DAXX and ATRX as drug targets: six ChEMBL rows and zero of them an affinity, against menin's 1,776 - but three crystal structures of the interface that would have to be blocked
also called is DAXX druggable · the DHB-ATRX interface
If a cell needs DAXX, the next question is whether anyone could make a drug against it. The honest answer is that nobody has tried: there is one chemistry record in the whole public database and it is not a binding measurement. But the shape of the join between DAXX and its partner has been solved to better than 2 angstroms, so the target is not structurally blind.
The same thing, for a scientist
**The chemistry does not exist.** ChEMBL holds one human DAXX target record (CHEMBL5724773) carrying **six activity rows, all of them `FC` (fold change) at 0.4-10 uM from a single document**, i.e. a proteomic or expression readout rather than an affinity [E1]. There is **no human ATRX target record at all**; the only ATRX record is murine (CHEMBL4879474) with two rows measuring reduction of phospho-sites in mouse C2C12 cells - an off-target observation inside a kinase-inhibitor phosphoproteomics study [E1]. For scale, menin carries **1,776** activity rows and the menin-MLL protein-protein interaction **48,291** [E1]. **Per this project's own rule about ChEMBL counts, that is a measure of how much medicinal chemistry has been DONE, not of tractability.** Zero rows means nobody tried. The correct statement is not "DAXX is undruggable" but "DAXX is unattempted", and the two are routinely confused. **The structural picture is better than the chemical one.** RCSB holds 14 human DAXX entries and 13 human ATRX entries, and among them are three crystal structures of the **DAXX-ATRX interface itself**: 5GRQ (DAXX DHB domain with an ATRX peptide, **1.58 A**), 5Y18 (DAXX helical bundle with ATRX, 2.20 A) and 5Y6O (DAXX N-terminal four-helix bundle with ATRX, 3.1 A) [E2], plus the DAXX/H3.3-H4 chaperone complexes 4H9N (1.95 A) and 4HGA (2.80 A) [E3]. **A peptide-in-groove interface solved at 1.58 A is the same structural class as the menin-MLL interface this project's whole drug programme is built on.** **And there is a specific selectivity problem that this graph already half-knows.** `protein-huca-complex` records that UBN1's HRD and DAXX read the same H3.3 G90 residue by *different* mechanisms and asks whether the two can be discriminated. That is the question a DAXX chaperone-pocket inhibitor would have to answer, and it is the difference between blocking heterochromatin H3.3 loading and blocking all of it.
Which surface would you actually drug, and what each one costs
| surface | structures | what blocking it does | the problem |
|---|---|---|---|
| DAXX-ATRX (DHB / 4HB) | 5GRQ 1.58 A, 5Y18 2.20 A, 5Y6O 3.1 A | separates the chaperone from the remodeller; should abolish the heterochromatin arm and spare HIRA gene-body deposition | ATRX has functions beyond DAXX; and this is the cleanest option |
| DAXX H3.3-binding cleft | 4H9N 1.95 A, 4HGA 2.80 A | blocks H3.3 handover directly | DAXX envelops the H3.3-H4 dimer - a large buried surface, the hard case for a small molecule, and the UBN1 selectivity problem sits here |
| DAXX SIM / SUMO | 2KQS (NMR) | detaches DAXX from PML bodies | DAXX has multiple SUMO-dependent partners; low selectivity by construction |
| ATRX ATPase | 13 human entries, none a full-length remodeller | inhibits the remodelling activity | ATPase inhibitors across the SNF2 family are notoriously non-selective, and ATRX loss is the ALT lesion |
| degradation | n/a | removes the protein | DAXX[tiab] AND (degrader OR PROTAC)[tiab] returns 0 records 4 |
The DAXX-ATRX interface is the one worth naming. It is a defined peptide-in-groove contact, it is solved to 1.58 A, it separates the two arms of H3.3 deposition rather than shutting the variant down globally, and it is the only surface here whose disruption has an obvious phenotypic prediction that this beat's experiment could test.
The bite-back is unusually strong and it is not speculative
For most targets in this graph the toxicity argument is an extrapolation. Here there is direct human evidence, in both directions:
- ATRX loss causes an inherited human syndrome. Germline ATRX loss of function produces alpha-thalassaemia with X-linked intellectual disability, and acquired somatic ATRX lesions in adults present as acquired alpha-thalassaemia myelodysplastic syndrome - so partial loss of this arm has a documented haematological phenotype in people.
- ATRX/DAXX loss is the canonical route to ALT, in 61% of one tumour type, every case carrying an ATRX or DAXX lesion. An inhibitor of this axis is administering the lesion that supplies telomerase-independent telomere maintenance elsewhere. A two-week viability screen cannot see that, and neither can any assay in the current DepMap evidence.
- The patients are infants. A drug that removes a chromatin remodeller with a developmental phenotype, given to a one-year-old, needs a therapeutic index nobody has begun to establish.
So the conclusion this entry reaches is: not obviously druggable, and the reason is not chemical tractability but that the best-characterised consequence of removing these proteins in humans is a second oncogenic mechanism. That is stated with evidence rather than by omitting the field.
| verdict | unattempted rather than undruggable. The chemistry is absent (6 rows, no affinity) and the structural biology is good (a 1.58 A interface). The blocking objection is not tractability but that the best-documented human consequence of losing this axis is alternative lengthening of telomeres. |
|---|---|
| relationship to existing entries | `gene-daxx` carries no bite_back field at all and `gene-atrx`'s bite_back names the ALT risk without the chemistry or the structures. Nothing in this corpus previously recorded a ChEMBL or PDB position for either protein. |
targets
[
{
"symbol": "DAXX",
"uniprot": "Q9UER7",
"uniprot_entry_name": "DAXX_HUMAN",
"gene_primary_readback": "DAXX",
"length_aa": 740,
"chembl_target": "CHEMBL5724773",
"chembl_activity_rows": 6,
"chembl_activity_types": [
"FC"
],
"chembl_best_affinity": null,
"chembl_source_documents": [
"CHEMBL5696440"
],
"pdb_human_entries": 14,
"pdb_ids": [
"2KQS",
"2KZS",
"2KZU",
"4H9N",
"4H9O",
"4H9P",
"4H9Q",
"4H9R",
"4H9S",
"4HGA",
"5GRQ",
"5KDM",
"5Y18",
"5Y6O"
]
},
{
"symbol": "ATRX",
"uniprot": "P46100",
"uniprot_entry_name": "ATRX_HUMAN",
"gene_primary_readback": "ATRX",
"length_aa": 2492,
"chembl_target_human": null,
"chembl_target_mouse": "CHEMBL4879474",
"chembl_activity_rows": 2,
"chembl_activity_types": [
"FC"
],
"chembl_best_affinity": null,
"pdb_human_entries": 13,
"pdb_ids": [
"2JM1",
"2LBM",
"2LD1",
"3QL9",
"3QLA",
"3QLC",
"3QLN",
"4W5A",
"5GRQ",
"5Y18",
"5Y6O",
"6G0O"
]
}
]comparator
{
"symbol": "MEN1",
"uniprot": "O00255",
"chembl_target": "CHEMBL1615381",
"chembl_activity_rows": 1776,
"chembl_ppi_target": "CHEMBL2093861",
"chembl_ppi_activity_rows": 48291,
"pdb_human_entries": 69
}candidate surfaces
[
{
"name": "DAXX-ATRX interface (DHB / four-helix bundle against an ATRX peptide)",
"structures": [
"5GRQ",
"5Y18",
"5Y6O"
],
"best_resolution_A": 1.584,
"rationale": "separates the heterochromatin arm of H3.3 deposition from the HIRA gene-body arm rather than shutting the variant down globally; the only surface here with a phenotypic prediction this beat's experiment can test",
"risk": "helical peptide-in-groove geometry, which is the PPI class that most often defeats small molecules; no pocket-detection analysis has been run"
},
{
"name": "DAXX H3.3-H4 chaperone cleft",
"structures": [
"4H9N",
"4HGA"
],
"best_resolution_A": 1.95,
"rationale": "blocks H3.3 handover directly",
"risk": "DAXX envelops the dimer - a large buried surface; and UBN1 reads the same H3.3 G90 residue, so selectivity between the two H3.3 routes is unsolved"
},
{
"name": "DAXX SIM-SUMO",
"structures": [
"2KQS"
],
"best_resolution_A": null,
"rationale": "would detach DAXX from PML nuclear bodies",
"risk": "SUMO-SIM contacts are promiscuous by construction; selectivity is implausible"
},
{
"name": "targeted degradation",
"structures": [],
"best_resolution_A": null,
"rationale": "reaches proteins with no ligandable pocket",
"risk": "no published DAXX degrader; 0 records for degrader or PROTAC"
}
]measurements — the numbers behind the chart
[
{
"quantity": "chembl_activity_rows",
"subject": "DAXX human target CHEMBL5724773",
"value": 6,
"units": "activity rows",
"n": 1,
"universe": "ChEMBL, fetched 2026-09-24; n is the number of source documents",
"null": 1776,
"note": "null column is menin CHEMBL1615381 as the within-project comparator; all 6 DAXX rows are fold-change, none an affinity"
},
{
"quantity": "chembl_activity_rows",
"subject": "ATRX - no human target record; murine CHEMBL4879474",
"value": 2,
"units": "activity rows",
"n": 1,
"universe": "ChEMBL, fetched 2026-09-24",
"null": 1776,
"note": "mouse C2C12 phospho-site reduction, i.e. an off-target observation in a kinase-inhibitor study"
},
{
"quantity": "structure_resolution",
"subject": "DAXX DHB domain bound to an ATRX peptide, PDB 5GRQ",
"value": 1.584,
"units": "angstrom",
"n": 1,
"universe": "RCSB human DAXX entries, fetched 2026-09-24",
"null": null,
"note": "the best-resolved candidate surface; two further DAXX-ATRX structures at 2.20 and 3.1 A"
}
]- Loss of ATRX or DAXX function is the canonical human route to alternative lengthening of telomeres - 61% of pancreatic neuroendocrine tumours are ALT-positive and every one carries an ATRX or DAXX lesion. An inhibitor of this axis administers the lesion that supplies telomerase-independent immortality elsewhere, and no viability screen can see it.
- Germline ATRX loss of function causes alpha-thalassaemia with X-linked intellectual disability, and acquired somatic ATRX lesions in adults present as acquired alpha-thalassaemia myelodysplastic syndrome. Partial loss of this arm has a documented human haematological and developmental phenotype, in a disease whose patients are infants.
- DAXX loss in mouse haematopoietic precursors pushes cells out of the B lineage into a Pu.1-driven myeloid programme. In a B-ALL where therapy-driven lineage switch to a myeloid state is an established escape route, a DAXX inhibitor may be pushing the cell down exactly that path.
- The DAXX H3.3 cleft and the UBN1 HRD read the same histone residue. A chaperone-pocket inhibitor that cannot discriminate them is not a heterochromatin drug, it is a global H3.3-deposition drug, with a toxicity profile nobody has estimated.
- There is no chemical matter to start from. Six fold-change rows and no affinity means any programme here begins with a screen, not with an optimisation - years, not months, before there is a tool compound to test any of the biology in this beat.
- Run a pocket-detection analysis on 5GRQ. A 1.58 A structure of the DAXX-ATRX contact either does or does not contain a druggable cavity, and that is a computational question answerable in an afternoon which nobody has asked. It is the single cheapest thing in this entry.
- Search the patent literature for DAXX and ATRX. ChEMBL's six rows distinguish 'unattempted' from 'attempted and abandoned' badly, and a failed industrial campaign would be far more informative than either number here.
- Can a ligand discriminate the DAXX and UBN1 routes to H3.3? `protein-huca-complex` poses this and it is the question that decides whether a chaperone-pocket inhibitor is a heterochromatin drug or a global H3.3 drug.
- What happens to a KMT2A-rearranged blast when DAXX is degraded acutely rather than knocked out? No degron reagent exists for DAXX in any leukaemia line, and the SPT6 precedent in this graph shows the two give different answers.
- E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database
As of 2026-09-24 ChEMBL holds one human DAXX target record (CHEMBL5724773) with 6 activity rows, every one a fold-change measurement at 0.4 to 10 uM from a single document (CHEMBL5696440) and none an affinity; there is no human ATRX target record, the only ATRX record being murine (CHEMBL4879474) with 2 rows measuring phospho-site reduction in mouse C2C12 cells; the same database holds 1,776 activity rows for human menin (CHEMBL1615381) and 48,291 for the menin-MLL protein-protein interaction (CHEMBL2093861).
ChEMBL target and activity records for DAXX (CHEMBL5724773), ATRX (CHEMBL4879474), menin (CHEMBL1615381) and the menin-MLL interaction (CHEMBL2093861) · ChEMBL, EMBL-EBI 2026What this evidence cannot show
Fetched live from the ChEMBL REST API on 2026-09-24; the release was not recorded by the endpoint used and should be if this is re-quoted. Per this project's standing caveat, an activity-row count measures how much medicinal chemistry has been performed, not how tractable a target is - zero rows means nobody tried. The DAXX rows are fold-change readouts of unstated origin ('DAXX (unknown origin)'), which is the pattern of a chemoproteomics or expression experiment where DAXX is an observed protein rather than the intended target; treating any of them as a DAXX binding measurement would be the off-target-pulldown error this graph already warns about.
Evidence that argues against this (1)
- ChEMBL under-represents chemistry that never left industry and it indexes patents unevenly. A target with no public rows can still have had a failed internal campaign, and the absence of rows is therefore weak evidence that nobody tried rather than strong evidence. A patent-space search - which was NOT run here - is the check that would distinguish 'unattempted' from 'attempted and abandoned', and the latter would be far more informative than either number in this entry. ChEMBL, EMBL-EBI 2026
- E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
The DAXX-ATRX contact is solved: RCSB holds 5GRQ (crystal structure of the DHB domain of DAXX in complex with an ATRX peptide, X-ray, 1.58 A), 5Y18 (DAXX helical bundle domain in complex with ATRX, 2.20 A) and 5Y6O (DAXX N-terminal four-helix bundle domain in complex with ATRX, 3.1 A), with 5Y18's primary citation being a structural study of the DAXX-ATRX interaction.
Structural basis for DAXX interaction with ATRX. · Wang X, Zhao Y, Zhang J, et al. · Protein & Cell 2017What this evidence cannot show
PDB identifiers, titles, methods and resolutions fetched from the RCSB data API on 2026-09-24; the paper title was taken from PubMed and not from RCSB, because the two differ in this project's experience. The PubMed record for 28875424 carries no abstract, so the title is the only content verified; the structural facts quoted here come from the deposits themselves rather than from the paper. 5GRQ and 5Y6O carry no primary-citation PubMed id in RCSB. These are domain constructs and peptides, not full-length proteins, and a solved interface is not a demonstration that the interface is ligandable - this project's own standing caveat is that a PDB entry existing is not a structure suitable for design.
Evidence that argues against this (1)
- Three structures of one interface solved by domain-and-peptide crystallography is also the signature of an interface that is large and helical rather than a pocket. The menin-MLL comparison flatters DAXX: menin's MBM1 site is a deep buried pocket that accepts a small molecule, and a four-helix-bundle-against-peptide contact is the protein-protein interaction class that has historically defeated small molecules. Nothing in the deposits says which case this is, and a pocket-detection analysis on 5GRQ - which would take an afternoon - has not been run. RCSB Protein Data Bank 2026
- E3established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
The DAXX chaperone surface is also solved, with DAXX enveloping an H3.3-H4 dimer: RCSB 4H9N is a 1.95 A X-ray structure of DAXX with H3.3 and H4 and 4HGA a 2.80 A structure of the variant H3.3-H4 heterodimer with its chaperone DAXX, and DAXX distinguishes H3.3 from H3.1 across only five differing residues.
Structure of the variant histone H3.3-H4 heterodimer in complex with its chaperone DAXX. · Liu CP, Xiong C, Wang M, et al. · Nature Structural & Molecular Biology 2012What this evidence cannot show
Abstract read from efetch 2026-09-24; PDB identifiers, methods and resolutions from the RCSB data API the same day. 4H9N's RCSB primary-citation record carries no PubMed id and its deposit title differs from the paper title, so the two are reported separately rather than merged. An enveloping chaperone surface is the least favourable geometry for a small-molecule inhibitor, so this structure is cited as evidence about the DIFFICULTY of that surface as much as its existence.
Evidence that argues against this (1)
- Blocking the DAXX H3.3 cleft would not be selective for the heterochromatin arm. This graph's `protein-huca-complex` already records that UBN1's HRD and DAXX both read H3.3 residue G90, by distinct mechanisms but on nearly the same surface, and asks whether the two routes can be discriminated pharmacologically. Until that is answered, a chaperone-pocket inhibitor is a whole-H3.3-deposition inhibitor, which is a different and much more toxic proposition than separating the two arms. Goldberg AD et al. 2010
- E4established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database
As of 2026-09-24 PubMed returns 0 records for 'DAXX[tiab] AND (degrader[tiab] OR PROTAC[tiab])' and 5 for 'DAXX[tiab] AND "small molecule"[tiab]', against 1,039 for bare DAXX[tiab] and 2,103 for bare ATRX[tiab]; every query returned a non-empty querytranslation.
Bounded PubMed absence for DAXX degraders and small molecules, run through scripts/pubmed_safe.py with operand controls · PubMed (NCBI E-utilities), via this project's pubmed_safe wrapper 2026What this evidence cannot show
Titles and abstracts only. The name list tried was `degrader` and `PROTAC`; NOT tried were `molecular glue`, `LYTAC`, `dTAG`, `SNIPER`, `bifunctional degrader` or the spelled-out 'proteolysis targeting chimera', and per this project's rule 72 a drug-class absence is bounded by its name list rather than by the query. The 5 'small molecule' records were not title-checked. A degrader reported only in a patent or a poster is invisible here.
Evidence that argues against this (1)
- This absence is the weakest kind and should not be quoted as evidence that DAXX is undegradable. Targeted degradation is the modality most likely to reach a protein with no ligandable pocket, and the absence of published DAXX degraders reflects that nobody has had a reason to build one - which is a statement about the field's attention, not about DAXX. The same query shape returns zero for most of the proteome. ChEMBL, EMBL-EBI 2026
6 ChEMBL activity rows, none an affinity; 14 human PDB entries; unattempted rather than undruggable
no human ChEMBL target record at all; the only record is murine with two off-target phospho rows
the UBN1-versus-DAXX H3.3 G90 selectivity question that entry poses is the blocking problem for a chaperone-pocket inhibitor
scale comparator: menin carries 1,776 activity rows and the menin-MLL PPI 48,291, against DAXX's 6
the dependency this entry asks whether anyone could act on
the ALT bite-back that blocks this target class comes from the panNET and glioma data in that entry
the menin-MLL interface is the worked precedent for drugging a peptide-in-groove contact in this disease, and the scale comparison against it is the point