On the absolute scale the H3K79me2 gradient is real and it runs the other way from the share: the most heavily marked decile retains 0.8% and the least marked retains 31%, a 41-fold gradient - and every decile, in both lines and under both drugs, still loses
also called 41-fold gradient, all of it downward · the peaks are stripped hardest · the shape survives the rescaling
Genes that carry the most of this chromatin mark lose almost all of it; genes that carry the least lose a smaller share. That gradient is real and not an accounting trick. But once a proper external yardstick is used, every group of genes loses - the ones that looked like they were gaining were simply losing less than average.
The same thing, for a scientist
**Statement, ABSOLUTE.** RS4;11 (genuine rows only), EPZ5676 day 9, stratified on DMSO day-3 signal - a stratifier external to the contrast, as a sibling beat established is mandatory - percentage of H3K79me2 **retained**: | stratum | retained | |---|---:| | TOP 0.1% | **0.75%** | | D1 (most marked decile) | 0.76% | | D2 | 0.88% | | D3 | 1.22% | | D4 | 2.05% | | D5 | 4.87% | | D6 | 16.25% | | D7 | 25.31% | | D8 | 28.01% | | D9 | 29.88% | | D10 (least marked decile) | **30.88%** | **A 41-fold gradient, monotone, and not one stratum above 1.0.** The same shape appears in MOLM13 and under compound 10, and in the cross-line decile table where the lowest-baseline decile retains 23.7% (RS4;11) and 8.0% (MOLM13) against 0.82% and 0.46% at the highest. **This settles a question this corpus has held open in two directions.** Cycle 110 first called the gradient an artefact of renormalisation, then corrected itself and called it real but read it as the sparse deciles *gaining*. **The gradient is real** - it is present in the absolute values, not only in the shares - **and nothing gains.** Both of cycle 110's readings were half right and the halves belong to different sentences.
The reading that is not safe
The tempting biological story is that lightly-marked genes resist DOT1L inhibition. The detection floor is an equally good explanation and this deposit cannot exclude it. A gene whose DMSO signal is near the bottom of the measurable range sits close to whatever non-specific signal the antibody and the sequencing depth produce, and that floor does not fall when the real signal does - so the ratio at low baseline is bounded below by the floor, mechanically.
The test that would settle it is a no-primary-antibody or IgG library at day 9, spike-in normalised. GSE149183 contains no such control - its only non-ChIP samples are two MOLM13 inputs and one RS4;11 input (GSM4491615, GSM4491644, GSM4491645), which are chromatin inputs and not antibody controls, and which carry no dm6 spike-in table of their own.
So the honest statement is: the top of the gradient is solid - heavily marked genes are stripped to under 1% - and the bottom of the gradient is not interpretable without a floor measurement. 1
Why the stratifier is external, and why that matters
The strata are defined on DMSO day 3, which is neither member of the day-9 contrast. A sibling analysis in this corpus showed that self-stratifying - ranking on one arm of the contrast being measured - manufactures a monotone gradient from two identical libraries through regression to the mean alone. Using an external stratifier is not optional and is the reason this gradient can be reported at all.
With a single library per condition there is still no way to put an interval on any individual stratum. The gradient is reported because it is monotone across eleven strata in four independent arms, which is a pattern argument, not because any one stratum is significant. 12
| script | analyses/absolute-h3k79me2-in-the-index-fusion/run.py |
|---|---|
| status | run |
| result summary | RS4;11 EPZ d9, stratified on external DMSO d3: TOP 0.75%, D1 0.76%, D5 4.87%, D10 30.88% retained; 41-fold gradient, monotone, no stratum at or above 100%. |
| stratifier | DMSO day 3, external to the day-9 contrast |
| gradient fold | 41 |
| strata retaining 100pct or more | 0 |
| no igg or no primary control in deposit | true |
inputs
[ "GSE149183 RS4_11 sitepro table, genuine rows only" ]
- If the fusion's highest-occupancy domains are the ones stripped most completely, then the chromatin effect of a DOT1L inhibitor is maximal exactly where the drug is supposed to act - and pinometostat still failed clinically. That is an argument that the limiting step is downstream of the mark, not at it.
- Is the low-baseline plateau at ~30% real residual H3K79me2 or the antibody floor? One no-primary library would answer it and none exists in this deposit or, as far as this beat could find, in any spike-in H3K79me2 deposit.
- Does the gradient track baseline signal or gene length? The two are correlated and the fixed -1 kb/+3 kb window makes them hard to separate from a TSS table alone.
- This corpus's eviction-gradient entries are all computed on shares. Re-running them against an absolute scale is impossible for want of spike-ins, but re-running THIS gradient with the share statistic and publishing the two side by side would give the corpus a calibration point it does not have.
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset
Stratified on an external DMSO day-3 signal, spike-in-scaled H3K79me2 retention at day 9 in RS4;11 runs monotonically from 0.75% in the top 0.1% of genes to 30.9% in the least-marked decile, with no stratum retaining 100% or more.
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
The low-baseline end of the gradient is not separable from the assay's detection floor, because the deposit contains no IgG or no-primary-antibody library. One library per condition, so no stratum has an interval. Restricted to the 19,214 genuinely-RS4;11 rows; the same gradient computed on the full contaminated table is shallower (TOP 0.79% to D10 23.5%), which is itself a reason to use the genuine rows.
Evidence that argues against this (1)
- This graph's own sibling result shows that a monotone gradient rising into the sparse strata is exactly what a pure global decrease produces in a SHARE statistic, with no redistribution at all. That warning applies to share-based gradients; this one is computed on spike-in-scaled values, where a global decrease moves every stratum down together and cannot generate a gradient on its own. The distinction is the whole point of using the spike-in, and it is why the same shape means something different here. Orlando DA et al. 2014
- E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
H3K79me2 is deposited by DOT1L across the bodies of actively transcribed genes and is most abundant at KMT2A-fusion target loci in rearranged leukaemia, so the top strata here are the fusion's own high-occupancy domains.
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; the identification of the top strata with fusion target domains is an inference from enrichment, not a demonstration in RS4;11, and this deposit carries no fusion occupancy track for RS4;11 at all.
Evidence that argues against this (1)
- The top strata being stripped hardest is also what a simple mass-action model predicts for any mark whose maintenance requires continuous re-deposition: the more there is, the more is lost per division. That would make the gradient a property of turnover kinetics rather than of fusion biology. Daigle SR et al. 2011
41-fold absolute retention gradient, monotone, every stratum losing
the same strata, the same conclusion: the shape is real and nothing gains
an absolute-scale analogue of the eviction gradient, in the index fusion's own line
the low-baseline end of the gradient is uninterpretable without a measured floor