One experiment that decides whether KDM4C is a target or the KDM4 family is: paralogue rescue of SD70 sensitivity in SEM
also called KDM4 paralogue rescue · SD70 rescue experiment
A drug that blocks three near-identical enzymes kills this leukemia, but deleting any one of them does nothing. Put each one back, one at a time, in cells treated with the drug: whichever ones rescue the cells tell you which enzyme the drug actually needs to hit.
The same thing, for a scientist
The KDM4C entry rests on a contradiction: SD70 preferentially kills SEM, a KMT2A::AFF1 line, while DepMap shows no KDM4C dependency in SEM (+0.009) or any other KMT2A::AFF1 line. The two candidate explanations — KDM4C is the target and CRISPR is blind to it because of paralogue buffering, versus SD70's effect is not through KDM4C at all — make opposite predictions under paralogue rescue, and the experiment is a single week of cell culture. It is proposed here as the gate KDM4C must clear before anything is spent on it, because every downstream decision (is selectivity achievable, is the therapeutic window set by the KDM4 triple-knockout HSC phenotype, is SD70 even the right tool) depends on the answer and none of them can be made without it.
Design
Cells. SEM and RS4;11 (KMT2A::AFF1 B-ALL, both FLT3 wild-type, which avoids the MV4-11 confounder), with a non-KMT2A-rearranged B-ALL line as the specificity control. MV4-11 optional as a bridge to the published tiling work.
Arms. Lentiviral overexpression of, separately: KDM4A, KDM4B, KDM4C, catalytically dead KDM4C (JmjC point mutant), and empty vector. Each arm treated with SD70 across a dose range around the published 0.8 uM used for human lines, plus vehicle.
Primary readout. Shift in SD70 GI50 relative to empty vector at 72 h and 7 days.
Secondary readouts. H3K9me3 by immunoblot (SD70's proximal pharmacodynamic mark) and HOXA9/MEIS1 mRNA by RT-qPCR at 72 h. The HOXA9 readout matters independently: no published work reports a transcriptional response to KDM4C perturbation in any KMT2A::AFF1 line, only viability, and that gap is why the KDM4C entry is moderate.
Controls. Catalytically dead KDM4C is the load-bearing control. If it rescues as well as wild-type, the rescue is not about demethylase activity and the whole catalytic-inhibitor premise is wrong — which is exactly what happened to KDM6A in CML, where both wild-type and demethylase-dead KDM6A restored imatinib resistance 3. Expression of each construct confirmed by immunoblot, since a failure to rescue is uninterpretable without it.
How the outcomes are read, decided before the experiment
| Result | Interpretation | Consequence for the register |
|---|---|---|
| KDM4A and/or KDM4B rescue; KDM4C also rescues | The target is the family, not KDM4C. The CRISPR null is explained. | KDM4C is not a gene-level candidate. The therapeutic window is set by the KDM4 triple-knockout HSC phenotype 2, which is a hard ceiling. Selectivity is not the goal; exposure control is. |
| Only KDM4C rescues | The target is KDM4C and the CRISPR null needs a different explanation. | KDM4C advances, and the paralogue-buffering hypothesis loses its best case. The next question is why a selective knockout does nothing. |
| Nothing rescues | SD70 is not acting through KDM4 in these cells. | KDM4C comes off the list. The SEM viability result becomes an off-target effect and the entry should be struck, with this experiment as the reason. |
| Catalytically dead KDM4C rescues as well as wild-type | The relevant function is non-catalytic. | A JmjC-site inhibitor is the wrong modality regardless of which paralogue matters. |
The third row is the one that makes this worth running rather than assuming: it is a real possible outcome that removes a candidate this project just added, and it costs a week to find out.
| design | Lentiviral overexpression of KDM4A, KDM4B, KDM4C, catalytically dead KDM4C (JmjC point mutant) or empty vector in SEM and RS4;11 (KMT2A::AFF1, FLT3 wild-type), with a non-KMT2A-rearranged B-ALL specificity control; SD70 dose-response around 0.8 uM plus vehicle. |
|---|---|
| readout | Primary: shift in SD70 GI50 versus empty vector at 72 h and 7 days. Secondary: H3K9me3 immunoblot as the proximal pharmacodynamic mark, and HOXA9/MEIS1 mRNA by RT-qPCR at 72 h — the fusion-output readout that no published work provides for KDM4C perturbation in any KMT2A::AFF1 line. |
| controls | Empty vector; catalytically dead KDM4C (distinguishes catalytic from scaffolding function, the KDM6A precedent); immunoblot confirmation of every construct's expression, without which a failure to rescue is uninterpretable; non-KMT2A-rearranged B-ALL line for specificity. |
| cost tier | low |
| timeline | 6-8 weeks including construct generation; 1 week of assay once lines are made |
| decision rule | Rescue by KDM4A or KDM4B means the target is the family and the window is bounded by the KDM4 triple-knockout HSC phenotype. Rescue only by KDM4C advances KDM4C and falsifies the paralogue-buffering explanation for its CRISPR null. No rescue by anything means SD70 is off-target in these cells and gene-kdm4c should be struck, citing this experiment. |
| why now | gene-kdm4c was created this cycle on a SEM viability panel. This is the cheapest experiment that could remove it again, and running it before any medicinal chemistry is the difference between a candidate and a sunk cost. |
- Overexpression rescue has a well-known failure mode in both directions: a non-physiological level of a paralogue can substitute for a function it never performs endogenously, producing a false family call, while insufficient expression produces a false KDM4C-specific call. The immunoblot control bounds the second and nothing bounds the first except interpreting a rescue as permissive rather than proof.
- A negative result here removes a target this project added one cycle earlier, and the incentive to explain such a result away will be real. The decision rule is written before the experiment for that reason and should not be renegotiated after it.
- Even the best outcome — KDM4C-specific rescue — leaves the target with no selective compound, since the JmjC pocket is the family's most conserved region. A positive result buys a chemistry problem, not a programme.
- Is SD70 potent enough on KDM4A and KDM4B for overexpression of either to be surmountable at achievable levels? If SD70 inhibits all three roughly equally, overexpression may not rescue even when the family is the target, and the experiment would need a resistant-allele design instead.
- Should the rescue be run at a fixed SD70 concentration with a competition readout rather than as a GI50 shift? A GI50 shift on a slow-growing B-ALL line over 7 days is a noisy endpoint.
- If the family is the target, is there any exposure schedule that separates the leukemia from the stem cell, and what assay short of competitive repopulation would show it?
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
SD70 at 0.8 uM for human cell lines preferentially suppressed leukemia lines carrying MLL fusions, naming SEM and THP1, over non-MLL lines, with induced apoptosis, differentiation, cell-cycle arrest and increased H3K9me3; no paralogue-rescue experiment and no transcriptional readout in SEM is reported.
Targeting Aberrant Epigenetic Networks Mediated by PRMT1 and KDM4C in Acute Myeloid Leukemia · Cheung N, Fung TK, Zeisig BB, et al. · Cancer Cell 2016What this evidence cannot show
Full text read (PMC4712026); the 0.8 uM human-cell-line concentration is from the paper's SD70 treatment methods. The authors did perform a rescue, but the reciprocal one: shRNA-resistant human KDM4C rescued the transformation defect of Kdm4c shRNA in mouse MLL-AF9, MLL-GAS7 and MOZ-TIF2 cells. That shows the shRNA phenotype is on-target for KDM4C in mouse; it does not address whether KDM4A or KDM4B could substitute, and it is not the SD70 arm.
Evidence that argues against this (1)
- The rescue design assumes SD70's relevant target is KDM4, and the paper it comes from does not establish that. The work is framed around an aberrant epigenetic network mediated by PRMT1 AND KDM4C jointly, so selective killing of MLL-fusion lines is not evidence that the KDM4 arm is the operative one. If SD70's leukaemia effect runs partly through PRMT1, a KDM4 paralogue add-back can fail to rescue for a reason that has nothing to do with paralogue redundancy, and the experiment will read as a clean negative. Cancer Cell 2016
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
The therapeutic window for a KDM4 family inhibitor is bounded by the family's requirement in normal haematopoiesis: conditional KDM4A/B/C triple-knockout mice lose HSC maintenance, with H3K9me3 accumulating at transcription start sites.
The KDM4/JMJD2 histone demethylases are required for hematopoietic stem cell maintenance · Agger K, Nishimura K, Miyagi S, Messling JE, Rasmussen KD, Helin K · Blood 2019What this evidence cannot show
Abstract read, not full text. Mouse, and complete genetic removal of all three paralogues rather than partial pharmacological inhibition, so it bounds the worst case rather than predicting a clinical window. The paper is explicitly framed as probing side effects of continuous inhibition of enzymes proposed as AML targets, which is the use made of it here.
Evidence that argues against this (2)
- A requirement in mouse HSC maintenance is weak evidence about the human therapeutic window, and this field has the counterexample in hand. DOT1L is likewise required in normal murine haematopoiesis, yet in the pinometostat phase 1 the maximum tolerated dose was NOT reached across 51 adults, with H3K79 methylation demonstrably reduced. The drug failed on efficacy, not on the window that mouse biology predicted would close. Treating the KDM4 triple-knockout phenotype as a bound on the window may discard a tractable target for the wrong reason. Blood 2018
- It cuts the other way too, and the entry should not take comfort from the pinometostat precedent. Pinometostat's tolerability came with only 2 complete remissions in 51 patients - a wide window around a drug that did not work. A KDM4 inhibitor could inherit exactly that outcome: tolerable, on-target by mark, and clinically inert. The paralogue-rescue experiment tests which enzyme the drug needs, not whether hitting it helps a patient. Blood 2018
- E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
A catalytically dead control is essential because the precedent exists in this gene family: in CML, re-introducing either wild-type OR demethylase-dead KDM6A restored imatinib resistance, so a histone demethylase's relevant function in leukemia can be entirely non-catalytic.
KDM6A promotes imatinib resistance through YY1-mediated transcriptional upregulation of TRKA independently of its demethylase activity in chronic myelogenous leukemia · See PubMed record PMID 33456567 for the full author list (not transcribed here) · Theranostics 2021What this evidence cannot show
Abstract read, not full text. KDM6 not KDM4, and CML not KMT2A-rearranged leukemia, so this is a methodological precedent for including the control rather than a prediction about KDM4C. It is a strong precedent precisely because nobody expected it there either.
Evidence that argues against this (1)
- The catalytically dead control is necessary but insufficient here. In the KDM6A precedent both wild-type and demethylase-dead protein restored the phenotype, which is informative because the readout was a single clean phenotype. In a three-paralogue add-back under drug, a dead-mutant rescue is ambiguous between a genuine non-catalytic scaffolding role and simple titration of the inhibitor by excess target protein. Distinguishing those needs an inhibitor-binding-deficient mutant as well, which this design does not specify. Blood 2019
- E4established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database
SEM and RS4;11 are KMT2A::AFF1 B-ALL lines without FLT3-ITD, whereas MV4-11 carries both KMT2A-AFF1 and FLT3-ITD, which is why the first two are specified as the primary models here.
Cellosaurus entries CVCL_0095 (SEM) and CVCL_0064 (MV4-11) · Bairoch A; SIB Swiss Institute of Bioinformatics · Cellosaurus 2026What this evidence cannot show
SEM and MV4-11 were fetched and read; RS4;11's FLT3 status is taken from general knowledge of the line and from its appearance in this graph's DepMap panels rather than from a fetched Cellosaurus record, so it should be confirmed before the experiment is run. If RS4;11 turns out not to be suitable, SEMK2 is the alternative second KMT2A::AFF1 line.
Evidence that argues against this (1)
- The experiment's whole design rests on line identity, and identity as recorded in a provenance database is not the property that matters. Ben-David et al. characterised 27 strains of one commonly used cell line and multiple strains of 13 others, found extensive clonal diversity and rapid genetic diversification driven by culture-condition-sensitive positive selection, and showed that at least 75% of 321 anti-cancer compounds that strongly inhibited some strains were completely inactive in others. SEM and RS4;11 are the right fusion on paper; whether this laboratory's SEM behaves like the SEM in which SD70 sensitivity was reported is an empirical question that a Cellosaurus record cannot answer. The experiment should include the published SD70 sensitivity as a reproduction arm before the paralogue add-back is interpreted, and should record strain provenance and passage. Ben-David U et al. 2018
the paralogue whose re-expression that experiment would test for rescue of SD70 sensitivity
the double knockdown here cannot separate KDM4B from KDM4C and leaves KDM4A intact, which is the same confound that experiment is designed to resolve
the decisive test: paralogue rescue of SD70 sensitivity separates family-level from gene-level dependency
the gate KDM4C must clear before anything is spent on it; a null result strikes the entry
run in SEM and RS4;11, KMT2A::AFF1 lines without the FLT3-ITD confounder
- analogous to →KDM6A / UTX — a commonly asserted KMT2A partner that belongs to a different complex and points the opposite way
the catalytically dead control exists because demethylase-dead KDM6A restored imatinib resistance in CML
the register records this as the gate KDM4C must clear first