Analysis: separate loss-of-protein from loss-of-occupancy by comparing degron-depletion and menin-inhibition transcriptional time courses
also called degron versus menin kinetics · occupancy-corrected response · fusion half-life inference
Two ways of shutting the fusion down — destroying it with a degron, or knocking it off DNA with a drug — should give different shaped response curves if the drug also destroys it. This analysis compares the shapes of published and in-house time courses to infer whether menin inhibition is partly a degrader, without needing a new experiment first.
The same thing, for a scientist
A degron system removes fusion **protein**; a menin inhibitor removes fusion **occupancy**. If menin inhibition were purely an occupancy effect, target-gene decay after drug should follow a first-order curve set by mRNA half-life and residual occupancy, and recovery after washout should be fast, gated by rebinding. If menin inhibition also destabilises the protein, the decay curve should acquire a second, slower component matching protein turnover, and recovery should be slow, gated by resynthesis — the signature of hysteresis. This analysis fits both models to paired time-series data and reports which the data support, with a bootstrap confidence interval on the inferred protein-loss fraction. It is runnable on data the lab already generates for `exp-degron-washout-recovery-kmt2a-aff1`, `exp-kmt2a-aff1-degron-occupancy-titration` and `exp-menin-pulse-chase-hysteresis`, and it produces the quantitative input that `hyp-evicted-fusion-is-degraded-not-displaced` needs.
What the script does
Inputs: (a) RNA-seq or nascent-RNA time courses after degron depletion (0, 1, 2, 4, 8, 24 h) and after menin inhibition at saturating dose over the same grid, from the same line; (b) matched CUT&RUN occupancy at the same timepoints for a fixed target-gene panel; (c) optionally, western-blot densitometry of the 240-kDa fusion species.
Model A (occupancy-only): target transcription is a monotone function of measured occupancy with a single first-order mRNA decay term. Model B (occupancy plus protein loss): adds a protein-decay term with its own rate constant, so that transcription declines further than occupancy alone predicts.
Fit both by maximum likelihood on the log-expression trajectories, compare by AIC and by a likelihood-ratio test, and bootstrap over genes to get a confidence interval on the extra protein-loss rate constant. Report the implied fusion half-life under drug and compare it with the directly measured cycloheximide-chase value when that becomes available — the analysis makes a falsifiable numerical prediction before the wet experiment is run, which is the whole point.
The key discriminating statistic is the residual: genes whose transcription falls more than their measured occupancy loss predicts. Under model A that residual is noise with mean zero; under model B it is systematically negative and grows with time.
Why this is worth doing before the imaging experiment rather than after
The imaging experiment costs 9-12 months. This analysis costs days and uses data that either exists or is being generated anyway for other entries in this graph. If it returns a clean model-A fit with no residual, the degradation hypothesis is weakened before anyone builds a knock-in line, and the imaging effort can be redirected to the compartment question. If it returns model B, the wet experiment has a pre-registered quantitative prediction to test, which is a far stronger publication than an exploratory time course.
It also has a built-in sanity check: menin's own degradation under these drugs is published, so menin protein levels in the same lysates serve as a positive control that the system reproduces a known effect.
| script | analyses/fusion-halflife-degron-vs-menin/run.py |
|---|---|
| status | planned |
| result summary | not yet run; outputs an AIC/likelihood-ratio comparison of occupancy-only versus occupancy-plus-protein-loss models, a bootstrap CI on the protein-loss rate constant, and an implied fusion half-life under drug |
| figure | /figures/fusion-halflife-model-comparison.svg |
| discriminating statistic | per-gene residual of transcription not explained by measured occupancy loss; mean zero under the occupancy-only model, systematically negative and time-growing under the protein-loss model |
| positive control | menin protein levels in the same lysates, whose drug-induced degradation is published |
| tests hypothesis | hyp-evicted-fusion-is-degraded-not-displaced |
inputs
[ "degron depletion RNA-seq or nascent-RNA time course (0,1,2,4,8,24 h)", "menin-inhibitor RNA-seq time course at saturating dose, same grid and line", "matched CUT&RUN fusion occupancy at the same timepoints for a fixed target panel", "optional western densitometry of the 240 kDa fusion species and of menin" ]
- Model selection on noisy time-series can favour the more complex model spuriously; the bootstrap and a permutation null on shuffled timepoints are required, not optional, and a marginal AIC difference should not be reported as support.
- If the two mechanisms operate simultaneously and with similar rate constants, the analysis will be underpowered and return ambiguity, which is a real possible outcome and should be pre-declared rather than resolved by choosing the more interesting fit.
- Is the degron system's own kinetics fast enough to separate the two models? If degron depletion takes hours rather than minutes, the protein-loss and occupancy-loss curves converge and the analysis loses power — a power calculation on the existing degron data should be the first step.
- Can the model be extended to fit the pulse-chase washout data directly, so that hysteresis magnitude becomes a second independent estimate of the same rate constant?
- Does the residual statistic behave differently for menin-dependent versus menin-independent target genes? If protein loss is real, both classes should respond; if only occupancy matters, menin-independent targets should be spared.
- E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper
Menin-MLL inhibitors reduce menin protein level and half-life through CHIP/Hsp70-mediated ubiquitination, and proteasome inhibition rescues both the degradation and the drug-induced reduction in target gene expression — establishing that a protein-loss component of the drug response exists for at least one complex member and that it is detectable in transcriptional readouts.
Disruption of the menin-MLL interaction triggers menin protein degradation via ubiquitin-proteasome pathway · Wu Y, Doepner M, Hojnacki T, Feng Z, Katona BW, He X, Ma J, Cao Y, Busino L, Zhou F, Hua X · American Journal of Cancer Research 2019What this evidence cannot show
Single-group cell-line work on menin, not on the fusion; provides the model's motivation and its positive control, not its conclusion.
Evidence that argues against this (1)
- The analysis is framed around the menin precedent for drug-induced degradation; the only direct measurement of MLL protein turnover points the other way for the fusion. Liu, Cheng and Hsieh report that MLL degradation initiates within the N-terminal approximately 1400 amino acids retained in every fusion, and that prevalent MLL fusions are resistant to the SCF(Skp2)/APC(Cdc20) degradation that clears wild-type MLL. A degron-tagged fusion therefore starts from an unusually stable baseline, and the comparison the analysis proposes (degron-driven loss versus menin-inhibitor-driven loss) may find a very large asymmetry for reasons that have nothing to do with menin. The half-life of the untreated fusion should be measured first and reported as the analysis's own control. Read: abstract only. Liu H et al. 2007
- E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
Menin inhibition is characterised in the pivotal literature by chromatin-occupancy and transcriptional readouts, so paired occupancy and expression time courses of the kind this analysis requires are the standard output of the field and are directly comparable across studies.
A Menin-MLL Inhibitor Induces Specific Chromatin Changes and Eradicates Disease in Models of MLL-Rearranged Leukemia · Krivtsov AV, Evans K, Gadrey JY, et al. · Cancer Cell 2019What this evidence cannot show
Published occupancy data are often at selected loci rather than genome-wide at matched timepoints, so cross-study application of the model will be weaker than an in-house paired design.
Evidence that argues against this (1)
- A design warning rather than a contradiction. The analysis reads fusion fate out of menin-inhibitor experiments whose published endpoints are occupancy and transcription. Risner et al. show that MLL-AF9 fusions carrying heterologous CXXC or MBD domains - including domains that do not bind the relevant DNA - are still delivered to Hoxa9 in primary progenitors, while differing sharply in leukemogenicity. Occupancy and function are therefore partly decoupled in this system, so a half-life comparison anchored to occupancy-based pharmacodynamics can move in the opposite direction to the biological effect. The analysis should report protein level, occupancy and transcription as three separate series rather than treating any one as a proxy for the others. Read: abstract only. Risner LE et al. 2013
supplies the reagent audit that analysis needs, and names the missing half-life number it currently assumes
that analysis fits washout kinetics whose published values come from an overexpressed construct
Counter-evidence sets the baseline: the untreated fusion's half-life must be measured first, because MLL fusions are reported abnormally stable relative to wild-type MLL
model comparison separating loss-of-protein from loss-of-occupancy
consumes the paired occupancy-expression time courses that entry generates
supplies the paired time-series data that analysis needs
replaces an inference from transcriptional time-course shape with a direct fraction-resolved measurement
- ← testsHypothesis: what the KMT2A half loses is control of its own abundance, not control of its activity
[tested_by] the head-to-head half-life measurement that would falsify or confirm it
supplies the mechanistic reason fusion half-life should be partner-set rather than KMT2A-set