REST / NRSF - the transcription factor that delivers G9a to chromatin through MED19 and MED26, is a paediatric cancer gene, and is the worst search token in this corpus

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.

• There is no REST-directed intervention and this entry does not propose one; the relevant intervention is on the enzyme REST delivers, and there the direction is against us. UniProt's own wording puts REST-CDYL recruitment of EHMT2 on the side of transformation SUPPRESSION, and mark-h3k9me3 argues H3K9 methylation restrains the MLL-fusion programme - so interfering with this axis is predicted to help the leukaemia.
• A G9a inhibitor given for the reasons in gene-ehmt2 would also disable this route, and nobody has measured whether the protective arm or the HoxA9-supporting arm dominates in a KMT2A-rearranged cell. That is a live, unmeasured trade-off inside one drug.
• The searchability problem is itself a risk to any claim made here, including the negative ones: this entry's central absence - no leukaemia, no haematopoietic literature - is bounded through two aliases that overlap by 11% of their union, so the net is demonstrably partial and the bound is correspondingly weak. It is stated as such, and any future beat quoting it must carry that sentence.
• The paediatric-cancer association is the least verifiable and most important thing in the record: Wilms tumor 6 is a UniProt DISEASE annotation with no primary paper retrieved here, and the literature that would confirm it is the single worst-colliding sub-literature for this symbol, because a nephrogenic REST is a pathological entity in those same papers.
• Compensation and control layers are dense: REST activity is set by proteasomal degradation (FBXW11, BTRC, USP7), by an inhibitory partner (ZFP90) and by an SRRM4-dependent splice switch that produces a competing isoform. A perturbation of REST protein level is not a perturbation of REST function, and three separate mechanisms can buffer it.

1. Is REST expressed and active in a KMT2A-rearranged blast? UniProt says the lymphocytic compartment expresses it at higher levels, on a single 1996 citation, and there is no REST measurement in any haematopoietic cell. One western blot and one RNA-seq lookup would settle whether this entry has a subject in this disease.
2. Does REST restrain lineage infidelity in this disease? REST is the canonical enforcer of lineage-inappropriate gene silencing and this disease's signature is lineage switch. That pairing has never been tested, and the graph's entire lineage-switch layer contains no REST measurement.
3. Is CDYL present in a t(4;11) line, and does the REST-CDYL-EHMT2 complex form there? The G9a route this graph half-describes through MED19 requires CDYL, CDYL has no entry here, and neither protein has been measured in a leukaemic cell.
4. Which Mediator subunit actually carries the route - MED19/MED26 or MED12? The same laboratory has published both, gene-med19 records the contest, and adding the CDYL requirement means the route has three open positions rather than one.
5. Is the REST-EHMT2 axis transformation-suppressing in leukaemia as UniProt's wording suggests for transformation generally? If so it is a protective axis and the therapeutic implication is negative - which matters because G9a inhibitors exist and this beat's EHMT2 entry records the opposite direction downstream.
6. How much REST literature is unreachable? NRSF and the full protein name overlap by only 81 of 709 records, so the union is demonstrably incomplete and the approved symbol cannot be counted at all. Any absence recorded for this gene - including this entry's - is an absence within a partial net, and nobody knows how partial.

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

   REST (Q13127, 1,097 aa) is a transcriptional repressor that binds the neuron-restrictive silencer element and represses neuronal gene transcription in non-neuronal cells, restricting neuronal gene expression by associating with the corepressors SIN3A and RCOR1 which recruit histone deacetylase; it carries nine C2H2-type zinc fingers with zinc finger 5 required for nuclear localization, interaction regions for SIN3A at 32-122, SIN3B at 43-57 and RCOR1 at 1009-1087, and an NRSE-binding region at 201-212; it is located in both Nucleus and Cytoplasm with multiple experimental citations for each; it is expressed in neurons and neural progenitors and ALSO at higher levels in the tissues of the lymphocytic compartment including spleen, thymus and peripheral blood lymphocytes; it is a substrate of FBXW11 and BTRC and interacts with USP7; SRRM4-dependent alternative splicing produces isoform 3, which binds NRSE with lower affinity and can oppose isoform 1; and it is listed as causative in Wilms tumor 6, gingival fibromatosis 5 and autosomal dominant deafness 27.

   UniProtKB Q13127 (REST_HUMAN), RE1-silencing transcription factor - FUNCTION, SUBUNIT, SUBCELLULAR LOCATION, TISSUE SPECIFICITY, DOMAIN, DISEASE and feature table, fetched 2026-09-22 · UniProt Consortium 2026
   ▸What this evidence cannot show

   A database record, not a measurement, and an unusually large one whose comments are split across isoforms - REST has multiple SUBCELLULAR LOCATION comment blocks, some scoped to isoform 3, and this entry collapses them into 'Nucleus and Cytoplasm with multiple citations for each' rather than attributing a compartment to an isoform, which the record does not cleanly support. None of the cited localisation papers was read, so rule 91 applies throughout. The Wilms tumor 6 disease association is a UniProt DISEASE annotation and no primary genetic paper for it was retrieved for this entry - it is recorded at database strength only and is flagged in the open questions rather than built on.

   ▸Evidence that argues against this (1)

   • The lymphocytic-compartment expression statement, which is the single most on-topic line in this record for a leukaemia graph, rests on ONE citation (PubMed:8568247) from 1996 in a comment that also says simply 'Ubiquitous', and every other tissue statement in the record is neuronal and carries protein-level evidence. So 'REST is lymphocyte-enriched' is a thirty-year-old mRNA-era assertion sitting inside a record whose weight is entirely neural, and it should not on its own motivate a haematopoietic hypothesis. That is also consistent with the literature: through safe search tokens, REST has no haematopoietic literature at all. UniProt Consortium 2026
2. E2moderate confidence 3 of 5: One strong study, or several smaller consistent ones.database

   The REST route to the H3K9 methyltransferases requires a cofactor this graph had not recorded: REST interacts with EHMT1 and EHMT2 ONLY in the presence of CDYL, a complex exists containing at least CDYL, REST, WIZ, SETDB1, EHMT1 and EHMT2, and transcriptional repression by REST-CDYL via recruitment of EHMT2 is described as potentially important in transformation SUPPRESSION.

   UniProtKB Q13127 and Q96KQ7 SUBUNIT and FUNCTION comments - 'Interacts with EHMT1 and EHMT2 only in the presence of CDYL (PubMed:19061646)', 'Part of a complex containing at least CDYL, REST, WIZ, SETB1, EHMT1 and EHMT2 (PubMed:19061646)' and 'Transcriptional repression by REST-CDYL via the recruitment of histone methyltransferase EHMT2 may be important in transformation suppression (PubMed:19061646)', read 2026-09-22 · UniProt Consortium 2026
   ▸What this evidence cannot show

   Database-strength only: this is a UniProt curation of a single paper, PMID 19061646, which was NOT retrieved for this entry, so the CDYL requirement should be read from the primary paper before it is built on. The record spells the fifth complex member 'SETB1', which is almost certainly SETDB1; that reading is an inference from context, not a database value. The underlying system is neuronal-gene silencing in non-neuronal cells, and 'may be important in transformation suppression' is a hedged statement about a general mechanism, not a result in any leukaemia.

   ▸Evidence that argues against this (1)

   • Which adaptor carries the G9a route is contested, and this graph already records the contest from the Mediator side: gene-med19 notes that Mediator's contribution to REST silencing has ALSO been placed on MED12 rather than on MED19/MED26, in work from the same laboratory. Adding a CDYL requirement gives the route three disputed or undocumented components - which Mediator subunit, whether CDYL is present in the relevant cell, and whether SETDB1 or EHMT1/2 is the operative methyltransferase - and none of the three has been measured in a KMT2A-rearranged cell. A route with that many open positions is a hypothesis, not a pathway. Oncofusion Knowledge Graph 2026
3. E3established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

   REST[gene] is a WORSE PubMed operand than the untagged form, not a better one: REST[gene] returns 216,469 records and its querytranslation expands to a disjunction including "rest"[All Fields] - the ordinary English word - against 196,013 for REST[tiab], while the only usable operands are NRSF[tiab] at 501 and "RE1-silencing transcription factor"[tiab] at 289; and a second disease-specific collision compounds it, because REST[gene] AND "Wilms tumor"[tiab] returns 74 records in the very disease UniProt lists REST as causing, where a nephrogenic REST is a standard pathological entity.

   NCBI E-utilities esearch, db=pubmed, every count with its querytranslation read back verbatim, run 2026-09-22: REST[tiab] = 196013; REST[gene] = 216469 with translation '"rest protein human"[Supplementary Concept] OR "rest protein human"[All Fields] OR "rest"[All Fields] OR "rest"[MeSH Terms]'; NRSF[tiab] = 501; "RE1-silencing transcription factor"[tiab] = 289; REST[gene] AND leukemia[tiab] = 676; REST[gene] AND hematopoi*[tiab] = 393; REST[gene] AND "Wilms tumor"[tiab] = 74 · NCBI PubMed 2026
   ▸What this evidence cannot show

   This measures operand behaviour on one date and does not audit the 216,469 records - the claim is about the translation string, which is verbatim from the API, and about the arithmetic that [gene] returns MORE than [tiab]. It also directly contradicts the dispatch brief that commissioned this entry, which recommended REST[gene] as one of three safe operands; the literature and the API win over the brief, per standing instruction. No claim is made that [gene] behaves this way for other symbols - it almost certainly does not, and the pathology here is that 'rest' is an indexed English word and a MeSH term in its own right.

   ▸Evidence that argues against this (1)

   • The opposite reading is that REST[gene] is behaving correctly and the expectation was wrong: PubMed's [gene] tag is documented to search a gene-symbol field AND to fall back through All Fields, so a symbol that is an English word will always explode. Which makes the useful statement narrower and sharper than 'the tag is broken': for a symbol that collides with ordinary vocabulary, NO field tag rescues it, because the tag expands rather than restricts. The only remedy is a synonym that is not a word - here NRSF - or the full protein name as a quoted phrase, and both must be used because neither reaches most of the literature alone. NCBI PubMed 2026
4. E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

   Through both safe operands and after title-checking every hit, REST has NO leukaemia literature and NO haematopoietic literature: NRSF[tiab] AND leukemia[tiab] = 1 and that record is a gene-regulatory-network inference algorithm paper; NRSF[tiab] AND hematopoi*[tiab] = 1 and that record is a review of MAPK and nuclear-receptor crosstalk in cancer immunotherapy; NRSF[tiab] AND MLL[tiab] = 0.

   NCBI E-utilities esearch, db=pubmed, counts with querytranslation and warninglist read back and each hit resolved to a title via esummary, run 2026-09-22; the zero run twice · NCBI PubMed 2026
   ▸What this evidence cannot show

   Titles and abstracts only; rule 80 applies and a REST measurement inside a supporting figure of a leukaemia paper is not bounded by this. The absence is also bounded by the SAFE operands only - NRSF and the full name - and those two reach a union of 709 records against an unknown true total, because the approved symbol cannot be counted at all. So this is an absence within a partial net, and it is stated that way. The two hits were resolved to titles rather than merely counted, because a control count is a list nobody opens; both turned out to be false positives, which is why the count and the title-check are both recorded.

   ▸Evidence that argues against this (1)

   • A zero in a partially reachable literature is weaker than a zero in a fully reachable one, and the expression data is the reason to keep the question open: UniProt records REST as expressed at higher levels in spleen, thymus and peripheral blood lymphocytes, it is listed as causative in a PAEDIATRIC malignancy, and its documented G9a-delivery route runs through two Mediator subunits one of which is curated against a KMT2A fusion construct. Every one of those is a reason a REST result in leukaemia could exist and be invisible - and the token pathology means it would also be invisible to anyone looking for it. UniProt Consortium 2026
5. E5strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

   REST is not a DepMap dependency: mean gene effect -0.2147 across 1,258 screened cell lines with 5.2% of lines depleted below -0.5, and a lymphoid-tissue mean of -0.1385 over 97 lines; CDYL, which its EHMT1/EHMT2 interaction requires, is likewise not a dependency (mean -0.0911, rank 4,136 in this project's KMT2A scan).

   DepMap CRISPR (Chronos) gene effect served through the Open Targets Platform GraphQL API, field depMapEssentiality, fetched 2026-09-22 with approvedSymbol read back as REST; CDYL values joined from analyses/depmap-kmt2a-selective-scan-full/results.json · Broad Institute DepMap via Open Targets Platform 2026
   ▸What this evidence cannot show

   Rule 83: the API does not state the DepMap release; 1,258 is REST's own screen count. In this project's KMT2A scan REST is at -0.1785 in KMT2A-rearranged lines against -0.217 in other leukaemia lines - a +0.0385 difference at rank 13,754 of 17,728, ranking noise between two near-zero arms, recorded only so the percentile is never quoted. Critically, the approvedSymbol readback matters more here than anywhere else in this beat: a symbol that is an English word is exactly the case where a resolver could silently return something else, and the readback confirmed REST.

   ▸Evidence that argues against this (1)

   • A viability screen is the wrong assay for a repressor whose documented role is lineage-inappropriate gene silencing, and the direction it would be expected to matter in is the opposite of a dependency. If REST-CDYL-EHMT2 repression is transformation-SUPPRESSING as UniProt's own wording suggests, then REST loss should make cells grow better, not worse - which a dropout screen registers as no effect. The relevant readout would be inappropriate neuronal-gene expression or lineage infidelity, and this graph has an entire lineage-switch layer in which REST has never been measured. UniProt Consortium 2026

• recruits →EHMT2 / G9a - the H3K9me1/2 writer that sets whether the KMT2A LOCUS ITSELF rearranges, and whose loss delays MLL-AF9 AML in a supplementary figure nobody can find

  REST recruits EHMT1/EHMT2 for H3K9me2 silencing - but ONLY in the presence of CDYL

• binds →MED19 - the Mediator subunit curated against an engineered MLL-AFF4 fusion protein, and the one REST uses to deliver G9a

  MED19 is one of the two Mediator subunits REST uses to deliver G9a

• binds →MED26 — a common-essential Mediator subunit with a solved 87-residue domain and no ChEMBL target entity at all

  the second Mediator subunit in the REST-to-G9a route

• related to →H3K9me3 - measured in KMT2A::AFF1 cells, in a supplementary figure, and it rises GENERALLY where H3K27ac falls SPECIFICALLY in the same experiment

  the mark family REST delivers is H3K9me1/me2 via EHMT1/EHMT2, not me3

• recruits →HDAC1 - a class I eraser that has no dependency claim of its own in this disease, because removing it alone does nothing until HDAC2 goes too

  REST recruits SIN3A/SIN3B and RCOR1/CoREST, which bring histone deacetylase

• related to →Lineage switch in KMT2A-rearranged leukemia — the same clone changes what kind of blood cell it is

  REST is the canonical enforcer of lineage-inappropriate gene silencing and has never been measured here

• related to →Hypothesis: KMT2A::AFF1 rewires splicing, because AFF1's own protein family are G-quadruplex RNA-binding splicing modulators and the fusion keeps AFF1

  SRRM4-dependent alternative splicing inactivates REST by producing isoform 3

• related to →Plain-English: chromatin and epigenetics — the sticky notes on the DNA, and which ones matter here

  a DNA-binding factor whose epigenetic readout is other enzymes marks

• binds →USP7 - the enzyme that holds PRC1.1 together, and the only pharmacological handle on this complex that exists today

  REST abundance is proteasomally controlled - FBXW11/BTRC substrate, USP7 interactor

• ← related toH3K9me2 - named sixty times in this graph with no entry of its own, the mark that defines the lamina-associated domains the fusion's targets sit beside, and the only H3K9 methylation state ever measured at a KMT2A fusion's own binding sites

  H3K9me2 is the repressive output of the REST-CDYL-G9a route recorded in that entry

• ← related toSIN3A (Q96ST3) - the scaffold of the OTHER HDAC1/2 corepressor complex, a 5' fusion partner in infant B-ALL and in myelodysplasia, and present in three MLL-fusion AP-MS columns but not the index fusion's

  REST docks the SIN3A PAH domains; this graph's gene-rest already carries REST's SIN3A-interaction region (REST coordinates)