ATF5 / ATFX (Q9Y2D1) - the most reproducible repression in the earlier dataset, and the three probe values the graph quotes are SEM, not RS4;11
also called ATF5 · ATFX · Cyclic AMP-dependent transcription factor ATF-5 · ATF5_HUMAN
ATF5 is a close relative of the cell's master stress switch and, like it, is made only when the cell is stressed. It falls sharply in KMT2A-rearranged leukaemia cells treated with a menin-blocking drug, but by very different amounts in the two cell lines tested, and the graph has been quoting the bigger line's numbers as if they came from the other one.
The same thing, for a scientist
ATF5/ATFX (Q9Y2D1, 282 aa, bZIP 208-271) is an ATF4-family bZIP factor that binds the amino-acid response element and, like ATF4, is itself translationally controlled by an upstream open reading frame. This graph records it as "the single most reproducible repression in the earlier dataset (-3.02 / -2.89 / -2.77 on three MI-2-2 probes, -2.43 under VTP-50469)". **Re-measured on all four GPL570 ATF5 probes in all three cell lines, the first three numbers are from SEM and not from RS4;11** [E1]: | probe | DMSO RS4;11 | **RS4;11 MI-2-2** | DMSO SEM | **SEM MI-2-2** | REH MI-2-2 | |---|---:|---:|---:|---:|---:| | 204998_s_at | 12.53 | **-0.74** | 11.94 | **-2.77** | +0.04 | | 204999_s_at | 11.36 | **-1.09** | 10.74 | **-3.02** | +0.18 | | 217389_s_at | 10.74 | **-1.12** | 10.03 | **-2.89** | +0.09 | | 230938_x_at | 8.07 | **-0.72** | 8.19 | **-1.43** | +0.48 | **In RS4;11 - the line the whole beat is scoped to - only two of four probes clear 1 log2, and the largest is -1.12.** The -2.43 VTP-50469 value is correct and is RS4;11. So the claim "most reproducible repression" survives in the sense that all four probes agree in direction in both t(4;11) lines and all four are flat-to-positive in the KMT2A-wild-type line REH - **but the effect size the graph quotes is a SEM effect size attached to an RS4;11 argument.** And the VTP-50469 value is **not menin-specific**: EPZ5676 gives -2.36 in the same samples [E1]. DepMap 26.06: mean gene effect -0.1727 across 1,257 screens, 1.0% of lines below -0.5 - not a dependency [E2]. PubMed returns **zero** records for ATF5 with any of eight fusion operands and zero with a thirteen-name menin-inhibitor list [E3].
This is a prediction, not a measurement. No one has seen this protein in this shape. The colours say how much the software trusts itself, and that is worth reading before anything else — the orange stretches are places the model is effectively saying this part has no fixed shape.
I have never read one of these pictures before
A protein is one long chain of amino acids that folds up. The ribbon traces that chain from one end to the other; it is not a wire, it is a simplification of where the backbone runs. Drag to rotate, scroll to zoom.
Domains are the stretches of chain that do a particular job, and they are what the coloured blocks above the picture mark out. Click one and the view flies to it. In this project the important question about any domain is usually whether the fusion still has it — the labels say lost where it does not.
Anything in yellow sticks is a bound small molecule, which is usually the drug.
What the four probes and three lines actually say
The table above is the whole result. Three readings of it, in decreasing strength:
Strong. All four probes move down in both t(4;11) lines and all four move flat-to-up in REH, and the four probes span 8.07 to 12.53 log2 RMA in baseline so this is not a single-probe or single-intensity-regime artefact. Four-probe internal agreement plus a clean wild-type control is the best-supported fusion-linkage in this beat's panel.
Moderate. RS4;11 and SEM differ by roughly 1.8 log2 on the same probes under the same compound at the same dose. Both are t(4;11) KMT2A::AFF1 B-ALL lines, so either the lines differ materially on this axis or one of the two arms is mis-estimated. SEM's MI-2-2 arm is n=2 and its DMSO arm is n=2; RS4;11's DMSO arm is also n=2. At those replicate numbers a 1.8 log2 discrepancy between lines is not resolvable.
Weak, and stated as a caution rather than a finding. Under the potent compound only RS4;11 was assayed, and there ATF5 falls -2.43 - i.e. the potent compound in RS4;11 reproduces roughly the magnitude the weak compound produced in SEM. That is a coincidence of numbers and nothing more; there is no SEM arm under VTP-50469 to check it against.
Why ATF5 matters more than its effect size: it is the second uORF-controlled node in the panel
ATF5, like ATF4, carries inhibitory upstream open reading frames in its 5-prime leader and is translationally de-repressed under eIF2alpha phosphorylation. So this beat's panel contains two uORF-controlled ISR transcription factors and they behave differently: ATF4's transcript is flat (-0.86 max anywhere) while ATF5's falls 1.1 to 3.0 log2 depending on line and compound 1.
That asymmetry is informative and it is the entry's main mechanistic content. If the perturbation acted purely on eIF2alpha-dependent translation, both uORF-controlled factors would show unchanged mRNA. ATF5's mRNA moves and ATF4's does not, which means at least part of the response is transcriptional and it is selective for ATF5 over ATF4. ATF5 is itself an ATF4 target in the canonical hierarchy, so the simplest reconciliation is: ATF4 activity falls (invisible in mRNA), and ATF5 - being downstream - loses its transcriptional input. Under that reading ATF5 belongs with CHAC1 and TRIB3 as an output gene, not with ATF4 as a regulator.
What would decide it: an ATF4 ChIP at the ATF5 locus, or an ATF5 measurement under a direct ISR manipulation in the same cells. Neither exists.
| uniprot | Q9Y2D1 |
|---|---|
| length AA | 282 |
| entry name | ATF5_HUMAN |
| protein name | Cyclic AMP-dependent transcription factor ATF-5 |
| ensembl gene ID | ENSG00000169136 |
| PDB note | UniProt Q9Y2D1 cross-references NO PDB entries. |
| alphafold ID | AF-Q9Y2D1-F1 |
| accession resolution | gene_exact:ATF5 AND organism_id:9606 AND reviewed:true returns EXACTLY ONE reviewed row, Q9Y2D1, ATF5_HUMAN, 282 aa, synonym ATFX; re-fetched alone with name and length compared back: PASS. |
| localization | CYTOPLASM, NUCLEUS and PERICENTRIOLAR MATERIAL. UniProt Q9Y2D1 carries Cytoplasm (ECO:0000269, PubMed 15358120), Nucleus (ECO:0000255 sequence model plus ECO:0000269 PubMed 15358120) and centrosome/pericentriolar material (ECO:0000269, PubMed 22528486 and 26213385), with a note that ATF5 is actively transported to the centrosome and accumulates in the pericentriolar material from G1 to M phase by a microtubule-dependent mechanism. RULE 91: none of these source papers was opened by this beat. No ATF5 localisation experiment exists in any haematopoietic or KMT2A-rearranged cell. |
| subnuclear localization | NOT ASSIGNED and NEVER LOOKED FOR. UniProt annotates Nucleus with no subnuclear qualifier; the only sub-compartment it names is CYTOPLASMIC (pericentriolar material), which is not a subnuclear answer. No nuclear body, speckle or chromatin address has been reported for ATF5 in any cell type. Recorded as a bounded absence; no compartment is invented. |
| cell type | RS4;11 and SEM (t(4;11) KMT2A::AFF1 B-ALL - and they DISAGREE by 1.8 log2 on this gene), REH (KMT2A wild type, clean negative control on all four probes) and MOLM13 (KMT2A::MLLT3 AML, only -0.74), measured here [E1]. UniProt records widest expression with higher levels in liver. No primary leukaemia sample, no normal haematopoietic cell. |
| cell state | Measured in the day-6/day-7 drug-arrested state. ATF5 is itself CELL-CYCLE-STATE-DEPENDENT in its localisation - transported to the pericentriolar material from G1 to M - so in a drug that arrests the cell cycle, a change in ATF5 could reflect the cycle state rather than the stress state, and neither dataset staged the cells. Its RNA-seq DMSO baseline rises 1.86 log2 from day 2 to day 7. |
| epigenetic readout | No writer, eraser or reader activity. ATF5's fall is reproduced by the DOT1L inhibitor EPZ5676 to within 0.07 log2 (VTP -2.43, EPZ5676 -2.36), so its expression tracks H3K79-methylation-dependent fusion output rather than the menin interface. No histone-mark measurement at the ATF5 locus exists in any KMT2A-rearranged cell, and no ATF4 ChIP at the ATF5 locus exists in any cell - which is the missing link that would place ATF5 as an ATF4 target rather than an ATF4 peer. |
| shown for fusion | MEASURED in KMT2A::AFF1 on four independent array probes in TWO lines (RS4;11 -0.72 to -1.12, SEM -1.43 to -3.02) with a clean KMT2A-wild-type REH control (+0.04 to +0.48), and in KMT2A::MLLT3 (MOLM13 -0.74). Transcript only. No ATF5 protein, no ChIP, no functional test in any fusion context. |
| absence class | literature-absence-bounded |
| absence class note | RECLASSIFIED 2026-09-23 (cycle 76). This field held free text, which `scripts/check-absence-class.py` rejects; the prose is preserved here VERBATIM and unsummarised because it is more informative than the token that replaced it. Token chosen: literature-absence-bounded - the two-part prose says MEASURED-AND-<something> for the data arm and BOUNDED-UNPUBLISHED for the prose arm. The data arm is a RESULT, not an absence; the absence is the bounded literature, and the bound's own limits (titles and abstracts only, per rule 80) are stated in the preserved prose. ORIGINAL absence_class, verbatim: «MEASURED-AND-LINE-DEPENDENT in transcript; BOUNDED-UNPUBLISHED in prose (0 fusion-scoped and 0 menin-drug-scoped records, both with valid querytranslation and 'No items found.'), with the caveat that ATF5's total literature is only 342 records so the zero is close to what marginal counts predict.» |
PDB IDS
[]
domains
[
{
"name": "bZIP (basic leucine zipper)",
"start": 208,
"end": 271
}
]the correction
{
"what_the_graph_says": "ATF5 was the single most reproducible repression in the earlier dataset: -3.02 / -2.89 / -2.77 on three MI-2-2 probes, -2.43 under VTP-50469",
"what_the_data_say": "Those three numbers are SEM, not RS4;11. The same three probes in RS4;11 give -1.09, -1.12 and -0.74. The -2.43 VTP-50469 value IS RS4;11 and is correct.",
"is_the_prior_entry_wrong": "Not strictly - it says 'in the earlier dataset' and does not name a line. But it sits inside an argument scoped to RS4;11 and reads as an RS4;11 result, and the brief dispatching this beat quoted it that way.",
"what_survives": "All four probes move down in BOTH t(4;11) lines and all four are flat-to-positive in KMT2A-wild-type REH, across a baseline range of 8.07 to 12.53 log2 RMA. Four-probe agreement plus a clean wild-type control is the best-supported fusion linkage in this beat's panel. Only the EFFECT SIZE was line-swapped."
}the measurement
{
"rs411_array_mi22_by_probe": {
"204998_s_at": -0.741,
"204999_s_at": -1.088,
"217389_s_at": -1.118,
"230938_x_at": -0.719
},
"sem_array_mi22_by_probe": {
"204998_s_at": -2.774,
"204999_s_at": -3.021,
"217389_s_at": -2.888,
"230938_x_at": -1.434
},
"reh_array_mi22_by_probe": {
"204998_s_at": 0.041,
"204999_s_at": 0.175,
"217389_s_at": 0.093,
"230938_x_at": 0.483
},
"rs411_rnaseq_d7_vtp_lfc": -2.432,
"rs411_rnaseq_d7_epz5676_lfc": -2.361,
"rs411_rnaseq_d2_vtp_lfc": -0.541,
"molm13_rnaseq_d7_vtp_lfc": -0.739,
"menin_specificity": "SHARED WITH DOT1L INHIBITION (VTP -2.43, EPZ5676 -2.36)"
}essentiality
{
"depmap_release": "26.06 (read back from meta{dataVersion})",
"depmap_screens": 1257,
"chronos_mean": -0.1727,
"pct_lines_below_minus_0_5": 1,
"leukaemia_screens": 57,
"leukaemia_mean_gene_effect": -0.1809,
"reading": "not a dependency in any lineage measured here"
}identifier traps
{
"registry_status": "NOT in data/symbol-collisions.json as of 2026-09-22 (157 rows read).",
"ATFX": "11 PubMed records against ATF5's 342, sharing 2. Fusion-scoped ATFX returns a valid 0, so nothing was missed.",
"the_real_trap_here_is_not_a_symbol": "It is a CELL LINE. Three effect sizes travelled from SEM into an RS4;11 argument. A fold change is not a fact until the line, the compound, the dose and the duration travel with it.",
"array": "GPL570 carries FOUR ATF5 probes and none whose own symbol field is ATFX. Reporting only the best-expressed probe, as the repo's earlier loader does, would have hidden the four-probe agreement that makes this the strongest fusion linkage in the panel.",
"uniprot": "clean - exactly one reviewed human record"
}- ATF5 is not a dependency in any lineage measured (-0.17 mean, 1.0% of lines below -0.5), so its repression is a readout and not a mechanism of killing - the same conclusion as for CHAC1, TRIB3 and STC2, and it applies to every member of the signature this graph has checked.
- ATF5 is required for survival in some lineages and ATF5-targeting agents exist, so any attempt to deepen this arm therapeutically runs into a factor whose loss is tolerated by leukaemia lines but not universally - the wrong combination of selectivity.
- If ATF5 protein is uORF-controlled like ATF4's, a 1.1 log2 mRNA fall could correspond to almost any protein change, and the graph should not treat ATF5 mRNA as a proxy for ATF5 activity any more than it should for ATF4.
- Why do RS4;11 and SEM differ by 1.8 log2 on ATF5 under the same compound at the same dose, when both are t(4;11) KMT2A::AFF1 B-ALL? A repeat with adequate replicates in both lines is the only way to tell a real line difference from an n=2 artefact, and it decides whether the graph's ATF5 claim is about the fusion or about SEM.
- Does ATF4 occupy the ATF5 locus in RS4;11? Without that ChIP, ATF5 cannot be placed as an ATF4 target rather than a parallel ATF4-family factor, and the two placements imply different mechanisms.
- Is ATF5's centrosomal pool relevant here? A drug that arrests the cell cycle changes the G1-to-M distribution ATF5's localisation depends on, and no experiment has staged the cells.
- Does ATF5 protein fall in RS4;11, and by how much relative to its 1.1 log2 mRNA fall? ATF5 is uORF-controlled like ATF4, so its protein and mRNA need not move together in either direction.
- E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.dataset
On all four GPL570 ATF5 probes, MI-2-2 gives -0.74/-1.09/-1.12/-0.72 in RS4;11 and -2.77/-3.02/-2.89/-1.43 in SEM, and +0.04/+0.18/+0.09/+0.48 in the KMT2A-wild-type line REH; VTP-50469 gives -2.43 in RS4;11 at day 7 and EPZ5676 gives -2.36 in the same samples.
What this evidence cannot show
THE CORRECTION THIS ENTRY EXISTS FOR: the values -3.02/-2.89/-2.77 quoted in ana-two-unrelated-menin-inhibitors-converge-on-the-atf4-stress-programme are SEM values, not RS4;11 values; RS4;11 gives -1.09/-1.12/-0.74 on the same probes. The prior entry says 'in the earlier dataset' without naming the line, so it is not strictly false - but it appears inside an argument scoped to RS4;11 and reads as an RS4;11 result. SEM's MI-2-2 arm is n=2 and both DMSO arms are n=2, so the 1.8 log2 inter-line discrepancy is not resolvable at these replicate numbers. No SEM arm exists under the potent compound.
Evidence that argues against this (1)
- The same collapse is produced almost exactly by EPZ5676, a potent DOT1L inhibitor, in the same cells at the same timepoint, so the effect marks disruption of the fusion's transcriptional output by either route and not the menin-KMT2A interface specifically. NCBI GEO, re-analysed here 2026
- E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database
In DepMap 26.06, ATF5 has a mean CRISPR gene effect of -0.1727 across 1,257 screens with 1.0% of lines below -0.5 and a leukaemia mean of -0.1809 across 57 leukaemia screens.
DepMap CRISPR (Chronos) gene effect via the Open Targets Platform GraphQL depMapEssentiality field; dataVersion year 26 month 06, read back from meta{dataVersion} in the same session · DepMap / Open Targets Platform 2026What this evidence cannot show
Release recorded per rule 83 (dataVersion year 26 month 06); gene-specific screen count of 1,257. Leukaemia screens identified by substring match on diseaseFromSource and their ids persisted per rule 90. A -0.17 mean with only 1% of lines below -0.5 is a non-dependency, not a weak dependency.
Evidence that argues against this (1)
- ATF5 has well-documented lineage-restricted roles - it is required for survival of glioblastoma and neural progenitor cells and ATF5-targeting peptides kill glioma cells - so a flat pan-cancer mean can conceal a real dependency in a few lineages. A leukaemia mean of -0.18 across 57 screens does not rule out a dependency in the eleven KMT2A-rearranged lines specifically, and those lines were not separated out here. UniProt Knowledgebase 2026
- E3established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database
PubMed returns zero records for ATF5 jointly with any of eight KMT2A-fusion operands and zero for ATF5 with a thirteen-name menin-inhibitor list, both with non-empty querytranslations and warninglist 'No items found.'; the alias ATFX reaches 11 records of which 2 also mention ATF5.
NCBI PubMed esearch, run 2026-09-22, with querytranslation and warninglist read back from the response for every query quoted · NCBI PubMed 2026What this evidence cannot show
Rule 3 satisfied: non-empty querytranslation AND 'No items found.' on both zeros, and the fusion-scoped ATFX query also returns a valid zero. Rule 94 form; rule 95 observed ('t(4;11)' translated to 't 4 11'[Title/Abstract], 736 records as control). Rule 80: titles and abstracts only - an ATF5 row in a supplementary expression table, which is exactly where the -2.43 came from, is invisible to this query.
Evidence that argues against this (1)
- A zero here is unsurprising and therefore carries little information: ATF5 has 342 PubMed records in total against ATF4's 5,480, so the gene is thinly studied and a joint zero with an eight-operand fusion list is close to what the marginal counts predict. The absence says more about ATF5's overall literature size than about the KMT2A field. NCBI PubMed 2026
- E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database
UniProt Q9Y2D1 is the single reviewed human ATF5 record, 282 aa with a bZIP domain at 208-271, synonym ATFX, annotated as a transcription factor binding the cAMP response element, an ATF5-specific response element and the amino acid response element, and localised to cytoplasm, nucleus and the pericentriolar material, to which it is actively transported from G1 to M phase by a microtubule-dependent mechanism.
UniProtKB - Q9Y2D1, ATF5_HUMAN, cyclic AMP-dependent transcription factor ATF-5 · UniProt Knowledgebase 2026What this evidence cannot show
Resolved live 2026-09-22: exactly one reviewed record (rule 62 satisfied), re-fetched alone with name, length and primary symbol compared back: PASS. RULE 91: the localisation calls carry ECO:0000269 to PubMed 15358120, 22528486 and 26213385, none of which was opened by this beat, so they are recorded as coded annotations whose source papers were not checked - a weaker standing than CHAC1's, where the coded citation is the gene's own founding paper. The Nucleus call also carries ECO:0000255 (sequence-model inference) alongside experimental codes.
Evidence that argues against this (1)
- UniProt gives ATF5 three compartments including a cell-cycle-dependent centrosomal pool, and the graph's entire use of ATF5 here is as a nuclear transcription factor. A factor that is substantially centrosomal from G1 to M is doing something other than transcription for part of the cycle, and the transcript-level measurement in this beat cannot distinguish loss of a nuclear pool from loss of a centrosomal one. UniProt Knowledgebase 2026
- related to →KMT2A::AFF1 (MLL-AF4), t(4;11)(q21;q23)
four independent array probes agree in direction in both t(4;11) lines and are flat-to-positive in KMT2A-wild-type REH; but the -3.02/-2.89/-2.77 values the graph quotes are SEM, not RS4;11 (RS4;11 gives -1.09/-1.12/-0.74)
two uORF-controlled ISR bZIP factors behaving differently: ATF4 mRNA flat at -0.86, ATF5 mRNA -2.43 in the same samples, which means part of the response is transcriptional and ATF5-selective
- ← regulatesPinometostat (EPZ-5676) — the DOT1L inhibitor, and the most instructive failure in this field
EPZ5676 -2.36 against VTP-50469 -2.43 in the same RS4;11 day-7 samples
the -3.02/-2.89/-2.77 ATF5 triple in the corpus is SEM; the same three probes in RS4;11 give -1.09/-1.12/-0.74