The same baseline-matched test that kills the target effect under EPZ5676 fires hard under compound 10 - at d6 and d9, in both cell lines, 4 of 4 contrasts (p = 0.0012 to <0.0001) - and in MOLM13 HOXA9 holds 24.5% of its H3K79me2 where the genome median holds 4.5%
also called compound 10 spares HOXA9 · two DOT1L inhibitors are not interchangeable · a drug-specific chromatin signature
Two drugs that block the same enzyme do not strip the chromatin mark the same way. The clinical-stage one removes it evenly from every gene. A newer one removes it everywhere too, but leaves several times more of it behind at exactly the genes a KMT2A fusion depends on - and this shows up in both cell lines and at both later timepoints.
The same thing, for a scientist
**Statement, RELATIVE.** The identical baseline-decile-matched permutation test was run in all twelve line-by-drug-by-day contrasts. It divides them cleanly: | contrast | observed median log2 | matched null | p | direction | |---|---:|---:|---:|---| | RS4;11 EPZ d3 | -0.414 | -0.400 | 0.896 | - | | RS4;11 EPZ d6 | -2.850 | -2.982 | 0.402 | - | | RS4;11 EPZ d9 | -3.874 | -4.046 | 0.331 | - | | MOLM13 EPZ d3 | -0.561 | -0.395 | 0.218 | - | | MOLM13 EPZ d6 | -3.018 | -2.847 | 0.341 | - | | MOLM13 EPZ d9 | -3.444 | -3.734 | 0.220 | - | | RS4;11 comp10 d3 | -0.548 | -0.547 | 0.986 | - | | MOLM13 comp10 d3 | -0.697 | -0.805 | 0.392 | - | | **RS4;11 comp10 d6** | **-1.319** | **-1.962** | **0.0012** | **spared** | | **RS4;11 comp10 d9** | **-1.672** | **-3.598** | **<0.0001** | **spared** | | **MOLM13 comp10 d6** | **-0.854** | **-1.835** | **0.0004** | **spared** | | **MOLM13 comp10 d9** | **-0.301** | **-2.622** | **<0.0001** | **spared** | **Zero of six EPZ5676 contrasts fire. Four of four compound-10 contrasts at d6 and d9 fire, in both cell lines, in the same direction.** Both compound-10 day-3 contrasts are null, so the effect is not present at the start and grows. **Statement, ABSOLUTE.** MOLM13, compound 10, day 9, percentage retained: HOXA9 **24.5%**, PBX3 18.2%, JMJD1C 11.3%, HOXA10 8.0%, MEIS1 6.2%, MYB 5.8%, CDK6 2.9%, HOXA7 1.9%, MEF2C 0.8%, HOXA5 0.4%. Genome median: **4.5%**. The decile-10 housekeeping comparators in the same library retain GAPDH 0.88%, RPL15 0.95%, RPLP0 0.87%, TUBB 0.79% - **an order of magnitude below HOXA9.** In RS4;11 the same pattern is weaker but present: HOXA10 6.9%, HOXA9 6.7%, JMJD1C 5.0% against a genome median of 2.0%.
The explanation that has to be excluded first, and is
The obvious reading is that compound 10 is simply a weaker drug at these genes, and weaker drugs spare things. The global numbers exclude it. Compound 10 at 100 nM produces a larger global collapse than EPZ5676 at 1 uM in RS4;11 (90.6x vs 52.4x) and a smaller one in MOLM13 (36.9x vs 63.5x). The target sparing appears in both lines. So it does not track global depth of inhibition in either direction.
The second candidate is background: if compound-10 libraries carried more non-specific signal, high-baseline genes would appear to hold a floor. But the null is matched on baseline decile, so every comparator gene is drawn from the same signal level as the target it stands in for; a uniform floor raises both equally and cancels.
The third is dose. EPZ5676 is given at 1 uM and compound 10 at 100 nM, so compound identity and concentration are confounded and cannot be separated in this deposit. That confound is real and is the main reason this entry's confidence is moderate rather than strong. 12
What this would mean if it holds, and what it would mean if it does not
If it holds, two inhibitors of the same catalytic site leave different residual H3K79me2 landscapes, and the one that spares the fusion's own target loci is the one that would be expected to work less well - which is the opposite of how a more potent compound is usually described. It would make 'residual H3K79me2 at HOXA9' a compound-discriminating pharmacodynamic readout rather than a generic exposure marker.
If it does not hold, the likeliest explanation is that these are four single libraries and the pattern is a property of the compound-10 d6/d9 material - a batch, a chromatin prep, a sequencing run. Nothing in this deposit can distinguish those, because there are no replicates at all.
The cheap test is specified in the companion experiment entry: two DOT1L inhibitors at matched target-occupancy, spike-in normalised, with replicates, read out at the HOXA cluster against baseline-matched controls. 13
| script | analyses/absolute-h3k79me2-in-the-index-fusion/run.py |
|---|---|
| status | run |
| result summary | Baseline-matched permutation test fires in 4 of 4 compound-10 d6/d9 contrasts (p 0.0012 to <0.0001) and 0 of 6 EPZ5676 contrasts and 0 of 2 compound-10 d3 contrasts. |
| hoxa9 absolute PCT retained molm13 comp10 d9 | 24.52 |
| genome median absolute PCT retained molm13 comp10 d9 | 4.47 |
| n permutations | 5000 |
| seed | 20260923 |
inputs
[ "GSE149183 RS4_11 and MOLM13 sitepro tables" ]
contrasts firing
[ "RS4;11 comp10 d6", "RS4;11 comp10 d9", "MOLM13 comp10 d6", "MOLM13 comp10 d9" ]
contrasts not firing
[ "RS4;11 EPZ d3", "RS4;11 EPZ d6", "RS4;11 EPZ d9", "MOLM13 EPZ d3", "MOLM13 EPZ d6", "MOLM13 EPZ d9", "RS4;11 comp10 d3", "MOLM13 comp10 d3" ]
- If residual H3K79me2 at the fusion's own targets is what matters, then the more potent compound on a global measure is the worse compound on the measure that counts - and potency screens would select against the right drug.
- The reverse bite is equally live: a compound that leaves 24% of H3K79me2 at HOXA9 while stripping the genome has a worse therapeutic ratio, since the normal-cell cost is paid genome-wide and the leukaemic benefit is not collected at the locus that matters.
- Either way, a single-locus H3K79me2 pharmacodynamic assay reads differently for different DOT1L inhibitors, so a PD threshold established with one compound cannot be carried to another.
- Does the sparing track a physical feature of the HOXA cluster - a retained DOT1L complex, a slower-replicating domain, a higher nucleosome turnover rate - or is it a property of the two compound-10 libraries?
- Compound 10 has no resolvable chemical identity in any public database this beat could reach, so the result cannot be connected to structure, selectivity or any later compound. Naming it is a prerequisite for anyone acting on this.
- Would the same matched-null test applied to a menin inhibitor split the same way? Every menin-inhibitor chromatin comparison in this graph uses an unmatched contrast.
- E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset
In GSE149183, the canonical KMT2A-fusion target genes retain significantly more H3K79me2 than baseline-matched genes under compound 10 at days 6 and 9 in both RS4;11 and MOLM13, and never under EPZ5676 at any timepoint.
GSE149183 series supplementary files: RS4_11 and MOLM13 H3K79me2 TSS-ChIPseq signal + dm6 spike-in sitepro tables · Perner F, Gadrey JY, Xiong Y, et al. · NCBI Gene Expression Omnibus 2020What this evidence cannot show
Confidence is moderate for four reasons and each is load-bearing. (1) NO REPLICATES - the four firing contrasts are four single libraries, and no noise floor can be measured or imported, so the p-values are permutation p-values over genes, not over experiments, and they cannot exclude a library-level batch effect. (2) DOSE IS CONFOUNDED WITH COMPOUND - EPZ5676 at 1 uM against compound 10 at 100 nM. (3) Ten genes. (4) Matching is on baseline signal only; gene length and the fixed -1 kb/+3 kb window are uncontrolled, and the HOXA cluster genes are short and tightly packed, so a window artefact specific to that locus is not excluded. What the design does support is that the pattern replicates across two cell lines, two fusion partners and two timepoints while being absent from six contrasts of the comparator drug.
Evidence that argues against this (1)
- The clinical-stage compound is EPZ5676, and it is the one that shows NO target sparing here. If residual H3K79me2 at fusion targets predicted failure, EPZ5676 should have been the better drug on this readout and it still produced only modest clinical activity - so this chromatin distinction has no demonstrated clinical correlate and may have none. Stein EM et al. 2018
- E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
Compound 10 is one of the novel DOT1L inhibitors reported in Perner et al., Blood 2020, tested at 100 nM against EPZ5676 at 1 uM in the H3K79me2 ChIP-seq time course.
Novel inhibitors of the histone methyltransferase DOT1L show potent antileukemic activity in patient-derived xenografts. · Perner F, Gadrey JY, Xiong Y, Hatton C, Eschle BK, Weiss A, et al. · Blood 2020What this evidence cannot show
The concentrations are read from the GEO sample titles (RS411_ChIP_H3K79me2_CST_EPZ5676_1uM_d9, ..._comp.10_100nM_d9), fetched 2026-09-23. Compound 10 has no ChEMBL entry or generic name that this beat could resolve, so it cannot be linked to an external chemistry record.
Evidence that argues against this (1)
- A tenfold lower concentration of a more potent compound is not necessarily a matched target occupancy, so the two arms may simply sit at different points on the same dose-response curve. The deposit contains no occupancy measurement that would settle it - the DOT1L occupancy ChIPs in the same series are MOLM13-only, at day 4, and carry no spike-in. NCBI GEO 2020
- E3established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
H3K79me2 deposited by DOT1L is enriched across KMT2A-fusion target loci including HOXA9 and MEIS1, and MLL-rearranged leukaemia depends on that aberrant methylation.
MLL-rearranged leukemia is dependent on aberrant H3K79 methylation by DOT1L. · Bernt KM, Zhu N, Sinha AU, Vempati S, et al. · Cancer Cell 2011What this evidence cannot show
Mouse MLL-AF9 model; the dependence is on the mark, not on a particular inhibitor, and the paper says nothing about compound-to-compound differences in residual landscape.
Evidence that argues against this (1)
- The companion paper in the same issue reports selective killing of MLL-rearranged cells by EPZ-5676's predecessor chemistry, establishing that near-complete H3K79me2 removal is achievable and lethal in vitro - so residual mark at target loci is not obviously the variable that decides efficacy. Daigle SR et al. 2011
same test, same genes, same null; the two drugs split 0-of-6 against 4-of-4
- related to →HOXA9 — required for survival of established MLL-rearranged leukemia, and dispensable for starting it
retains 24.5% of its H3K79me2 in MOLM13 under compound 10 at d9, against a genome median of 4.5%
- related to →Pinometostat (EPZ-5676) — the DOT1L inhibitor, and the most instructive failure in this field
EPZ5676 is the arm in which the sparing is absent at every timepoint in both lines
two inhibitors of the same catalytic site leave different residual landscapes
the residual mark at fusion target loci is compound-dependent
the observation this hypothesis generalises