Oncofusion

start / graph / gene-pax5

Genestrong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.Q46-cooperating-lesions

PAX5 - the most frequently hit non-KMT2A gene in infant KMT2A-rearranged ALL, and it is a deletion in one allele, not a mutation

also called PAX5 · Paired box protein Pax-5 · BSAP · B-cell-specific transcription factor · Q02548

PAX5 is the master switch that makes a blood cell become a B cell, and it is the gene most often damaged - besides KMT2A itself - in infant KMT2A leukaemia. The damage is a deleted piece of one copy, not a spelling change, which is why studies that only look for mutations report this disease as having none.

The same thing, for a scientist

PAX5 (Q02548, 391 aa) is the B-lineage commitment transcription factor and is, by a clear margin, the **most frequently altered non-KMT2A gene** in infant KMT2A-rearranged ALL. In whole-genome sequencing of 22 infant cases, 133 structural variants and copy-number alterations affected annotated genes; about half (67) were direct consequences of the KMT2A rearrangement itself, and within the remaining 66 the most frequently affected gene was **PAX5, in 5 cases**, ahead of CDKN2A/CDKN2B (3) and DLEU1/2 (3). These events were predicted to cause **loss of function of a single allele** [E1]. Two features make this more than a frequency entry. **It is specific to the infant disease.** None of 20 non-infant KMT2A-rearranged leukaemias in the same study harboured a focal PAX5 lesion, against **5 of 22 (23%)** infant KMT2A-rearranged ALL cases [E1]. Whatever is producing PAX5 loss is a property of the infant disease, not of the fusion class. **And it is a lesion class that mutation-centric analyses cannot see.** PAX5 alterations here are structural: deletions and, in one case, a **PAX5-KANK1 fusion** detected by RNA-seq that had **no supporting WGS reads at all**, placing it in a minor subclone [E1]. A cohort screened by a mutation panel, or by RNA-only variant calling, reports this gene as untouched. The functional interpretation is not settled and this entry does not settle it. Monoallelic PAX5 loss is the canonical dosage lesion of B-ALL generally, and in a disease whose fusion already imposes a differentiation block on a pro-B/mixed-lineage cell, whether a further half-dose of PAX5 changes anything has never been tested in a KMT2A-rearranged model.

What it looks like
PAX5391 aa
Paired box (DNA-binding)Paired box (DNA-binding) · residues 16–1420100200300
loading AF-Q02548-F1…
AlphaFold AF-Q02548-F1 ↗
colourshape

This is a prediction, not a measurement. No one has seen this protein in this shape. The colours say how much the software trusts itself, and that is worth reading before anything else — the orange stretches are places the model is effectively saying this part has no fixed shape.

very high (90+)confident (70–90)low (50–70)very low (<50)
I have never read one of these pictures before

A protein is one long chain of amino acids that folds up. The ribbon traces that chain from one end to the other; it is not a wire, it is a simplification of where the backbone runs. Drag to rotate, scroll to zoom.

Domains are the stretches of chain that do a particular job, and they are what the coloured blocks above the picture mark out. Click one and the view flies to it. In this project the important question about any domain is usually whether the fusion still has it — the labels say lost where it does not.

Anything in yellow sticks is a bound small molecule, which is usually the drug.

Go deeper
The numbers, and what they are a denominator of
observationvaluesource
SVs/CNAs affecting annotated genes, 22-case infant cohort1331
of which a direct consequence of the KMT2A rearrangement671
remaining independent SVs/CNAs661
most frequently affected gene among thosePAX5, 5 cases1
nextCDKN2A/CDKN2B, 3 cases; DLEU1/2, 3 cases1
PAX5 focal lesions, infant KMT2A-r ALL5/22 (23%)1
PAX5 focal lesions, non-infant KMT2A-r leukaemia0/201
mean CNAs per case, infant KMT2A-r1.01
mean CNAs per case, non-infant KMT2A-r2.6 (P = 0.0234)1

The last two rows are the ones that make the PAX5 contrast surprising rather than trivial. Non-infant KMT2A-rearranged leukaemias have more copy-number alterations overall and none of them are focal PAX5 lesions. So the infant excess is not a consequence of a generally more rearranged genome; if anything it is the opposite.

Why this belongs in a cooperating-lesions beat at all

The standing description of this disease is that it has almost no cooperating lesions, and the numbers behind that - a mean of 1.3 non-silent mutations in the dominant clone - are counts of point mutations and indels. PAX5 is the clearest demonstration that a point-mutation count is not a lesion count.

On a strict reading, infant KMT2A-rearranged ALL has:

  • about 1.3 non-silent SNV/indel events in the dominant clone,
  • about 1.0 copy-number alteration per case,
  • and the single most frequent recurrent target of the second category is PAX5.

This does not overturn the quiet-genome claim; a mean of one CNA per case is still extraordinarily quiet. It does mean that the quiet-genome claim, when used to argue that the fusion has no partner lesion, has to account for a recurrent monoallelic loss of the B-lineage master regulator in nearly a quarter of cases.

And it changes which absences are credible. Every counted absence in this beat that was established from mutation data alone - for TP53, for IKZF1, for DNMT3A - inherits the same blind spot PAX5 exposes, and each of those entries says so.

Accession resolution

gene_exact:PAX5 AND organism_id:9606 AND reviewed:true with gene_primary and length returns exactly one reviewed record:

accessionprimarysynonymsprotein namelength
Q02548PAX5nonePaired box protein Pax-5 (B-cell-specific transcription factor) (BSAP)391

One row, primary name matches, no gene synonyms. The alias to be aware of in the literature is BSAP, which is how this protein was named through the 1990s and which is still used - and BSAP is also used in other fields for bone-specific alkaline phosphatase, so an alias search on that string needs title-checking. The paired-box DNA-binding domain is annotated at residues 16-142 in the same record.

Facts and figures
uniprotQ02548
gene symbolPAX5
length AA391
alphafold IDAF-Q02548-F1
cytoband9p13.2
resolution notegene_exact:PAX5 AND organism_id:9606 AND reviewed:true with fields=accession,gene_primary,gene_synonym,protein_name,length returned one row: Q02548, primary gene name PAX5, no gene synonyms, 391 aa. Fetched from the UniProt REST API on 2026-09-20; the paired DNA-binding domain coordinates come from the same record's feature table the same day.
druggabilityNot druggable. PAX5 is a monoallelically lost lineage transcription factor; the therapeutic content of this entry is interpretive, in that it constrains what the quiet-genome argument can claim.
domains
[
  {
    "name": "Paired box (DNA-binding)",
    "start": 16,
    "end": 142,
    "pfam": null
  }
]
cooperating lesion profile
{
  "disease": "infant KMT2A-rearranged B-ALL",
  "lesion_class": "focal structural variant / copy-number loss, predicted monoallelic loss of function; one RNA-only PAX5-KANK1 fusion in a minor subclone",
  "frequency": "5 of 22 (23%) in the whole-genome discovery cohort; the most frequently affected non-KMT2A gene among the 66 SVs/CNAs not caused by the rearrangement itself",
  "age_specificity": "0 of 20 non-infant KMT2A-rearranged leukaemias had a focal PAX5 lesion",
  "detectability": "invisible to SNV-only calling; requires copy-number or structural-variant analysis",
  "verdict": "the strongest candidate for a genuinely recurrent cooperating lesion in this disease, and the reason a point-mutation count understates the lesion burden"
}
depmap kmt2a selectivity
{
  "chronos_mean_kmt2a_rearranged": -0.1848,
  "chronos_mean_other_leukaemia": -0.2565,
  "difference": 0.0717,
  "pan_essentiality_all_lines": -0.088,
  "selectivity_rank": 16057,
  "rank_denominator": 17728,
  "provenance": "Joined 2026-09-21 from analyses/depmap-kmt2a-selective-scan-full (generated 2026-09-19T10:12:06Z), the 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 here, including that the KMT2A-rearranged arm is a small number of lines drawn from nine individuals, that the rank denominator is 17,728 genes, and that a difference is not a dependency. Added because only 26 of 308 gene entries carried this number, which made any KMT2A-selectivity question answerable for 8% of the corpus.",
  "how_to_read": "A negative difference means MORE required in KMT2A-rearranged lines than in other leukaemia lines. A rank near 17,728 is ANTI-selective. Neither is evidence of a fusion-driven dependency on its own: check pan_essentiality_all_lines before calling anything selective, and note that a paralogue-redundant gene can score near zero for reasons that have nothing to do with this disease - see ana-the-paralogue-redundancy-excuse-is-not-tested-and-my-proxy-decided-the-answer."
}
fusion construct association
{
  "what": "PAX5 is curated in IntAct against THREE KMT2A fusion CONSTRUCTS - mll_af9_human_protein (EBI-2890930), mll_enl_human_protein (EBI-2620068) and mll_ell_fusion_protein (EBI-9513644) - anti-tag co-IP, physical association, HEK293T, interactions EBI-9513622 / EBI-9513632 / EBI-9513647, all from PMID 24735925 (IntAct publication ac EBI-9512250).",
  "why_it_is_unusual": "Of 21 proteins curated against any of the six known KMT2A fusion constructs, 7 appear on more than one, and PAX5 is tied with AFF1 for the most at three each. The other six are AFF1, CDC73, CTR9, DOT1L, LEO1 and PAF1 - SEC, PAF1 complex and the H3K79 methyltransferase. PAX5 is the only lineage transcription factor among them.",
  "the_unresolved_tension": "This entry otherwise frames PAX5 as a cooperating LESION, a one-allele deletion. See E4 counter-evidence: the co-IP and the deletion come from unrelated systems and neither explains the other.",
  "not_independent_replication": "Three distinct interaction records, ONE publication, ONE cell line, one laboratory. Corrected from an earlier statement in this entry that the rows shared a single interaction id; EBI-9512250 is the publication ac.",
  "source": "analyses/intact-fusion-construct-interactors/results.json, re-run 2026-09-22 after analyses/intact-fusion-construct-census/run.py found the sixth construct",
  "recorded_because": "The fact previously lived only in ana-intact-holds-forty-seven-fusion-side-records-our-instrument-drops-and-every-partner-is-already-written, whose own open question asked whether gene-pax5 records it."
}
How this could bite us back
  • PAX5 is the B-lineage commitment factor, and this disease's documented escape route from CD19-directed therapy is lineage switch to myeloid. A pre-existing monoallelic PAX5 loss is exactly the background that would make that switch easier, which would make an immunotherapy-first strategy worse in precisely the patients who carry the commonest secondary lesion.
  • Nothing restores a deleted tumour-suppressor allele, so identifying PAX5 loss provides no intervention and risks becoming a stratification variable with no arm attached to it.
  • The lesion is invisible to the assays most commonly run. A trial that stratifies on 'genomic risk' using a mutation panel will misclassify these patients as lesion-free.
Still open
  1. Does monoallelic PAX5 loss change anything in a KMT2A-rearranged cell? The fusion already enforces a pro-B or mixed-lineage arrest, and no KMT2A-rearranged model has been tested on a Pax5 heterozygous background.
  2. Why is focal PAX5 loss present in 23% of infant KMT2A-rearranged ALL and 0% of non-infant KMT2A-rearranged leukaemia, when the non-infant cases carry more copy-number alterations overall?
  3. Is PAX5 loss in these cases clonal or subclonal? The one PAX5 event with clonality information - the PAX5-KANK1 fusion with no WGS reads - was subclonal, and no allele-specific copy-number analysis of the deletions has been published.
  4. Does PAX5 dosage influence the lineage switch seen under CD19-directed therapy in this disease? A half-dose of the B-lineage master regulator is a plausible predisposing background and nobody has looked.
  5. What is PAX5 doing in the MLL-AF9 and MLL-ENL co-IPs of PMID 24735925 -- a co-purifying protein, a functional mediator of the pro-B tumour-suppression programme, or a control? The abstract does not say and the full text was not read. It matters because this graph otherwise frames PAX5 as a deleted cooperating lesion, and the two readings do not sit together.
  6. Is PAX5 curated against KMT2A::AFF1 anywhere? The two construct rows are MLL-AF9 and MLL-ENL even though PMID 24735925's title names MLL-AF4, so the index fusion may be in the paper without being in the database.
  7. Does the pro-B-specific tumour suppression of PMID 24735925 interact with PAX5 dosage? PAX5 is monoallelically deleted in this disease and is the master pro-B lineage factor, so a programme that fires in pro-B but not myeloid cells is a programme running in cells with half the normal PAX5. Nothing here tests that and it should not be asserted.
  8. Why is PAX5 on three MLL constructs in PMID 24735925 - AF9, ENL and ELL - while the paper's title names MLL-AF4? Either the authors tested a construct panel and IntAct curated three of them, or the AF4 construct's rows are curated under an interactor this enumeration has not reached. The census recovered only 4 of the 5 constructs already known by hand, so its recall is demonstrably below 100% and a missing AF4 row cannot be excluded.
Every source behind this entry (4)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    In whole-genome sequencing of 22 infant MLL-rearranged ALL cases, 133 structural variants and copy-number alterations affected annotated genes, of which 67 were direct consequences of the MLL rearrangement; among the remaining 66, PAX5 was the most frequently affected gene (5 cases), ahead of CDKN2A/CDKN2B and DLEU1/2 (3 each), with the lesions predicted to cause single-allele loss of function; a PAX5-KANK1 fusion in case INF016 was detected by RNA-seq with no supporting WGS reads, indicating a minor subclone; and no focal PAX5 lesion was found in any of 20 non-infant MLL-rearranged leukaemias against 5 of 22 (23%) infant cases.

    The landscape of somatic mutations in infant MLL-rearranged acute lymphoblastic leukemias. · Andersson AK, Ma J, Wang J, et al. · Nature Genetics 2015
    PMID 25730765DOI
    What this evidence cannot show

    5 of 22 cases; a 23% figure from a 22-case cohort has wide binomial uncertainty. 'Predicted to result in loss of function of a single allele' is the authors' prediction from breakpoint position, not a functional measurement - no PAX5 protein or transcript quantification is reported for these cases. The comparator arm (20 non-infant MLL-R leukaemias) mixes ALL and AML, so the infant-versus-non-infant contrast is partly a lineage contrast. Full text read via efetch db=pmc (PMC4553269) after Europe PMC fullTextXML returned HTTP 500.

    Evidence that argues against this (1)
    • Monoallelic PAX5 deletion is one of the commonest lesions in B-ALL generally, so its presence in 23% of infant KMT2A-rearranged cases is not by itself evidence of a fusion-specific cooperation; the informative comparison would be against non-KMT2A infant ALL, and in the same study infants lacking MLL rearrangement had MORE copy-number alterations overall (2.2 vs 1.0 per case). Andersson AK et al. 2015
  2. E2moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    Targeted duplex sequencing of infant and childhood KMT2A-rearranged ALL identified PAX5 among the genes recurrently altered in this disease, within a set of 87 cancer-implicated mutations of which 62% were below the resolution of standard next-generation sequencing.

    PMID 36204688DOI
    What this evidence cannot show

    This is a POINT-MUTATION finding for PAX5 and does not corroborate the deletion lesion that is PAX5's principal alteration class in this disease - a duplex sequencing panel detects sequence variants, not copy number. The abstract names PAX5 in a list without a per-gene count or prevalence, so no frequency is quoted here. 27 patients. Abstract read.

    Evidence that argues against this (1)
    • That PAX5 appears in both a deletion-based and a point-mutation-based enumeration of this disease is consistent with two independent lesion classes hitting the same gene, but neither study measures both classes in the same patients, so no combined PAX5 alteration frequency for this disease exists. Pilheden M et al. 2022
  3. E3established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    UniProtKB lists exactly one reviewed human record with primary gene name PAX5: Q02548, 391 aa, protein names Paired box protein Pax-5, B-cell-specific transcription factor and BSAP, with a paired DNA-binding domain at residues 16-142 and no gene synonyms.

    What this evidence cannot show

    Fetched 2026-09-20. A database fact about the reference isoform; PAX5 has multiple annotated isoforms and the 391-residue length is the canonical sequence, so a paper numbering residues from a different isoform would disagree. This entry records no residue-level claims that depend on isoform choice beyond the DNA-binding domain boundary.

    Evidence that argues against this (1)
    • The alias BSAP is shared with bone-specific alkaline phosphatase in the clinical-chemistry literature, so a literature search on that alias returns off-target records and any count built on it must be title-checked. UniProt Consortium 2026
  4. E4moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    PAX5 (Q02548, taxid 9606) is curated in IntAct as a PHYSICAL ASSOCIATION with THREE KMT2A fusion CONSTRUCTS, by anti-tag co-immunoprecipitation in Homo sapiens HEK293T embryonic kidney cells: mll_af9_human_protein (EBI-2890930, interaction EBI-9513622), mll_enl_human_protein (EBI-2620068, interaction EBI-9513632) and mll_ell_fusion_protein (EBI-9513644, interaction EBI-9513647). All three trace to one publication, PMID 24735925 (IntAct publication ac EBI-9512250). Of the 21 proteins IntAct curates against any of the six known KMT2A fusion constructs, only 7 appear on more than one, and PAX5 is tied with AFF1 for appearing on the most - three each. The other six multi-construct partners are AFF1, CDC73, CTR9, DOT1L, LEO1 and PAF1: super elongation complex, PAF1 complex and the H3K79 methyltransferase. PAX5 remains the only lineage transcription factor in that set. The source paper's own headline is a separate and striking claim: MLL-fusion proteins are rarely present at high levels, and when proteasome inhibition (bortezomib, carfilzomib) reduces their turnover, the ACCUMULATED fusion induces a latent tumour-SUPPRESSION programme - apoptosis via caspase-8 cleavage of BID and cell-cycle arrest via p27 - in MLL pro-B LYMPHOID but not myeloid leukaemias.

    Proteasome inhibitors evoke latent tumor suppression programs in pro-B MLL leukemias through MLL-AF4 · Liu H, Westergard TD, Cashen A, Piwnica-Worms D, et al. · Cancer Cell 2014
    PMID 24735925DOI
    What this evidence cannot show

    CORRECTED, cycle 55: this item first said TWO constructs and was superseded within the same cycle by the systematic census, which found a sixth construct, mll_ell_fusion_protein, whose only curated partner is PAX5. The construct list this figure rests on was assembled by hand until that census ran, so treat 3-of-6 and 7-of-21 as current-best rather than settled. Abstract-level only: the full text was not read, so what PAX5's role in PMID 24735925 IS - a co-purifying protein, a functional mediator of the pro-B programme, or a control - is NOT established here, and the IntAct rows assert association without direction, stoichiometry or a reported negative control. The system is a TRANSFECTED TAGGED CONSTRUCT in HEK293T, neither a leukaemia cell nor a pro-B cell, so this is evidence about the construct rather than the patient lesion - the objection this graph applies to the Skucha AP-MS network, and with more force, since that network at least used isogenic human leukaemia lines with a control. The three constructs are MLL-AF9, MLL-ENL and MLL-ELL, NOT the index fusion KMT2A::AFF1, even though the paper's title names MLL-AF4. Title, journal, first author and DOI verified live by efetch 2026-09-22; all three IntAct rows and their interaction acs re-read from analyses/intact-fusion-construct-interactors/results.json the same day.

    Evidence that argues against this (3)
    • This does not resolve against the rest of this entry and should not be read as if it did. This graph's PAX5 record is about a COOPERATING LESION -- a monoallelic deletion, the most frequently hit non-KMT2A gene in infant KMT2A-rearranged ALL. A gene deleted in the disease is an odd thing to find physically associating with the oncoprotein, and the two observations come from unrelated systems: patient sequencing on one side, a tagged construct in HEK293T on the other. Neither licenses a statement about the other, and an entry that presented the co-IP as an explanation of the deletion -- or the deletion as evidence the association matters -- would be wrong.
    • The source paper's direction of effect cuts against reading any fusion-construct association as straightforwardly oncogenic: its finding is that ACCUMULATED MLL-fusion is tumour-SUPPRESSIVE in pro-B lymphoid cells, which is the lineage of this project's index disease. If there is a fusion dose above which pro-B cells die, then partner associations measured under overexpression -- as a transfected tagged construct necessarily is -- were measured in a regime the leukaemia does not occupy.
    • The 'only lineage transcription factor among the seven' framing is a statement about what IntAct HOLDS, not about biology. Six elongation-machinery proteins recurring across constructs is the expected signature of a database in which the same complexes are re-assayed against every new fusion. And PAX5's three rows are NOT three independent observations: they are three distinct interaction records (EBI-9513622, EBI-9513632, EBI-9513647) all from a SINGLE publication, PMID 24735925, in a single cell line. One laboratory testing one protein against three of its own constructs is a within-study consistency check, not replication. NOTE the correction: an earlier version of this item claimed the rows shared one interaction id and were therefore one experiment double-counted. That was wrong - EBI-9512250 is the PUBLICATION ac, not the interaction ac - and the error under-stated the curation while over-stating the redundancy.
Connected to