Where else in the body each target is expressed, and the two controls that say the method works
also called normal tissue exposure · GTEx target filter · off-tumour expression
Every drug target here was picked for what it does in leukaemia cells, but none of them exists only there. This checks all 54 normal human tissues for each one, so a child cured at one is not injured at another.
The same thing, for a scientist
GTEx v8 median TPM across 54 tissues for every gene this graph names as a target, **47 of 98 symbols resolved**. The result is validated by two controls pointing opposite ways: **CD19**, the one target with proven tolerable toxicity in infants, comes out at median **0.26 TPM** with brain 0.19 and heart 0.14 against 166 in blood; and **MCL1**, whose cardiac liability this graph already records from the clinical literature, comes out at **98 TPM in heart**. The analysis recovers a known-safe and a known-toxic profile without being told either.
The controls
| gene | brain | heart | blood | liver/kidney | endocrine | peak tissue |
|---|---|---|---|---|---|---|
| CD19 | 0.2 | 0.1 | 166.2 | 0.4 | 12.6 | Spleen (166) |
| PLK1 | 11.2 | 0.8 | 11.3 | 2.1 | 74.8 | Cells_EBV-transformed_lymphocytes (103) |
| DOT1L | 37.7 | 6.3 | 25.4 | 17.4 | 111.7 | Testis (112) |
| MEN1 | 45.1 | 7.2 | 35.0 | 16.6 | 43.2 | Brain_Cerebellum (45) |
| MLLT1 | 141.7 | 33.5 | 50.5 | 44.5 | 107.8 | Brain_Cerebellum (142) |
| MAT2A | 194.6 | 88.1 | 117.0 | 89.8 | 474.4 | Thyroid (474) |
| MCL1 | 75.5 | 98.0 | 376.6 | 123.7 | 218.6 | Lung (475) |
CD19 is the positive control for tolerability: median 0.26 TPM across all 54 tissues, essentially absent from brain, heart, liver and kidney, present only in blood and spleen. That is what a target behind a well-tolerated infant therapy looks like, and the method finds it unprompted.
MCL1 is the control in the other direction. The graph already holds, from clinical reports, that MCL1 inhibitors produced cardiac toxicity signals that interrupted development of more than one compound. GTEx independently puts MCL1 at 98 TPM in heart. An expression table recovering a known clinical liability is the strongest evidence available here that the ranking means something.
What it says about targets in play
PLK1 is better placed than a mitotic kinase has any right to be. Median 2.85 TPM, heart 0.82, brain 11.2, peaking in EBV-transformed lymphocytes. The Q17 panel flagged it on a 99.9% depletion rate and the standing worry about a mitotic kinase is every dividing tissue; GTEx says the resting body barely expresses it. The toxicity to expect is proliferative, which is also the kind that recovers.
MLLT1/ENL peaks in cerebellum at 142 TPM, against a whole-body median of 52. This graph holds a phase-2 ENL YEATS dossier. The cerebellum is the brain region that grows most in the first two years of life and this is an infant disease. Expression is not dependency and nothing here shows harm — but of every target in the set, this is the profile that sits most awkwardly against the developmental stage of the population.
MAT2A is broadly high (median 173, brain 195, endocrine 474), which is the expected cost of an upstream metabolic lever.
MEN1 is unremarkable and that is the useful part: median 20, cerebellum 45, heart 7. Menin sits above its own whole-body median in the cerebellum and menin inhibitors are approved and tolerated. High expression is not a veto, and any use of this table that treats it as one is misusing it.
- Do not use this table to drop a target. mRNA abundance is not dependency, and the one target here with proven infant tolerability (CD19) still causes neurotoxicity in CAR-T at a brain expression of 0.19 TPM.
- Do not use it to clear one either. MEN1 sits above its whole-body median in cerebellum and menin inhibitors are approved; PLK1's low resting expression says nothing about the proliferating tissues an infant depends on to grow.
- The brain rows are the least trustworthy in the table and the most consequential. GTEx is adult tissue; the cerebellum of a treated infant is still growing, and no row here measures that.
- Any go/no-go that cites this analysis must cite the developmental caveat with it, or it is quoting an adult atlas about a one-year-old.
- Fix the symbol extractor. 51 of 98 candidates were not genes and AURKA - a real target in this graph - was missed. The gene list should come from the entries' structured fields, not from regex over titles.
- Get a developmental atlas. GTEx cannot speak to infant cerebellum, which is where the ENL result would matter. BrainSpan or the Human Developmental Cell Atlas would be the instrument.
- Pair this with dependency rather than expression. The question is not where a target is present but where its loss is not tolerated; a DepMap-style normal-tissue dependency map does not exist, but tissue-restricted essentiality from knockout mouse phenotypes partly substitutes.
- Follow MLLT1 specifically. A 142 TPM cerebellar peak under an active phase-2 dossier, in a disease of the first year of life, is the single row in this table worth a literature search of its own.
- E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database
GTEx v8 median expression across 54 tissues resolves for 47 of 98 candidate target symbols drawn from this graph's target, drug and gene entries. CD19 has a median of 0.26 TPM with brain 0.19, heart 0.14 and liver/kidney 0.4 against 166 in whole blood; MCL1 has 98 TPM in heart; PLK1 has a median of 2.85 with heart 0.82; MLLT1 peaks at 142 TPM in cerebellum; MEN1 has a median of 20 with 45 in cerebellum.
GTEx v8 median gene expression (TPM) across 54 tissues, gtexportal.org API v2 · GTEx Consortium / Broad Institute 2026What this evidence cannot show
Queried 2026-09-19 through the GTEx API v2 with datasetId gtex_v8. The 51 unresolved symbols are overwhelmingly not genes — drug codes and acronyms scraped from entry titles (AZD4573, BMF219, CTL019, CRISPR) — so the resolution rate is a property of the symbol extractor, not of GTEx coverage; AURKA, which is a real gene in this graph, did not resolve and that is an extractor miss worth fixing. Cells_EBV-transformed_lymphocytes and Cells_Cultured_fibroblasts are cell lines rather than tissues and top the list for several genes.
Evidence that argues against this (2)
- **mRNA is not protein and neither is dependency.** GTEx cannot distinguish a tissue that expresses a transcript from one that needs it, and toxicity turns entirely on the second. A target can be abundant in a tissue that tolerates its loss completely, and scarce in one that does not. GTEx Consortium / Broad Institute 2026
- **GTEx is adult post-mortem tissue and this project treats infants.** A gene required during a developmental window and a gene merely present in adult tissue are different facts, and only the second is measured here. For every brain row in this analysis — which is exactly where the developmental concern lies — the instrument is close to silent, and an infant-tissue atlas would be needed to say anything real. Oncofusion knowledge graph (internal analysis) 2026
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.analysis
The analysis is validated by two controls pointing in opposite directions. CD19, the one target in this graph with demonstrated tolerable toxicity in infants, scores as the most tissue-restricted target in the set. MCL1, whose cardiac toxicity signals are already recorded here from the clinical literature, scores highest in heart of any target except NPM1. Neither outcome was supplied to the analysis.
Normal-tissue exposure of the graph's candidate targets, GTEx v8 across 54 tissues · Oncofusion knowledge graph (internal analysis) 2026What this evidence cannot show
Two controls is two, and both were chosen after the results were seen rather than pre-specified, which is the weaker form of this argument. They are nonetheless the only two targets in the set whose clinical tolerability is independently known, so the choice was not wide.
Evidence that argues against this (2)
- Post-hoc controls chosen from the output are close to circular, and a set of 47 contains enough targets that some pair would have fitted some story. The honest version is that the two known answers happen to be right and no third known answer exists to test. Oncofusion knowledge graph (internal analysis) 2026
- CD19 may be the wrong kind of control entirely. Blinatumomab's tolerability is usually attributed to B-cell aplasia being survivable with immunoglobulin replacement rather than to CD19's tissue distribution, and CAR-T against CD19 does produce neurotoxicity despite brain expression of 0.19 TPM. A target's expression profile predicted neither of those. GTEx Consortium / Broad Institute 2026