Oncofusion

start / graph / gene-ppp1r15a

Genemoderate confidence 3 of 5: One strong study, or several smaller consistent ones.Q106-isr-stress-programme

PPP1R15A / GADD34 (O75807) - the eIF2alpha phosphatase adaptor, and KMT2A fusion proteins were shown to BIND it and block its apoptosis in 1999, a paper this graph has never cited

also called PPP1R15A · GADD34 · Protein phosphatase 1 regulatory subunit 15A · PR15A_HUMAN

GADD34 is the switch that turns the stress response back off, by removing the chemical tag that started it. In 1999 three different KMT2A leukaemia fusion proteins were shown to bind GADD34 directly and to block the cell death it causes - the first gain-of-function ever shown for these fusions - and this project had never recorded it.

The same thing, for a scientist

PPP1R15A/GADD34 (O75807, 674 aa) recruits the catalytic phosphatase PPP1CA to dephosphorylate eIF2alpha, **reversing the translational shut-off that starts the integrated stress response** - it is the negative-feedback terminator of the exact mechanism this beat argues is the operative one for ATF4. **And it has a direct, physical, published link to KMT2A fusion proteins that this graph has never cited.** Adler 1999 showed by yeast two-hybrid and co-immunoprecipitation in human cells that **HRX (MLL/KMT2A) proteins interact directly with GADD34**, that GADD34 overexpression increases apoptosis after ionising radiation via its amino-terminal third, and that **three different HRX fusion proteins - HRX-ENL, HRX-AF9 and HRX-ELL - abrogate GADD34-induced apoptosis while wild-type HRX increases it**; the authors present this as **the first gain of function demonstrated for leukaemic HRX fusion proteins** [E1]. GADD34 also binds hSNF5/INI1. **The graph missed this because the query missed it**: `PPP1R15A[tiab]` AND an eight-operand fusion list returns **0**, while `GADD34[tiab]` AND the same list returns **2**, including this paper - a rule-71 false absence on the single most fusion-relevant gene in this beat's panel [E3]. Transcriptionally PPP1R15A barely moves: RS4;11 array MI-2-2 -0.84 and -0.65 on two probes, VTP-50469 -0.29 at day 7, and **+1.14 in MOLM13 at day 2** - discordant across lines and timepoints [E2]. DepMap 26.06: -0.1127 across 1,258 screens, 1.6% below -0.5 [E4].

What it looks like
loading AF-O75807-F1…
AlphaFold AF-O75807-F1 ↗
colourshape

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.

very high (90+)confident (70–90)low (50–70)very low (<50)
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.

Go deeper
The 1999 result, and why it is the most important thing in this beat

Adler 1999, Molecular and Cellular Biology 19(10):7050-60, PMID 10490642 1. From the abstract, read in full:

  • Direct interaction. "Using the yeast two-hybrid system and human cell culture coimmunoprecipitation experiments, we show here that HRX proteins interact directly with the GADD34 protein." HRX is MLL/KMT2A/ALL-1/HTRX.
  • GADD34 is pro-apoptotic and the domain is mapped. Transfected cells overexpressing GADD34 show significantly increased apoptosis after ionising radiation, and "the amino-terminal third of the GADD34 protein was necessary for this observed increase".
  • The fusions and the wild type do OPPOSITE things. "Coexpression of three different HRX fusion proteins (HRX-ENL, HRX-AF9, and HRX-ELL) had an anti-apoptotic effect, abrogating GADD34-induced apoptosis. In contrast, expression of wild-type HRX gave rise to an increase in apoptosis."
  • The authors' own framing. "These studies demonstrate, for the first time, a gain of function for leukemic HRX fusion proteins compared to wild-type protein."
  • And a second partner: GADD34 binds human SNF5/INI1, a SWI/SNF subunit.

Why this matters to this beat specifically. GADD34 is the adaptor that returns phospho-eIF2alpha to its unphosphorylated state, i.e. it is the OFF switch for the translational mechanism that controls ATF4. This beat's central inference is that menin inhibition removes ATF4 activity rather than ATF4 transcript, and the only mechanism that does that in the ISR is eIF2alpha phosphorylation. So the one ISR component with a demonstrated physical contact with a KMT2A fusion protein is the component that sets eIF2alpha phosphorylation. That is either a coincidence or the mechanism, and nobody has looked since 1999.

Three things this does NOT establish, stated plainly.

  1. The fusions are wrong. HRX-ENL, HRX-AF9 and HRX-ELL are KMT2A::MLLT1, KMT2A::MLLT3 and KMT2A::ELL. KMT2A::AFF1, the index fusion, was not tested.
  2. The system is transfection and overexpression, in 1999, before the taspase1 cleavage products were known, and the apoptosis readout is ionising-radiation-induced in a non-leukaemic host.
  3. Nothing connects it to the ISR. GADD34's role as the eIF2alpha phosphatase adaptor was established in the same era but the Adler work frames GADD34 as a DNA-damage apoptosis gene, not an ISR gene. The ISR reading is this entry's inference, not the paper's claim, and it is graded accordingly.
Why the graph could not see it: a rule-71 false absence

Measured 2026-09-22, every querytranslation read back 3:

querynverdict
PPP1R15A[tiab] AND the eight fusion operands0non-empty translation, "No items found." - a well-formed, false absence
GADD34[tiab] AND the same eight operands2PMID 10490642 (the Adler paper) and PMID 20858700 (MDA-7/IL-24 apoptosis in myeloid leukaemia)
PPP1R15A[tiab] alone181-
GADD34[tiab] alone57853 shared with PPP1R15A, Jaccard 0.0751

The canonical HGNC symbol reaches 181 records; the old name reaches 578, three times as many, and the fusion-relevant paper is only in the second set - the paper predates the PPP1R15A symbol entirely. This is rule 71 in its purest form: a query that is accurate, correctly formed, controlled and still false, because the name list was wrong.

And it is the case that matters most. Of the fourteen genes this beat queried against the fusion operand list, thirteen alias-widenings added nothing or added only noise. This one added the only direct physical KMT2A-fusion interaction in the entire panel. The lesson is not "always widen" - widening DDIT3 to CHOP added twelve off-topic records - but that a zero on a gene whose canonical symbol postdates the relevant literature must be re-run under the historical name before it is recorded.

The transcript, which is the least interesting part
datasetcellarmDMSOlog2FC
array 202014_atRS4;11MI-2-26.31-0.84
array 37028_atRS4;11MI-2-26.49-0.65
RNA-seqRS4;11 d7VTP-504696.90-0.29
RNA-seqRS4;11 d7EPZ56766.90-0.86
RNA-seqRS4;11 d2VTP-504695.85+0.32
RNA-seqMOLM13 d2VTP-504694.54+1.14
RNA-seqMOLM13 d7VTP-504695.19+0.54

PPP1R15A is the one panel member that moves in OPPOSITE directions in the two cell lines - down in RS4;11 on the array, up 1.14 log2 in MOLM13 at day 2. It is not one of the 24 concordant genes and would not be under any threshold. This entry does not rest on the transcript; it rests on the 1999 protein interaction, and the transcript is recorded because a flat, discordant transcript is exactly what a post-translationally regulated feedback node should look like.

And it matters for the ISR argument: if GADD34 transcript is flat while ATF4 output collapses 5 log2, the feedback arm is not being transcriptionally reconfigured, which again points at phospho-eIF2alpha itself as the quantity to measure.

Facts and figures
uniprotO75807
length AA674
entry namePR15A_HUMAN
protein nameProtein phosphatase 1 regulatory subunit 15A
ensembl gene IDENSG00000087074
PDB noteUniProt O75807 cross-references four PDB entries and NONE was fetched. No chain, resolution, residue range or interface is claimed from them.
alphafold IDAF-O75807-F1
domains noteUniProt O75807's feature table carries NO Domain features, so neither the N-terminal amphipathic intramembrane region nor the PP1-binding region has annotated integer boundaries here and none is invented. Adler 1999 maps the pro-apoptotic activity to 'the amino-terminal third' - a verbal description, not residue numbers, so it is NOT converted into a domain lane. Lane deliberately EMPTY per the integer-residue rule.
accession resolutiongene_exact:PPP1R15A AND organism_id:9606 AND reviewed:true returns EXACTLY ONE reviewed row, O75807, PR15A_HUMAN, 674 aa, synonym GADD34; re-fetched alone with name and length compared back: PASS.
localizationENDOPLASMIC RETICULUM MEMBRANE and MITOCHONDRION OUTER MEMBRANE, associated via an N-terminal amphipathic intramembrane region - and BOTH STATEMENTS ARE UNCODED. UniProt O75807's subcellular-location comment carries NO evidence codes at all on either location. Per rule 91 the uncoded statements are recorded rather than discarded, and they are labelled as uncoded. No PPP1R15A localisation experiment exists in any haematopoietic cell.
subnuclear localizationNOT APPLICABLE AND NOT ASKABLE. PPP1R15A has no annotated nuclear pool, coded or uncoded - it is a membrane-associated protein. There is no subnuclear question. BUT NOTE THE TENSION: Adler 1999 reports a direct interaction with KMT2A, which is nuclear, and with SNF5/INI1, which is nuclear. Either the interaction occurs in a pool nobody has annotated, or it is an artefact of the transfection system - and resolving that is the single cheapest experiment this entry points at.
cell typeRS4;11 (t(4;11) KMT2A::AFF1) and MOLM13 (KMT2A::MLLT3) measured here, with SEM and REH on the array [E2]. The 1999 interaction work is in yeast and in unnamed transfected human cell culture [E1] - not leukaemia, not a fusion-positive cell. No primary leukaemia sample, no normal haematopoietic cell, and no KMT2A::AFF1 cell anywhere in this protein's interaction record.
cell stateThe transcript was measured in the day-2 and day-7 drug-arrested states and is discordant between them and between lines. The 1999 functional result is in the IONISING-RADIATION-DAMAGED state, which is a DNA-damage state and not the amino-acid or ER-stress state the ISR responds to - so the one functional KMT2A-fusion result on this gene is in a different cell state from the one this beat is about, and that gap is the reason its ISR interpretation is graded moderate.
epigenetic readoutNo writer, eraser or reader activity of its own. Its chromatin connection is by INTERACTION: GADD34 binds human SNF5/INI1 (SMARCB1), a core SWI/SNF subunit, and binds KMT2A itself - both from PMID 10490642, both in transfection systems. This graph already holds SWI/SNF entries, so a GADD34-SMARCB1-KMT2A triangle is a specific, unexplored structure. No histone-mark measurement at the PPP1R15A locus exists in any KMT2A-rearranged cell.
shown for fusionSHOWN, for the WRONG FUSIONS: a direct interaction with HRX/KMT2A and abrogation of GADD34-induced apoptosis by KMT2A::MLLT1, KMT2A::MLLT3 and KMT2A::ELL, in transfected cells in 1999, with wild-type KMT2A acting oppositely. KMT2A::AFF1 was NOT tested. Never reproduced. The transcript was measured here in KMT2A::AFF1 and KMT2A::MLLT3 and is flat and discordant.
absence classnever-measured
absence class noteRECLASSIFIED 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: never-measured - the prose absence was FALSE and is now CLOSED (the historical-symbol query returned the paper). The four absences that remain - no reproduction in 27 years, no KMT2A::AFF1 test, no phospho-eIF2alpha in any fusion context, no endogenous co-IP - are all experiments nobody has done. ORIGINAL absence_class, verbatim: «FALSE ABSENCE, NOW CLOSED. The canonical-symbol query returned a well-formed zero; the historical-symbol query returned the paper. The remaining genuine absences are: no reproduction of the 1999 interaction in twenty-seven years, no KMT2A::AFF1 test, no phospho-eIF2alpha measurement in any fusion context, and no endogenous co-IP.»
PDB IDS
[
  "4XPN",
  "7NXV",
  "7NZM",
  "8QZZ"
]
domains
[]
the kmt2a fusion interaction
{
  "source": "PMID 10490642, Adler 1999, Molecular and Cellular Biology 19(10):7050-60",
  "methods": "yeast two-hybrid; co-immunoprecipitation in human cell culture",
  "direction": "direct interaction between HRX (MLL/KMT2A) proteins and GADD34",
  "functional_result": "HRX-ENL, HRX-AF9 and HRX-ELL abrogate GADD34-induced apoptosis; wild-type HRX increases apoptosis",
  "authors_framing": "the first gain of function demonstrated for leukaemic HRX fusion proteins",
  "second_partner": "GADD34 also binds human SNF5/INI1 (SMARCB1), a SWI/SNF subunit",
  "index_fusion_tested": "NO. KMT2A::AFF1 was not among the three fusions tested.",
  "citation_status_in_this_graph": "NOT CITED ANYWHERE before this cycle (scripts/check-citation-reuse.py, 2026-09-22)",
  "why_it_was_missed": "PPP1R15A[tiab] AND the fusion operand list returns a well-formed ZERO with a valid querytranslation and 'No items found.'; GADD34[tiab] AND the same list returns 2 including this paper. The paper predates the PPP1R15A symbol.",
  "the_inference_that_is_NOT_the_paper_s": "That this connects the KMT2A fusion to the ISR. Adler frames GADD34 as a DNA-damage apoptosis gene. The eIF2alpha-phosphatase reading is this entry's, and is graded moderate."
}
the measurement
{
  "rs411_array_mi22_lfc_202014_at": -0.844,
  "rs411_array_mi22_lfc_37028_at": -0.654,
  "rs411_rnaseq_d7_vtp_lfc": -0.294,
  "rs411_rnaseq_d7_epz5676_lfc": -0.856,
  "rs411_rnaseq_d2_vtp_lfc": 0.316,
  "molm13_rnaseq_d2_vtp_lfc": 1.139,
  "molm13_rnaseq_d7_vtp_lfc": 0.539,
  "verdict": "DISCORDANT between lines and timepoints; not one of the 24 concordant genes under any threshold. No directional claim is made. A flat, post-translationally regulated feedback node is what this looks like."
}
essentiality
{
  "depmap_release": "26.06 (read back from meta{dataVersion})",
  "depmap_screens": 1258,
  "chronos_mean": -0.1127,
  "pct_lines_below_minus_0_5": 1.6,
  "leukaemia_screens": 57,
  "leukaemia_mean_gene_effect": -0.0789,
  "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). NEW, and it is the highest-value trap in this beat.",
  "GADD34": "578 PubMed records against PPP1R15A's 181, sharing 53 (Jaccard 0.0751). The historical name reaches THREE TIMES as many records as the canonical symbol, and the only direct KMT2A-fusion interaction paper in this beat's entire panel is reachable ONLY under GADD34.",
  "the_generalisable_rule": "A well-formed PubMed zero - non-empty querytranslation, warninglist 'No items found.' - is still FALSE if the gene's canonical symbol postdates the relevant literature. Rule 3's test passed and the answer was wrong. Always re-run a fusion-scoped zero under the gene's HISTORICAL symbol.",
  "array": "GPL570 carries two PPP1R15A probes and no probe whose own symbol field is GADD34; both sit near the detection floor at 6.31 and 6.49 log2 RMA.",
  "uniprot": "clean - exactly one reviewed human record"
}
How this could bite us back
  • If KMT2A fusion proteins hold GADD34 and block its apoptotic function, then the fusion has already reset the cell's eIF2alpha-dephosphorylation feedback before any drug is given - so a menin inhibitor is acting on a stress axis that is not in its default state, and the response may not generalise to a cell without the fusion.
  • GADD34 is the target of the only ISR-directed drug class with real clinical history: ISRIB and related compounds act on eIF2B to override eIF2alpha phosphorylation, and guanabenz and sephin1 were developed as GADD34 inhibitors. Any attempt to combine an ISR-axis drug with a menin inhibitor is entering a system where the fusion may already bind the feedback node.
  • The apoptosis this gene controls in the 1999 paper is ionising-radiation-induced. If KMT2A fusions block GADD34-dependent apoptosis, that is a radioresistance mechanism, and total body irradiation is part of some KMT2A-rearranged transplant conditioning - a connection nobody has drawn and which this entry cannot support beyond noting it.
Still open
  1. Does KMT2A::AFF1 bind GADD34, and does it abrogate GADD34-induced apoptosis as KMT2A::MLLT1, KMT2A::MLLT3 and KMT2A::ELL do? The index fusion was never tested, and this is the only published gain-of-function contact in the ISR panel.
  2. Does the 1999 interaction reproduce at endogenous levels, with a modern co-IP or proximity labelling, and which taspase1 cleavage product is involved? The experiment predates the N320/C180 cleavage model entirely.
  3. Is phospho-eIF2alpha altered in KMT2A-rearranged cells relative to a wild-type line, and does it change under a menin inhibitor? If GADD34 is held by the fusion, the resting eIF2alpha phosphorylation state of a fusion-positive cell is not the default one.
  4. What is the GADD34-SMARCB1-KMT2A triangle? GADD34 binds both, both are in this graph, and nobody has asked whether the three are in one complex or in two separate pairwise contacts.
  5. In which compartment does a membrane-associated protein meet a nuclear one? Either GADD34 has an unannotated nuclear pool or the 1999 interaction is a transfection artefact, and one immunofluorescence experiment distinguishes them.
Every source behind this entry (4)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    HRX (MLL/KMT2A) proteins interact directly with GADD34 by yeast two-hybrid and by co-immunoprecipitation in human cell culture; GADD34 overexpression increases apoptosis after ionising radiation and requires its amino-terminal third for that effect; coexpression of HRX-ENL, HRX-AF9 or HRX-ELL abrogates GADD34-induced apoptosis while wild-type HRX increases apoptosis, which the authors present as the first gain of function demonstrated for leukaemic HRX fusion proteins; and GADD34 also binds human SNF5/INI1.

    Leukemic HRX fusion proteins inhibit GADD34-induced apoptosis and associate with the GADD34 and hSNF5/INI1 proteins. · Adler HT, Chinery R, Wu DY, Kussick SJ, et al. · Molecular and Cellular Biology 19(10):7050-60 1999
    PMID 10490642DOI
    What this evidence cannot show

    ABSTRACT READ IN FULL; the figures were not. THE FUSIONS ARE THE WRONG ONES: HRX-ENL (KMT2A::MLLT1), HRX-AF9 (KMT2A::MLLT3) and HRX-ELL (KMT2A::ELL) were tested and KMT2A::AFF1, the index fusion of this project, was NOT. TRANSFECTION AND OVEREXPRESSION throughout, in 1999, before taspase1 cleavage of KMT2A was described, so which species of HRX was in the co-IP is unknown. The apoptosis readout is ionising-radiation-induced in a non-leukaemic host cell. NOT PREVIOUSLY CITED IN THIS GRAPH (checked with scripts/check-citation-reuse.py, 2026-09-22: 'NOT CITED HERE - genuinely new to this graph'). The ISR interpretation of this result is THIS ENTRY'S INFERENCE and not the paper's claim - Adler frames GADD34 as a DNA-damage apoptosis gene.

    Evidence that argues against this (1)
    • A 1999 yeast two-hybrid plus overexpression co-IP is the weakest tier of interaction evidence and this one has, as far as a fusion-scoped PubMed query can tell, never been independently reproduced in twenty-seven years. Yeast two-hybrid against a protein as large and as sticky as KMT2A is exactly the setting that produces non-physiological interactions, and the absence of any follow-up is itself weak evidence against the interaction being important. NCBI PubMed 2026
  2. E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.dataset

    PPP1R15A moves -0.84 and -0.65 log2 on two GPL570 probes under MI-2-2 in RS4;11 but only -0.29 under VTP-50469 at day 7, +0.32 at day 2, and +1.14 in MOLM13 at day 2 - discordant in direction between the two cell lines - and it is not one of the 24 concordant genes under any threshold.

    What this evidence cannot show

    A count inside named cohorts. The values are small and the direction is inconsistent across lines and timepoints, so NO directional claim is made - the finding is that the transcript is uninformative, which is what a post-translationally regulated feedback node should look like. RS4;11 array DMSO arm n=2. Both array probes sit near the 6.0 log2 detection floor used here, so they are the least reliable intensity regime on the platform.

    Evidence that argues against this (1)
    • A flat transcript is equally consistent with the feedback arm being irrelevant to this drug. Reading 'flat GADD34 mRNA' as support for a post-translational mechanism is unfalsifiable from these data: any result would have been compatible with the hypothesis, which is the definition of a non-test. NCBI GEO, re-analysed here 2026
  3. E3established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    PubMed returns 0 records for PPP1R15A jointly with an eight-operand KMT2A-fusion list, with a non-empty querytranslation and warninglist 'No items found.', while the historical alias GADD34 returns 2 records against the same list including PMID 10490642; GADD34 alone reaches 578 records against PPP1R15A's 181, sharing 53 (Jaccard 0.0751).

    What this evidence cannot show

    Rule 3 satisfied on the zero and it was STILL FALSE - this is the clean demonstration that a well-formed zero with a valid querytranslation and 'No items found.' is not a true absence when the name list is wrong (rule 71). Both GADD34-scoped titles were fetched and read. Rule 94 form; rule 95 observed. Rule 80 still applies: titles and abstracts only. NOT in data/symbol-collisions.json (157 rows read 2026-09-22).

    Evidence that argues against this (1)
    • Widening a name list is not uniformly beneficial and this entry should not be read as licensing it: in the same session, widening DDIT3 to CHOP added twelve records of which zero were about the gene. The defensible rule is narrower - re-run under the HISTORICAL symbol when the canonical symbol postdates the relevant literature - and even that would have to be applied gene by gene. NCBI PubMed 2026
  4. E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

    UniProt O75807 is the single reviewed human PPP1R15A record, 674 aa, synonym GADD34, which recruits the phosphatase PPP1CA to prevent excessive phosphorylation of eIF2alpha and thereby reverses the stress-induced shut-off of protein synthesis and facilitates recovery from stress; it associates with the endoplasmic reticulum and mitochondrial outer membranes via an N-terminal amphipathic intramembrane region. DepMap 26.06 mean gene effect -0.1127 across 1,258 screens, 1.6% of lines below -0.5, leukaemia mean -0.0789.

    What 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. The membrane-association statements carry NO evidence codes at all in the location comment block - they are UNCODED, and per rule 91 that is recorded rather than discarded. The DepMap figures come from the same session's Open Targets query with the release read back (rule 83). The eIF2alpha-phosphatase function is cited from a curated database line whose primary papers (PubMed 26095357, 26742780) were not opened.

    Evidence that argues against this (1)
    • If GADD34 is the eIF2alpha-phosphatase adaptor and this beat's hypothesis is that eIF2alpha phosphorylation controls the observed ATF4-output loss, then GADD34's near-zero DepMap gene effect and flat transcript are both unsupportive: a cell that has lost ATF4 output should not be indifferent to the state of the feedback arm. The convergence of the 1999 interaction and the ISR mechanism may be a coincidence of one protein having two unrelated jobs. DepMap / Open Targets Platform 2026
Connected to