The noncoding dependency screen five hypotheses in this graph are waiting on - and the KMT2A-rearranged screen that already exists and bounds it
also called noncoding CRISPR screen KMT2A::AFF1 · CRISPRi tiling screen infant ALL
Most of the genome is not genes, and almost every dependency screen in this leukaemia has only tested genes. This is the screen that would test the rest of it - the switches, the repeats and the structural boundaries - in the fusion that causes infant leukaemia.
The same thing, for a scientist
Five hypothesis entries (`hyp-nuclear-speckle-paraspeckle-kmt2a-r`, `hyp-r-loop-landscape-kmt2a-r`, `hyp-splicing-landscape-kmt2a-r`, `hyp-transposable-element-reactivation-kmt2a-r`, `hyp-trna-ribosome-biogenesis-kmt2a-r`) list this id in `data.planned_experiments` and it did not exist. It is written here with one correction to the premise those entries imply: **a noncoding CRISPR screen in KMT2A-rearranged leukaemia has already been done.** A dropout screen against **229 HOXA9-bound peaks** identified reproducible noncoding hits including a distal *FLT3* enhancer and a *CDK6* intronic element, validated by Cas9 editing and dCas9-KRAB silencing in vitro and in NSG transplants [E1]. That screen was **HOXA9-anchored, in MLL-r lines, and did not cover the element classes the five hypotheses are about**. This entry is therefore an extension with a stated prior, not a first-of-kind.
What already exists, and exactly what it does not cover
| existing | this design adds |
|---|---|
| 229 HOXA9-bound ChIP-seq peaks, dropout, MLL-r lines 1 | elements selected by fusion occupancy and by H3K79me2/3 dependence, not by a downstream transcription factor |
| HOXA-locus CTCF boundaries in a focused library | boundaries plus the element classes below |
| - | KEEs - H3K79me2/3-dependent enhancer elements, the one element class whose dependency is created by this disease |
| - | R-loop-forming regions at fusion target gene bodies |
| - | lncRNA promoters and gene bodies |
| - | transposable-element and ERV loci in the hypermethylated t(4;11) genome |
| - | tRNA genes and rDNA-proximal elements |
| - | speckle- and paraspeckle-associated loci |
And the system. The existing screen is in MLL-r lines anchored on HOXA9 - which is precisely the module that a HOXA-low infant subgroup may not run. A KMT2A::AFF1 screen that inherits a HOXA9-anchored element list inherits that blind spot.
Design
Modality: two, not one. dCas9-KRAB CRISPRi for elements that can be silenced, and Cas9 saturating mutagenesis for elements whose function is sequence-encoded. The two disagree systematically - CRISPRi spreads and can silence a neighbour; Cas9 tiling resolves minimal features but cannot test an element inside an essential gene body 23. Running one modality is the commonest way this screen fails.
Library construction, which is the real work. Elements are enumerated from fusion CUT&RUN, H3K79me2/3 CUT&RUN under DOT1L inhibition (to define KEEs by dependency rather than by mark), ATAC, and a repeat annotation - then tiled at ~20 guides per element with non-targeting, safe-harbour and promoter-positive-control guides in every pool.
Lines. SEM and RS4;11 (KMT2A::AFF1) as the index arm; a KMT2A::MLLT3 line as a partner control; a KMT2A-germline B-ALL line as the fusion-negative control. The comparator is the point - a noncoding element essential in every leukaemia line is not a fusion dependency, and gene-level screens in this disease have repeatedly been misread that way.
Readout. Dropout at 14 and 21 days; hits re-tested individually with transcriptional readout of the nearest genes; a subset taken to NSG transplant, as 1 did.
Naming, and why it is the hard part (brief rule 70)
Noncoding elements have no HGNC. The same element appears under an acronym, a spelled-out phrase, a target-gene-plus-enhancer description, or no name at all - this graph's own PRC1.1/MYC beat found the founding paper for one blood enhancer describing it only by coordinate. Consequences for this screen:
- Enumerate names from full texts before searching, not after.
- Search the genomic coordinate as free text; it is often the only stable identifier.
- Title-check every four-letter acronym; in biology they are almost always taken by something else.
- Record the name list in the caveat of every hit, because a reader cannot judge a noncoding claim without knowing which strings were tried.
A screen that reports element IDs without coordinates produces results nobody can re-find, which is the specific reason this design fixes hg38 coordinates as the primary key for every library member.
| design | Paired-modality noncoding dependency screen in human KMT2A::AFF1 lines: dCas9-KRAB CRISPRi plus Cas9 saturating mutagenesis, ~20 guides per element, over KEEs, fusion-bound intergenic and intronic elements, HOXA-locus CTCF boundaries, lncRNA loci, TE/ERV loci, tRNA genes and speckle-associated loci; index arm SEM and RS4;11, partner control KMT2A::MLLT3, fusion-negative control KMT2A-germline B-ALL. |
|---|---|
| readout | 14- and 21-day dropout; individual re-test with nearest-gene transcriptional readout; NSG transplant for a hit subset |
| controls | non-targeting, safe-harbour and promoter-positive-control guides in every pool; fusion-negative comparator line; both modalities on every element class |
| cost tier | high |
| timeline | 12-18 months |
| primary key for library members | hg38 coordinate, not element name (brief rule 70 - noncoding elements have no HGNC) |
tests
[ "hyp-nuclear-speckle-paraspeckle-kmt2a-r", "hyp-r-loop-landscape-kmt2a-r", "hyp-splicing-landscape-kmt2a-r", "hyp-transposable-element-reactivation-kmt2a-r", "hyp-trna-ribosome-biogenesis-kmt2a-r" ]
prior art
{
"pmid": "38016946",
"what": "dropout screen over 229 HOXA9-bound peaks in MLL-r lines",
"what_it_does_not_cover": [
"KEEs defined by H3K79me2/3 dependence",
"R-loop-forming regions",
"lncRNA loci",
"TE and ERV loci",
"tRNA genes",
"speckle- and paraspeckle-associated loci",
"a fusion-negative comparator arm"
],
"title": "Systematic characterization of the HOXA9 downstream targets in MLL-r leukemia by noncoding CRISPR screens.",
"authors": "Wright S, Zhao X, Rosikiewicz W, et al.",
"venue": "Nat Commun",
"year": 2023,
"title_provenance": "Title/authors/venue backfilled from PubMed esummary on 2026-09-20 because this citation stored a bare PMID with no title, which made it unscoreable by verify-citations.py. BACKFILLING IS NOT VERIFICATION: it confirms the PMID resolves to a real paper, not that this is the right paper for the claim. That still requires reading."
}element classes
[ "KEE", "fusion-bound intergenic enhancer", "fusion-bound intronic enhancer", "HOXA CTCF boundary", "lncRNA promoter and body", "TE/ERV locus", "tRNA gene", "speckle-associated domain" ]
- A noncoding dependency is not a drug target: an essential enhancer identifies a gene worth drugging, and the element itself is undruggable, so the screen's output must be read as target discovery at one remove.
- Element-level hits in a cell line are the most transfer-fragile results this graph collects, because enhancer usage is cell-state dependent and cell lines have drifted; a hit that does not reproduce in a PDX should be assumed to be a line artefact.
- A screen seeded on DOT1L-inhibitor response will preferentially find elements whose loss phenocopies a drug the field already has, and will systematically under-find dependencies orthogonal to it.
- Does a fusion-anchored element list recover the HOXA9-anchored hits, or are FLT3-enhancer and CDK6-intron dependencies invisible to a screen that does not start from HOXA9?
- In the HOXA-low infant subgroup, does the noncoding dependency map look like the HOXA-high one at all? No infant-derived material has ever been screened at the noncoding level.
- Can CRISPRi and Cas9 tiling be reconciled at KEEs, which lie inside gene bodies where cutting is confounded and silencing spreads?
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
Dropout CRISPR screens against 229 HOXA9-bound peaks in MLL-rearranged leukaemia identified reproducible noncoding hits including a distal FLT3 enhancer and a CDK6 intronic element; Cas9 editing and dCas9-KRAB silencing of those sites reduced the corresponding gene's transcription and impaired proliferation in vitro and in NSG transplants.
Systematic characterization of the HOXA9 downstream targets in MLL-r leukemia by noncoding CRISPR screens. · Wright S, Zhao X, Rosikiewicz W, Mryncza S, et al. · Nat Commun 2023What this evidence cannot show
Abstract read 2026-09-20. This is the prior art that makes this design an extension rather than a first. Its element list is anchored on HOXA9 ChIP-seq peaks, so it inherits whatever HOXA9 does not bind - and in the HOXA-low infant t(4;11) subgroup that may be most of the fusion's programme. Cell lines plus xenograft; no infant-derived material.
Evidence that argues against this (1)
- Anchoring on a downstream transcription factor rather than on the fusion means a hit is a HOXA9 dependency, not necessarily a fusion dependency, and the screen carries no fusion-negative comparator arm that would separate the two. Faber J et al. 2009
- E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
CRISPR interference across more than one megabase around MYC and GATA1 discovered nine distal enhancers controlling gene expression and cellular proliferation, and chromatin-state and conformation features distinguished the seven MYC-regulating enhancers from elements that did not.
Systematic mapping of functional enhancer-promoter connections with CRISPR interference. · Fulco CP, Munschauer M, Anyoha R, Munson G, et al. · Science 2016What this evidence cannot show
Abstract read 2026-09-20. Methodological citation establishing the CRISPRi tiling modality and its scale. In a leukaemia line MYC is itself a fusion-relevant locus, so the assay is demonstrated on relevant biology - but the paper is not about KMT2A-rearranged disease.
Evidence that argues against this (1)
- dCas9-KRAB silencing spreads over kilobases, so a CRISPRi hit can be an artefact of silencing a neighbouring element or promoter; this is why the design pairs it with Cas9 saturating mutagenesis rather than running it alone. Canver MC et al. 2015
- E3established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
Pooled CRISPR-Cas9 guide libraries performing in-situ saturating mutagenesis of the human and mouse BCL11A erythroid enhancers revealed critical minimal features and discrete vulnerabilities, with the crucial human sequences appearing to be primate-specific despite conserved composite function.
BCL11A enhancer dissection by Cas9-mediated in situ saturating mutagenesis. · Canver MC, Smith EC, Sher F, Pinello L, et al. · Nature 2015What this evidence cannot show
Abstract read 2026-09-20. Methodological citation for the second modality. The primate-specificity finding is the reason this design does not propose a mouse validation arm for element function: a murine result about a human enhancer can be conserved in function and divergent in sequence.
Evidence that argues against this (1)
- Saturating mutagenesis cannot test an element lying inside an essential gene body, because cutting there kills the cell for an unrelated reason - which is exactly where KEEs in this disease lie, intragenic in genes such as PROM1. Godfrey L et al. 2019
- E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
DOT1L inhibition defines H3K79me2/3-dependent enhancer elements (KEEs) at which the mark is required to maintain chromatin accessibility, histone acetylation and transcription-factor binding, while non-KEE enhancers are unaffected.
DOT1L inhibition reveals a distinct subset of enhancers dependent on H3K79 methylation. · Godfrey L, Crump NT, Thorne R, Lau IJ, et al. · Nat Commun 2019What this evidence cannot show
Abstract read 2026-09-20. MLL-AF4 cells. This defines the one element class whose dependency is created by this disease and is therefore the highest-priority library component; it also defines KEEs by drug response, which means the library is partly a pharmacological rather than a purely genomic construct.
Evidence that argues against this (1)
- An element defined by its response to a DOT1L inhibitor may be a drug-response element rather than a fusion-dependency element, and a screen seeded from that definition cannot distinguish the two; the fusion-negative comparator arm is what carries that burden. Godfrey L et al. 2021
R-loop-forming regions as a library element class
TE and ERV loci as a library element class
tRNA genes and rDNA-proximal elements as a library element class
named in that entry's planned_experiments
speckle- and paraspeckle-associated loci as a library element class
KEEs are the one element class whose dependency this disease creates
the other genuine gap among the multi-entry dangling references