Where else in a child is the target switched on? (run)
also called HPA normal-tissue filter · tissue exposure ratio · normal-tissue safety axis
A drug reaches the whole body, not just the cancer. This asks, for every protein on our target list, how much of it normal organs make compared with bone marrow, which is the tissue we are actually trying to treat. Two candidates turn out to be made far more in the brain and the pancreas than in marrow.
The same thing, for a scientist
🟠 **RE-EXAMINED 2026-09-24 (cycle 137): the numbers below reproduce, and nine of them are inflated by lymphocytes - but none of the ones this entry leads with.** A re-run on the identical panel reproduced **13 of 13** of the ratios below within tolerance. It then recomputed the maximum with **spleen, lymph node and thymus excluded**, because a high value there can be the lymphocytes inside the organ rather than the organ itself (see `concept-the-leukocyte-artefact-inflates-a-haematopoietic-antigens-tissue-ratio-it-does-not-deflate-it`). **Unchanged, because their top tissue is parenchymal:** MEIS1 **17.21** (pancreas), MEF2C **11.57** (cortex), MLLT1 **6.56** (cortex), MEN1 **3.26** (liver), DPY30 2.67, DOT1L 2.38. **Every headline in this entry stands.** **Inflated, and the corrected figures are:** **PSIP1 2.07 -> 1.62**, RBBP5 1.38 -> 1.30, **KMT2A 0.97 -> 0.92**, BOD1L1 0.98 -> 0.92, **KDM6A 0.79 -> 0.56**, **KDM4C 0.77 -> 0.59**, CCNT1 0.44 -> 0.40. 🔴 **And the control moved: ACTB 1.21 -> 0.94**, because its top tissue was lymph node. This entry's `controls` field uses ACTB to say what an unremarkable ratio looks like; corrected, an unremarkable gene is **below 1.0**, so the band this table is read against sits lower than it states. **The list of most-marrow-selective genes below is therefore conservative rather than wrong.** --- Human Protein Atlas consensus tissue RNA for 22 candidates plus two controls, scored as the ratio of the highest-expressing normal tissue to bone marrow. MEIS1 is 17.2x higher in pancreas and MEF2C 11.6x higher in cerebral cortex; MLLT1 is 6.6x higher in cerebral cortex. MEN1 itself is 2.6x higher in pancreas than in marrow and equal in parathyroid, which puts a number on the endocrine liability the graph already records. CCNT1, MLLT3, KDM4C, KDM6A, KAT7 and KMT2A are the most marrow-selective.
The axis this project was missing
Target safety here has been assessed two ways: gnomAD constraint (how intolerant the human population is to losing a gene) and DepMap pan-essentiality (how many cell lines die without it). Neither asks what a clinician asks first — where in a normal child is this protein present, and what are those tissues for?
That question acquired a specific edge this month. The graph now carries a risk entry arguing that menin inhibitors occupy the pocket binding both MLL1 and JUND, and that menin-JUND repression is the tumour-suppressor function whose loss causes MEN1 syndrome in parathyroid, pancreatic islet and pituitary. Whether MEN1 is actually present in those tissues, relative to the marrow we intend to treat, is a checkable fact rather than an argument 1.
The sixth identifier trap, and the guard against it
The HPA search endpoint does substring matching. A query for MEN1 returns KREMEN1 alongside it, with entirely different tissue values.
This is the sixth identifier trap this project has hit, after the catfish paper, the porcine bone sialoprotein paper, the Toxoplasma paper, the telomerase paper, the nucleosome paper standing in for an HIV-integration paper, and gene_exact:TCF3 resolving to TCF7L1. Every row here is therefore filtered on an exact gene-symbol match and then confirmed against the Ensembl id we asked for, in the same spirit as scripts/verify-accessions.py. All 24 genes resolved with zero mismatches.
The result3 deeper
nTPM, consensus normal tissue. The ratio is the highest-expressing normal tissue divided by bone marrow: higher means the drug is acting more on something we are not treating.
| Gene | bone marrow | highest normal tissue | ratio | parathyroid | pituitary | pancreas |
|---|---|---|---|---|---|---|
| MEIS1 | 2.4 | pancreas 41.3 | 17.2 | 4.2 | 8.1 | 41.3 |
| MEF2C | 10.5 | cerebral cortex 121.5 | 11.6 | 5.6 | 8.8 | 1.6 |
| MLLT1 | 8.2 | cerebral cortex 53.8 | 6.6 | 12.9 | 27.7 | 9.5 |
| MEN1 | 6.8 | liver 22.2 | 3.3 | 6.8 | 12.9 | 17.7 |
| DPY30 | 29.6 | liver 79.1 | 2.7 | — | — | 54.8 |
| DOT1L | 20.7 | testis 49.2 | 2.4 | — | — | 14.4 |
| PRMT1 | 61.3 | heart muscle 139.6 | 2.3 | — | — | 94.1 |
| PSIP1 (LEDGF) | 58.5 | thymus 120.9 | 2.1 | — | — | 48.8 |
| ELL2 | 45.6 | pancreas 89.0 | 1.9 | — | — | 89.0 |
| CDK9 | 44.3 | ovary 85.2 | 1.9 | — | — | 64.7 |
| CTR9 | 22.4 | parathyroid 28.3 | 1.3 | 28.3 | — | — |
| KMT2A | 16.0 | thymus 15.5 | 1.0 | — | — | 10.1 |
| KAT7 | 48.3 | testis 44.7 | 0.9 | — | — | 35.9 |
| KDM6A / KDM4C | 17.9 / 28.4 | thymus | 0.8 / 0.8 | — | — | — |
| MLLT3 | 35.1 | thyroid 25.5 | 0.7 | — | — | 11.9 |
| CCNT1 | 27.8 | thymus 12.1 | 0.4 | — | — | 10.8 |
| ACTB (control) | 3397.8 | lymph node 4107.2 | 1.2 | — | — | — |
| ALB (control) | 0.0 | liver 198,523.8 | undefined | — | — | — |
The controls behave. ACTB, a housekeeping gene, sits at 1.2 — present everywhere in similar amounts. ALB is absent from marrow entirely (0.0) and astronomically high in liver, so its ratio is undefined by division by zero, which is exactly what a perfectly tissue-restricted gene should produce.
MEN1 is at least as abundant in the syndrome tissues as in the marrow we are treating
Pancreas 17.7, pituitary 12.9, parathyroid 6.8, against bone marrow 6.8. So the target of the one approved drug in this disease is 2.6x more abundant in pancreas and equally abundant in parathyroid compared with the tissue the drug is meant to act on.
This does not show harm. It removes one possible reassurance — that the endocrine tissues might simply not express much menin — and replaces it with a number. The risk entry can now say the exposure is at least proportionate rather than resting on the pocket argument alone.
MEF2C and the brain: a paediatric flag that supports the redefinition already made
MEF2C is 11.6x higher in cerebral cortex than in bone marrow, and MEF2C haploinsufficiency is the accepted cause of the 5q14.3 deletion syndrome phenotype — intellectual disability, epilepsy, hypotonia, brain malformations and facial dysmorphism 2. For a drug given to infants, reducing MEF2C dosage would be a poor idea on this evidence alone.
This independently supports the redefinition made in the cycle-23 target audit. That audit concluded, from mouse genetics, that the handle is the upstream MARK/SIK kinase acting on the S222 phospho-state, not MEF2C abundance — a phospho-dead knock-in has normal haematopoiesis yet resists leukemogenesis. Two unrelated lines of evidence now say the same thing: do not lower MEF2C, change its modification.
MLLT1 (ENL) carries a weaker version of the same flag at 6.6x cerebral cortex.
CCNT1 is the most marrow-selective candidate in the set, which sharpens an earlier doubt
CCNT1 has the lowest ratio of any candidate at 0.4 — more abundant in marrow than in any normal tissue tested. It was struck in the cycle-23 rewrite, and that strike has already been questioned once: the alvocidib mortality argument turned out not to support it, since that arm had the best response of three and the authors recommended further development.
So CCNT1 was struck on a reason that failed, and now scores best in the graph on an axis nobody had measured. That does not make it a good target — its strike rests on co-dependency, which this analysis does not address, and CDK9/P-TEFb inhibition has a real toxicity history. But the reopening condition recorded against the CCNT1 strike deserves an owner.
What this cannot do
mRNA is not protein. HPA consensus values are transcript abundance. Protein level, turnover and post-translational state can all diverge.
Expression is not dependence. A tissue can express a gene abundantly and be unharmed by inhibiting it, and can express it sparsely and be exquisitely sensitive. This ranks where to look for toxicity. It does not predict toxicity, and treating a low ratio as a safety clearance would be a serious misreading.
These are adult tissues. The population this project exists to serve is infants, and developmental expression can differ sharply from adult. MEF2C is the obvious case: a gene whose haploinsufficiency causes a neurodevelopmental syndrome is one whose developmental expression matters more than the adult value measured here.
A single tissue panel. Sixteen tissues, chosen to include the MEN1-syndrome organs and the usual sites of dose-limiting toxicity. A gene could be concentrated somewhere not on the list.
| script | analyses/hpa-normal-tissue-target-filter/run.py |
|---|---|
| status | run |
| run date | 2026-09-18 |
| source | Human Protein Atlas consensus tissue RNA (nTPM) |
| n genes | 24 |
| n tissues | 16 |
| n identifier problems | 0 |
| identifier guard | exact gene symbol plus Ensembl id confirmation; HPA search is substring-based (MEN1 returns KREMEN1) |
| figure | analyses/hpa-normal-tissue-target-filter/results.md |
men1 ntpm — the numbers behind the chart
{
"bone marrow": 6.8,
"parathyroid gland": 6.8,
"pituitary gland": 12.9,
"pancreas": 17.7,
"liver": 22.2
}worst ratios — the numbers behind the chart
{
"MEIS1": 17.2,
"MEF2C": 11.6,
"MLLT1": 6.6,
"MEN1": 3.3,
"DPY30": 2.7,
"DOT1L": 2.4
}most marrow selective — the numbers behind the chart
{
"CCNT1": 0.4,
"MLLT3": 0.7,
"KDM6A": 0.8,
"KDM4C": 0.8,
"KAT7": 0.9,
"KMT2A": 1
}controls
{
"ACTB": 1.2,
"ALB": "undefined; bone marrow 0.0, liver 198523.8"
}parenchyma only cycle 137
{
"PSIP1": 1.62,
"RBBP5": 1.3,
"KMT2A": 0.92,
"BOD1L1": 0.92,
"KDM6A": 0.56,
"KDM4C": 0.59,
"CCNT1": 0.4,
"ACTB": 0.94,
"unchanged": {
"MEIS1": 17.21,
"MEF2C": 11.57,
"MLLT1": 6.56,
"MEN1": 3.26
},
"note": "maximum recomputed with spleen, lymph node and thymus excluded; 13 of 13 published ratios reproduced first. The ACTB control also moved, so the reference band is lower than this entry states."
}- The most dangerous use of this table would be to treat a low ratio as a safety clearance. ACTB scores 1.2 and is among the most essential genes in the genome. The ratio says where a drug is distributed, not what breaks.
- MEN1 being abundant in the endocrine tissues does not demonstrate endocrine harm from menin inhibitors, and no such harm has been published. It removes one reassurance and supplies a number; it is a reason to measure.
- Adult tissue standing in for infant tissue is the same substitution this project criticises in its DepMap work. For MEF2C specifically, where the concern is developmental, the adult value is the wrong measurement.
- HPA consensus values are adult. What does MEF2C expression look like in infant and fetal brain, where the 5q14.3 phenotype originates, and does the 11.6x ratio understate the concern?
- MEIS1 at 17.2x pancreas is the worst ratio in the set and its statistical case has flipped three times. Is there any remaining reason to keep it on the list?
- Does protein-level HPA data (immunohistochemistry) agree with the transcript ranking, particularly for MEN1 in parathyroid and pituitary?
- CCNT1 scores best on this axis and was struck on a reason that failed. Its reopening condition has no owner. Who checks it?
- Should this ratio enter the phase-2 go/no-go register as a gate, or only as context? Given that ACTB scores 1.2, it cannot be a gate on its own.
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset
In Human Protein Atlas consensus tissue RNA, MEN1 is expressed at 6.8 nTPM in bone marrow, 6.8 in parathyroid gland, 12.9 in pituitary gland and 17.7 in pancreas. Across 22 candidate targets scored as highest normal tissue divided by bone marrow, MEIS1 is highest at 17.2 (pancreas 41.3 versus marrow 2.4), followed by MEF2C at 11.6 (cerebral cortex 121.5 versus 10.5) and MLLT1 at 6.6 (cerebral cortex 53.8 versus 8.2); the most marrow-selective are CCNT1 (0.4), MLLT3 (0.7), KDM6A and KDM4C (0.8), KAT7 (0.9) and KMT2A (1.0). Controls behave: ACTB 1.2, and ALB is absent from marrow (0.0) with liver at 198,524.
Human Protein Atlas consensus tissue RNA expression (nTPM) via the HPA search_download API; analysis in analyses/hpa-normal-tissue-target-filter/ · Human Protein Atlas; analysis run by this project · Database and derived analysis 2026What this evidence cannot show
mRNA is not protein and expression is not dependence; this ranks where to look for toxicity rather than predicting it. HPA consensus values are adult tissue, and the population of concern is infants, so developmental expression may differ - most importantly for MEF2C. Sixteen tissues only. The HPA search endpoint matches substrings, so a query for MEN1 also returns KREMEN1; every row was filtered on exact gene symbol and confirmed against the requested Ensembl id, and all 24 resolved cleanly.
Evidence that argues against this (1)
- A low ratio is not a safety clearance and a high one is not a toxicity prediction. The clearest counter-case in this table is its own control: ACTB scores a benign-looking 1.2 while being one of the most essential genes in the genome. Ranking by tissue ratio would place it among the safest candidates, which is obviously wrong, and the same reasoning error would apply to any gene that is uniformly expressed and uniformly required. Human Protein Atlas; analysis run by this project 2026
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
5q14.3 deletion syndrome (MIM#613443) is characterised by intellectual disability, epilepsy, hypotonia, brain malformations and facial dysmorphism, and haploinsufficiency of MEF2C is considered responsible for the distinctive phenotype, although cases without MEF2C involvement have been documented.
Cleft Lip Palate in a Patient with 5q14.3 Deletion Syndrome: A Possible Unreported Feature? · Fernández Hernández L, Alcántara Ortigoza MA, Ramos Angeles SE, González-Del Angel A · Cytogenetic and Genome Research 161:556-563 2022What this evidence cannot show
Read as abstract only; this is a case report whose framing of the syndrome is background rather than its own finding. It states that at least six reported cases lack MEF2C involvement, so MEF2C haploinsufficiency is the accepted but not exclusive cause. Germline haploinsufficiency through development is also a very different exposure from pharmacological inhibition after birth.
Evidence that argues against this (1)
- The same report notes at least six documented 5q14.3 cases that do not involve MEF2C yet share the clinical features, so attributing the phenotype solely to MEF2C haploinsufficiency overstates what the genetics shows. The neurodevelopmental concern for a MEF2C-lowering drug is real but should not be stated as though MEF2C dosage alone determined the syndrome. Fernández Hernández L et al. 2022
second use of Human Protein Atlas expression as an outside control in this project
extends that filter to the surface antigens it omitted, on the identical panel and metric
ranking by ratio alone mixes parenchymal liability with tissue cellular composition
reproduces 13 of 13 of its ratios and supplies parenchyma-only figures for the nine lymphoid-topped genes
- supports →The same pocket suppresses JUND: menin inhibition's unmeasured endocrine liability in infants
puts a number on the exposure: MEN1 is 2.6x more abundant in pancreas and equal in parathyroid compared with bone marrow
a third safety axis beside constraint and pan-essentiality: where in a normal child the target is present
11.6x higher in cerebral cortex than marrow, and MEF2C haploinsufficiency causes the 5q14.3 neurodevelopmental syndrome
worst tissue ratio in the set at 17.2x pancreas, on top of a statistical case that has flipped three times
most marrow-selective candidate at 0.4, which sharpens the doubt about a strike whose stated reason failed
candidate context for the register, but not a gate: ACTB scores a benign 1.2 while being pan-essential
normal-tissue distribution of the target of the one approved drug in this disease
- supports →Dataset: Human Protein Atlas - the protein-level and single-cell check that GTEx cannot give
executes the access route that dataset audit described, and records the substring-matching trap it did not
the MEF2C cortical-expression flag bears directly on a SIK2/3 programme aimed at MEF2C-high disease