Oncofusion

start / graph / gene-aldh1l2

Genemoderate confidence 3 of 5: One strong study, or several smaller consistent ones.Q109-serine-one-carbon

ALDH1L2 (Q3SY69) - the second-largest fall in the serine/one-carbon panel (-3.40), named nowhere in this graph, zero of 1,258 DepMap lines require it, and a tumour SUPPRESSOR where it has been studied

also called mitochondrial 10-formyltetrahydrofolate dehydrogenase · AL1L2_HUMAN · mtFDH

ALDH1L2 burns off spare one-carbon units inside mitochondria and makes NADPH while doing it. A menin inhibitor switches it off about ten-fold in KMT2A-rearranged leukaemia cells, which nobody had noticed - and losing it raises damaging oxygen radicals, so this is a fall that could be helping the cancer rather than hurting it.

The same thing, for a scientist

ALDH1L2 (Q3SY69, 923 aa, exactly one reviewed human record, no UniProt synonyms) is the **mitochondrial** 10-formyltetrahydrofolate dehydrogenase, converting 10-formyl-THF to THF and CO2 with NADP+ [E4]. **It is named in no entry of this graph and its literature is 64 records total.** **Measured here:** RS4;11 day 7, VTP-50469 330 nM, **ALDH1L2 -3.40 log2** from a detected baseline of 7.08 log2 CPM - **the second-largest fall in the whole serine/one-carbon panel after CBS (-4.07) and larger than PHGDH (-3.24)** [E1]. In MV4;11 from an independent laboratory it is **-2.06 under VTP-50469 against -0.16 under the DOT1L inhibitor SGC0946, a margin of -1.89 that ties PHGDH for the largest in that arm** - yet in RS4;11 its margin is only -0.49 (EPZ5676 reproduces 85%), so **its compound-selectivity is line-discordant and cannot be asserted** [E1]. In MOLM13 it is **+0.08**, flat [E1]. **The dependency figure is the most extreme in the panel: mean +0.0065, 0.0% of 1,258 screens below -0.5, leukaemia -0.0207. Not one cell line needs it** [E2]. **And the only mechanistic study of ALDH1L2 in cancer makes its loss oncogenic**: ALDH1L2 is an NADPH-producing mitochondrial enzyme that LIMITS ROS and formate production in pancreatic acinar cells, its expression falls progressively during acinar-to-ductal metaplasia and is absent in ductal cells, and **ALDH1L2 loss elevates ROS, promotes metaplasia and accelerates tumour progression** [E3]. So the largest unnoticed fall in this beat is, on the only available mechanism, a change in the tumour-favouring direction.

Go deeper
Why this is a genuine find and not a panel artefact

ALDH1L2 was declared in this beat's pre-registered panel as a member of the serine/one-carbon core on textbook grounds, before any value was read. It then produced the second-largest effect in the panel, and the corpus contains no mention of it at all - it is not in the hypothesis entry's 16-gene measurement table, not in the ATF4-panel analysis, and not in the MV4;11 replication, whose own panel was copied from the hypothesis entry 1.

This is a worked instance of the point the pre-registration rule exists to make: the earlier serine panel was assembled after a concordance list pointed at ATF4, so it inherited that list's membership. A panel declared from pathway anatomy rather than from a previous result found a member the previous result had no way to contain.

The detection floor is reported beside the number, as it must be: 7.08 log2 CPM in RS4;11 DMSO, 7.17 in MOLM13, 4.32 in MV4;11, all above the declared 1.0 floor, and no floor collision. On the GPL570 array only one of three probes (231202_at) is above the 6.0 floor in RS4;11 (6.47, MI-2-2 -0.91) and it is at background in SEM (3.02) and REH (3.14), so ALDH1L2 has no usable KMT2A-wild-type array control and the fusion-dependence of its fall is untested.

Its compound-selectivity contradicts itself between lines, and that is reported rather than resolved
linemenin inhibitorDOT1L inhibitormargin
RS4;11 d7VTP-50469 -3.40EPZ5676 -2.91-0.49 (85% reproduced)
MV4;11VTP-50469 -2.06SGC0946 -0.16-1.89 (8% reproduced)
MOLM13 d7VTP-50469 +0.08EPZ5676 -0.04-0.12

Two KMT2A::AFF1 lines give margins differing by 1.4 log2 on the same gene. The lines differ in laboratory, series, DOT1L compound, dose and normalisation behaviour (MV4;11's housekeeping set is not flat), so the discordance has at least four candidate explanations and this entry picks none. ALDH1L2 therefore contributes to the serine arm's median margin but is not evidence for menin-selectivity on its own, and the beat's verdict does not rest on it.

Identifier and literature bounds

gene_exact:ALDH1L2 AND organism_id:9606 AND reviewed:true returns exactly one reviewed record, Q3SY69 / AL1L2_HUMAN, 923 aa, gene_primary read back as ALDH1L2, no synonyms, and no PDB cross-references 4.

PubMed 2026-09-22, one tag per term with translations read back 5:

queryrecordsverdict
ALDH1L2[tiab]64the entire literature
ALDH1L2[tiab] AND the 8-operand fusion set0non-empty translation, warninglist: ["No items found."] - a correctly formed zero
ALDH1L2[tiab] AND the 9-operand menin set0same signature
ALDH1L2[tiab] AND leukemia[tiab]3all three titles opened; none is about ALDH1L2 in leukaemia - a scoping review of ALDH enzymes in haematological disease, an ALDH2/TGF-beta stromal paper, and a benzene fatty-acid-oxidation study in mice
ALDH1L2[tiab] AND ATF4[tiab]2one bioRxiv TGF-beta glycine paper, one ccRCC ATF4/ISR remodelling paper
mtFDH[tiab]3the abbreviation is not used in the indexed literature

So ALDH1L2 has never been studied in leukaemia of any kind, and the alias adds nothing. This is the cleanest bounded absence in the beat: a tiny literature, a correctly formed zero on the fusion operand set, and every leukaemia-scoped title read.

Facts and figures
gene symbolALDH1L2
uniprotQ3SY69
length AA923
entry nameAL1L2_HUMAN
protein nameMitochondrial 10-formyltetrahydrofolate dehydrogenase
ensembl gene IDENSG00000136010
ensembl readbackresolved live through the Ensembl symbol endpoint with display_name read back as ALDH1L2 before use; the readback procedure is the one that rejected VSIG2 for CTH in this project
PDB noteQ3SY69 has NO PDB cross-references. No structure is claimed.
accession resolutiongene_exact:ALDH1L2 AND organism_id:9606 AND reviewed:true returns EXACTLY ONE reviewed row (rule 62 asserted): Q3SY69, AL1L2_HUMAN, gene_primary ALDH1L2, no synonyms, 923 aa.
localizationMITOCHONDRION, experimental. UniProt Q3SY69 carries one SUBCELLULAR LOCATION statement, Mitochondrion, with ECO:0000269 support from PubMed 20498374 - and per rule 91 the cited paper IS about the fact (its title is 'ALDH1L2 is the mitochondrial homolog of 10-formyltetrahydrofolate dehydrogenase'), which is the well-behaved case rather than the usual mismatch. No localisation experiment on ALDH1L2 has been performed in any KMT2A-rearranged cell, and no co-localisation with any fusion oncoprotein exists.
subcellular localizationMITOCHONDRIAL, and this is the real content of the subcellular dimension for this beat: ALDH1L2, SHMT2 (P34897) and MTHFD2 (P13995) are mitochondrial while PHGDH, PSAT1 and GPT2 have no curated compartment at all. The menin-inhibitor response therefore crosses a membrane - it is not a single-compartment programme - and the mitochondrial members (ALDH1L2 -3.40, SHMT2 -1.09, MTHFD2 -1.16) are not the most menin-selective ones, so compartment does not predict compound-selectivity.
subnuclear localizationNONE, answered negative rather than left blank: a mitochondrial matrix enzyme with no reported nuclear pool. Unlike SHMT1 and SHMT2 in the same panel, which both carry ECO:0000269 Nucleus statements, ALDH1L2 has no nuclear annotation of any kind.
cell typeRS4;11 (KMT2A::AFF1 B-ALL), SEM (KMT2A::AFF1 B-ALL, at array background), REH (KMT2A wild type, at array background), MOLM13 (KMT2A::MLLT3 AML) and MV4;11 (KMT2A::AFF1 myeloid) measured here. Pancreatic acinar and ductal cells, murine pancreatitis and PDAC models in PMID 41922744. Renal cell carcinoma organoids and small cell lung cancer in the opposing literature. Highly expressed in pancreas, heart, brain and skeletal muscle per UniProt. NO haematopoietic cell of any kind has been studied for ALDH1L2 function.
cell stateMeasured in the drug-treated arresting state at day 7 of continuous menin-inhibitor exposure, under uncontrolled medium serine and glycine supply. At day 2 ALDH1L2 carries 38% of its day-7 fall (-1.31 of -3.40), mid-panel. The state in which ALDH1L2 has actual published function is METAPLASTIC TRANSITION: its expression falls progressively during acinar-to-ductal metaplasia and is absent in ductal cells, so its own literature makes it a state marker of a differentiation transition, which is suggestive in a disease treated by a differentiation-inducing drug and is not evidence of anything.
epigenetic readoutNONE MEASURED, anywhere, for this gene. No histone mark, no H3K4me3 or H3K79me2 ChIP, no DNA-methylation measurement at the ALDH1L2 locus exists in any system in the menin or KMT2A literature, and PMID 29672864's H3K4me3 measurement is at the PHGDH promoter only. This is a bounded absence and not an oversight of this entry: the gene has 64 records in total and none is chromatin-scoped. It is the cheapest thing to add to any ChIP this beat's other entries request, because the locus would already be in the sequencing.
shown for fusionREADOUT ONLY, and only in KMT2A::AFF1. The -3.40 fall is in RS4;11 and the -2.06 in MV4;11, both KMT2A::AFF1; it is FLAT in KMT2A::MLLT3 (MOLM13, +0.08 from a detected baseline). No mechanism, no dependency and no chromatin measurement has been shown for ALDH1L2 in any fusion. Nobody has looked at ALDH1L2 in any leukaemia at all: the fusion-scoped and menin-scoped literature queries both return a correctly formed zero.
array coverage is poorthree GPL570 probes; only 231202_at is above the 6.0 log2 RMA floor in RS4;11 (6.47, MI-2-2 -0.91), and it reads 3.02 in SEM and 3.14 in REH. 1556841_a_at and 1559393_at are at background in all three lines. So there is NO usable KMT2A-wild-type control for ALDH1L2 and its fusion-dependence is untested.
named in entries when written0
absence classmeasured-not-searched
absence class noteThe signal was in a public archive (GSE127506/GSE127508, deposited 2019) and was unfindable because no search space included ALDH1L2 - the previous serine panel in this graph was inherited from an ATF4-concordance list that could not contain it. Closing it required no new data, only a panel declared from pathway anatomy. The literature absence is separately 'never-measured': no ALDH1L2 experiment exists in any leukaemia.
PDB IDS
[]
the measurement rs411 d7 log2fc
{
  "vtp50469_330nM": -3.4,
  "epz5676_1uM": -2.91,
  "vtp_minus_epz": -0.489,
  "epz_fraction_of_vtp": 0.85,
  "dmso_baseline_log2cpm": 7.08,
  "detected": true,
  "floor_collision": false,
  "d2_vtp": -1.306,
  "d2_share_of_d7": 0.38,
  "rank_in_panel": "second-largest absolute fall after CBS -4.07, ahead of PHGDH -3.24"
}
molm13 kmt2a mllt3
{
  "vtp": 0.077,
  "epz": -0.043,
  "dmso_baseline_log2cpm": 7.17,
  "detected": true,
  "reading": "well expressed and completely unresponsive in a KMT2A::MLLT3 AML line whose positive control collapses (MEIS1 -2.96)"
}
mv411 independent lab
{
  "menin_vtp50469_500nM": -2.06,
  "dot1l_sgc0946_5uM": -0.16,
  "margin": -1.889,
  "dmso_baseline_log2cpm": 4.32,
  "note": "ties PHGDH for the largest menin-minus-DOT1L margin in that arm, and CONTRADICTS the RS4;11 margin of -0.49; housekeeping is not flat in this matrix"
}
essentiality
{
  "depmap_release": null,
  "depmap_release_note": "Rule 83: the Open Targets payload carries no release identifier. Fetched 2026-09-22.",
  "depmap_screens": 1257,
  "depmap_tissue_blocks_flattened": 29,
  "chronos_mean": 0.0065,
  "chronos_median": 0.0107,
  "pct_lines_below_minus_0_5": 0,
  "leukaemia_screens": 57,
  "leukaemia_mean_gene_effect": -0.0207,
  "depmap_common_essential": "not asserted",
  "reading": "ZERO of 1,257 screens fall below -0.5. This is the most extreme non-dependency in the panel and it sits on the second-largest fold change, which is the single clearest illustration in this beat that a fold change is not a dependency.",
  "paralogue_confound_not_measured": "ALDH1L1 is the cytosolic paralogue catalysing the same reaction and was NOT in the pre-registered panel, so redundancy is unmeasured and the null cannot distinguish 'not needed' from 'covered by ALDH1L1'."
}
How this could bite us back
  • **The only mechanistic study of ALDH1L2 in cancer makes its LOSS oncogenic.** ALDH1L2 limits ROS and formate in pancreatic acinar cells; its loss elevates ROS, promotes acinar-to-ductal metaplasia and accelerates tumour progression. A menin inhibitor lowers it 3.40 log2. If that direction transfers at all, the drug is removing an NADPH-generating ROS brake, which is a mechanism for genotoxic stress in the surviving cells rather than a therapeutic gain - and this population already receives anthracyclines, etoposide and cytarabine.
  • **ALDH1L2 is not a target and must not become one on the strength of a fold change.** Zero of 1,257 DepMap screens fall below -0.5. There is no inhibitor, no trial, and no leukaemia experiment of any kind.
  • **Its direction in cancer is actively contested** - tumour-suppressive in pancreas, ferroptosis-suppressing and therapy-resistance-promoting in renal and small-cell lung cancer. Anyone citing this entry for a therapeutic direction is citing one of two opposite literatures.
  • **A gene with no curated human disease is not a gene with no essential function.** ALDH1L2 has no Mendelian phenotype in UniProt, which is compatible with embryonic lethality or with nobody having looked, and one-carbon enzyme loss elsewhere in this panel causes neural tube defects (MTHFD1) and lethal malformation syndromes (PHGDH, PSAT1).
Still open
  1. Does ALDH1L2 protein and formate output change in KMT2A-rearranged cells under a menin inhibitor, and does ROS rise? PMID 41922744 makes ALDH1L2 a ROS brake and a formate source. A -3.40 log2 transcript fall predicts ROS up and formate down, both measurable in a day, and if ROS rises the fall is a liability of the drug rather than part of its mechanism.
  2. Is ALDH1L2's menin-selectivity real? The two KMT2A::AFF1 lines disagree by 1.4 log2 (-0.49 in RS4;11, -1.89 in MV4;11) with four confounded differences between them. A dose-matched menin/DOT1L comparison in one line settles it, and it is the same experiment the rest of this beat needs.
  3. Why is ALDH1L2 unresponsive in MOLM13 while falling 3.40 log2 in RS4;11, both well expressed? This is the largest single-gene line discordance in the panel and the most extreme instance of the beat's second finding.
  4. Is ALDH1L1, the cytosolic paralogue, reciprocally induced? It was not in the declared panel and is the standard confound for every one-carbon enzyme. A paralogue that rises when its partner falls would make the fall physiologically silent, and SHMT1 already does exactly that in these samples (+0.78 while SHMT2 falls -1.09).
Every source behind this entry (5)
  1. E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.dataset

    On a panel declared to disk before any value was read, ALDH1L2 falls 3.40 log2 in RS4;11 at day 7 under VTP-50469 330 nM from a detected DMSO baseline of 7.08 log2 CPM - the second-largest fall in the serine/one-carbon panel after CBS (-4.07) and larger than PHGDH (-3.24) - with EPZ5676 giving -2.91 (margin -0.49); at day 2 it is -1.31, 38% of the day-7 effect; in MOLM13 it is +0.08 from a detected baseline of 7.17; in MV4;11 it is -2.06 under VTP-50469 500 nM against -0.16 under SGC0946 5 uM (margin -1.89); and on the GPL570 array only probe 231202_at is above the 6.0 log2 RMA floor in RS4;11 (6.47, MI-2-2 -0.91), the same probe reading 3.02 in SEM and 3.14 in REH.

    What this evidence cannot show

    One line per platform, single doses, no dose-response, n=2-3 per arm, day 7 so arrest is a competing explanation. The compound-selectivity margin is DISCORDANT between the two KMT2A::AFF1 lines (-0.49 in RS4;11, -1.89 in MV4;11) and this entry does not resolve which is right. There is NO usable KMT2A-wild-type control for this gene: two of three array probes are at background in RS4;11 and the one usable probe is at background in REH, and the RNA-seq series contains no wild-type line. So the fusion-dependence of the fall is untested for ALDH1L2 specifically.

    Evidence that argues against this (1)
    • A gene at background on two of three array probes in the very line where the RNA-seq reads it at 7.08 log2 CPM is a gene whose measurement is platform-dependent, and ALDH1L2 has a documented SEQUENCE CAUTION comment in UniProt plus alternative products. A 3-prime-end assay reading one transcript isoform while the array probes read others is a live explanation for both the platform disagreement and the line discordance, and nothing here excludes it. UniProt Consortium 2026
  2. E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    ALDH1L2 is required by no cell line in DepMap: gene effect mean +0.0065, median +0.0107, 0.0% of 1,257 screens below -0.5, leukaemia mean -0.0207 over 57 screens, fetched with all 29 tissue blocks flattened and the resolved Ensembl id's approvedSymbol read back as ALDH1L2.

    What this evidence cannot show

    Rule 83: the payload carries no release identifier, so depmap_release is null. A POSITIVE mean gene effect means knockout is very slightly growth-promoting on average, which is what a tumour-suppressive function predicts but is well within noise and is not asserted as an effect here. 0.0% is an absolute-threshold fraction over the whole screen set, not a median split. Leukaemia screens are matched on the payload's own diseaseFromSource string and not split by KMT2A status.

    Evidence that argues against this (1)
    • A flat-to-positive CRISPR profile is exactly what a redundant enzyme looks like as well as what a tumour suppressor looks like, and ALDH1L1 is the cytosolic paralogue performing the same reaction. This beat did NOT include ALDH1L1 in its pre-registered panel, so paralogue redundancy - the standard confound for every one-carbon enzyme - is unmeasured here and the DepMap null cannot distinguish 'not needed' from 'covered by ALDH1L1'. Krupenko NI et al. 2010
  3. E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    ALDH1L2, an NADPH-producing mitochondrial enzyme of the one-carbon pathway, limits reactive oxygen species and formate production in pancreatic acinar cells; its expression decreases progressively during acinar-to-ductal metaplasia and is completely absent in pancreatic ductal cells; ALDH1L2 loss elevates ROS and promotes acinar-to-ductal metaplasia in a model of pancreatitis and accelerates tumour progression in models of pancreatic cancer; and formate increases during pancreatic ductal adenocarcinoma progression in mice and humans.

    ALDH1L2 regulates reactive oxygen species and acinar-to-ductal metaplasia in the pancreas. · Hennequart M, Mervant L, Stockis J, et al., Vousden KH · Nat Metab 2026
    PMID 41922744DOI
    What this evidence cannot show

    Abstract read this session; full text not retrieved. **The system is pancreas - acinar cells, pancreatitis models and PDAC models - and has nothing to do with leukaemia or with any fusion oncoprotein.** It is cited because it is the only mechanistic characterisation of ALDH1L2 in a cancer context that exists, and because its direction is the opposite of what a reader would assume from a large drug-induced fall. A ROS-limiting NADPH source in acinar cells need not behave the same way in a B-lymphoblast. 2026 paper, no independent replication.

    Evidence that argues against this (1)
    • A tumour-suppressive role in one epithelial context does not transfer, and there is a direct opposing literature in other tumours: ALDH1L2 suppresses ferroptosis-associated responses and reduces sunitinib sensitivity in renal cell carcinoma organoids (PMID 42304489), and induces chemotherapy resistance in small cell lung cancer by inhibiting ferroptosis (PMID 41764940). Metadata for both was seen and neither abstract was read. So ALDH1L2 is reported as tumour-suppressive in pancreas and tumour-supporting in kidney and lung, and its direction in leukaemia is simply unknown. Hu X et al. 2026
  4. E4established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    UniProt reviewed human record Q3SY69 (AL1L2_HUMAN) has primary gene name ALDH1L2 with no synonyms, length 923 aa, no PDB cross-references, is a mitochondrial 10-formyltetrahydrofolate dehydrogenase catalysing the NADP+-dependent conversion of 10-formyltetrahydrofolate to tetrahydrofolate and carbon dioxide, is annotated Mitochondrion with ECO:0000269 experimental support from PubMed 20498374, is highly expressed in pancreas, heart, brain and skeletal muscle, and records NO disease.

    What this evidence cannot show

    Resolved 2026-09-22 with gene_exact:ALDH1L2 AND organism_id:9606 AND reviewed:true; exactly one reviewed row, gene_primary and length read back. Rule 91 applies to the localisation: the single ECO:0000269 citation is PubMed 20498374, whose title is 'ALDH1L2 is the mitochondrial homolog of 10-formyltetrahydrofolate dehydrogenase' - so the cited paper IS about the fact, which is the well-behaved case and is worth recording as such. That paper's abstract was not read; only its title and bibliographic record were verified. The absence of a disease comment means no Mendelian phenotype is curated, not that loss is tolerated.

    Evidence that argues against this (1)
    • No human disease is recorded for ALDH1L2, which removes the strongest bite-back available for PHGDH, PSAT1, SHMT2 and GPT2 - there is no germline phenotype to warn from. That absence is not reassurance: it can equally mean that biallelic loss is embryonic-lethal or that no cohort has looked, and this entry cannot distinguish those. UniProt Consortium 2026
  5. E5established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    PubMed 2026-09-22, one tag per term with translations read back: ALDH1L2[tiab] returns 64 records in total; ALDH1L2[tiab] AND the eight-operand KMT2A-fusion set returns 0 with a non-empty querytranslation and warninglist 'No items found.'; the same against a nine-operand menin/menin-inhibitor set returns 0 with the same signature; ALDH1L2[tiab] AND leukemia[tiab] returns 3 and all three titles were read and none is about ALDH1L2 in leukaemia; ALDH1L2[tiab] AND ATF4[tiab] returns 2; and the alias mtFDH[tiab] returns 3 with 0 fusion-scoped.

    What this evidence cannot show

    A PubMed zero is bounded by TITLES AND ABSTRACTS, not by papers: a measurement in a supporting figure or a supplementary table would not appear, and for a gene this obscure that is the likely place for it. Supporting figures were not searched and no full-text index was queried for ALDH1L2, which is the cheapest remaining check and was not done. The three leukaemia-scoped titles were read; the 64-record total literature was not read.

    Evidence that argues against this (1)
    • The 64-record literature includes the possibility that ALDH1L2 appears in a KMT2A-fusion expression dataset under a probe id or an Ensembl id rather than a symbol, which no literature query reaches. This beat found ALDH1L2 precisely that way - by declaring it in a panel and reading a count matrix - so the bounded absence is an absence of REPORTING, and the archive check that matters is the one this entry itself performed. this repository 2026
Connected to