Day 3 of a DOT1L inhibitor has already removed 85-90% of H3K79me2 absolutely, and the collapse becomes progressively LESS uniform: the p10-p90 spread of per-gene retention widens from 2.4-fold at d3 to 8-44-fold at d9 in all four arms
also called 85% gone by day 3 · the collapse disperses · D3 fires - neither uniform decay nor kinetic order is scoreable here
Three days of the drug already strips most of the chromatin mark, and the loss is very even across genes at that point. By day nine there is far less left, and what is left is spread wildly unevenly - some genes keep thirty times more of it than others. The collapse does not flatten out; it fans out.
The same thing, for a scientist
**Statement, ABSOLUTE.** Global fold-down of H3K79me2 at TSS windows, spike-in scaled, by day: | arm | d3 | d6 | d9 | |---|---:|---:|---:| | RS4;11 EPZ5676 | **10.8x** | 24.3x | 52.4x | | RS4;11 compound 10 | **6.5x** | 45.3x | 90.6x | | MOLM13 EPZ5676 | **7.6x** | 12.6x | 63.5x | | MOLM13 compound 10 | **7.3x** | 32.9x | 36.9x | **The decline is monotone in all four arms**, so the pre-registered D3 worry about non-monotonicity being a scale-factor artefact does not arise at the global level. And **85-90% of the mark is already gone at day 3** - the first timepoint in the deposit is well past half-maximal. **Statement, ABSOLUTE, and it is the more interesting one.** The per-gene retention distribution **fans out** as the collapse deepens. p10 to p90 of percentage retained: | arm | d3 | d6 | d9 | |---|---|---|---| | RS4;11 EPZ | 6.9-16.3% (**2.4x**) | 2.1-58.4% (27x) | 0.7-32.2% (**44x**) | | MOLM13 EPZ | 8.5-26.5% (3.1x) | 2.8-72.7% (26x) | 0.6-8.1% (14x) | | RS4;11 comp10 | 11.9-37.9% (3.2x) | 1.3-12.1% (9.1x) | 0.5-12.9% (26x) | | MOLM13 comp10 | 10.0-41.1% (4.1x) | 1.6-12.9% (8.1x) | 1.2-10.0% (8.5x) | Every arm widens. At day 3 the mark is removed almost evenly from every gene; by day 9 the surviving fraction varies by one to nearly two orders of magnitude between genes. **Pre-registered branch D3 fires, for a reason the branch anticipated in kind but not in detail.** The rank-stability statistic that would distinguish D1 (uniform decay) from D2 (kinetic order) is **not scoreable here**: Spearman of per-gene ratio is 0.49-0.78 between d3 and d6 but 0.78-0.92 between d6 and d9, and the low early value is exactly what dynamic-range attenuation produces when the d3 contrast spans only 2.4x. **A correlation cannot separate 'the genes reordered' from 'there was nothing yet to order.'**
What an absolute time course buys that a share never could
A within-sample share at day 3 would report that the distribution barely moved - because it barely did, in share terms: with the loss nearly uniform, every gene's share of the total is almost unchanged. The spike-in says 85-90% of the material is gone.
And at day 9 the share would report a dramatic redistribution, with sparse genes apparently gaining. The spike-in says every gene lost; what changed is how unevenly.
So the share and the absolute scale disagree most severely at exactly the two ends of this time course, and in opposite directions. That is the practical case for the spike-in, measured rather than argued. 1
The one anomaly, named rather than smoothed
The EPZ5676 day-6 libraries in both cell lines carry a heavy right tail that neither the day-3 nor the day-9 libraries share and that neither compound-10 day-6 library shares: p90 retention of 58.4% (RS4;11) and 72.7% (MOLM13) against medians of 4.5% and 14.6%. At the loosest threshold 122 and 900 genes respectively appear to hold or gain at d6, and zero do at d9.
With one library per condition there is no way to decide whether that tail is a real transient - a subset of loci that resist until between day 6 and day 9 - or a property of those two libraries. It is recorded here so that a future replicated experiment has a specific thing to look for, and it is not used to support any claim. 1
| script | analyses/absolute-h3k79me2-in-the-index-fusion/run.py |
|---|---|
| status | run |
| result summary | Global absolute fold-down d3/d6/d9: RS4;11 EPZ 10.8/24.3/52.4; RS4;11 comp10 6.5/45.3/90.6; MOLM13 EPZ 7.6/12.6/63.5; MOLM13 comp10 7.3/32.9/36.9. Monotone in all four. p10-p90 retention spread widens from 2.4-4.1x at d3 to 8.5-44x at d9. |
| monotone in all arms | true |
| prereg branch | D3 - the rank-stability statistic is confounded by dynamic range and cannot separate D1 from D2 |
| named anomaly | EPZ5676 d6 in both lines carries a heavy right tail (p90 58.4% and 72.7%) absent at d3, d9 and in both compound-10 d6 libraries |
inputs
[ "GSE149183 RS4_11 and MOLM13 sitepro tables" ]
spearman d3 d6 range
[ 0.493, 0.784 ]
spearman d6 d9 range
[ 0.781, 0.915 ]
- If 85-90% of H3K79me2 is gone by day 3 and the cells are still alive at day 9, then the mark's abundance is a poor proxy for the dependency the drug is meant to exploit, and a PD assay reading total H3K79me2 saturates long before the biology responds.
- A widening spread of retention means late resistance could be seeded by a small set of loci that hold the mark, which a genome-average PD readout would never see.
- The first timepoint is already past 85% loss, so the deposit cannot see the fast phase at all. Hours, not days, is the sampling this question needs, and no public dataset with a spike-in does it.
- Does the widening spread survive a measured detection floor? An IgG or no-primary library at day 9 would settle in one experiment whether the low tail is biology or background, and it is the single cheapest missing control in this deposit.
- RS4;11 doubles in roughly 35-60 hours, so day 9 is about 4-6 divisions. Passive dilution alone predicts 16-64x - strikingly close to the measured 52-91x. Does that arithmetic hold at day 3 and day 6 too, and if it does, is any active removal needed at all?
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset
Spike-in scaled H3K79me2 in GSE149183 falls 6.5-10.8x by day 3 and 36.9-90.6x by day 9, monotonically in all four arms, while the p10-p90 spread of per-gene retention widens from 2.4-4.1x at day 3 to 8.5-44x at day 9.
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
Each timepoint's DMSO control is its own library, so the day-to-day global figures rest on three separate unreplicated scale factors rather than one; the monotonicity is therefore a property of three independent pairs and is the more robust for it, but no individual figure has an error bar. Retention deciles are computed on genes with that day's DMSO baseline at or above 25, so the gene set differs slightly between days. The widening spread is partly expected: as the surviving signal falls towards the detection floor, relative variation must grow. That floor cannot be measured because the deposit has no IgG or no-primary control library.
Evidence that argues against this (1)
- Because H3K79me2 has no well-established demethylase, loss is generally modelled as replication-coupled dilution plus turnover, which predicts a roughly uniform exponential decay - i.e. the opposite of a widening spread. Either the dilution model is incomplete or the widening reflects the detection floor rather than biology; this deposit cannot distinguish them. Bernt KM et al. 2011
- E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
The GSE149183 design measured H3K79me2 deposition at days 3, 6 and 9 after DOT1L-inhibitor treatment in MOLM13 and RS4;11 with Drosophila spike-in chromatin, one library per condition per timepoint.
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 one-library-per-condition design is read directly from the GEO sample list: GSM4491626-GSM4491634 (MOLM13) and GSM4491635-GSM4491643 (RS4;11), nine each, no replicate labels, fetched 2026-09-23.
Evidence that argues against this (1)
- The DOT1L and MEN1 occupancy samples in the same series DO carry REP1 labels, which shows the depositors used replicate labelling where replicates existed - so the absence of such labels on the H3K79me2 samples is evidence that there are none, not an omission. NCBI GEO 2020
absolute-scale time course: 85-90% gone by day 3, 37-91x by day 9
4-6 divisions by day 9 predicts 16-64x by passive dilution; measured 52-91x
three timepoints on an external ruler, but the first is already past 85% loss so the fast phase is invisible
the day-9 endpoint of the same scaled series
- tests →Pinometostat (EPZ-5676) — the DOT1L inhibitor, and the most instructive failure in this field
EPZ5676 1 uM time course, absolute scale: 10.8x / 24.3x / 52.4x in RS4;11