GPT2 (Q8TD30) - the EARLIEST-moving and third-most menin-selective gene in the serine arm, and the one REH cannot control because it is below the array floor there

This is real experimental data — atoms placed by crystallography or cryo-EM, deposited by the people who did the experiment.

• **Germline biallelic GPT2 loss causes a neurodevelopmental disorder with severe psychomotor delay, poor or absent speech, spastic paraplegia and PROGRESSIVE microcephaly.** 'Progressive' is the word that matters for a paediatric intervention: unlike a fixed malformation, this is a phenotype that worsens postnatally, which is exactly the window in which a drug would act.
• **This gene's fusion-dependence has never been tested and cannot be tested in the available control line.** It is one of three genes carrying the beat's menin-selectivity signal, and for it the fusion-selectivity argument rests on nothing. Any use of GPT2 as evidence for a fusion-selective serine programme must say so.
• **GPT2 is not a target and has no inhibitor.** 0.5% of 1,258 DepMap lines fall below -0.5, its paralogue was not measured, and the fusion-scoped and menin-scoped literature are both zero.
• **A transaminase's transcript is the panel's most likely reporter of metabolism rather than of chromatin.** If GPT2's early, menin-selective fall reflects a glutamate-pool change, it is a consequence of the serine programme rather than a co-regulated member of it - and it would then be evidence that a metabolic shift is happening fast, which is interesting and is not the two-input mechanism this beat proposes.

1. How can GPT2's fusion-dependence be tested at all? REH does not express it above the array floor and the RNA-seq series has no KMT2A-wild-type line, so one of the beat's three menin-selective genes has no available wild-type control. A second wild-type B-ALL line that does express GPT2 - NALM-6 is the obvious candidate and is already used in the PSAT1 literature - would close it, and a chromatin measurement at the locus would substitute.
2. Is GPT2 mitochondrial? UniProt carries no compartment statement, the literature consensus says mitochondrial, and no primary source was opened here. If it is, this beat's observation that compartment does not predict compound-selectivity has to be restated, because GPT2 is the third-most menin-selective gene in the arm.
3. Is GPT2's transcript responding to chromatin or to metabolite pools? A transaminase reports on glutamate and 2-oxoglutarate by design, and GPT2 is the earliest mover in the arm. A metabolite measurement at day 1 and day 2 under a menin inhibitor would separate these, and it is the same tracing experiment the rest of the beat needs.
4. Does the paralogue GPT compensate? It was not in the declared panel. The SHMT1/SHMT2 case in this same panel shows a paralogue can hide a requirement completely, and GPT2's 0.5% DepMap fraction is uninterpretable without it.

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

   On a pre-registered panel GPT2 falls 1.54 log2 in RS4;11 at day 7 under VTP-50469 330 nM from a detected baseline of 7.65 log2 CPM against 0.51 under EPZ5676 - a margin of -1.031 with 33% reproduced, third-largest in the serine arm - with 58% of the day-7 effect already present at day 2 (-0.902), the earliest share of any serine gene and inside the menin-target range; it is -2.34 against SGC0946 -1.29 in MV4;11 and -0.21 in MOLM13 from a detected baseline of 7.48; and on the GPL570 array its single probe 224839_s_at reads 10.78 log2 RMA in RS4;11 (MI-2-2 -0.43) and 10.35 in SEM (-0.87) but 5.17 IN REH, BELOW the 6.0 log2 RMA floor, so REH cannot serve as the KMT2A-wild-type control for this gene and the RNA-seq series contains no wild-type line at all.

   analyses/q109-is-the-serine-collapse-downstream-of-atf4-or-parallel - panel declared to disk before any expression value was read and loaded by run.py; GSE127506+GSE127508, GSE63664 (every probe, own baseline) and GSE260456 · this repository 2026
   ▸What this evidence cannot show

   One line per platform, single doses, NO dose-response, n=2-3 per arm, day 7 of continuous exposure so arrest is a competing explanation. No series records the amino-acid composition of the medium, and this is an amino-acid-response panel. MI-2-2 in the array arm is the WEAK compound, so the REH wild-type contrast has little dynamic range. GPT2 has NO usable KMT2A-wild-type control in this beat: REH is below the array floor and no wild-type line exists in the RNA-seq series. Its fusion-dependence is therefore untested, which is a significant gap for one of the three genes carrying the beat's menin-selectivity signal. GPT2 has only ONE probe on GPL570, so the every-probe discipline yields a single measurement and there is no internal probe agreement to check.

   ▸Evidence that argues against this (1)

   • An early, menin-selective fall in a transaminase is equally consistent with GPT2 being a direct menin-complex target and with GPT2 responding rapidly to any change in glutamate or 2-oxoglutarate pools - and a menin inhibitor at day 2 is already changing the transcriptional programme broadly. A transaminase reports on metabolite pools by design, so it is the panel member whose transcript is most likely to be responding to metabolism rather than to chromatin. this repository 2026
2. E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

   GPT2 DepMap gene effect is mean -0.0882, median -0.0858, with 0.5% of 1,258 screens below -0.5 and a leukaemia mean of -0.0406 over 57 screens, with all 29 tissue blocks flattened and approvedSymbol read back as GPT2.

   Open Targets Platform GraphQL target.depMapEssentiality, all 29 tissue blocks flattened, panel-maximum payload guard PASS at 1,258 screens; analyses/q109-serine-one-carbon-depmap-dependency/results.json · Open Targets / DepMap 2026
   ▸What this evidence cannot show

   Rule 83: the payload carries NO release identifier, so depmap_release is recorded as null and absolute values drift quarterly; the cross-gene contrast is the durable part. Leukaemia screens are matched on the payload's own diseaseFromSource string against four substrings and are NOT split by KMT2A status. Rule 84: no median-split proportion is reported - the percentage is against an absolute -0.5 threshold. A CRISPR knockout in cysteine- and serine-replete medium understates a biosynthetic dependency. GPT2's leukaemia mean (-0.0406) is SHALLOWER than its pan-lineage mean (-0.0882), i.e. leukaemia lines need it slightly less than average - the opposite direction from MTHFD1, GLS and MTHFD2 in the same fetch. With values this close to zero that difference is not interpretable as an effect and is reported only to prevent it being quoted as one.

   ▸Evidence that argues against this (1)

   • The paralogue GPT (alanine aminotransferase 1) was NOT in this beat's pre-registered panel, so redundancy is unmeasured for GPT2 exactly as it is for ALDH1L2 - and the SHMT1/SHMT2 case in this same panel shows how completely a paralogue can hide a requirement. A 0.5% fraction on a gene with an unmeasured paralogue is not evidence of dispensability. Pikman Y et al. 2022
3. E3established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

   UniProt reviewed human record Q8TD30 (ALAT2_HUMAN) has primary gene name GPT2 with synonyms AAT2 and ALT2, length 523 aa, one PDB cross-reference, catalyses the reversible transamination between alanine and 2-oxoglutarate to form pyruvate and glutamate, is expressed at high levels in muscle, adipose tissue, kidney and brain and at lower levels in liver and breast, carries NO SUBCELLULAR LOCATION comment of any kind, and records a neurodevelopmental disorder with spastic paraplegia and microcephaly - an autosomal recessive syndrome of severe psychomotor developmental delay, dysarthria, walking difficulties, moderately to severely impaired intellectual development, poor or absent speech and progressive microcephaly.

   UniProtKB entry Q8TD30 (ALAT2_HUMAN), Alanine aminotransferase 2 · UniProt Consortium · UniProt 2026
   ▸What this evidence cannot show

   Resolved 2026-09-22 with gene_exact:GPT2 AND organism_id:9606 AND reviewed:true; exactly one reviewed row, gene_primary and length read back. The absent SUBCELLULAR LOCATION comment is a fact about the database - GPT2 joins PHGDH and PSAT1 as a panel enzyme with no curated compartment at all, which is notable because GPT2 is widely described as mitochondrial in the literature and UniProt records nothing. No compartment is asserted here on the strength of that literature because no primary source was opened for it. The single PDB entry was not fetched.

   ▸Evidence that argues against this (1)

   • GPT2 is commonly described as the MITOCHONDRIAL alanine aminotransferase, opposite the cytosolic GPT - and UniProt carries no compartment statement at all, so the compartment most readers would assume is unsupported in the database this project resolves against. If GPT2 is mitochondrial, the beat's observation that compartment does not predict compound-selectivity would need restating, because GPT2 is the third-most menin-selective gene in the arm and would then be a mitochondrial one. UniProt Consortium 2026
4. E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

   Biallelic GPT2 mutations cause developmental encephalopathy with microcephaly and features of complicated hereditary spastic paraplegia; and a separate report extends the clinical phenotype and estimates population prevalence. PubMed 2026-09-22, one tag per term with translations read back: GPT2[tiab] returns 326 records; GPT2[tiab] AND the eight-operand KMT2A-fusion set returns 0 with a non-empty querytranslation and warninglist 'No items found.'; GPT2[tiab] AND the nine-operand menin/menin-inhibitor set returns 0 with the same signature; GPT2[tiab] AND leukemia[tiab] returns 2, both titles read - an IGF2BP2 glutamine-metabolism AML paper and a Chinese-language study of GPT2 in HL-60 cells; and the alias ALT2[tiab] returns 119 with 0 fusion-scoped.

   GPT2 mutations cause developmental encephalopathy with microcephaly and features of complicated hereditary spastic paraplegia. · Hengel H, Keimer R, Deigendesch W, et al., Schols L · Clin Genet 2018
   PMID 29882329DOI
   ▸What this evidence cannot show

   The GPT2 disease claim is taken from this paper's TITLE and from UniProt's curated DISEASE comment [E3]; the abstract was NOT read and neither was the prevalence paper (PMID 31471722, metadata verified). The literature counts are bounded by titles and abstracts only, with supporting figures not searched. The two leukaemia-scoped titles were read but not their abstracts; the HL-60 paper is in Chinese and was not read at all, which is a bounded rejection rather than a judgement.

   ▸Evidence that argues against this (1)

   • The one leukaemia paper that could bear on GPT2's role here is an IGF2BP2 glutamine-metabolism study in AML (PMID 36306790, Cancer Cell 2022) - and glutamine feeds the glutamate pool GPT2 transaminates, while this graph already holds gene-igf2bp3 from the same protein family and a serine-biosynthesis AML paper driven by IGF2BP3 (PMID 40328743). Both were left unread, and together they are the likeliest place for an existing GPT2-adjacent measurement in leukaemia. Weng H et al. 2022

• part of →PSAT1 (Q9Y617) - a DIRECT ATF4 target, and the gene that shows the brief's menin-selectivity claim was PHGDH's and not the pathway's

  PSAT1 consumes glutamate to transaminate 3-phosphohydroxypyruvate and GPT2 regenerates it; GPT2 is the earliest mover in the arm at 58% of its day-7 effect by day 2

• depends on →MEN1 / menin — the adaptor the fusion cannot work without

  third-largest menin-selective margin in the panel (-1.03, only 33% reproduced by EPZ5676) and the earliest kinetics

• related to →IGF2BP3 — an RNA-binding protein required for MLL-Af4 leukaemogenesis in mice that CRISPR barely touches in human lines

  the nearest leukaemia literature to GPT2 is m6A-reader-driven glutamine and serine metabolism in AML, and this graph holds gene-igf2bp3; both papers left unread

• confers resistance to →Pinometostat (EPZ-5676) — the DOT1L inhibitor, and the most instructive failure in this field

  EPZ5676 reproduces only 33% of the VTP-50469 effect on GPT2, the lowest fraction of any detected serine gene

• ← supportsThe serine/one-carbon arm is NOT a pure ATF4 output: it is menin-selective where the ISR arm is compound-independent (median margin -0.55 vs -0.02, p=0.011), and it is ABSENT in KMT2A::MLLT3 AML where the ISR arm is present

  third-largest margin and earliest kinetics; REH cannot control it