Mediator's KMT2A signal is in its pan-essential core, CDK8 is expressed and not required at all, and MED13L is rank 9 while its paralogue MED13 is rank 13,569
also called mediator module split · MED13L paralogue · CDK8 negative
A large protein machine scored highly for this leukaemia, so we asked which part of it mattered. Almost all of it is needed by every cell, which makes it a poor target - but one swappable piece stands out, and its near-twin does nothing.
The same thing, for a scientist
The 719-complex sweep put core Mediator 11th, the one large-complex hit not exposed to the small-complex objection, and the entry recording it admitted it got no attention because it fit no story. Scored subunit by subunit: **the signal is in the core** (median rank 445 against the kinase module's 4,346) **but 25 of 26 core subunits are broadly essential** (median pan-essentiality -0.908), so the complex-level rank is largely pan-essentiality. **CDK8 is rank 11,802 and expressed at 7.6 nTPM** - present and not required, a measured negative against CDK8 inhibition here. The exception is **MED13L: rank 9 of 17,728, clears both a dependency and a pan-essentiality bar, and its paralogue MED13 sits at rank 13,569 while expressed at 12.6 nTPM in every panel line** - a 13,560-rank split with no expression explanation, the largest in this project.
Which half of Mediator, and the answer is the unhelpful one
Membership taken from the Complex Portal rather than written from recall: CPX-3227 for the 26-subunit core, CPX-3232 and CPX-3263 for the two kinase-module variants.
| group | n | median rank | best | median pan-essentiality | broadly essential |
|---|---|---|---|---|---|
| core | 26 | 445 | 16 (MED23) | -0.908 | 25 of 26 |
| CDK8 kinase module | 5 | 4,346 | 89 (CCNC) | -0.075 | 2 of 5 |
The signal is in the core, and the core is scaffolding every cell needs. A median pan-essentiality of -0.908 means these subunits kill most lines when removed; a somewhat larger effect in KMT2A-rearranged lines is close to uninformative. That is the outcome the script named in advance as the one that would make the complex-level rank a poor lead, and it is what happened.
The druggable half is the half that does nothing. CDK8 inhibitors exist and the kinase module is the part of Mediator that docks reversibly and is dispensable in many cells - so a signal there would have been target-shaped. Instead CDK8 ranks 11,802 with a mean effect of -0.019, CDK19 ranks 4,346, MED13 ranks 13,569. Only CCNC (89) and MED12 (131) score, and both are broadly essential.
MED13L, and why this one survives every control I have
| gene | rank | mean in KMT2A-r | pan-essentiality | median nTPM | verdict |
|---|---|---|---|---|---|
| MED13L | 9 | -0.916 | -0.156 | 27.0 | clears both bars |
| MED13 | 13,569 | -0.139 | -0.059 | 12.6 | expressed, not required |
| MED12 | 131 | -1.172 | -0.556 | 17.8 | pan-essential |
| MED12L | 2,376 | -0.019 | +0.051 | 0.10 | not expressed |
| CDK8 | 11,802 | -0.019 | -0.075 | 7.6 | expressed, not required |
| CDK19 | 4,346 | -0.019 | +0.020 | 17.2 | expressed, not required |
MED13L is the 9th most selective gene in a 17,728-gene scan, is not broadly essential, and is a real dependency in rearranged lines. It is one of only 34 genes in the whole scan that clear both bars. Its paralogue MED13 is transcribed in every panel line at 12.6 nTPM and ranks 13,569 - a spread of 13,560 ranks, larger than JADE2/JADE3 (11,597) and the largest this project has measured.
This analysis contains its own positive and negative control, which is why I believe it. The same table shows MED12L at 0.10 nTPM - not expressed - so the MED12/MED12L spread of 2,245 ranks is an artefact of exactly the kind that destroyed the ALAS1/ALAS2 control earlier in this cycle. One paralogue pair in this complex is real and one is not, distinguished by a measurement rather than by which one I preferred.
MED16 also clears both bars at rank 229, pan-essentiality -0.166. It is the only core subunit that does.
The literature check, and the reason this cannot be treated as a target
This entry's first open question asked whether MED13L is connected to KMT2A-rearranged leukaemia anywhere in the literature. Run 2026-09-20, with every query translation read:
| query | records |
|---|---|
MED13L | 114 |
MED13L AND leukemia | 4 |
MED13L AND (KMT2A OR MLL) | 7 |
MED13L AND cancer | 26 |
MED13L haploinsufficiency syndrome | 19 |
Not one of the 4 or the 7 is about MED13L in KMT2A-rearranged leukaemia. The leukaemia hits are an ETV6::RUNX1 optical-genome-mapping paper, a congenital-heart-disease variant study, a Cornelia de Lange series and the MED13L syndrome delineation itself. The KMT2A hits are Cornelia de Lange, an ASD neural-development platform, an exome pool-seq study and a syndromic intellectual disability panel. The connection does not exist in the literature at all.
That cuts both ways and the second way is more important than the first. It means rank 9 is genuinely unexplored — and it means what MED13L is actually known for is a developmental syndrome. Heterozygous loss of function causes MED13L haploinsufficiency syndrome: intellectual disability, congenital heart defects, facial dysmorphism. One functional copy is not enough in a developing human.
The patients in this disease are infants. A dependency whose germline heterozygous loss produces intellectual disability and heart defects is not a target to pursue casually in a population under one year old, whatever its CRISPR rank. That is not a reason to drop the finding — the biology may still explain something about how the fusion works — but it is a reason the finding should never be written up as a therapeutic lead without this sentence attached.
- Is the MED13L requirement for the protein or for the CDK8 module it nucleates? MED13 is expressed and dispensable while MED13L is required, so the two modules are not interchangeable in these cells, and that difference is the mechanistic question worth asking before any therapeutic one.
- Does a dual CDK8/CDK19 knockout or a dual inhibitor do anything in KMT2A-rearranged lines? Single knockout of either is uninformative because they are expressed paralogues.
- MED13 and MED13L differ in the CDK8 module they nucleate. Does the MED13L-containing module have a distinct composition or target set in these cells, and is that where the requirement comes from?
- Separate the core into head, middle and tail modules and re-score. A median over 26 subunits may be hiding a submodule the way the HBO1 median hid ING4.
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.analysis
Across Mediator, KMT2A-selectivity sits in the core rather than the CDK8 kinase module - core median rank 445 of 17,728 against the module's 4,346 - but 25 of the 26 core subunits are broadly essential, with a median pan-essentiality of -0.908.
Core Mediator against its CDK8 kinase module, subunit by subunit, with pan-essentiality and expression controls · Oncofusion knowledge graph (internal analysis) 2026What this evidence cannot show
Computed 2026-09-20 from the cached genome-wide scan; membership from the Complex Portal. Every rank rests on **11 KMT2A-rearranged lines that are nine patients, seven myeloid, none infant-derived**. A median over 26 core subunits mixes head, middle and tail modules that this analysis does not separate, and Complex Portal's core record excludes subunits some other curations include. **Pan-essentiality does not disqualify a target** - it changes the question to therapeutic index - so 'uninformative' here means uninformative as a *selectivity* signal.
Evidence that argues against this (1)
- The core-versus-module split may be an artefact of what CRISPR can measure. The kinase module docks reversibly and its loss is often compensated by the core, so a near-zero gene effect for CDK8 is consistent with a module that matters but is buffered - which single-gene knockout cannot see. The negative below is therefore about CDK8 the gene, not about CDK8 kinase activity. Oncofusion knowledge graph (internal analysis) 2026
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.analysis
MED13L ranks 9 of 17,728 for KMT2A-selectivity, is a real dependency in rearranged lines (mean -0.916) and is not broadly essential (pan-essentiality -0.156), making it one of 34 genes clearing both bars. Its paralogue MED13 ranks 13,569 while being expressed at a median 12.6 nTPM across all seven panel lines held by the Human Protein Atlas.
Core Mediator against its CDK8 kinase module, subunit by subunit, with pan-essentiality and expression controls · Oncofusion knowledge graph (internal analysis) 2026What this evidence cannot show
The expression check covers **7 of the 9 patients** in the panel and uses HPA's own cultures rather than the DepMap isolates, so it is a different laboratory's stock of a line with the same name. **RNA is not protein.** MED13L is expressed about twice as highly as MED13 (27.0 against 12.6 nTPM), so a dosage explanation is weakened but not excluded - unlike ING4, which is transcribed *more* than its selective sibling and is still dispensable.
Evidence that argues against this (1)
- MED13L has no published connection to KMT2A-rearranged leukaemia that this analysis has checked, and a rank of 9 in a scan built on eleven lines is exactly the kind of result that has dissolved before in this project - the acetyltransferase block, the coherence mechanism and the ALAS2 control all looked strong and did not survive their controls. **This has survived the controls available from cached data; it has not survived a literature check or an experiment.** Oncofusion knowledge graph (internal analysis) 2026
- E3moderate confidence 3 of 5: One strong study, or several smaller consistent ones.database
CDK8 is expressed at a median 7.6 nTPM across the panel lines and ranks 11,802 of 17,728 with a mean gene effect of -0.019 in KMT2A-rearranged lines. CDK19 is expressed at 17.2 nTPM and ranks 4,346. Neither is a dependency.
Human Protein Atlas, RNA expression by cell line (rna_celline.tsv, nTPM, 1,206 cell lines) · Human Protein Atlas 2026What this evidence cannot show
This is a CRISPR knockout result, not a pharmacology result. CDK8 and CDK19 are paralogues that can substitute for each other, so single-gene knockout of either is the weakest possible test of the module - **a dual CDK8/19 inhibitor is not predicted to be inactive by these data**, and no double-knockout was available here. The eleven-line panel limitation applies.
Evidence that argues against this (1)
- CDK8 and CDK19 buffering each other is the obvious reading and it is not excluded: both are expressed, both score near zero, and that is precisely the pattern reciprocal buffering produces. The honest claim is that **neither gene alone is required**, which is weaker than 'the kinase module does not matter'. Oncofusion knowledge graph (internal analysis) 2026
- E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.analysis
MED12L is not expressed in the panel lines (median 0.10 nTPM), so the 2,245-rank spread between MED12 and MED12L is an expression artefact rather than a functional paralogue split - in the same complex where the MED13L/MED13 spread is not.
Expression control for the paralogue splits and the ALAS1/ALAS2 comparison, against Human Protein Atlas cell-line RNA · Oncofusion knowledge graph (internal analysis) 2026What this evidence cannot show
0.10 nTPM is HPA's floor for these lines rather than a measured trace, so the accurate statement is 'at or below detection'. The point of recording it is methodological: two paralogue pairs in one complex, one real and one artefactual, separated by a measurement rather than by preference.
Evidence that argues against this (1)
- Having a negative control land correctly in the same analysis is reassuring but it is not independent evidence that the positive is real - both rest on the same HPA table and the same seven lines. If HPA's line identities or its nTPM normalisation were wrong, both results would move together. EMBL-EBI Complex Portal 2026
- E5strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database
MED13L has no published connection to KMT2A-rearranged leukaemia. `MED13L AND leukemia` returns 4 records and `MED13L AND (KMT2A OR MLL)` returns 7, and none of either set concerns MED13L in KMT2A-rearranged disease; they are developmental and syndromic studies. What MED13L is known for is haploinsufficiency syndrome - heterozygous loss of function causing intellectual disability and congenital heart defects.
PubMed enumeration of MED13L against leukaemia and KMT2A, run 2026-09-20 with query translations read · NCBI E-utilities 2026What this evidence cannot show
Searched 2026-09-20 with query translations read for each term, which is how this project now guards against a parser-rewritten zero. **A PubMed enumeration is not a literature review**: MED13L could appear in a supplementary table or a screen hit list without being indexed to these terms, and this project has already found one case where a gene appeared only in a deposition and not in a paper's body text. Absence here means absence of an indexed, titled connection.
Evidence that argues against this (1)
- An absent literature is weak evidence either way. It is consistent with MED13L being a real and overlooked dependency, and equally consistent with rank 9 being the kind of artefact an eleven-line panel produces - and this project has watched three strong-looking results dissolve under their controls in two cycles. The paralogue and expression controls it has passed are real but they are all computational and all from one dataset. NCBI E-utilities 2026
follows up the one large-complex hit that entry flagged as unexamined; core Mediator's 11th-of-719 rank is largely pan-essentiality
a second real paralogue split (MED13L/MED13) and a second artefactual one (MED12/MED12L) in a single complex
MED12 ranks 131 but is broadly essential (pan -0.556); MED12L is not expressed
MED23 is the best-ranked core subunit at 16, and is broadly essential
MED26 ranks 555 and is broadly essential (pan -0.944)
the systematic subunit scan behind the single-subunit verdict reached earlier
CDK8 and CDK19 are both expressed and both score near zero, which is what reciprocal buffering looks like
every rank here comes from that panel
MED19 is one of the 25 of 26 broadly essential core subunits