Oncofusion

start / graph / gene-rad50

Genestrong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.Q37-archive-sweep

RAD50 (Q92878, 1312 aa) - the third MRN subunit, and the one that settles whether the complex is incoherent in DepMap: mean -0.114 with 5.2% of 1,257 lines below -0.5, so it sits with NBN and not with MRE11, yet it covaries with MRE11 across cell lines as tightly as BRCA1 does with BARD1

also called hRAD50 · RAD50 double strand break repair protein · the MRN zinc-hook scaffold

RAD50 is the long, hinged arm of a three-protein machine that finds broken DNA and makes the first cut, and losing it in a person causes a rare syndrome with bone-marrow failure. In the big gene-knockout screens it looks almost dispensable, like one partner and unlike the other, which turns out to say more about how the screens are scaled than about the machine.

The same thing, for a scientist

**RAD50 is the ATPase-and-coiled-coil scaffold of the Mre11-Rad50-Nbs1 (MRN) complex**, carrying a 100-residue zinc-hook (Q92878 residues 635-734) at the apex of a long antiparallel coiled coil that dimerises two MRN complexes across a break [E1]. Biallelic hypomorphic RAD50 variants cause a Nijmegen-breakage-syndrome-like disorder - microcephaly, growth retardation, no immunodeficiency in the index family [E2] - and a later cohort adds **bone marrow failure and immunodeficiency** to the phenotype, which puts RAD50 in the same clinical territory as NBN [E3]. **This entry exists to answer a question two entries in this graph asked and neither could close.** `gene-mre11` and `gene-nbn` both record the same paradox - **MRE11 mean gene effect -0.542 with 56.1% of 1,258 lines below -0.5, against NBN -0.047 with 1.3%, one release, one call, eleven-fold apart** - and both name RAD50 as the discriminator, in their own `open_questions`, as unfetched. **Fetched, Open Targets Platform 26.06, 2026-09-22** [E4]: | gene | n screens | mean Chronos | % below -0.5 | |---|---:|---:|---:| | MRE11 | 1,258 | **-0.542** | 56.1% | | **RAD50** | **1,257** | **-0.114** | **5.2%** | | NBN | 1,258 | -0.047 | 1.3% | **RAD50 sits with NBN.** So the complex is not split one-against-two: **MRE11 alone scores as a dependency and both of its obligate partners do not.** That kills the reading in which the nuclease matters and the scaffold does not, because RAD50's ATPase is required for MRE11's endonuclease in every reconstitution, and it kills the reading in which NBN is the odd one out. **And then the covariation reverses the verdict** [E5]. Across the 1,208 cell lines screened for all of them, **MRE11 and RAD50 correlate at Pearson r = 0.554** - statistically indistinguishable from **BRCA1 against BARD1, r = 0.560**, this analysis's obligate-heterodimer positive control - against a background of essentially zero for unrelated pairs (MRE11~BRCA1 -0.004, MRE11~MEN1 -0.004, CBX1~CBX3 -0.019). **The complex behaves like one machine in how its members' dependencies move together and like three unrelated genes in where their means sit.** Both readings are correct; they answer different questions, and the entries that recorded the eleven-fold gap were reading the second one as though it were the first. See `ana-q37-the-mrn-complex-covaries-like-one-machine-and-scores-like-three-unrelated-genes`. **In haematopoietic and lymphoid lines (n = 141) the ordering is preserved and RAD50 moves toward MRE11**: MRE11 -0.586 (58.9% below -0.5), **RAD50 -0.240 (18.4%)**, NBN +0.003 (0.7%) [E4]. **At the *KMT2A* locus RAD50 is not merely unmeasured, it is the least-measured protein in this whole layer** [E6]. `RAD50[tiab] AND MLL[tiab]` = 0, `RAD50[tiab] AND "11q23"[tiab]` = 0, `"MRN complex"[tiab] AND MLL[tiab]` = 0 and `"MRN complex"[tiab] AND KMT2A[tiab]` = 0, all genuine zeros with non-empty translations; `RAD50[tiab] AND KMT2A[tiab]` = 2 and both title-check off-topic; the rule-67 positive control `RAD50[tiab] AND leukemia[tiab]` = 45 through the identical shape. **The complex form, which rule 64 requires and which `gene-mre11` and `gene-nbn` both flag as unrun, is run here and is zero.** In the archives, RAD50 has **an ENCODE target record and no ENCODE experiments, zero ChIP-Atlas SRX, zero ReMap datasets, and exactly one GEO ChIP-seq pair** - GSM5387834/5, an IP and an input in a DEAF1-overexpressing myeloma line (GSE178328).

What it looks like
DNA repair protein RAD50 (hRAD50)1,312 aa
Zinc-hookZinc-hook · residues 635–734Coiled coil 1Coiled coil 1 · residues 228–359Coiled coil 2Coiled coil 2 · residues 401–598Coiled coil 3Coiled coil 3 · residues 789–107905001,000

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.

Facts and figures
gene symbolRAD50
uniprotQ92878
uniprot entry nameRAD50_HUMAN
gene primary readbackRAD50
length AA1312
protein nameDNA repair protein RAD50 (hRAD50)
how the accession was resolvedgene_exact:RAD50 AND organism_id:9606 AND reviewed:true returned exactly ONE reviewed record whose gene_primary reads back as RAD50 (rules 62/65). UniProt lists NO gene synonyms for RAD50, which is unusual in this layer and is why the retired-symbol trap that hides MRE11 (as MRE11A) and NBN (as NBS1) does not apply to it. Accession and length are both asserted so verify-accessions.py can catch a swap; the nearest length in this tranche is MRE11 at 708 aa, so the length discriminator is safe here.
domains sourceUniProt Q92878 feature table, fetched 2026-09-22. The table declares one Domain (Zinc-hook 635-734) and five Coiled coil features (228-359, 401-598, 635-673, 706-734, 789-1079); the two short coiled coils at 635-673 and 706-734 lie inside the zinc-hook domain and are omitted here to avoid drawing three overlapping lanes for one feature. The N-terminal and C-terminal Walker A/B ABC-ATPase lobes are NOT annotated as Domain features in this record and are therefore NOT listed, rather than given invented boundaries; the record does carry nucleotide-binding residues at 13, 38-44, 67, 69 and 159 and zinc-coordinating residues at 681 and 684.
localizationNucleus; Chromosome; Chromosome, telomere (UniProt Q92878, ECO:0000269 from PubMed 10783165, 15916964, 26215093, 10888888, 21757780)
subnuclear localizationdiscrete nuclear foci after genotoxic treatment; localises to double-strand breaks; telomeres (UniProt Q92878, experimentally coded)
localization shown for fusionNO - nobody has looked in any fusion. RAD50 is the least-measured protein in this layer: an ENCODE target record with zero experiments, zero ChIP-Atlas SRX, zero ReMap 2022 datasets, and one GEO ChIP-seq IP/input pair (GSM5387834/5, GSE178328) in a DEAF1-overexpressing myeloma line.
shown for fusionNO - nobody has looked in any KMT2A fusion, and the complex form is zero too: "MRN complex"[tiab] AND MLL[tiab] = 0 and "MRN complex"[tiab] AND KMT2A[tiab] = 0
cell typenot measured in any haematopoietic cell at this locus; the clinical phenotype includes bone marrow failure and immunodeficiency, which is haematopoietic
cell statedamage-conditional by construction - RAD50 acts at a break, so a resting-state occupancy measurement would be uninformative even if one existed
epigenetic readoutnone directly. RAD50 acts on DNA ends, not on histone marks, and MRN recruitment is downstream of the H3K9/HP1 chromatin state this graph's CTCF-primed cascade describes. No mark has been measured in a RAD50-null human cell in this graph's bibliography.
relationship to existing entriesWritten to close the first open question of both `gene-mre11` and `gene-nbn`, each of which names RAD50 as the unfetched discriminator for the MRN DepMap asymmetry. It does not restate their UniProt or clinical content.
domains
Coiled coil 1228.0Coiled coil 1: 228Coiled coil 2401.0Coiled coil 2: 401Zinc-hook635.0Zinc-hook: 635Coiled coil 3789.0Coiled coil 3: 789
gene synonyms in uniprot
[]
domains — the numbers behind the chart
[
  {
    "name": "Zinc-hook",
    "start": 635,
    "end": 734
  },
  {
    "name": "Coiled coil 1",
    "start": 228,
    "end": 359
  },
  {
    "name": "Coiled coil 2",
    "start": 401,
    "end": 598
  },
  {
    "name": "Coiled coil 3",
    "start": 789,
    "end": 1079
  }
]
depmap
{
  "source": "Open Targets Platform, meta.dataVersion year 26 month 06, fetched 2026-09-22 (rule 83: the release, not just the date)",
  "endpoint": "https://api.platform.opentargets.org/api/v4/graphql, target(ensemblId).depMapEssentiality",
  "ensembl": "ENSG00000113522",
  "ensembl_resolution": "HGNC primary ensembl_gene_id via rest.genenames.org, matched on the PRIMARY symbol only (rule 78)",
  "approvedSymbol_readback": "RAD50",
  "gene_effect_metric": "Chronos",
  "n_screens_this_gene": 1257,
  "mean_gene_effect": -0.1141,
  "median_gene_effect": -0.09,
  "n_lines_below_minus_0_5": 65,
  "fraction_below_minus_0_5": 0.0517,
  "min": -1.501,
  "max": 0.692,
  "haematopoietic_and_lymphoid_subset": {
    "n": 141,
    "mean": -0.24,
    "fraction_below_minus_0_5": 0.184
  },
  "line_identities_persisted": "yes - analyses/q37-mrn-depmap-adjudication/results.json stores every depmapId, cell line name, tissue and gene effect, not only the summary (rule 90)",
  "controls_in_the_same_call": "low-dependency OR4F5 (n=5, mean -0.043) and OR5A1 (n=1,096, mean -0.143); pan-essential RPL13A (n=394, mean -2.074, 100% below -0.5). POLR2A returned zero DepMap blocks and is recorded as a failed control rather than dropped."
}
bounded absence at the kmt2a locus
{
  "instrument": "scripts/pubmed_safe.py esearch, db=pubmed, 2026-09-22; every querytranslation read back; every zero carries a non-empty translation",
  "query_form": "one [tiab] tag per term (rule 94); the complex form run as a quoted phrase (rule 64)",
  "genuine_zeros": {
    "RAD50[tiab] AND MLL[tiab]": 0,
    "RAD50[tiab] AND \"11q23\"[tiab]": 0,
    "\"MRN complex\"[tiab] AND MLL[tiab]": 0,
    "\"MRN complex\"[tiab] AND KMT2A[tiab]": 0
  },
  "nonzero_and_title_checked": {
    "RAD50[tiab] AND KMT2A[tiab]": {
      "35117846": "pulmonary granular cell tumour, three cases with gene analysis - off topic",
      "41549603": "Pathogenic Germline PALB2 and RAD50 Variants in Patients With Relapsed Ewing Sarcoma - a germline panel, not a KMT2A-locus measurement",
      "count": 2
    }
  },
  "rule_67_positive_control_identical_shape": {
    "RAD50[tiab] AND leukemia[tiab]": 45
  },
  "name_list_tried": [
    "RAD50",
    "\"MRN complex\""
  ],
  "name_list_NOT_tried_and_this_is_the_limit": "MRE11-RAD50-NBS1 spelled out, hRAD50, RAD50L, and the phrase 'Mre11 complex' which is how much of the older literature names it. This project has undercounted a literature tenfold by keying to a current symbol, so the incompleteness of this list is the largest weakness of the absence.",
  "bound_class": "titles and abstracts only (rule 80): a RAD50 measurement in a supporting figure is invisible to all of these."
}
archive census 2026 09 22
{
  "ENCODE": "target record /targets/RAD50-human/ returns HTTP 200; the experiment search returns HTTP 404, i.e. zero released experiments of any assay type. This pair is the cleanest demonstration in this beat that an ENCODE 404 is about the query and not about the target.",
  "ChIP-Atlas": "0 distinct SRX (hg19 + hg38), checked with an alias regex as well as exact match",
  "ReMap 2022": "HTTP 401 with body 'Sorry. There is no experiment for target RAD50 and tax_id 9606 on remap2022' - ReMap's spelling of zero",
  "GEO": "2 GSMs with RAD50 in the title and ChIP in the record - GSM5387834 (input) and GSM5387835 (IP), library_strategy ChIP-Seq, hg38, series GSE178328 'Genome-wide maps of chromatin state in DEAF1 overexpression (OE in Multiple Myeloma)', no linked PMID",
  "absence_class": "measured-not-analysed, thinly. One IP and one input in an overexpression model is not a usable occupancy resource, but the honest class is not never-measured."
}
How this could bite us back
  • RAD50 is not a drug target here and the bite-back is on the inference rather than on an intervention: this entry makes MRE11 look like the anomalous member of MRN, and the tempting next step is to treat MRE11 as 'the dependency' and design around it. If the offset is guide toxicity at 11q21 then MRE11 inhibition would be far less selective than the Chronos mean suggests, and mirin-class MRE11 inhibitors would fail for exactly that reason.
  • The complex is required for normal haematopoiesis - RAD50 loss causes bone marrow failure in humans - so any therapeutic attack on MRN in a leukaemia is attacking a pathway whose loss in a child produces marrow failure. That is an essential-function toxicity with a human phenotype attached, which is stronger evidence than a mouse knockout.
Still open
  1. Is the MRE11 offset a guide-cutting artefact or a real difference in knockout tolerance? The covariation says the three subunits share a dependency structure and the means say they do not share a scale, and the measurement that separates those - guide-level log-fold-change against the copy number at 11q21, 5q31 and 8q21 - has not been run for any MRN subunit.
  2. Does RAD50 haploinsufficiency appear in any infant KMT2A-rearranged cohort's germline data? The complex's other two subunits have leukaemia-predisposition literatures and RAD50 has bone marrow failure, yet RAD50 x 11q23 is a genuine zero; the cohorts that would answer it are sequenced and the question has not been asked in these words.
  3. Should the one existing RAD50 ChIP-seq (GSM5387835, GSE178328) be read at the KMT2A locus? It is a single IP with a single input in a DEAF1-overexpressing myeloma line, which is a weak instrument, but it is the only RAD50 occupancy file that exists and it was not read by this beat.
Every source behind this entry (6)
  1. E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    RAD50 is UniProt Q92878, 1312 aa, primary gene name RAD50, protein name 'DNA repair protein RAD50', with a Zinc-hook domain at residues 635-734 and coiled-coil segments at 228-359, 401-598 and 789-1079; it localises to the nucleus, to chromosomes and to telomeres, and forms discrete nuclear foci after genotoxic treatment.

    What this evidence cannot show

    Rule 91 applies: the ECO:0000269 localisation citations tell you a curator saw an experiment in those papers, not that any of them is the founding observation for RAD50's localisation, and none of them was opened here. The record annotates no ABC-ATPase Walker A/B domain features, so the N- and C-lobe boundaries that the structural literature uses are absent from data.domains rather than invented.

    Evidence that argues against this (1)
    • UniProt lists no gene synonyms for RAD50, which makes an alias search look unnecessary. The same record type lists MRE11A for MRE11 and NBS1/NBS/P95 for NBN, and in both of those cases the retired symbol carries data the current one cannot reach. A reader should not infer from an empty synonym list that RAD50's literature is fully reachable from the current symbol. UniProt Consortium 2026
  2. E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    Biallelic hypomorphic RAD50 variants cause a Nijmegen-breakage-syndrome-like disorder in humans - microcephaly, growth retardation and a defective ATM-dependent damage response - establishing that RAD50 loss is survivable in a whole human and that the phenotype overlaps NBN's.

    Human RAD50 deficiency in a Nijmegen breakage syndrome-like disorder. · Waltes R, Kalb R, Gatei M, et al. · American Journal of Human Genetics 2009
    PMID 19409520DOI
    What this evidence cannot show

    A single index patient in the founding report. These are hypomorphic alleles, not nulls, so the case establishes that residual RAD50 function is compatible with life and says nothing about complete loss - which is the condition a CRISPR knockout approximates and is exactly the disanalogy that makes the DepMap comparison in E4 difficult to interpret. The abstract was read; the full text was not opened.

    Evidence that argues against this (1)
    • The index family had no immunodeficiency and no reported malignancy, which distinguishes the RAD50 phenotype from Nijmegen breakage syndrome proper, where lymphoid malignancy is the defining risk. A reader moving from 'RAD50 is the third MRN subunit' to 'RAD50 is a leukaemia-predisposition gene' would be overreaching on this citation. Ragamin A et al. 2020
  3. E3moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    A later series extends the human RAD50 phenotype into the haematopoietic compartment, reporting bone marrow failure and immunodeficiency associated with RAD50 variants - which is the clinical feature that makes RAD50, like NBN, relevant to a graph about a blood cancer.

    Bone Marrow Failure and Immunodeficiency Associated with Human RAD50 Variants. · Takagi M, Hoshino A, Bousset K, et al. · Journal of Clinical Immunology 2023
    PMID 37794136DOI
    What this evidence cannot show

    Read from the title and abstract only; the number of patients, the allele spectrum and whether the marrow failure is attributed to RAD50 or to a co-segregating variant were not established here, so the claim is recorded at moderate confidence and should be re-read before it is built on. Bone marrow failure is not leukaemia, and no KMT2A rearrangement is reported in any RAD50-deficient patient anywhere in this literature.

    Evidence that argues against this (1)
    • The counter-position is the silence rather than a paper: across 45 records returned by RAD50[tiab] AND leukemia[tiab], none reports a KMT2A rearrangement in a RAD50-deficient patient, and RAD50[tiab] AND "11q23"[tiab] is a genuine zero. So the haematopoietic phenotype does not connect RAD50 to this graph's disease by any published route. National Center for Biotechnology Information 2026
  4. E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

    RAD50's DepMap profile sits with NBN and not with MRE11: Open Targets Platform release 26.06, fetched 2026-09-22, gives RAD50 mean Chronos gene effect -0.1141 over 1,257 screens with 65 lines (5.2%) below -0.5, against MRE11 -0.5423 over 1,258 with 56.1% and NBN -0.0471 over 1,258 with 1.3%. In the 141 haematopoietic and lymphoid lines the ordering is preserved and RAD50 moves toward MRE11: MRE11 -0.586 (58.9%), RAD50 -0.240 (18.4%), NBN +0.003 (0.7%).

    What this evidence cannot show

    The three denominators (1,257 / 1,258 / 1,258) are gene-specific and nearly identical, so these three are directly comparable - but the same call returns DNA2 with only 80 screens at mean -2.712 and RPL13A with 394 at -2.074, and POLR2A with none at all. The screen count returned by this endpoint is therefore NOT simply 'lines screened', its variation is unexplained, and no conclusion in this entry rests on a gene with a small n. Chronos gene effect from a CRISPR knockout is not a model of the human hypomorphic alleles in E2 and E3. Rule 84 applies: 5.2% below -0.5 is not a median-split proportion, but it is still a threshold statistic and the mean and the full distribution are given beside it.

    Evidence that argues against this (1)
    • A CRISPR gene effect conflates the fitness cost of losing the protein with the cutting toxicity at that locus, and RAD50 (5q31), MRE11 (11q21) and NBN (8q21) sit in different copy-number contexts across a panel enriched for aneuploid cancer lines. The guide-level decomposition that would separate these is the design already written for CBX1 in this graph and it was not run for RAD50. Q35-encode-reanalysis 2026
  5. E5strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset

    Across the 1,208 cell lines screened for all of them, MRE11 and RAD50 gene effects correlate at Pearson r = 0.554 (Spearman 0.551), indistinguishable from the obligate-heterodimer positive control BRCA1 against BARD1 at r = 0.560 (Spearman 0.549), while unrelated pairs run through the identical code path give MRE11~BRCA1 -0.004, MRE11~MEN1 -0.004, BRCA1~ATM -0.019 and CBX1~CBX3 -0.019. RAD50~NBN is 0.283 and MRE11~NBN is 0.321.

    What this evidence cannot show

    No p-values or confidence intervals were computed; at n = 1,208 a Pearson r of 0.554 against a background of -0.02 is not in doubt as a signal, but 'indistinguishable from 0.560' is an eyeball judgement and not a test, and the difference between 0.554 and 0.283 was likewise not tested. Correlated gene effects can arise from shared copy-number and shared screen batch as well as from shared biology, and no such correction was applied.

    Evidence that argues against this (1)
    • The per-line dependency CALL tells the opposite story and it is equally true: 292 of 1,208 lines (24.2%) have MRE11 below -0.5 while both RAD50 and NBN sit above -0.2, and exactly ONE line of 1,208 has all three below -0.5. Anyone classifying lines as MRN-dependent would conclude the complex is completely incoherent, and they would be right about the question they asked. Q37-archive-sweep 2026
  6. E6strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

    No published title or abstract places RAD50, or the MRN complex by name, at the KMT2A locus or at 11q23. Run live on 2026-09-22 through scripts/pubmed_safe.py with every querytranslation read back: RAD50[tiab] AND MLL[tiab] = 0, RAD50[tiab] AND "11q23"[tiab] = 0, "MRN complex"[tiab] AND MLL[tiab] = 0 and "MRN complex"[tiab] AND KMT2A[tiab] = 0, all with non-empty translations and warninglist 'No items found.'; RAD50[tiab] AND KMT2A[tiab] = 2 and both records title-check off-topic; the positive control RAD50[tiab] AND leukemia[tiab] = 45 through the identical shape.

    NCBI E-utilities esearch over PubMed via scripts/pubmed_safe.py, six queries with querytranslation read back, 2026-09-22 · National Center for Biotechnology Information · NCBI Entrez Programming Utilities 2026
    What this evidence cannot show

    Titles and abstracts only (rule 80). The name list is two strings and the older literature calls this machine 'the Mre11 complex', which was not searched - and `gene-mre11`'s own bounded absence has the same hole. This zero is an absence in the current-vocabulary literature, not in the literature, and the worked demonstration that those differ is TOP2B[tiab] AND KMT2A[tiab] = 0 in this very graph's most-cited relationship.

    Evidence that argues against this (1)
    • The complex form being zero is the part that closes rule 64 for this subunit, and it is also the part most likely to be wrong for the vocabulary reason above. A paper framing its work as 'Mre11 complex' or 'MRN' without the word 'complex' would be invisible to both of the quoted zeros. Oncofusion knowledge graph coordinator 2026
Connected to