Strand-resolved transcription across the AFF1 / AFF1-AS1 / C4orf36 promoter - the only measurement that can say which gene the signal belongs to
also called Q155 stranded readout · which transcript at 4q21.3
An occupancy measurement looks at protein sitting on DNA, and a piece of DNA shared by three genes has no owner. To find out which gene is actually being switched on, you have to measure the RNA and which direction it is going. Nobody has.
The same thing, for a scientist
**Why occupancy cannot answer it, stated first because it is the reason this experiment is necessary.** A divergent promoter is **one piece of DNA**. CUT&RUN coverage is unstranded and reports protein on that DNA, so asking *"is the fusion at AFF1's promoter or AFF1-AS1's?"* is not a well-posed question for an occupancy assay at any resolution. Base resolution helps only insofar as the summit's **position** and the direction of its decay are informative, and this beat showed that at this locus the summit is on neither of the two candidate TSSs it was sent to adjudicate. **The well-posed question is about transcription**, and it is answerable with strand-resolved RNA, in three tiers of increasing cost: **Tier 1 - free, no wet work, no new data.** Query an existing **stranded** RNA-seq deposit from KMT2A-rearranged material for coverage on each strand over `chr4:86,924,893-86,934,850` (AFF1-AS1's body, minus strand), `chr4:86,936,204-86,946,161` (AFF1 body downstream of every non-AFF1 TSS, plus strand) and `chr4:86,887,308-86,934,000` (C4orf36's body, minus strand). **Three numbers per sample.** If AFF1-AS1's body carries minus-strand reads, the transcript exists in this disease and the whole attribution question becomes biological. If it carries none, AFF1-AS1 is an annotation with no expression here and the promoter signal has two candidates rather than three. **This has never been run and is the cheapest open item this beat leaves.** **Tier 2 - public data, more work.** A 5'-end assay (CAGE, or GRO-cap / PRO-cap) over the same interval places **initiation sites** rather than gene bodies, and is the only thing that can say whether the three annotated TSSs within 300 bp are three promoters or one initiation region modelled three ways. FANTOM5 CAGE covers many haematopoietic primary cells; none is KMT2A-rearranged, so this tier answers the annotation question and not the disease question. **Tier 3 - wet work.** PRO-seq or TT-seq in a KMT2A::AFF1 line (SEM, RS4;11, MV4-11), with and without a menin inhibitor. This is the only design that measures **nascent, strand-resolved, drug-responsive** transcription at the locus and therefore the only one that can say whether the fusion drives AFF1, AFF1-AS1, C4orf36, or none of them. **Readout and pre-declared interpretation.** Plus-strand reads through AFF1's body and no minus-strand reads through AFF1-AS1's ⇒ the locus is an AFF1 promoter with an annotated but silent antisense neighbour, and this project's AFF1 promoter measurements become attributable in practice though not in principle. Minus-strand reads through AFF1-AS1's or C4orf36's body comparable to the plus-strand reads through AFF1's ⇒ the window is genuinely shared and every promoter-scale AFF1 statement keeps its caveat permanently. **Both outcomes are publishable and the second is the one this beat expects.**
Controls, and the ones that have already burned this project
- Strandedness must be verified, not assumed. A library labelled stranded can be unstranded, and the check is a housekeeping gene with a known orientation: GAPDH (plus strand, chr12) should give a strand ratio far from 1 in the same pipeline. Without that control the whole experiment returns a plausible 50:50 that means nothing.
- The three intervals must be disjoint under the CURRENT annotation, not the 2020 one. C4orf36's long transcripts run from 86,936,204 down past AFF1-AS1's entire body, so 'AFF1-AS1's gene body' is not C4orf36-free and a minus-strand read there does not identify which minus-strand gene produced it. The only C4orf36-free minus-strand interval is inside AFF1-AS1's exon 1,
chr4:86,934,486-86,934,850, which is 365 bp and will be read-poor. - Width-match or normalise. AFF1's body is 206 kb and AFF1-AS1's is 10 kb; a raw read count comparison is a comparison of lengths.
- Report the deposit's own strand convention. dUTP-based protocols report read 1 antisense to the transcript, and getting this backwards inverts the entire result while producing a clean, confident answer. This is the same family as the NCBI
chrstart/chrstopstrand trap that this beat's brief warns about. - Do not use GSE252378. It is CUT&RUN, it is unstranded, and it cannot answer this. That is the point.
| design | Tier 1: strand-resolved coverage over three disjoint intervals in an existing stranded RNA-seq deposit from KMT2A-rearranged material, no new data and no wet work. Tier 2: public 5'-end data (CAGE/GRO-cap) over chr4:86,933,000-86,937,000 to place initiation sites. Tier 3: PRO-seq or TT-seq in SEM, RS4;11 or MV4-11 with and without a menin inhibitor. |
|---|---|
| readout | per-strand coverage over AFF1-AS1 exon 1 (chr4:86,934,486-86,934,850, minus, the only C4orf36-free minus-strand interval), AFF1's body downstream of every non-AFF1 TSS (chr4:86,936,204-86,946,161, plus) and C4orf36's body (chr4:86,887,308-86,934,000, minus); width-normalised, with a GAPDH strand-ratio control asserting the libraries really are stranded. |
| controls | GAPDH strand ratio to verify strandedness; width normalisation; explicit statement of the deposit's dUTP/strand convention; intervals defined against the CURRENT annotation so they are genuinely disjoint; GSE252378 explicitly excluded because it is unstranded CUT&RUN. |
| cost tier | low |
| timeline | Tier 1: hours, one analyst, no new data. Tier 2: days. Tier 3: months. |
pre declared interpretation
{
"plus_only": "AFF1 is transcribed and AFF1-AS1 is not in this disease -> the promoter window's signal has one plausible owner and this project's AFF1 promoter statements become attributable in practice, with the caveat retained in principle.",
"both_strands": "the window is genuinely shared -> every promoter-scale AFF1 statement keeps its caveat permanently and AFF1-AS1 becomes a biological object worth its own beat.",
"minus_dominant": "the locus is primarily an antisense or C4orf36 promoter -> the 20 of 27 occupancy call is about a gene nobody has studied, and the autoregulation reading is dead twice over.",
"neither": "no transcription either way in KMT2A-r material -> the occupancy is at a silent bidirectional CpG island, consistent with the coordinator's finding that the shared-window enrichment is accessibility rather than fusion targeting."
}measurements
[]
- A tier-1 negative must not be reported as 'AFF1-AS1 is not expressed'. Poly-A selection, exosome degradation and low depth all produce the same zero, and this project's own rule is that an absence of testing is not a negative result. The design has to state which of those it can exclude.
- If tier 3 is ever run, note that a menin inhibitor removes the fusion from promoters generally, so a fall in transcription at this locus under drug is not evidence that the locus is a specific fusion target - the partner breakdown already says it is not.
- Which deposit? This beat did not enumerate the stranded RNA-seq datasets available for KMT2A-rearranged primary material, and picking the wrong one - unstranded, or poly-A selected in a way that loses an unstable antisense RNA - would produce a clean false negative. An antisense RNA degraded by the exosome is invisible to poly-A RNA-seq and visible to nascent or total RNA-seq, and that choice decides the answer.
- Is AFF1-AS1 poly-adenylated? NR_038841 is a three-exon spliced transcript, which suggests it is, but nothing confirms it and the tier-1 design silently assumes it.
- Should tier 1 also be run at IRX2/IRX2-DT? Q153 left the identical question open there and the code is the same.
- E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.dataset
This project's GSE252378 measurements are unstranded CUT&RUN coverage, so no measurement in this deposit at any resolution can assign a promoter signal to one of two divergently transcribed genes; the gene-body route is the only discriminator available in it, and this beat ran it.
Q155 interval deconvolution over all 133 GSE252378 libraries, with the intervals and their rationale · Iqbal Lab oncofusion knowledge graph, beat Q155 · this project 2026What this evidence cannot show
'Unstranded' is a property of the bigWig coverage tracks the deposit provides, which is what this project reads; the underlying sequencing may or may not have been stranded, and the raw reads were not examined. If the raw data are stranded, tier 1 could in principle be run inside GSE252378 itself - but CUT&RUN measures protein occupancy and not transcription, so strandedness there would not answer the transcription question either.
Evidence that argues against this (1)
- The gene-body route did produce a directional signal - AFF1's width-matched body carries a median 2.247x the AFF1-AS1 body and exceeds it in 25 of 27 samples. Presenting occupancy as wholly uninformative would understate that. What it cannot do is separate a promoter's ownership, which is the narrower and correct claim. Core LJ et al. 2008
- E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
Divergent initiation from a single nucleosome-depleted region is pervasive at human promoters, so an annotated antisense TSS 151 bp from a coding gene's TSS is expected to share initiation machinery with it rather than to have a separable promoter.
Nascent RNA sequencing reveals widespread pausing and divergent initiation at human promoters. · Core LJ, Waterfall JJ, Lis JT · Science 2008What this evidence cannot show
A 2008 GRO-seq study in human fibroblasts (IMR90), not in leukaemia and not at this locus. It establishes the general phenomenon and licenses the experimental design; it says nothing specific about AFF1. Pervasive divergent initiation also means the mere existence of an antisense transcript at AFF1 is unremarkable - the question is whether it is regulated, not whether it exists.
Evidence that argues against this (1)
- If divergent initiation is pervasive and mostly unregulated, then AFF1-AS1 may be a by-product of AFF1's own promoter rather than an independent gene - in which case the attribution problem is real for the measurement and empty for the biology, because there is only one regulatory unit. That is a possible outcome of this experiment and would be the most useful result: it would license this project to treat the AFF1 window as AFF1's. Trinklein ND et al. 2004
the only design that can say whether this transcript is made in KMT2A-rearranged material
would settle which of the three genes the 20/27 promoter call belongs to
Q153 found the gene-body route for IRX2 and that it had never been run; this beat ran the AFF1 equivalent and found the route is not sufficient for a promoter question