    "AT-hooks 169-309",
    "CXXC 1147-1195"
  ],
  "mbm_retained": true,
  "cxxc_retained": true,
  "in_leukaemic_breakpoint_cluster_region": true,
  "partner_residues_predicted": "none translated in the partner's own frame - exon 24 is aa 1535-1651 in the NM_015531 frame but the junction is predicted frameshifted",
  "partner_modules_predicted_retained": [],
  "predicted_frame": "OUT OF FRAME (C2CD3 exon 24 begins 4,603 nt after the ATG; 4603 % 3 == 1, while KMT2A exon 8 ends at a complete codon)",
  "frame_derivation": "computed 2026-09-22 from NM_001197104.2 and NM_015531.6 exon coordinates - a derivation, not an observation; it changes the predicted LESION CLASS for this case, from chimera to KMT2A N-terminal fragment, and the KMT2A-side conclusions do not depend on it",
  "source": "Supplemental Table 2 of PMID 32461620",
  "fusion_protein_observed": false,
  "reciprocal_transcript_reported": false,
  "functional_assay_done": false,
  "authors_own_verdict": "of unclear significance"
}
localization
{
  "CONSTITUTIVE": "CYTOPLASM, CYTOSKELETON, MICROTUBULE ORGANIZING CENTRE, CENTROSOME, CENTRIOLE - specifically the DISTAL END of the centriole, where it colocalises with OFD1 (PMID 24997988). Two ECO:0000269s (PMIDs 23769972, 24997988). RULE 91, MIXED: the OFD paper genuinely reports colocalisation at the distal centriole, but the other coded citation is a genome-wide siRNA screen whose own wording is that the result 'suggests that it functions in the basal body to template primary cilia' - a functional inference standing behind a coded localisation. The separate 'cilium basal body' comment is ECO:0000250, by similarity, and is not a human measurement at all.",
  "systems": "human cell lines in a centriole screen; patient cells and families with orofaciodigital syndrome; the avian talpid2 mutant. No haematopoietic and no mesenchymal or sarcoma measurement."
}
depmap kmt2a selectivity
{
  "chronos_mean_kmt2a_rearranged": 0.0643,
  "chronos_mean_other_leukaemia": 0.0286,
  "difference": 0.0357,
  "pan_essentiality_all_lines": 0.02,
  "n_all_lines": 1208,
  "selectivity_rank": 13487,
  "rank_denominator": 17728,
  "provenance": "Joined 2026-09-22 from analyses/depmap-kmt2a-selective-scan-full/results.json (generated 2026-09-19T10:12:06Z), this project's own genome-wide scan over 11 KMT2A-rearranged and 43 other leukaemia DepMap lines. NOT re-fetched and NOT independent of that scan: every caveat it carries applies, including that the 11 KMT2A-rearranged lines are drawn from nine individuals and that the arm is myeloid-enriched. The scan records a generation timestamp and NOT a DepMap RELEASE, which is rule 83 exposure carried forward rather than fixed here.",
  "how_to_read": "A rank is not an effect size and a difference is not a dependency (this graph's DepMap rules). Check pan_essentiality_all_lines before reading anything as selective, and note the anti-selective half is 50% of any gene set by construction, with this graph's own base rate at 45.3% - so a tally of genes in that half is the null (rule 84). The decisive limitation here is lineage: the screen is leukaemia lines, and for a sarcoma fusion that is the wrong tissue. Cellosaurus holds 592 KMT2A-commented cell lines and ZERO are sarcoma or any mesenchymal tumour, so no cell-line screen can address these cases at all. Every mean here is POSITIVE, so C2CD3 is not a dependency anywhere. And a specific blindness: primary cilia are lost in most cultured cancer cell lines, so a screen across 1,208 such lines is close to blind to a ciliary dependency by construction."
}
queries run 2026 09 22
[
  "C2CD3[tiab] -> 42",
  "C2CD3[tiab] AND fusion[tiab] -> 0",
  "C2CD3[tiab] AND translocation[tiab] -> 0",
  "C2CD3[tiab] AND KMT2A[tiab] -> 0",
  "C2CD3[tiab] AND MLL[tiab] -> 0",
  "C2CD3[tiab] AND histone[tiab] -> 1 (PMID 38892227, ON topic - SMYD3 binds the C2cd3 promoter)",
  "C2CD3[tiab] AND chromatin[tiab] -> 1 (PMID 27799912, talpid2 GLI review - transcription, not chromatin)",
  "C2CD3[tiab] AND cancer[tiab] -> 2, both off topic",
  "controls: EWSR1[tiab] AND fusion[tiab] -> 1208; EWSR1[tiab] AND translocation[tiab] -> 462; AFF1[tiab] AND KMT2A[tiab] -> 104; MLLT3[tiab] AND MLL[tiab] -> 84; KMT2A[tiab] AND histone[tiab] -> 218"
]
graph neighbours already present
[
  "gene-kmt2a",
  "disease-kmt2a-r-sarcoma",
  "concept-kmt2a-fusion-polarity-is-inverted-in-sarcoma"
]
How this could bite us back
  • This entry must not be read as a menin-inhibitor rationale. Menin is a TUMOUR SUPPRESSOR in mesenchyme, and the only non-haematologic menin trial with a result, NCT05731947, was terminated by the sponsor after phase 1b for not meeting pre-defined efficacy signals. An MBM-retaining KMT2A fusion in a sarcoma is a test of a hypothesis, not an indication.
  • If menin inhibition were tried here off the gene name alone, the predicted direction of effect is not obviously anti-tumour: displacing menin from chromatin in a mesenchymal cell removes a tumour suppressor from its targets. See risk-a-menin-inhibitor-prescribed-off-a-kmt2a-gene-name-in-a-solid-tumour.
  • Every quantitative statement about the fusion in this entry is a coordinate derivation from transcript records. If the panel's exon numbering differs from RefSeq's, the frame prediction, the retained-residue count and the MBM claim all move together - one systematic error, not several independent ones.
  • There is no C2CD3 intervention to propose and the reason is structural: the gene is not required in any DepMap arm (every mean positive), its loss of function in humans causes a severe developmental ciliopathy with microcephaly and cerebral malformations, and the fusion is predicted not to translate it.
  • If the junction is out of frame, then reading this case as a chimeric oncoprotein is wrong in a way that would propagate: it would be an argument for targeting a partner the tumour does not express, built on an exon number in a supplementary table.
  • The isoform trap generalises beyond this entry. Any partner-domain-retention claim in this cohort that is derived from an exon number and read against UniProt coordinates is unsafe unless the two frames were aligned, and the corpus's own accession checker cannot detect the error.
Still open
  1. Is the KMT2A::C2CD3 junction actually out of frame? The prediction changes the lesion class for this case - a KMT2A N-terminal fragment rather than a chimera - and it is settled by one RT-PCR across the junction, or by the panel's own transcript model.
  2. Which C2CD3 isoform does an MPNST express? UniProt lists five and the cited transcript is 390 residues shorter than the canonical, so even the question of what the partner allele encodes is unanswered.
  3. Does the KMT2A 1-1362 fragment predicted here do anything on its own? Three of the seven cases are predicted to give KMT2A N-terminal fragments rather than chimeras, which makes the behaviour of an isolated MBM-plus-CXXC module a question this beat raises three times and cannot answer.
  4. Is SMYD3 worth an entry in this graph? It binds the C2CD3 promoter, drives fatty-acid oxidation through CPT1A in leukaemia stem cells, and is a histone lysine methyltransferase with no node here - two of these seven partners reached it independently.
  5. Do KMT2A-rearranged tumours of any lineage retain primary cilia? C2CD3's entire biology is ciliary, most cultured tumour cells have lost the cilium, and nobody has looked in this disease - which also means DepMap cannot answer a ciliary dependency question at all.
Every source behind this entry (11)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    C2CD3 is a positive regulator of centriole elongation that colocalises with OFD1 at the distal end of centrioles and physically associates with it: OFD1 deletion causes centriole hyperelongation while loss of C2CD3 gives short centrioles without subdistal or distal appendages, C2CD3 overexpression triggers hyperelongation, and OFD1 antagonises that activity; distinct mutations in C2CD3 cause a new subtype of oral-facial-digital syndrome with severe microcephaly and cerebral malformations.

    The oral-facial-digital syndrome gene C2CD3 encodes a positive regulator of centriole elongation. · Thauvin-Robinet C, Lee JS, Lopez E et al · Nature Genetics 2014
    PMID 24997988DOI
    What this evidence cannot show

    Human genetics in two affected families plus cell-biological experiments in human cells; a developmental disorder, not a tumour, and no haematopoietic or mesenchymal system. Abstract read from the eutils record; full text not read, so the imaging modality behind 'colocalises at the distal end' and the overexpression construct are taken as described. This is one of the two ECO:0000269 citations behind UniProt's centriole localisation comment and rule 91 passes for it - it genuinely reports a localisation experiment.

    Evidence that argues against this (1)
    • An overexpression phenotype (hyperelongation) and a loss-of-function phenotype (short centrioles) bracket a dose-dependent structural role, which is not the same as showing that C2CD3 measures length. The paper's own framing - that regulation of centriole length is an 'emerging' pathogenic mechanism - is appropriately tentative, and the OFD1 antagonism is inferred from opposing phenotypes plus an association rather than from a reconstituted activity. Balestra FR et al. 2013
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    A high-content genome-wide siRNA screen for centriole formation in human cells identified 44 genes required for centriole formation and 32 needed to restrict centriole number, and found that C2CD3 is required for distal centriole formation, which the authors read as suggesting it functions in the basal body to template primary cilia.

    Discovering regulators of centriole biogenesis through siRNA-based functional genomics in human cells. · Balestra FR, Strnad P, Flueckiger I et al · Developmental Cell 2013
    PMID 23769972DOI
    What this evidence cannot show

    Human cells, genome-wide siRNA with automated GFP-centrin focus detection plus manual follow-up - a screen, so C2CD3 is one hit among 76 and the follow-up characterisation is the paper's own secondary work. IMPORTANT FOR RULE 91: this is the second ECO:0000269 citation behind UniProt's 'centrosome, centriole' localisation comment, and its statement about WHERE C2CD3 acts is explicitly a SUGGESTION ('suggests that it functions in the basal body'), not a measurement - so part of UniProt's coded localisation for this protein rests on a functional inference. The separate 'cilium basal body' comment on the same record is ECO:0000250, by similarity, and is not a human measurement at all. Abstract read; full text not read.

    Evidence that argues against this (1)
    • siRNA screens carry off-target and seed-based artefacts, and a requirement for centriole formation does not localise a protein - the inference from phenotype to compartment is exactly the move rule 89 warns about in this graph's own coverage metrics. The distal-centriole localisation is better supported by the OFD paper's direct colocalisation with OFD1. Thauvin-Robinet C et al. 2014
  3. E3established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    C2CD3 (Q4AC94, canonical isoform 2,353 aa) is built from six C2 domains with extensive disordered regions and no catalytic or chromatin-binding module; it promotes assembly of the centriolar distal appendage and is required for recruiting CEP83, SCLT1, CEP89, FBF1 and CEP164, for primary cilium formation, for sonic hedgehog signalling and for proteolytic processing of GLI3, while being dispensable for centriolar satellite integrity and RAB8 activation.

    What this evidence cannot show

    Accession resolved this cycle by gene_exact query with gene_primary read back: exactly one reviewed human record, primary symbol C2CD3, entry C2CD3_HUMAN (rules 62, 65, 66). THE LENGTH ASSERTED HERE IS THE CANONICAL ISOFORM 5 AT 2,353 aa, AND IT IS NOT THE ISOFORM THE FUSION SOURCE CITES: NM_015531.6/NP_056346.3 encodes 1,963 aa, 390 residues shorter, and UniProt lists five isoforms. All six C2-domain boundaries are annotated on the 2,353 frame. The SUBUNIT interactions with IFT88, BBS4 and PCM1 are ECO:0000250 (by similarity), as is the 'cilium basal body' location; the SHH and GLI3 statements are carried without a coded human citation in the comment. Domain calls are UniProt feature-table annotations; there is no experimental structure of C2CD3.

    Evidence that argues against this (1)
    • A C2 domain is a calcium- and phospholipid-binding module, and annotating six of them does not establish what C2CD3 binds at the centriole - the protein's functional assignments all come from loss-of-function phenotypes rather than from a demonstrated molecular activity. And with five isoforms and a 390-residue spread, a domain-content claim is isoform-specific and this entry cannot say which isoform the tumour expressed. NCBI RefSeq 2026
  4. E4moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    In case S14 of a pan-sarcoma KMT2A-rearrangement series, a malignant peripheral nerve sheath tumour carried KMT2A exon 8 fused to C2CD3 exon 24 (NM_015531), one of seven cases with KMT2A 5-prime and one of the three whose breakpoint lies inside the leukaemic breakpoint cluster region.

    PMID 32461620DOI
    What this evidence cannot show

    This is ONE patient (case S14, malignant peripheral nerve sheath tumour) in a commercial targeted RNA-panel series enriched for advanced disease. The source authors call the non-recurrent KMT2A partners of UNCLEAR SIGNIFICANCE and several may be passengers. No fusion protein has been observed for this or any other sarcoma KMT2A fusion by any method, no reciprocal transcript was reported, no functional experiment has been done, and the panel targets specific KMT2A exons, which biases which breakpoints are detectable at all. Supplemental Table 2 reports EXON NUMBERS from the panel's own annotation - the junction nucleotide, the transcript sequence and the reading frame are not given - and it misspells the index gene as KTM2A in two of eighteen rows, so a reader grepping for KMT2A silently loses cases. This graph's parse is at analyses/sarcoma-nonrecurrent-partners/results.json. Naming this case makes it auditable; it does not promote the partner to a driver and it is not a treatment rationale.

    Evidence that argues against this (1)
    • The cohort's recurrent, morphology- and methylation-defined KMT2A sarcoma arms are YAP1-KMT2A-YAP1 and VIM::KMT2A, and those are what the paper's title and conclusions are about. The seven 5-prime non-recurrent cases are explicitly set aside by the authors as of unclear significance, and a targeted panel run in advanced disease will also detect passenger rearrangements. A single MBM-retaining case is a lead, not an entity. Massoth LR et al. 2020
  5. E5strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

    KMT2A exons 1-8 encode aa 1-1362 and therefore retain the menin-binding motif (6-25), LEDGF IBM1 (123-134) and IBM2 (147-152), all three AT-hooks and the CXXC domain (1147-1195); the exon-8 break lies inside the leukaemic breakpoint cluster region (exons 8-14), so the KMT2A portion of this fusion is not distinguishable from that of an infant-leukaemia KMT2A fusion.

    What this evidence cannot show

    KMT2A exon-to-residue map, re-derived this cycle from RefSeq NM_001197104.2 (CDS 22..11940, 36 exons, NP_001184033.1) and independently reproducing this graph's own ana-kmt2a-exon-to-domain-map: exon 1 = aa 1-144 (MBM 6-25, LEDGF IBM1 123-134), exon 2 = aa 145-168 (IBM2 147-152), exon 3 = aa 168-1052 (all three AT-hooks), exon 4 = 1053-1112, exons 5-6 = 1112-1212 (CXXC 1147-1195). Exon 3 ends at aa 1052, exon 8 at 1362, exon 9 at 1406, exon 15 at 1668, exon 20 at 1888, exon 22 at 1987. All six KMT2A donor exons in this cohort end at a COMPLETE CODON (phase 0), which is why KMT2A can fuse in frame to so many acceptors. Note NM_001197104.2 encodes 3,972 aa while UniProt Q03164 is 3,969 aa; the three-residue difference lies downstream of every coordinate used here. This is a COORDINATE DERIVATION from a transcript record, not a measurement of a fusion protein: no sarcoma KMT2A fusion protein has ever been observed by any method and no menin-binding assay has been run on one.

    Evidence that argues against this (1)
    • Retaining a motif is not the same as presenting it. The chimera's folding, stability and nuclear import are all unmeasured, and a fusion protein can retain a binding motif in primary sequence while failing to engage its partner. This graph's queued test exp-menin-pulldown-on-yky-versus-kmt2a-aff1 exists precisely because retention has never been shown to imply binding for any sarcoma arm. Massoth LR et al. 2020
  6. E6moderate confidence 3 of 5: One strong study, or several smaller consistent ones.database

    C2CD3 exon 24 in NM_015531.6 (CDS 211..6102) spans nt 4814-5161 and therefore begins 4,603 nucleotides after the start codon, which is not a multiple of three; since every KMT2A donor exon in this cohort ends at a complete codon, the reported KMT2A exon 8 :: C2CD3 exon 24 junction is predicted to be OUT OF FRAME, giving a KMT2A 1-1362 fragment with a short frameshifted tail rather than a chimeric protein.

    What this evidence cannot show

    A derivation computed this cycle from two RefSeq records, not an observation, and this is the one case in the beat where the derivation changes the lesion CLASS rather than a detail. It can be wrong if the commercial panel's exon numbering for C2CD3 differs from RefSeq's, if the breakpoint is intra-exonic, or if the tumour expressed a different C2CD3 isoform - and UniProt lists five, so the last possibility is not remote. The junction nucleotide is not reported. NOTE ALSO that an out-of-frame prediction does not mean no product: a KMT2A N-terminal fragment terminating shortly after the junction is still a protein, and in the leukaemia literature KMT2A N-terminal fragments are not inert. The KMT2A-side conclusions at E5 do NOT depend on this calculation.

    Evidence that argues against this (1)
    • The same calculation says four of the seven cases ARE in frame, so it is not a method that produces a uniform answer - but that cuts both ways, because a systematic error in exon numbering would flip several cases together. And the source table demonstrably contains annotation errors: it misspells the index gene as KTM2A in two of eighteen rows. Massoth LR et al. 2020
  7. E7established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    The transcript the fusion source cites for C2CD3, NM_015531.6 / NP_056346.3, encodes 1,963 amino acids, while UniProt's canonical C2CD3 (Q4AC94, isoform 5) is 2,353 amino acids - a 390-residue, 20% difference within the same gene - so the C2-domain coordinates annotated on the UniProt sequence cannot be mapped onto an exon number from the RefSeq transcript without an alignment.

    What this evidence cannot show

    Both lengths were read programmatically this cycle from the records themselves - the RefSeq GenBank /translation string and the UniProt sequence length field - not from recall. UniProt lists five C2CD3 isoforms with isoform 5 displayed and NM_015531.6 cross-referenced as one of two RefSeq entries on the record, so this is ordinary isoform diversity rather than a database error; the hazard is in combining coordinate systems. THE DEFENCE THIS PROJECT RELIES ON DOES NOT CATCH IT: verify-accessions.py asserts accession plus length and therefore catches a wrong GENE, not a wrong ISOFORM of the right gene. No alignment was performed here, so this entry deliberately makes NO claim about which C2 domains an in-frame product would retain.

    Evidence that argues against this (1)
    • Choosing the UniProt canonical as the reference frame is itself a convention, not a fact about the tumour: the clinically reported transcript is arguably the more relevant frame, and which isoform an MPNST actually expresses is unmeasured. The honest position is that neither frame is privileged and the entry should carry both, which is what it does. UniProt Consortium 2026
  8. E8moderate confidence 3 of 5: One strong study, or several smaller consistent ones.review

    The avian ciliopathic mutant talpid2, which carries a C2CD3 lesion, is an established model for dissecting GLI-mediated transcription during craniofacial development and disease.

    PMID 27799912DOI
    What this evidence cannot show

    A review in an avian and craniofacial-development context, cited as a map rather than as primary evidence (the brief's rule on reviews), and the system is chicken embryonic development - about as far from infant KMT2A-rearranged leukaemia or a human sarcoma as this beat gets. It is included because it is the only route from C2CD3 to transcriptional output in the literature, and because it is one of the two records returned by C2CD3[tiab] AND chromatin[tiab] - where it is about TRANSCRIPTION rather than about chromatin, which is worth recording so a future beat does not count it as an epigenetics hit.

    Evidence that argues against this (1)
    • GLI processing is a cytoplasmic, cilium-dependent proteolytic event; calling it an epigenetic or chromatin link would be a category error of the kind this graph flags for IGF2BP3 and m6A. C2CD3's route to transcription runs through an organelle, not through a mark. UniProt Consortium 2026
  9. E9moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    SMYD3, a histone/lysine methyltransferase, binds the promoter region of C2cd3 and regulates the transcription of cilia genes including C2cd3, Cep164, Ttbk2, Dync2h1 and Cp110; SMYD3 is itself a novel component of the centriolar distal appendage, is required for appendage assembly, and its loss reduces the proportion of ciliated cells and produces stumpy cilia.

    SMYD3 Controls Ciliogenesis by Regulating Distinct Centrosomal Proteins and Intraflagellar Transport Trafficking. · Agborbesong E, Zhou JX, Zhang H et al · International Journal of Molecular Sciences 2024
    PMID 38892227DOI
    What this evidence cannot show

    The gene names are written in mouse style (C2cd3, Cep164, Cp110) and the abstract does not state the cell system, so the species is not resolvable from the record read here - treat the promoter-binding result as murine or of unstated system until the full text is read (this graph's own unstated-systems problem, which it measures at 39% of load-bearing evidence). It is a ciliogenesis study, not a cancer study, and the promoter binding is one assay among several with no occupancy coordinates given in the abstract. Abstract read; full text not read. Cited as the single inbound epigenetic link for this gene, which is real: the C2CD3 promoter is a direct target of a histone lysine methyltransferase.

    Evidence that argues against this (1)
    • PROBABLY A COINCIDENCE, AND THE NULL IS STATED: SMYD3 is also the methyltransferase upstream of CPT1A in CML leukaemia stem cells, so two of these seven unrelated partners are downstream of the same enzyme. Nothing was pre-specified, SMYD3 has a large and heterogeneous target literature, and 2 of 7 is not a rate that distinguishes itself from chance - so this is recorded as an observation and explicitly NOT as a pattern, in the same spirit as this graph's own SF3A3 coincidence, which has no clean null either. Massoth LR et al. 2020
  10. E10moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    C2CD3 has essentially no cancer literature and is not a reported fusion partner in any tumour: the whole C2CD3 title/abstract literature is 42 records, C2CD3 with 'fusion' returns zero, with 'translocation' zero, with KMT2A zero, with MLL zero, and the two records returned with 'cancer' are both off topic.

    PMID 32461620DOI
    What this evidence cannot show

    Bounded 2026-09-22 with one [tiab] per term and querytranslation read back (rules 94, 95): C2CD3[tiab] = 42; C2CD3[tiab] AND fusion[tiab] = 0; AND translocation[tiab] = 0; AND KMT2A[tiab] = 0; AND MLL[tiab] = 0; AND cancer[tiab] = 2, title-checked and both off topic (PMID 36587566, a salivary-microbiome in-silico study; PMID 34176789, a vitamin-A and COVID paper). Positive controls through the IDENTICAL one-tag-per-term query shape, run 2026-09-22 (rule 67, rule 94): EWSR1[tiab] AND fusion[tiab] = 1,208; EWSR1[tiab] AND translocation[tiab] = 462; AFF1[tiab] AND KMT2A[tiab] = 104; MLLT3[tiab] AND MLL[tiab] = 84; KMT2A[tiab] AND histone[tiab] = 218. Every query was written with one [tiab] per term and its querytranslation read back from the esearch response, so no parenthesised group silently expanded to [All Fields]. Bounded by TITLES AND ABSTRACTS, not by papers (rule 80): supporting figures were not searched. This is the cleanest set of zeros in the beat, which is exactly why the controls matter: a gene with a 42-record literature will return zero for almost any conjunction, so the absence is weak evidence about biology and strong evidence only about the literature.

    Evidence that argues against this (1)
    • A 42-record literature cannot bound much: C2CD3 is a 31-exon gene and the absence of a reported rearrangement in a gene nobody studies is close to uninformative. The strongest version of the claim is that no partner-agnostic fusion survey has named it, and no such survey of sarcoma has been run here. Balestra FR et al. 2013
  11. E11weak confidence 2 of 5: Preliminary, small, indirect, or not yet peer reviewed.dataset

    C2CD3 is not required in any arm of this project's own DepMap scan: mean Chronos +0.064 in 11 KMT2A-rearranged lines, +0.029 in 43 other leukaemia lines, difference +0.036, pan-essentiality +0.020 across 1,208 lines, selectivity rank 13,487 of 17,728.

    DepMap CRISPR (Chronos) gene-effect scan of 11 KMT2A-rearranged versus 43 other leukaemia lines (this project's own scan) · Broad Institute DepMap; joined locally from analyses/depmap-kmt2a-selective-scan-full · analyses/depmap-kmt2a-selective-scan-full 2026
    What this evidence cannot show

    Joined 2026-09-22 from analyses/depmap-kmt2a-selective-scan-full/results.json (generated 2026-09-19T10:12:06Z), this project's own genome-wide scan over 11 KMT2A-rearranged and 43 other leukaemia DepMap lines. NOT re-fetched and NOT independent of that scan: every caveat it carries applies, including that the 11 KMT2A-rearranged lines are drawn from nine individuals and that the arm is myeloid-enriched. The scan records a generation timestamp and NOT a DepMap RELEASE, which is rule 83 exposure carried forward rather than fixed here. Every mean is POSITIVE, so this gene is not a dependency anywhere and no selectivity claim is possible in either direction. A rank is not an effect size and a difference is not a dependency (this graph's DepMap rules). Check pan_essentiality_all_lines before reading anything as selective, and note the anti-selective half is 50% of any gene set by construction, with this graph's own base rate at 45.3% - so a tally of genes in that half is the null (rule 84). The decisive limitation here is lineage: the screen is leukaemia lines, and for a sarcoma fusion that is the wrong tissue. Cellosaurus holds 592 KMT2A-commented cell lines and ZERO are sarcoma or any mesenchymal tumour, so no cell-line screen can address these cases at all.

    Evidence that argues against this (1)
Connected to