EIF2AK4 / GCN2 (Q9P2K8) - the amino-acid-sensing eIF2alpha kinase: transcript flat, and NOTHING has ever been published about it in any KMT2A-rearranged system under either name

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.

• ASNS falls 1.72 log2 under a menin inhibitor, and asparagine synthetase level is the main determinant of asparaginase sensitivity. That predicts a menin inhibitor SENSITISES to asparaginase - a rational, cheap, untested combination in exactly the disease where asparaginase is already backbone therapy. It also predicts the reverse risk: if the ISR arm recovers, ASNS recovers and asparaginase resistance returns.
• GCN2 carries a 9.9% tail of dependent DepMap lines, the only panel member other than ATF4 to reach double digits. If any of those lines are KMT2A-rearranged, a flat transcript would have hidden a real dependency - and nobody has looked, because nothing has ever been published about this kinase in this disease.
• The amino-acid arm splits between DOT1L-shared genes (CHAC1, TRIB3, ASNS) and menin-selective ones (PHGDH), and a single amino-acid sensor cannot produce both patterns. Part of what this graph calls the ISR arm is probably a fusion-programme arm wearing ISR clothing, and treating it as one mechanism will make any intervention argument built on it wrong in a way that is hard to detect.

1. Is GCN2 active in a KMT2A-rearranged cell, and does a menin inhibitor change it? Nothing has ever been published about this kinase in this disease under either name, GCN2 inhibitors and activators are both in development, and this is a clean unoccupied question in the index disease.
2. Does asparaginase, the canonical GCN2-activating stimulus and a backbone agent in KMT2A-rearranged ALL, interact with a menin inhibitor on this axis? ASNS falls 1.72 log2 under VTP-50469, which by itself predicts increased asparaginase sensitivity, and no experiment has combined the two agents.
3. Which DepMap lines make up EIF2AK4's 9.9% dependent tail, and are any of them KMT2A-rearranged? A 9.9% tail on a near-zero mean is the signature of a context-specific dependency, and this beat did not identify the lines.
4. Was the amino-acid state of the medium the same in the DMSO and drug arms at day 6-7? The drug arrests growth, so it changes medium exhaustion, and an amino-acid-response programme measured under differing amino-acid supply has an alternative explanation that no reported metadata can exclude.

1. E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.dataset

   EIF2AK4's transcript does not move under either inhibitor in any line or at either timepoint: RS4;11 array MI-2-2 +0.12 and +0.05 on two well-expressed probes (DMSO 8.96 and 8.07), SEM +0.31, REH -0.03, RNA-seq RS4;11 day 7 VTP-50469 +0.10 and EPZ5676 -0.07, day 2 +0.23, MOLM13 day 7 -0.38 - while in the same RS4;11 day-7 samples the amino-acid-response programme it senses falls 1.09 to 5.11 log2 (PHGDH -3.24, PSAT1 -2.40, ASNS -1.72, SLC7A5 -1.34, SLC3A2 -1.26, SHMT2 -1.09, CHAC1 -5.11, TRIB3 -3.10).

   Q106 pre-registered ISR panel run over the cached GSE63664 (MI-2-2 / EPZ004777, SEM / RS4;11 / REH, GPL570 RMA) and GSE127506+GSE127508 (VTP-50469 330 nM / EPZ5676 1 uM, RS4;11 and MOLM13, d2 and d7, 3-prime-end RNA-seq) matrices. Script analyses/does-atf4s-own-transcript-move-under-a-menin-inhibitor/run.py, results.json, panel frozen in panel-preregistration.json BEFORE any expression value was read. EIF2AK4 probes 225164_s_at and 237145_at; RNA-seq row EIF2AK4; the amino-acid arm rows. · NCBI GEO, re-analysed here 2026
   ▸What this evidence cannot show

   THE MOST SOLIDLY FLAT RESULT IN THE PANEL - two well-expressed array probes, a well-expressed RNA-seq row, no sign flip, no floor effect, platforms agreeing on the baseline. But the bound is TRANSCRIPTIONAL ONLY: GCN2 is activated by dimerisation and autophosphorylation in response to uncharged tRNA, entirely post-translationally, so a flat transcript is compatible with GCN2 activity having risen, fallen or not moved. AND THE CONFOUND BITES HARDEST HERE: GCN2's physiological input is amino-acid supply, and neither public experiment controlled or reported the amino-acid composition or exhaustion state of the medium at day 6-7, in cultures of differing density. An amino-acid sensor was measured under uncontrolled amino-acid supply. RS4;11 array DMSO arm n=2.

   ▸Evidence that argues against this (1)

   • The sensor-untouched/programme-collapsed argument assumes the amino-acid-response genes here are being driven by GCN2 at all. In a fusion-driven leukaemia they may be direct or indirect targets of the fusion complex - PHGDH and PSAT1 are named in the menin-ATF4 serine-biosynthesis paper, and PHGDH's 3.24 log2 fall is strongly menin-selective (EPZ5676 only -1.52), which looks more like a fusion-programme gene than an amino-acid-sensing output. If the arm is fusion-driven rather than GCN2-driven, a flat GCN2 explains nothing. Jimenez JA et al. 2021
2. E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

   PubMed returns zero records for EIF2AK4 and zero for GCN2 jointly with an eight-operand KMT2A-fusion list, and zero for each against a thirteen-name menin-inhibitor list - four independent zeros, every one with a non-empty querytranslation and warninglist 'No items found.'; GCN2[tiab] reaches 1,170 records against EIF2AK4's 180, sharing 54 (Jaccard 0.0417), so the absence holds under both the canonical symbol and the common name.

   NCBI PubMed esearch, run 2026-09-22, with querytranslation and warninglist read back from the response for every query quoted · NCBI PubMed 2026
   ▸What this evidence cannot show

   Rule 3 satisfied on all four zeros. THE ALIAS CHECK WAS RUN AND THE ZERO HELD, which in this beat is the exception: the same procedure broke the zero for PPP1R15A (GADD34 -> 2 records, including a direct KMT2A-fusion interaction) and for EIF2AK3 (PERK -> 5 records, including the only functional paper). Rule-67 controls in the same session were non-empty. Rule 94 form; rule 95 observed. RULE 80 IS DOING REAL WORK HERE: a GCN2 or phospho-GCN2 western in a supporting figure of a menin-inhibitor paper would be invisible to all four queries, and a phospho-kinase western is exactly the kind of panel that never reaches an abstract.

   ▸Evidence that argues against this (1)

   • The operand list contains no amino-acid-starvation, asparaginase, serine, one-carbon or ISR-pathway terms, and GCN2's relevance to this disease would most plausibly be reported in an asparaginase-resistance paper - asparaginase is a backbone agent in KMT2A-rearranged ALL and asparagine depletion is the canonical GCN2-activating stimulus. So this zero bounds the GCN2-named literature scoped to fusion operands, and does NOT bound the asparaginase literature, which was not searched and is substantial. NCBI PubMed 2026
3. E3established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

   In DepMap 26.06, EIF2AK4 has a mean CRISPR gene effect of -0.1168 across 1,258 screens with 9.9% of lines below -0.5 and a leukaemia mean of -0.1045 across 57 leukaemia screens - the second-deepest gene effect in this beat's panel after ATF4 (-0.7118, 71.9%), and the only panel member other than ATF4 with a double-digit percentage of dependent lines.

   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 2026
   ▸What this evidence cannot show

   Release recorded per rule 83; gene-specific screen count; leukaemia screens by substring match with ids persisted per rule 90. A mean of -0.12 is still not a dependency - what distinguishes EIF2AK4 is the 9.9% TAIL, i.e. a minority of lines that do depend on it, and this beat did NOT identify which lines or whether any are KMT2A-rearranged. The eleven KMT2A-rearranged DepMap lines were not separated out, and a previous cycle's failure to persist those line identifiers is recorded in this graph as an unresolved gap.

   ▸Evidence that argues against this (1)

   • A 9.9% tail on a gene with a mean near zero is the signature of a context-specific dependency, and the context is almost certainly amino-acid or nutrient stress rather than KMT2A rearrangement - GCN2 dependency is documented in amino-acid-auxotrophic and asparagine-synthetase-low cancers. Reading the tail as potentially fusion-related, without identifying the lines, would be exactly the kind of unlisted-denominator inference this graph has been burned by. DepMap / Open Targets Platform 2026
4. E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

   UniProt Q9P2K8 is the single reviewed human EIF2AK4 record, 1,649 aa with an RWD domain at 25-137 and two protein-kinase domains at 296-539 and 590-1001, synonyms GCN2 and KIAA1338, eight PDB entries, annotated as a metabolic-stress-sensing protein kinase that phosphorylates eIF2alpha in response to low amino-acid availability and acts as an activator of the integrated stress response required for adaptation to amino-acid starvation; its only subcellular-location statement is Cytoplasm, carrying ECO:0000250 by similarity to mouse Q9QZ05.

   UniProtKB - Q9P2K8, E2AK4_HUMAN, eIF-2-alpha kinase GCN2 · UniProt Knowledgebase 2026
   ▸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. RULE 91 FLAG: like PERK, the human localisation is ECO:0000250 BY SIMILARITY to the mouse orthologue with no human experimental code anywhere in the location block - recorded as an inference, not a measurement. The eight PDB entries were NOT fetched and nothing is claimed from them. The amino-acid-starvation function line carries ECO:0000269 citations (PubMed 25329545, 32610081) plus a 'By similarity' clause for the ISR-activator role; those papers were not opened.

   ▸Evidence that argues against this (1)

   • GCN2's canonical role was established largely in yeast and mouse, and the human record's ISR-activator statement is partly 'By similarity'. For a 1,649-residue kinase with two kinase domains - one of which is degenerate - the mechanistic detail that would matter for a drug argument is not settled by a database line, and this beat did not open the primary papers. UniProt Knowledgebase 2026

• regulates →ATF4 / CREB2 (P18848) - the hub this graph inferred from its targets and never wrote, and its own transcript does not move under either menin inhibitor

  GCN2 transduces low amino-acid availability into eIF2alpha phosphorylation and ATF4 induction; its transcript is the most solidly flat in the panel (max 0.38 log2) while the amino-acid-response programme it senses falls 1.09-5.11

• regulates →ASNS - the resistance gene for a backbone drug, never measured in this disease

  ASNS is the amino-acid-response output most relevant clinically here - it falls 1.72 log2 under VTP-50469, which predicts increased asparaginase sensitivity, and GCN2 is its sensor

• related to →Native E. coli asparaginase - the second-largest in-vitro resistance signal in infant ALL, and a different product from the pegylated one four doses later

  asparagine depletion is the canonical GCN2-activating stimulus and asparaginase is backbone therapy in KMT2A-rearranged ALL; no experiment has combined it with a menin inhibitor, and ASNS falls 1.72 log2 under one

• related to →KMT2A::AFF1 (MLL-AF4), t(4;11)(q21;q23)

  BOUNDED ABSENCE: four independent PubMed zeros (EIF2AK4 and GCN2, each against an eight-operand fusion list and a thirteen-name menin-drug list), all with valid querytranslations and 'No items found.'; the only fusion-context data are this beat's flat transcripts

• ← supportsATF4's own transcript does NOT move under either menin inhibitor - and the whole ISR arm is reproduced by a DOT1L inhibitor, so it is not a menin-interface readout

  the amino-acid-sensing kinase is the most solidly flat result in the panel (max 0.38 log2 on two well-expressed probes) while its downstream programme falls 1.09-5.11