KDM2B, the PRC1.1 CpG-island recruiter, sits at the 99.7th percentile of its own distribution on the fusion's strongest sites in DMSO and does not gain share under either a menin inhibitor or a DOT1L inhibitor - it holds position while comparable bins drift down
also called Q117 KDM2B does not move · KDM2B ceiling at fusion sites
The protein that tows the repressive PRC1.1 complex to unmethylated CpG islands is already sitting on the fusion's favourite sites before any drug is given, so it has nowhere to climb. What changes under treatment is not where it is but what it is allowed to do there.
The same thing, for a scientist
KDM2B is the CxxC-ZF subunit that addresses PRC1.1 to unmethylated CpG islands and is a documented member of the fusion interactome. In GSE260742's murine MLLAF9 line, at the 1,228 2 kb bins that carry the top 0.1% of FLAG-MLLAF9 signal in DMSO, **KDM2B's own within-sample percentile in DMSO has a median of 99.685** - it is already essentially maximal [E1]. Under VTP50469 its within-sample share of those bins changes by **+0.9%** (ratio 1.009) and under SGC0946 by **+5.0%** (1.050); in the +/-10 kb flank the ratios are **0.96 and 0.98** [E1]. Stratifying by how far the fusion fell makes almost no difference: 1.09 / 1.01 / 0.99 / 0.99 across quartiles under VTP50469 [E1]. At the locus level the only appreciable movements are **Six1 1.87x, Hoxa10 1.31x and Hoxa9 1.25x**, with everything else between 0.81 and 1.13 [E2]. **This is a partial dissent from the depositing paper's prose**, which states that *"following treatment, MLLAF9 target genes acquired strong KDM2B and BCOR occupancy"* [E3]. The BCOR half reproduces here (1.33 pooled, 1.44 in the most-displaced quartile). The KDM2B half does not, on this statistic at this resolution. The paper's own Fig. 4d statement supplies the reconciliation: **"KDM2B and H2AK119ub were enriched at active, H3K4me3-marked genes before treatment"** - so KDM2B was already there, and *acquired* is doing less work in that sentence than it appears to [E3]. **One measurement points the other way and is reported rather than dropped.** Against a class-matched null - bins matched on decile of DMSO KDM2B percentile crossed with decile of input percentile - the fusion-held bins hold flat (median shift +0.002 percentile points) while the matched bins fall 8.4 points (Mann-Whitney p = 5.4e-201) [E1]. **Relative to comparable chromatin, KDM2B is retained at fusion sites.** It does not gain; it fails to lose. That is a real and different statement from either 'acquired strong occupancy' or 'nothing happens'.
The ceiling, stated so nobody re-derives it as a finding
A rank statistic cannot detect a gain at a site that is already at the 99.7th percentile, and this beat's pre-registered secondary test was a rank statistic. That is why the share statistic was computed as well: a share has no ceiling, and it agrees - 1.009. The two independent statistics agreeing is the reason this entry states the negative rather than reporting the rank test as uninformative.
The general form is the project's own floor-compression rule turned upside down. A gene depleted in 100% of DepMap lines has no room to show selectivity; a factor at the 99.7th percentile has no room to show recruitment. Before running a gain test on a chromatin factor, look at where its baseline sits in its own distribution, and if it is at the ceiling use a statistic that has none.
What this implies for the PRC1.1 recruitment model
If KDM2B is already present at the fusion's sites before treatment, then treatment is not recruiting PRC1.1 by recruiting its addressing subunit. The candidates that remain are: licensing of an already-docked complex (the paper's model, in which H3K79me2 directly antagonises PRC1 ubiquitin-ligase activity), assembly of the catalytic subunits onto a pre-positioned KDM2B, or stabilisation of a complex that was present but exchanging rapidly. The RING1B and BCOR data in this beat are consistent with assembly: KDM2B flat, BCOR up 1.33, RING1B up 1.38 where the fusion is most completely lost.
That is a testable, cheap distinction - a KDM2B ChIP is not a PRC1.1 ChIP, and the deposit contains all three subunit readouts in the same cells under the same conditions, which is unusual and is why this question can be asked here at all.
| script | analyses/q117-gained-fusion-occupancy/run117.py + locus117.py |
|---|---|
| status | run |
| result summary | KDM2B is already at the 99.685th percentile of its own distribution at the fusion's top sites in DMSO. Share ratios 1.009 (VTP50469) and 1.050 (SGC0946) at those sites, 0.96 and 0.98 in the flank, flat across fusion-loss quartiles. Relative to a decile-matched null it holds position while matched bins fall 8.4 points. |
| gene | KDM2B |
| complex | PRC1.1 (variant PRC1, PCGF1-containing) |
| role | CxxC-ZF subunit that addresses PRC1.1 to unmethylated CpG islands |
| epigenetic readout | occupancy by ChIP-seq; the complex's catalytic product H2AK119ub was not measured in this beat |
| median DMSO percentile at fusion held bins | 99.685 |
| replicates | n = 1 per condition |
| shown for fusion | KMT2A::MLLT3 (MLLAF9) in a murine myeloid line; the KDM2B-fusion relationship has not been measured this way in human KMT2A::AFF1 |
| cell state | murine MLLAF9 myeloid leukaemia line, 48 h drug exposure, differentiation not yet overt at this timepoint in the depositing paper's own flow data |
inputs
[ "GSM8123392", "GSM8123393", "GSM8123394", "GSM8123359", "GSM8123361", "GSM8549823" ]
share ratio at fusion held bins
{
"VTP50469": 1.009,
"SGC0946": 1.05
}share ratio in 10kb flank
{
"VTP50469": 0.962,
"SGC0946": 0.981
}quartile by fusion loss VTP
[ 1.087, 1.007, 0.986, 0.987 ]
class matched null
{
"median_shift_fusion_held": 0.002,
"median_shift_matched_null": -8.425,
"delta_percentile_points": 8.427,
"pre_registered_threshold": 10,
"threshold_met": false,
"mannwhitney_p": 5.4e-201,
"n_matched": 1228,
"n_unmatched": 0,
"n_null_draws": 24560
}- If KDM2B is already resident, then any therapeutic strategy that aims to recruit PRC1.1 to fusion targets is aiming at the wrong step - the complex's address is already written, and what is being manipulated is its permission to act. A KDM2B-recruiting agent would do nothing.
- The converse is the risk: KDM2B is resident at active CpG islands genome-wide, not only at fusion targets, so a drug that licenses PRC1.1 catalysis rather than its recruitment would deposit H2AK119ub wherever KDM2B already sits. That is a large and largely untargeted footprint, and it is what a DOT1L inhibitor appears to do in this deposit.
- Restricted to unmethylated CpG islands - where KDM2B actually binds - does the picture change? This beat used genome-wide 2 kb bins and a 24-locus table; a CGI-restricted analysis is the right unit for a CxxC-ZF protein and was not done.
- Is KDM2B's occupancy at fusion sites exchanging rapidly in DMSO? A flat ChIP signal is compatible with a stably docked complex and with a fast-exchanging one, and the licensing model needs to know which.
- Does KDM2B occupancy at the fusion's sites depend on the fusion at all? The series contains a PCGF1 knockout arm (NTvKO) but no KDM2B ChIP in it, and no arm in which the fusion is degraded rather than pharmacologically displaced.
KDM2B is at the 99.685th percentile of its own distribution at the fusion's top sites in DMSO and its share does not move under either drug
KDM2B is the CxxC-ZF addressing subunit of PRC1.1
KDM2B flat while BCOR and RING1B rise is the pattern that distinguishes assembly-on-a-resident-address from recruitment
- related to →KMT2A / MLL1 — the gene that breaks
KDM2B is a documented member of the KMT2A-fusion interactome and this is its chromatin relationship to the fusion under drug
KDM2B at the 99.685th percentile is the worked example of the rank statistic's ceiling