Anti-AF9 antibodies see three proteins, not two - and the third is an internal-promoter isoform discovered in 2025
also called AF9-C antibody · anti-MLLT3 attribution · YEATS-domain-devoid AF9
In a KMT2A::MLLT3 leukemia an antibody against the back half of AF9 picks up normal AF9 and the fusion. It also picks up a third, shorter AF9 protein that a 2025 paper found is made from a separate internal start site and is heavily expressed in blood stem cells and in the worst AMLs - and nobody designing these experiments has counted it.
The same thing, for a scientist
KMT2A::MLLT3 retains the AF9 C-terminus, so an anti-AF9-C antibody binds the fusion and wild-type AF9 alike - the standard two-species problem, and the one authors of a 2024 occupancy study state explicitly when explaining why they had to build a genetic workaround [E1]. What is new is a **third** species. A strong internal promoter in MLLT3 drives massive expression of **YEATS-domain-devoid AF9 transcripts in haematopoietic stem cells and in the most lethal AML** [E2], and the equivalent promoter exists in MLLT1 [E3]. A YEATS-devoid AF9 lacks the N-terminal reader module and retains the C-terminal region - which is precisely the region KMT2A::MLLT3 retains and precisely the region every anti-AF9-C antibody is raised against. So in a KMT2A::MLLT3 cell an AF9-C reagent may be reading wild-type full-length AF9, the internal-promoter truncated isoform, and the fusion. This is the **exact structural analogue of KMT2A\*** on the partner side of the junction, it is invisible to junction RT-PCR for the same reason, and this graph had no entry for it before this beat.
The workaround that does work, and the assumption inside it
A 2024 study wanting to compare MLL1 and MLL-AF9 occupancy in the same cells recognised the problem and solved it genetically: knock out AF9 in MOLM-13, a line carrying one copy of t(9;11), so that the only AF9-epitope-bearing protein left is the fusion, then do CUT&Tag with an AF9-C antibody 1. That is a real attribution - the best one this beat found for any KMT2A fusion in a human patient-derived line - and it produced a substantive result: 16,691 MLL1 peaks against a much smaller MLL-AF9 set, with 87% of MLL-AF9 peaks co-occupied by MLL1, and no difference in the location or the width of occupancy sites, contrary to the prevailing model.
The assumption inside it is stated in the paper: AF9 was judged safe to delete because "its expression in most of the hematopoietic cell lines, including MLLr cell lines, are so low that its loss may be of little consequence." The 2025 internal-promoter result complicates that in two ways. First, MLLT3 expression measured at the whole-gene level may be dominated by the truncated isoform rather than by full-length AF9, so a low full-length figure is not the same as a low locus output. Second, whether the knockout guides removed the internal-promoter isoform depends on where they cut, and if they did not, the AF9-C CUT&Tag in the knockout line still has two targets rather than one.
Neither point undoes the result. Both are checkable in a day by anyone with the cell line, and neither has been checked.
Why this matters beyond MLLT3
The graph already holds protein-kmt2a-star-truncated-product-of-the-internal-promoter: an internal promoter inside the KMT2A breakpoint region producing a truncated, menin-blind KMT2A protein that no assay can see. The MLLT3 finding is the same phenomenon on the other side of the junction, discovered independently, and the pair generalises to a rule this graph should carry:
A fusion gene has two parents, and either parent's locus may have an internal promoter producing a truncated protein that shares the fusion's retained epitopes and has no junction.
That rule predicts where to look next - AFF1, MLLT1 (confirmed 3), MLLT10, ELL - and it says something uncomfortable about counted absences built on junction RT-PCR, which is structurally blind to every such species.
| reagent class | antibodies against the AF9/MLLT3 C-terminal (retained) region |
|---|---|
| structural analogue | protein-kmt2a-star-truncated-product-of-the-internal-promoter |
| generalisation | either parent locus of a fusion gene may have an internal promoter producing a junctionless, epitope-sharing truncated protein |
| mllt1 internal promoter active in haematopoiesis | false |
species recognised
[ "wild-type full-length AF9", "YEATS-domain-devoid internal-promoter AF9 isoform (unquantified at protein level)", "KMT2A::MLLT3 fusion" ]
worked attribution in the literature
{
"design": "AF9 knockout in MOLM-13 (t(9;11), one fusion copy), then AF9-C CUT&Tag",
"citation_pmid": "39002676",
"result": "16,691 MLL1 peaks; 87% of MLL-AF9 peaks co-occupied by MLL1; no difference in peak location or width",
"unchecked_assumption": "whether the knockout removes the internal-promoter YEATS-devoid isoform"
}- If a truncated AF9 isoform competes for SEC binding, a therapeutic that blocks the AF9 AHD interface would act on wild-type and truncated AF9 as well as the fusion, in haematopoietic stem cells where the isoform is most abundant.
- Any counted absence built on junction RT-PCR - including every clinical MRD assay - is blind to internal-promoter products on both sides of the junction, so 'fusion-negative' after therapy does not exclude a persisting epitope-sharing species.
- Do the AF9-knockout guides used in MOLM-13 remove the internal-promoter YEATS-devoid isoform? If not, the best attributable CUT&Tag in this field still has two targets, and the check is a single western blot with a C-terminal antibody.
- Is the YEATS-devoid AF9 isoform present as protein in KMT2A::MLLT3 blasts, and does it compete with the fusion for SEC components? A truncated AF9 retaining the AHD would be a stoichiometric competitor for exactly the interface the fusion uses.
- Do AFF1, MLLT10 and ELL have analogous internal promoters? The MLLT1/MLLT3 pair suggests the architecture is not unique, and nobody has looked on the AFF1 side.
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
The authors of a 2024 occupancy study state that no MLL-AF9-specific antibody exists and that an AF9-fragment antibody also recognises wild-type AF9; they generated an AF9-knockout MOLM-13 line so that AF9-C CUT&Tag reports only the fusion, and found 87% of MLL-AF9 peaks co-occupied by MLL1 with no difference in peak location or width.
MLL-AF9 regulates transcriptional initiation in mixed lineage leukemic cells. · Yang Z, Zhang G, Zhao R, et al. · J Biol Chem 2024What this evidence cannot show
Read in full via PMC11345648. The authors note their MLL-AF9 peak number is much smaller than the MLL1 peak number and offer antibody efficiency as one of two explanations, so the comparison is not signal-matched. Their justification for deleting AF9 is an assumption about its expression, not a measurement in the line used.
Evidence that argues against this (1)
- An earlier study using an MLL-C antibody that cannot detect MLL-AF9 reported that MLL1 and MLL-AF9 occupy distinct chromatin regions - the opposite conclusion from a design that is also attributable, by a different route. Two valid attribution strategies give opposite answers, so the occupancy question is not settled by attribution alone. Xu J et al. 2016
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
A strong internal promoter in MLLT3 drives massive expression of YEATS-domain-devoid AF9 transcripts in haematopoietic stem cells and in the most lethal AML.
A strong internal promoter drives massive expression of YEATS-domain devoid MLLT3 transcripts in HSC and most lethal AML. · Bessière C, Zamani A, Pfeifer R, et al. · Cancer Commun (Lond) 2025What this evidence cannot show
Title and record verified from PubMed; the claim as used here is transcript-level. Whether the truncated AF9 protein is made, and whether it carries the epitope of any particular commercial anti-AF9-C antibody, was not established in this beat - the epitope of the AF9-C reagent used in PMID 39002676 is not stated in its retrievable text. The consequence for antibody attribution is therefore an inference from architecture, flagged as such.
Evidence that argues against this (1)
- A highly expressed transcript is not necessarily a highly expressed protein, and if the truncated AF9 isoform is not translated or is rapidly degraded it adds nothing to an AF9-C immunoprecipitate. The magnitude of this third source is unmeasured, exactly as for KMT2A*. Dailhau S et al. 2026
- E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
A similar internal promoter exists in the MLLT1 locus, inactive in the haematopoietic lineage and in AML but functional in pituitary, brain, thyroid, uterus and lung, producing a comparable YEATS-domain-devoid ENL isoform.
Tissue- and tumor-type-specific expression of internal-promoter-driven YEATS-domain-devoid isoforms of MLLT1 and MLLT3. · Dailhau S, Zamani A, Pfeifer R, et al. · Biomark Res 2026What this evidence cannot show
Abstract read verbatim. The MLLT1 internal promoter being **inactive** in the haematopoietic lineage and AML is the load-bearing detail and it cuts the other way: it means the third-species problem is an MLLT3 problem and not an MLLT1 problem in leukaemia, which makes anti-ENL reagents cleaner than anti-AF9 reagents in this disease.
Evidence that argues against this (1)
- Tissue-level promoter activity in bulk data does not exclude activity in a rare or transformed subpopulation, and KMT2A::MLLT1 leukaemias were not specifically profiled for this isoform, so 'inactive in AML' is a statement about the cohorts examined. Bessière C et al. 2025
- related to →MLLT3 / AF9 — a famous fusion partner that is not a dependency, and a hub domain that is
an internal promoter in this locus makes a YEATS-devoid isoform sharing the fusion's retained epitopes
- related to →KMT2A::MLLT3 (MLL-AF9), t(9;11)(p21.3;q23.3)
AF9-C antibodies read up to three species in these cells
- analogous to →KMT2A*: a menin-blind, LEDGF-blind, SET-competent KMT2A protein made from out-of-frame reciprocal alleles
the same internal-promoter, junctionless, epitope-sharing architecture on the partner side of the fusion
MLLT1 carries the same internal promoter but it is inactive in the haematopoietic lineage, making anti-ENL reagents cleaner than anti-AF9 in leukaemia
why no endogenous image exists
the YEATS-devoid isoform is a third AHD-bearing competitor