CARM1 / PRMT4 (Q86X55) - the H3R17me2a writer, and the enzyme whose product is read by the KMT2A C-terminus that every fusion deletes

No three-dimensional model is available for this protein, so the map above is the whole picture. That is itself informative: AlphaFold does not model chains this long, and the parts that have been crystallised are held under the entries for the individual domains.

• A selective small-molecule CARM1 inhibitor exists and was validated in leukaemia cells in vivo (PMID 29894694), so this entry touches an intervention and must be read with that in mind. The result is in MOUSE conditional knockouts and in MYELOID AML, not in infant KMT2A::AFF1 B-ALL, and the abstract names no specific fusion.
• This graph's own dependency data does not support CARM1 as a KMT2A-selective target: -0.3222 across 1,258 lines, 22.3% below -0.5, and very slightly LESS dependent in KMT2A-rearranged leukaemia lines (-0.4004) than in other leukaemia lines (-0.45). It is not pan-essential and it is not selective.
• The strongest caution is mechanistic and comes from the same literature: loss of CARM1 lets cells TOLERATE higher replication stress (PMID 33412112). If this disease's vulnerability is transcription-replication conflict, inhibiting CARM1 could move a fusion-positive cell in the wrong direction. And because the fork function is methyltransferase-INDEPENDENT, a catalytic inhibitor may not reproduce the genetic loss at all.
• 'Dispensable for normal hematopoiesis' is a mouse genetic statement. CARM1 is required in normal development and its loss affects cell-cycle progression and differentiation, and nothing here bounds toxicity in a human infant.

1. Does KMT2A::AFF1 lack the arginine-methyl reader domain? The residue boundaries of the C-terminal MLL reader are not in Kawabe's abstract, and the fusion breakpoints sit at KMT2A 1362/1363, 1406/1407 and 1444/1445 - so the inference needs the reader mapped against those coordinates before it is more than a shape argument.
2. Is PAX5 arginine-methylated by CARM1? PAX7's paralogue PAX5 is central to this disease, gene-pax5 exists and gene-pax7 does not, and no arginine-methylation site map for PAX5 was found.
3. Which of CARM1's pools matters in a fusion-positive cell - the promoter H3R17me2a pool, the replication-fork PARP1 pool, or the non-histone substrate pool? Two of its three documented functions do not require the histone mark.
4. Was Greenblatt's AML model KMT2A-rearranged? The abstract says only 'driven by oncogenic transcription factors'. This decides whether the therapeutic-window result is about this disease class at all, and it needs the full text.
5. Does the UniProt TCF7L2 annotation have a source? It is By-similarity and the indexed intersection is a genuine zero under both tokens, while this graph's own entry on the DOT1L TCF7L2 partner names CARM1.

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

   CARM1 is UniProt Q86X55, CARM1_HUMAN, 608 aa, Histone-arginine methyltransferase CARM1, with the alternative names Coactivator-associated arginine methyltransferase 1 and Protein arginine N-methyltransferase 4 and the gene synonym PRMT4. It carries one annotated domain, SAM-dependent MTase PRMT-type at 146-453. It performs mono- and asymmetric dimethylation of guanidino nitrogens of arginyl residues in proteins involved in DNA packaging, transcription regulation, pre-mRNA splicing and mRNA stability, and is recruited to promoters on gene activation together with histone acetyltransferases of the EP300/p300 and p160 families, methylating histone H3 at Arg-17 to form mainly H3R17me2a. Its subcellular location record lists Nucleus, Cytoplasm and Chromosome on ECO:0000269, and states it is mainly nuclear during G1, S and G2, cytoplasmic during mitosis after breakup of the nuclear membrane, and localises to replication forks.

   UniProtKB entry Q86X55 (CARM1_HUMAN), retrieved by gene_exact:CARM1 AND organism_id:9606 AND reviewed:true with fields accession, id, protein_name, gene_names, length, cc_subcellular_location, ft_domain, cc_function, cc_tissue_specificity; fetched 2026-09-22 with the record count asserted as exactly one before use
   ▸What this evidence cannot show

   Accession, entry name, length, synonyms and the single domain's coordinates were read from the JSON of an exact-gene query asserted to return one reviewed record. The FUNCTION comment is a curated summary over several primary citations, none read here, and parts of it are annotated By similarity rather than experimentally - the TCF7L2/beta-catenin clause in particular, whose literature this entry measures as empty. The three-compartment localisation is a union over experiments and cell-cycle stages, not a description of any one cell at one time.

   ▸Evidence that argues against this (1)

   • A methyltransferase with three compartments, a replication-fork pool, non-histone substrates across four different processes and a documented methyltransferase-INDEPENDENT activity is a poor object for a clean mechanistic claim. Any statement of the form 'CARM1 does X at the fusion' has to specify which pool and whether catalysis is even involved, and nothing in this entry does that for a KMT2A-rearranged cell.
2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

   The C-terminal region of MLL is an arginine-methyl reader domain, and CARM1 makes the mark it reads. Kawabe et al. identified Carm1 as a Pax7-interacting protein that specifically methylates multiple arginines in the N terminus of Pax7; methylated Pax7 directly binds the C-terminal cleavage forms of the trithorax proteins MLL1/2, resulting in recruitment of the ASH2L:MLL1/2:WDR5:RBBP5 histone H3K4 methyltransferase complex to regulatory enhancers and the proximal promoter of Myf5; Carm1 is required for de novo Myf5 transcription after asymmetric satellite stem cell division. The authors state that they defined the C-terminal MLL region as a reader domain for the recognition of arginine-methylated proteins such as Pax7.

   Carm1 regulates Pax7 transcriptional activity through MLL1/2 recruitment during asymmetric satellite stem cell divisions. · Kawabe Y, Wang YX, et al. · Cell Stem Cell 2012
   PMID 22863532DOI
   ▸What this evidence cannot show

   Abstract-level; no figure or methods section was read, so the reader-domain boundaries, the affinities and the controls are not established here. The system is mouse skeletal-muscle satellite stem cells and the substrate is Pax7 - not haematopoietic, not PAX5, not a KMT2A-rearranged cell. The inference that KMT2A::AFF1 therefore lacks this reader is MINE and not the authors': they make no claim about any fusion, and the mapping of 'C-terminal cleavage forms of MLL1/2' onto the region a fusion deletes needs the residue boundaries this abstract does not give. Uncited in this graph before now.

   ▸Evidence that argues against this (2)

   • The subtraction inference may be the wrong way round. If the C-terminal MLL fragment is the reader, then wild-type MLL retains it and the fusion does not - but this graph's central model is that KMT2A fusions act by GAINING elongation machinery, and a lost reader domain predicts loss of a wild-type function rather than anything oncogenic. The interesting version of the claim requires that losing the reader matters for leukaemogenesis, and there is no evidence for that at all.
   • Taspase-1 cleavage complicates the mapping. MLL is cleaved into N- and C-terminal fragments that heterodimerise, and 'the C-terminal cleavage form' is a specific species rather than simply 'the C-terminus'. A separate paper establishes that human HCF-1 maturation is taspase-1 independent while MLL is taspase-1 cleaved and that these pathways are species-selective, so a mouse satellite-cell result about cleavage forms may not transfer to a human B-cell precursor.
3. E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

   CARM1 is required for myeloid leukaemia and largely dispensable for normal blood formation in mice. Using conditional knockout mice, loss of CARM1 had little effect on normal haematopoiesis, while knockout of Carm1 abrogated both the initiation and the maintenance of acute myeloid leukaemia driven by oncogenic transcription factors. CARM1 knockdown impaired cell-cycle progression, promoted myeloid differentiation and ultimately induced apoptosis, and a selective small-molecule CARM1 inhibitor was used to validate the efficacy of CARM1 inhibition in leukaemia cells in vitro and in vivo.

   CARM1 Is Essential for Myeloid Leukemogenesis but Dispensable for Normal Hematopoiesis. · Greenblatt SM, Man N, et al. · Cancer Cell 2018
   PMID 29894694DOI
   ▸What this evidence cannot show

   Abstract-level, and the disease is MYELOID AML in mouse conditional knockouts - not infant B-ALL and not human. The abstract says 'driven by oncogenic transcription factors' and does NOT name MLL-AF9 or any specific fusion, so this entry does not claim the model was KMT2A-rearranged; that would need the full text. 'Dispensable for normal hematopoiesis' is a mouse genetic statement, not a human safety finding, and a conditional knockout is not a drug. The selective inhibitor is named only as such in the abstract.

   ▸Evidence that argues against this (2)

   • This graph's own dependency data pulls the other way for the index disease. CARM1's Chronos gene effect is -0.3222 across 1,258 lines with 22.3% below -0.5, and in KMT2A-rearranged leukaemia lines it is -0.4004 against -0.45 in other leukaemia lines - very slightly LESS dependent in the index lines. A mouse AML requirement and a non-selective human CRISPR dependency are not contradictory, but nothing here supports CARM1 as a KMT2A-selective vulnerability.
   • The therapeutic-window reading is exactly the inference this project has been burned by. 'Essential in leukaemia, dispensable in normal haematopoiesis' in a mouse conditional knockout has been claimed for many chromatin enzymes and has repeatedly failed to transfer; and this entry's own Genois evidence shows CARM1 loss lets cells TOLERATE more replication stress, which is not the direction a cytotoxic strategy wants.
4. E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

   CARM1 has a replication-fork function that does not require its catalytic activity and that runs through PARP1. CARM1 associates with replication forks and reduces fork speed independently of its methyltransferase activity; the speeding of forks in CARM1-deficient cells requires RECQ1, which resolves reversed forks, and RAD18, which promotes translesion synthesis. Loss of CARM1 reduces fork reversal and increases single-stranded DNA gaps but allows cells to tolerate higher replication stress. Mechanistically CARM1 interacts with PARP1 and promotes PARylation at replication forks, and in vitro stimulates PARP1 activity by enhancing its DNA binding, acting jointly with HPF1.

   CARM1 regulates replication fork speed and stress response by stimulating PARP1. · Genois MM, Gagné JP, et al. · Molecular Cell 2021
   PMID 33412112DOI
   ▸What this evidence cannot show

   Abstract-level. The cell systems are not named in the abstract, so this is not established in any haematopoietic or KMT2A-rearranged cell. The connection to this project is by topic - this graph holds a transcription-replication conflict model for the KMT2A breakpoint and a menin-PARP experiment - and no experiment links CARM1's fork function to the KMT2A locus. Uncited in this graph before now.

   ▸Evidence that argues against this (2)

   • This is the strongest single argument against targeting CARM1 in this disease, and it comes from the same entry that reports the therapeutic window. If loss of CARM1 lets cells TOLERATE higher replication stress, then inhibiting CARM1 in a fusion-positive cell whose vulnerability this graph models as transcription-replication conflict could make it MORE stress-tolerant, not less. The catalytic independence makes it worse: a SAM-competitive methyltransferase inhibitor might not touch the fork function at all, so the two effects could be separated pharmacologically in an unhelpful direction.
   • It also undercuts reading CARM1 through its mark. Two of the three functional results in this entry - the fork/PARP1 axis and the reader-domain recruitment - do not require H3R17me2a at chromatin, so an H3R17me2a-centred account of CARM1 in this graph would describe the minority of its documented activity.
5. E5strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

   The alias split for this gene runs opposite to the recent pattern in this project, and the alias still matters. CARM1[tiab] returns 564 records and PRMT4[tiab] returns 157, intersecting in 73, for a union of 648 - so the approved symbol is the MAJORITY token, unlike HCFC1, B2M, NUMA1 and DHX9. But the alias is still not redundant: CARM1[tiab] AND (MLL[tiab] OR KMT2A[tiab]) returns 3 while the union returns 4, so querying the approved symbol alone would have missed one of the four fusion-context records. Separately, UniProt's By-similarity claim that CARM1 acts in TCF7L2/TCF4 activation has an empty indexed literature: (CARM1[tiab] OR PRMT4[tiab]) AND TCF7L2[tiab] returns 0 as a genuine zero, with a parsed querytranslation and warninglist 'No items found.'

   NCBI esearch against PubMed with ID sets compared at retmax=1200, 2026-09-22; every zero checked for a non-empty querytranslation and a warninglist entry before being recorded
   ▸What this evidence cannot show

   ID-set comparison at retmax=1200 is complete for every set here, the largest being 564, so nothing was truncated. Title-and-abstract scope only. The TCF7L2 zero bounds indexed titles and abstracts and would not see a result inside a figure or a supplementary file, and UniProt's By-similarity annotation may rest on a non-human or non-indexed source rather than on nothing.

   ▸Evidence that argues against this (1)

   • A single gene going the other way does not establish that the alias question has no default direction - four of five measured genes still had the approved symbol in the minority, so on this evidence the prior should still favour checking the alias. The honest statement is that the direction is unpredictable per gene, which is an argument for always measuring rather than for treating the two tokens as equivalent.

• supports →Five new arginine-axis entries in one day, and not one of their genes is a selective dependency - in an instrument that recovers MEN1 at p=5e-05 and TDRD3 is not a dependency at all

  CARM1 was in that panel and comes back non-selective: -0.3222 over 1,258 lines, 22.3% below -0.5, and -0.4004 in KMT2A-r leukaemia against -0.45 in other leukaemia lines; re-measured independently and identical

• related to →KMT2A / MLL1 — the gene that breaks

  CARM1 makes the mark that the C-TERMINAL MLL region reads - Kawabe 2012 defined that region as a reader domain for arginine-methylated proteins such as Pax7, and every KMT2A fusion deletes it

• recruits →WDR5 — the partner that is not a partner: it binds the half of KMT2A the fusion throws away

  methyl-Pax7 bound to the MLL1/2 C-terminus recruits the ASH2L:MLL1/2:WDR5:RBBP5 H3K4 complex - so CARM1 sits upstream of WRAD by this route, in mouse satellite cells

• recruits →ASH2L — the WRAD subunit that has DPY30's essentiality profile and none of its co-dependency signal

  same ASH2L:MLL1/2:WDR5:RBBP5 recruitment, via arginine-methylated Pax7 read by the MLL C-terminus

• recruits →RBBP5 — a WRAD subunit that binds the one part of KMT2A every fusion throws away

  same complex; recorded per subunit rather than as one roster entry, per the PI's one-entry-per-thing rule

• binds →PARP1 - the target whose inhibitors this disease was specifically shown to resist, and which three later papers combine anyway

  CARM1 interacts with PARP1 and promotes PARylation at replication forks, stimulating PARP1 by enhancing its DNA binding and acting jointly with HPF1 - and this is INDEPENDENT of CARM1's methyltransferase activity

• refutes →Transcription–replication conflict as the candidate reason KMT2A-rearranged cells need the Fanconi pathway — a chain with two measured links and three inferred ones

  the uncomfortable direction: loss of CARM1 lets cells TOLERATE higher replication stress, so inhibiting it in a cell whose vulnerability this graph models as transcription-replication conflict may help the cell rather than harm it

• related to →PAX5 - the most frequently hit non-KMT2A gene in infant KMT2A-rearranged ALL, and it is a deletion in one allele, not a mutation

  CARM1 methylates multiple arginines in the N-terminus of PAX7; PAX5 is its paralogue and is central to this disease, and no PAX5 arginine-methylation site map was found

• analogous to →PRMT1 — an MLL-fusion cofactor whose first-in-human inhibitor was terminated for lack of efficacy and toxicity

  both asymmetric-dimethylarginine writers in this graph's arginine beat, with opposite dependency profiles: PRMT1 -1.3984 at 98.7% of lines against CARM1 -0.3222 at 22.3%, measured in the same pull

• analogous to →TDRD3 (Q9H7E2) - the reader the new H4R3me2a entry could not name, and it brings TOP3B to the MYC promoter and keeps R loops off it

  two halves of the arginine layer that this graph had only half-written: TDRD3 is a methylarginine READER delivering effectors, and CARM1 is a WRITER whose product has a reader in KMT2A's own deleted C-terminus